Based on the theory of probability,with a small cloth bag in the catalyst bound as the study object,the model of tracer concentration distribution in the fixed bed was established using the traditional Galton board model,and the expression of the concentration distribution model in the fixed bed was obtained.The experimental setup was established using KCl solution as the tracer and the experimental results were compared with the model predicted values.The experimental value fitted well with the model predicted value.The influences of different parameters on concentration distribution in the fixed bed were analyzed.The trace concentration distribution curve gradually became gentle with increasing catalyst particle size and increasing height of the fixed bed but decreasing liquid spray quantity.

Complex process fault prognosis is a key scientific issue that ensures the security of the process and reliable operation,however complex systems work state is often determined by multivariate delayed sequence.It contains the relationship between the variables and the time delay information,so it has information completeness.So a complex process fault prognosis approach based on multivariate delayed sequences is proposed.First this method construct the Time Delay Signed Direct Digraph (TD-SDG) to get multi-delayed sequence,then combine Independent Component Analysis and ELM neural network to get the independent component of the multi-delayed sequence,and finally realize the purpose of fault prognosis of complex system.The simulation results on Tennessee Eastman process illustrate that the proposed method can predict fault earlier 15 min,increase operator's reaction time and detect the fault.

For the purpose of achieving automatic recognition and objective judgment of the production states of the froth phase,statistical modeling of froth images is introduced in the flotation process monitoring.Firstly,the Weibull distribution is applied to model the statistical distribution of the edge response of the froth images of all-round orientations,which leads to the effectively extraction of the spatial structure features of the froth images.Successively,a Mixture of Gaussian (MoG) model of the statistics of the froth images under each typical froth state is obtained by the statistical learning of the structure features of the froth image samples.Consequently,the froth states can be inferred effectively by the Bayesian inference.The froth state recognition results indicate that the proposed method can monitor the on-line spatial structural changes of the froth images,which achieves more accurate recognition results of the froth states comparing to the other froth image perception based froth state recognition.

An ethylene plant employs multiple cracking furnaces in parallel to convert various hydrocarbon feedstocks to smaller hydrocarbon molecules.The continuous operational performance of cracking furnaces gradually decays because of coke formation in the reaction coils,which requires each furnace to be periodically shut down for decoking.Given multiple feeds and different cracking furnaces as well as various product prices and manufacturing costs,the operational scheduling for the entire furnace system should be optimized to achieve the best economic performance.In this paper,a new MINLP (mixed-integer nonlinear programming) model is developed to obtain cyclic scheduling strategies for cracking furnace systems.Compared to previous studies,the new model has more capabilities to address operation profitability of multiple feeds cracked in multiple furnaces.Meanwhile,solving the problem of the choice for the right time of changing feedstock in the process of cracking furnace,case studies demonstrate the efficacy of the developed methodology.

Increasing twins pressure of environmental and legislation have resulted in changing the approach of traditional design to consider the multi-objective sustainable design in chemical processes.This paper focuses on the phenol-ammonia wastewater stripping section in a coal to synthetic natural gas (SNG) plant.Comparison between the single pressurized tower with side-draw design and double pressurized tower design shows the first design benefits from lower economic cost at the expense of low sustainability.On such basis,three key decision variables are selected,including total number of stages,side-draw position and heat feed stream temperature to minimize the total annualized cost (TAC) and potential environment impact (PEI) metric.The optimal solution reveals that the single pressurized tower with side-draw design has potential to achieve low economic costs and high environmental performance simultaneously by changing the operating conditions.

For the distillation process with heat exchange between the bottoms and the feed,the analysis based on both the first principle and the flowsheet simulation software are proceeded to reveal the couple among the reboiling duty,the liquid level of the bottom and the temperature of the feed,and then a novel thermal-coupling-oriented integrated control and optimization strategy is proposed.In this integrated strategy,the reboiling duty is used to compensate the change of the temperature of the raw material and ensure the quality of the products.The strategy is applied on a practical distillation process,and the coordinated control problem between the flow rate of the bottoms and the reboiling duty is solved with great decrease of the standard errors of key variables.

Having time-dependent streams makes the heat exchanger network (HEN) synthesis of batch process much more complicated than its continuous partner.In this paper,heat storages are employed to overcome the time-dependent limit and heat storages and heat exchangers are combined for sake of equipment cost reduction.To obtain the optimal solution,a nonlinear programming (NLP) with the objective of minimize total annual cost is established for the overall HEN synthesis,including the optimization of heat storage temperatures.At last an example is used to show the application and effectiveness of the proposed methods.

Aimed to obtain reaction path with benign HSE performance during early phase of chemical process,the multi-objective reaction path synthesis method based on fuzzy HSE assessment is proposed. The HSE index framework is provided based on the information of HSE impact.Through setting membership function of each index,establishing fuzzy inference system,and determining the weighting vector by AHP,the fuzzy HSE assessment method is proposed.Based on the fuzzy assessment system,the benign reaction path will be obtained through multi-objective optimization treating HSE as objective functions.By analyzing the application of reaction path synthesis for carbaryl production,reaction paths with their HSE objective function values are obtained quantificationally,which provide data for decision making in early stage.

Simulated moving bed (SMB) is an important technology in xylene isomers separation.The reflux ratios of SMB in each zone are of key importance in determining the purity and recovery of PX.On the basis of the true moving bed (TMB) modeling scheme and industrial data,the SMB mechanism model is established.The operation interval for the reflux ratio is obtained via analyzing the effect of reflux ratio on the purity and recovery of PX product.The results show that the TMB modeling scheme performs great in describing the industrial SMB,and the analysis based on the mechanism model facilitates the design and operation of a SMB for PX separation.

The key technical index,namely,alkaline solution concentration of alumina evaporation process,can not be measured on-line.Moreover,it is usually difficult to describe the dynamics characteristic between the techniques indices and the control loops by using an accurate mathematical model.Thus,the existing optimal methods can not solve the optimal control for operation of the alumina evaporation process.In this paper,a hybrid intelligent optimal control method is developed.The proposed method is comprised of a presetting model based on case-base reasoning (CBR),soft sensor model based on recursive partial least squares (RPLS),feedforward and feedback compensators based on expert rules.Simulation is researched by the actual data of the alumina evaporation process.The simulation results show that the proposed methods can control the alkaline solution concentration within their scopes.

The extreme learning machine (ELM) is inefficient in high-dimensional data modeling during the chemical process,where the data is always strongly coupled and with much noise.Aiming at dealing with this problem,a data attributes decomposition based hierarchical ELM neural network (DHELM) is proposed.In the modeling process of DHELM,the data attributes decomposition (DAD) method is used to cluster the high-dimensional inputs and build the auto-associative subnets,and then the extracted characteristic components yielded by auto-associative subnets are inputted to extreme learning machine.Meanwhile,the effectiveness of DHELM is verified by the UCI standard data sets and an industry application object.Through the verification and comparison,the proposed DHELM model has the advantages of fast convergence speed with high modeling accuracy and strong network stability. Furthermore,it provides a new way for neural network development and its application to chemical processes.

Traditionally,plant simulation model is classified as macroscopic or microscopic. An ERP-level model and a PCS-level model belong to macro and micro simulation model respectively while an MES-level model lies in-between.Simulation models of certain level established by former researchers usually have low reusability and poor scalability.To this end,a novel four-layered approach,which is composed of model entity structure/model base,model abstraction,plant modeling and IPS simulation systems layer,has been proposed for plant modeling and simulation.Configurable and reusable simulation model can be easily obtained by adopting the abstraction modeling and simulation method with respect to plant network. In this article,a multi-level intelligent plant simulation system is constructed with the aforementioned method and the effectiveness and reliability are verified through case studies.Therefore,it is expected to be widely used in plant modeling and simulation in practice.

In ethylene cracking process,the changes of feed have many kinds,and due to its expensive feed analyzer,little industrial site equips with it,so online recognition of oil property is important to achieve cracking online optimization.As the traditional fuzzy C-means algorithm is based on the membership of the strike Euclidean distance,the algorithm contains only the mean center,bringing the unity of clustering results.To take full advantage of effective information of cracking feed,this paper proposes a fuzzy membership set method based on hybrid probabilistic model,namely through the establishment of Gaussian mixture model to achieve describing the probability distribution of clustering sample's affiliation,and use EM algorithm to estimate the model parameter's pole maximum likelihood.The algorithm can not only consider mean center of the sample,but also effectively use sample covariance and the weight coefficient information for mode discrimination.Finally,the simulation is based on classic IRIS data clustering and ethylene cracking feedstock identification,verifying the method described in this paper in the index of dunn and Xiebieni is better than fuzzy C-means clustering algorithm,showing that the method is effective.

Syngas based co-feed and co-production (COCO) system is one of the important parts of sustainable development for energy and chemical industries.Co-feed and co-production factors are two key elements to affect the performance of COCO system.The coal/natural gas co-feed and electricity/methanol co-production system is proposed as the case study.Co-feed factor is defined as the separate ratio of unreacted syngas from methanol reactor unit to gas-steam combined cycle power generator; co-production factor is defined as the exergy ratio between natural gas and coal.The effect of the proposed factors on the methanol yield,exergy loss rate,and product cost was investigated systematically.The effect of the prices of coal and natural gas on product cost was also investigated.The corresponding effect relations were discussed.

Molecular weight distribution (MWD) of polymer is an important performance index of quality.Due to the MWD can't be measured online,direct quality control is difficult to achieve.The combined of Legendre orthogonal polynomial and neural network had been used to build the model of the MWD,then the MWD tracking control in three-dimensionals space were decomposed the tracking control of its distribution moment in two-dimensionals time domain. The mothod,which employed the moments of distribution to predict the MWD,was proposed. The optimal objective function was based on the sum of squared errors (SSE) of the moments under certain constraints,meanwhile the correction term of the lower order moments was also introduced,in which part of the closed-loop feedback control of MWD can be obtained.The proposed control method was tested on styrene polymerization reaction in CSTR,and perfect control performance shows the effectiveness of the work.

Traditionally,real-time optimization needs to reach steady state before being implemented. While,regarding multivariable and high dimensional systems global optimization is always time-consuming and too long to adjust the control parameters in real time.To solve these problems,this paper proposes a real-time evolution approach towards multilevel process system,which divides a multilevel process system into several interconnected subsystems.When the system is disturbed,the process waiting for steady state is divided into several pseudo-steady states.In each steady state,a particle swarm optimization algorithm is adapted to optimize each sub-system in turn,while the rest of sub-systems are regarded as the static system.Then coordinate the shared and internal variables between the sub-systems,and put the finally optimal solution into the controller to change the system set-point continuously.This paper gives the detailed algorithm steps and the two stage series reactor is carried out to demonstrate the benefits of the proposed methodology.

While doing an engineering project,engineers always use standard analysis,design methods and notations to assist them to communicate,discuss and document the related results and decisions.In the computer dynamic simulation field,the most frequently used analysis,design method and notation is the information flow diagram.Based on the previous research,an improved,more perfect information flow diagram for modeling and simulation process was presented.Such notation contains three broad categories (variable,directed line segment and boundary) and nine small classes elements,which could completely cover all kinds of mathematical formulas and other model expression ways in the computer dynamic modeling and simulation process.Besides,common combination illustrations and corresponding numerical methods were also been introduced in this paper.And at the end of this paper,dynamic modeling and simulation process of a distillation column is used to illustrate the using process of such improved information flow diagram detailedly.Besides,a large number of applications of such information flow diagram in research issues and engineering projects proves that it could not only provides a perfect communication medium for the simulation related staffs,but also makes people to reuse requirements and design components from the previous simulation project more easily,which could greatly improves the efficiency of the modeling and simulation process and the quality of the simulation model consequently.

Zinc electrowinning process is a typical large time-delay system,so zinc ion and acid concentration are difficult to control.To overcome the fluctuations of current density and zinc ion concentration in the feed,feedforward-feedback control strategy based on the mechanism model of zinc electrolytic is used.First,based on the model given by Scott,the key parameters such as the mass transfer coefficient are estimated and the energy consumption model is built.Then,a more complete zinc electrowinning process simulation model is established by adding the dynamic balance model of zinc ion concentration in the cell.For the situation such as the changes of current density and fluctuations of feed zinc ion concentration that affect the zinc ion concentration in the cell,feedforward-feedback control strategy is put forward,where the feedback loop uses PID,while the feedforward gain is calculated directly from the mechanistic models.Finally,the control strategy is verified using the simulation model,and the results show that this method is simple and effective.

Most of the basic multi-objective evolutionary algorithm is one kind of similar random search algorithm based on the concept of Pareto optimization,which with the slowly speed,especially for dynamic multi-objective problems.Accordingly,the hybrid optimization algorithm based on single and multi-objective gradient information (HSMGOA) is proposed.The algorithm confirms the direction of variation on each individual by using the gradient information.Firstly,the negative gradient direction information of each target in the population is calculated,those operation guarantees the individual species moving to the optimization declining direction for each of the single objective value effectively.Due to the conflict between each objective of multi-objective problem,it may cause the increase of the other objective function value if only considering the fall direction of one target.Therefore,this article also joins in the random weighted integration method,which fusing the gradient direction information of multiple goals to one search direction.Also,based on the traditional crowding distance selection method,this paper proposes a new select scatter point method to further speed up the optimization algorithm and provide the best initial population.Through the simulation of ZDT series test function and the analysis results with the NSGA2,it can be seen that the performance of the proposed algorithm is much better than the NSGA2 algorithm with less run times.It also shows that this algorithm has faster convergence.Finally a new algorithm was proposed by mixing the algorithm with NSGA2,and it is applied to dynamic multi-objective optimization of fed-batch bioreactor,the preferable Pareto optimal solution set is obtained.Compared with NSGA2 and MOPSO,the new algorithm shows better performance.

Strategy based on interval selection is widely used in the near-infrared spectroscopy analysis. Inspired by interval partial least-squares method (iPLS),the present paper proposed a new wavelength method combining interval selection strategy with least-squares support vector machine (LSSVM).By overcoming the shortcomings of traditional interval selection methods whose predictive ability totally depend on the linear model,this new algorithm,named as iLSSVM (interval LSSVM),can select the optimal informative interval more reasonably to significantly improve the model prediction accuracy with less modeling variables.Two real near-infrared datasets were applied to this new approach and the prediction performance was compared to the other interval selection methods.The experimental results demonstrated that the root mean square error of prediction (RMSEP) of this new method is 20% and 4% smaller than that of full-spectrum PLS modeling method respectively,and is 28% and 2% smaller than that of the traditional iPLS (interval partial least-squares) method respectively.

Steady-state detection is to determine whether the chemical process is operating in a state of steadiness or not.It is an essential step in the field of process optimization and so on.Based on wavelet packet analysis for denoising,a steady state detection scheme is constructed in this study.The scheme includes three steps,removing coarse error,denoising and identifying steady state periods,and is featured with simple implementation and good accuracy.In the denoising step with wavelet packet transform,a compensating factor is introduced for the threshold,making it flexible and easy to achieve the right balance between removal of the noise and preservation of the true value of a signal.Two simulation tests are presented to demonstrate the effectiveness of the scheme of this study.

The group search optimizer (GSO),which is inspired by animal searching behaviour and group living theory,is a novel optimization algorithm.In this paper,a novel group search optimizer called global group search optimizer (GGSO) is proposed to improve the performance of standard GSO.In the optimizing,the initial population of GGSO is generated uniformly in the search space.Early in the algorithm,GSO evolutionary strategy is retained and PSO evolutionary strategy is adopted during the later computation period.The main approaches included introducing crossover operation in each iteration to increase the diversity of individuals,breaching the restrictions of local optimization points with a new chaotic disturbance mechanism and mutation operation during the later computation period.Tests are carried out through four standard test functions on GSO,LDWPSO and GGSO independently,the results shows that GGSO has a preferable convergence rate and accuracy.The application of gasoline blending online shows that GGSO is effective.

In order to improve the estimation accuracy of a soft sensor in the process of chemical production,a combination model for soft sensor is presented based on Gaussian process and Bayesian committee machine.The original data are classified into several subclasses,and then,the sub-models are built by Gaussian process regression.In order to get a global probabilistic prediction,Bayesian committee machine is used to combine the outputs of the sub-estimators.Finally,the algorithm is applied to a soft sensor model for a production plant of bisphenol A.Simulation results show that the integration algorithm can make full use of sample information in the actual production,and the estimated accuracy of model is improved,and the generalization ability is better,comparing to the traditional switch or a weighted combination of multiple model.

Energy consumption analysis of products is of great significance for evaluating the sustainability of processes in chemical processing systems.A reasonable evaluation of energy consumption for streams in the system is crucial to ensure the accuracy and rationality of energy allocation to products.From the view of energy consumption assessment of waste streams,a reasonable analysis of streams is examined by distinguishing the types of energy consumption and their allocating coefficients.A device-level MFA-EFA energy consumption assessment approach of the product is proposed based on mass flow analysis (MFA) and exergy flow analysis (EFA).The proposed method is used to evaluate energy consumption of five products in the vacuum gas oil hydrotreating unit in a refinery.The consistency of the proposed method is exemplified from three aspects including devise or process scales,energy consumption intensity and CO₂ emission intensity of the products.

Parameter setting plays an important role in the performance of optimization solver.Hence,the potential solving performance can be full employed by tuning the parameters.The increase of complexity and scale of process model has a great influence on the efficiency of parameter auto-tuning algorithm.In this paper,according to the independence of the parameter setting selection and evaluation of the random sampling based parameter auto-tuning algorithm,the efficiency is improved by utilizing parellel technology. The numerical experiment shows that the parallel computing based parameter auto-tuning algorithm has an enhanced tuning efficiency and it is suitable to online operation optimization for operation run of production process with sufficient hardware support.

In consideration of the many-to-one raw materials procurement and pricing problems of base oil supply chain the refinery dominants,a revenue sharing contract game model is proposed using Stackelberg game methods and decentralized decision.The existing pricing model in the context of base oil supply chain is also introduced.Then the PSO algorithm is applied to estimate the model parameters.Finally,a numerical simulation is given,in which effects of various parameters on profit of supply chain members are analyzed and the comparison between the proposed revenue sharing contract game model and the existing pricing model is also discussed.It indicates that the revenue sharing contract mechanism coordinates the profit distribution among the members of the supply chain and increases the overall profit of raw materials suppliers and refiners.Moreover,a general procurement and pricing model is proposed,which is more applicable.

In liquid-chlorine pipeline leak detection,leak signals are difficult to obtain for feature extraction and modeling,a pipeline leak detection method with none leakage samples based on power spectrum comparison is proposed.In view of the same type of sensors have the similar output signal characteristics used in different transportation medium,by analyzing and comparing the frequency domain features of existing leak signals and normal signals of gas pipeline,frequency domain feature extraction method based on double moving windows and power spectrum comparison is proposed.Taking the normal signals of liquid-nitrogen pipeline as the target object,leak diagnose model for liquid-chlorine pipeline leak detection based on SVDD with none leakage samples is completed,and reliable leak detection is implemented.The results of online application show that this method can detect pipeline leak effectively.

On the basis of the collector pressure control characteristics,a large number of historical data in production process the operating experiences,an operation pattern extraction and migrating reconfiguration method is proposed based on the subtractive clustering algorithm for controlling coke oven collector pressure.The subtractive clustering algorithm is used to realize the pattern discovery,the pattern rule extraction,eventually forming the optimized operation pattern database in order to optimize the pressure set-points.The pattern reconfiguration strategy based on the model migration ideological is used to carry out the operation pattern revision.The industrial application results demonstrate the effectiveness of the proposed method.

There exist two different strategies in model predictive control (MPC),i.e.,set-point control and zone control.In comparison,zone MPC is thought to be a better strategy.However,there exists an improved set-point control strategy,i.e.,double-layered MPC,which computes set points through the steady-state target calculation in the same control cycle.In this paper,from the aspect of process steady state,the principles of zone control and double-layered MPC are analyzed.We discuss the differences and similarities between the two strategies both qualitatively and quantitatively,and prove that the two strategies are consistent in a certain context.Simulation results are given to validate the conclusion.

Aimed to obtain inherently safer chemical reaction path during early phase of chemical process,fuzzy safety assessment is integrated to reaction path synthesis to provide effective method.Indices are selected based on process information in the reaction path synthesis phase.Through setting membership function of each index,establishing fuzzy inference system,and determining the weighting vector by AHP,the fuzzy assessment method is proposed.To eliminate the impact of medium variable,the fuzzy inference system of single input variable and double input variables are provided respectively.Integrating the method to reaction path synthesis,the benign reaction path will be obtained through material screening rules,fuzzy safety assessment,and optimization treating safety as objective function.By analyzing the application of reaction path synthesis for carbaryl production,reaction paths with their objective function value is obtained.The results for two kinds of fuzzy system are compared finally.

Water is a key element for the normal functioning of the industry and the society,which is now gravely concerned about.Though the purpose of water-using network (WUN) synthesis is to reduce the freshwater consumption and the wastewater discharge,the network reliability is worth considering for the sake of reducing the repair and maintenance cost so as to reduce the total annual cost.The WUN design has a significant impact on reliability issues.So the issue discussed in this paper is to integrate the reliability analysis into the WUN synthesis problem.The first step is to determine the reliability of the water-using system.Divide the whole system into possible subsystems,which means that the whole system maybe include one or more different subsystems.The reliability of WUN is evaluated in terms of the subsystems.For subsystem whose degree of reliability fails to meet the process specification,the dividing of the system is not accepted.Then the minimum freshwater requirement should be considered.A non-linear programming mathematical model is established to solve the problem.Take all the accepted situations into consideration.Choose the one that consumes the minimum freshwater.At last,an example is used to illustrate the application of the proposed method.The degree of reliability of the system is 0.904 which meets the process specification and the minimum freshwater is 142.88 t·h^-1 with 2.27% increasing compared to the situation which takes no consideration of reliability.This result means this method is effective and could be accepted by industry.

Boiler combustion process modeling is critical for the accuracy and reliability of combustion optimization.Firstly,a steady-state detection (SSD) algorithm is used to extract steady-state samples for building steady-state combustion process model.Considering the unbalance of samples over power load,a new method of data division,which divides the available data into training subset and test subset according to load,is proposed to improve model generalization.Then,a single factor graph analysis is conducted to determine searching range of three SVM model structural parameters.After choosing the model parameters by combining grid search with cross validation,four multiple inputs single output SVM models,including boiler efficiency,NO_x emission,flue gas temperature and the unburned carbon in fly ash,are established based on the divided data.The results are demonstrated that four models all have a good generalization capability.

The traditional principal component analysis (PCA) similarity factor method does not make full use of the higher-order statistics of the process data,which results in degraded fault identification performance.In order to solve this problem,a statistics pattern analysis (SPA) similarity factor method is proposed in this paper.Firstly,the original process data are transformed into the statistics space by using the SPA.Then,the PCA is adopted to obtain the principal component directions in the statistics space.Finally,the similarity between the principal components is calculated to identify faults.Simulation results on the continuous stirring tank reactor (CSTR) process show that the proposed SPA similarity factor method is more effective than the traditional PCA similarity factor method in terms of identifying faults.

The multi-period operation of boiler steam system with decaying performance was formulated as a mixed-integer nonlinear program (MINLP).The objective function accounted for both the cost of fuel consumption and the cost of boiler cleaning and maintenance.To solve this problem,a method called steepest descent-approximate linear programming (SDALP) is proposed,which is based on the thought of purposeful search and rapid convergence from steepest descent method.The algorithm use the thought of the steepest descent to replace the role of the restricted step and the reduced coefficient in approximate linear programming (ALP).The algorithm also introduces the concept of the move-into vector and move-out vector and positive redundancy,and adjust the boundary of variables by the compound of this two vectors.By redefining the judgment conditions and the way of boundary adjustment of the algorithm,it can excludes the optimal solution out of the boundary condition,which is caused by the subjective choices of initial feasible point,step restriction and reduction coefficient of traditional approximate linear programming.After several test cases,numerical results shows that not only both of the accuracy and the speed of the algorithm are better than the previous one,but also the global solution.Apply the method in a real thermal power plant,the computation time for the SDALP method was proportional to the number of periods,but usefulness of the method shows that it also can achieves remarkable energy saving when compared with the original approaches.

Batch process plays an important role in the processing of specialty chemical,semi-conductor,food and biology industries for producing high-value-added products to meet rapidly changing market,which causes their monitoring and control to emerge as essential techniques.A novel method for uneven-length-stage batch process was proposed by using a new stage segmentation strategy based on k-means clustering method.First,the three-dimension data array is unfolded in variable ways,resulting in two-dimension forms.Then according to data correlations,two-dimension training data set is divided into many clusters by applying the conventional k-means clustering method,and sequentially principal component analysis (PCA) is performed for each separated cluster.For online monitoring,a similarity index is introduced to choose the most suitable local model.The effectiveness and utility of the proposed method was validated through the simulation benchmark of fed-batch penicillin production.

Proton exchange membrane (PEM) fuel cells are electrochemical devices that covert the chemical energy to electricity directly.This study was concentrated on the air supply system.This paper presented an improved four order fuel cell air supply system model,analyzing its constraints.And the linear parameter varying (LPV) model was proposed to deal with the non-linear characteristics of the dynamic model.To cope with the problem of immeasurable states,the Kalman filter was used to estimate the states and the output variables which should be measured were selected.Additionally,the state space model predictive control algorithm satisfying the constraints is designed based on LPV model to control the compressor voltage to ensure an adequate response to hydrogen fuel.Simulation results showed that model predictive controller based on LPV model is able to effectively control the fuel cell air system,and to meet the compressor surge and blocking boundary conditions and other constraints,Comparing with single model predictive control,better control performance was obtained with the proposed LPV-MPC.

The primary objective of the model verification is to establish a comprehensive test scenario set.In order to solve the problems of current existed model verification methods,a model verification method based on qualitative trend and signed directed graph (SDG) is proposed in this paper.First of all,by summarizing various SDG modeling methods studied in previous research,a modeling method of the SDG verification model is proposed.Secondly,the model verification method based on qualitative trend and SDG is presented,including the establishment of SDG verification model,generation of test scenarios,reasoning of standard trend,extraction and identification of simulation model data trend,and comparison of qualitative trends.Finally,this paper takes Tennessee Eastman process as an example to prove the validity of the aforementioned method.

Introducing purification devices in hydrogen systems can improve the hydrogen reuse effectively and economically,which is widely used in hydrogen systems.When a purification process is introduced,the position of the hydrogen pinch point may change and also multiple pinch points of the system may occur.For the pinch point has a considerable role for analyzing the performance of a system and optimizing it,the change of pinch point will influence the hydrogen distribution system.The pinch point change rules and tendencies for hydrogen systems with purification reuse processes are analyzed in different purification product concentration cases,based on the pinch technology through graphical methods.The work has great signification for hydrogen systems integration with purification reuse processes.

Sludge volume index (SVI),a key sludge sedimentation performance evaluation index,is difficult to be obtained accurately online and the conventional approaches are time-consuming,tedious and complicated.A new recurrent fuzzy neural network (HRFNN) method is proposed in this paper to predict the evolution of the sludge volume index (SVI).HRFNN is constructed by adding feedback connections with the internal variable in the third layer of the fuzzy neural network,so it achieves output information feedback.Finally,the results of simulation indicate the efficiency of the modeling method.And compared with other fuzzy neural networks,the scale of network can be simplified and its capability of dealing with dynamic information can be strengthened,it also has better accuracy.

Iron precipitation is the key process for iron removal in zinc hydrometallurgy.In iron precipitation process,it is very important to stabilize the pH value of the solution.However,the pH value fluctuates strongly in practice because of the nonlinearity and coupling in iron precipitation process. At first,the reaction kinetics mechanism of iron precipitation is analyzed.Then,the relationship between the pH value of the solution and the amount of zinc calcine (AZC) added into the solution is researched and the AZC added into the solution is calculated according to the technical requirement of pH value. At last,the parameter self-tuning fuzzy controller is proposed to adjust AZC added into the solution. Simulation results show that the parameter self-tuning fuzzy controller is very effective and it is a feasible method to stabilize pH value in iron precipitation process.

Self-adaptive differential evolution algorithm (SaDE) is an improved version of differential evolution (DE) with strategies and parameters changed automatically.SaDE can be used in optimization of chemical process whose model is always computation expensive,and the computation cost can be saved. However,SaDE doesn't perform well when considering convergence rate and precision cause optimization of chemical process is difficult.A Kriging model based SaDE algorithm was proposed to improve performance of SaDE.Using data from SaDE which means the population and objective value measuring searching space during given generations,an approximation model of Kriging can be built. Using gradient of the model,the local optimal point can be obtained which then competes with the best individual of population.By adding the local optimal,the population is perturbed and the whole evolution process is changed if the new local optimal point is better than current optimal individual.This strategy makes the Kriging based SaDE perform better than the original SaDE.Ten benchmark functions from CEC2005 were chosen for testing this new method,and the results showed that Kriging based SaDE performs better than SaDE and DE.Kriging based SaDE was used in optimization of the typical styrene plant which is a computation expensive plant-wide model,and less operation cost was obtained.

In this paper,a novel calculating method on relative permeability curve is proposed based on improved RBF neural network.In this method,the hybrid RNA genetic algorithm (HRGA) with the position displacement idea of bare bones particle swarm optimization (PSO) changing the mutation operator is proposed.The HRGA is applied to optimize the value of radial basis function centers in the hidden layer of RBF neural network.This method is used in the calculation of relative permeability curve.By comparing and analyzing the accuracy of relative permeability curve calculated by HRGA-RBF and standard RBF,the experimental result indicated that HRGA-RBF can improve the calculating accuracy obviously.

Batch processes,where a great number of products are produced to meet human demands in daily life,have become significant in chemical manufacturing.However economy globalization has resulted in growing serious competitions in traditional chemical process industry.In order to keep competitive in the global market,each company must optimize the production technique and management.And scheduling is the core of production management in chemical processes.Considering the character of such processes,this paper studies the multi-objective no-wait multi-task scheduling problem of batch processes.Model and optimization methods are introduced and the problem is decomposed into two sub-problems,the sequencing and timetabling problem and used hybrid non-order strategy and modified complete local search with memory to solve the two problems separately.In addition SPEA2-based multi-objective selection is present.A large number of experiments on benchmark problems proved the feasibility and effectiveness of this algorithm.

For uncertainties of parameter detection in penicillin fermentation process,penicillin concentration prediction scheme by chaos least squares support vector machine is put forward.The LSSVM parameters were optimized by chaos optimization algorithm to set up Chaos-LSSVM model. Firstly,simulation is conducted for two kinds of nonlinear function curve.The results show that the algorithm has good precision of modeling.Secondly,taking the data of Pensim simulation platform to model penicillin concentration curve,predicting the product of the penicillin fermentation process.The results show that chaos optimization algorithm has a good global optimization performance,which prevents parameters from falling into local minimum,improving the prediction accuracy of the model.

Aiming at the time-variant and nonlinear characteristics of polypropylene production process,a discount moving window recursive PLS based on data blocks (DMW-RPLS) is proposed in the paper. Based on PLS,the amount of data blocks is controlled through bringing in the moving window and data blocks are discounted through bringing in the forgetting factor,and then updates the soft model in order to predict the measuring variables in real time.Through the study for the polypropylene melt index,the result shows the discount moving window recursive PLS can reduce the calculation and overcome the data saturation,and then use the process information effectively to reflect the process characteristic quickly and accurately.

Ultrasonic guided-wave detection of pipeline corrosion is an important pipeline inspection technology.Conventional ultrasonic guided-wave echo signal processing is based on A-scan, which easily misses defects and is unable to locate defects, such as circumferential position.L (0,2) modal of center frequency 48 kHz guide-wave was used to detect pipeline.According to comparison of A-scan result and artificial defects, it was suggested that wavelet transform was used to process data, then log linearization was made, and finally C-scan was used to display pipeline corrosion.Since echo signal of gradational or large corrosion defect was gentle, and could be easily missed, so correlation analysis was used to choose sym5 wavelet transform processing.It could clearly show the defect location and determine the basic direction of defects.Because guided-wave propagation energy decayed exponentially, linear representation of the extent of corrosion defects could be achieved by log linearization of echo amplitude.Based on A-scan collected from the pipe cross-sectional distribution of echo signals from 120 probes,the pipeline was circumferentially unfolded 360°, and colors represented the extent of corrosion.C-scan could not only accurately scan the axial position of defect but also display the distribution of defects in the circumferential direction.The experiments on corrosion detection of a ±6m long pipeline showed that the proposed method could accurately detect and locate the locations, sizes, and shapes (rectangles, circles, ellipses) of pipeline corrosion defects.C-scan ultrasonic guided-wave detection provided a simple and effective method for application of guided-wave to locate defects.

According to the shortcoming of total projection to latent structure (T-PLS) algorithm which has a higher false alarm rate and a higher missing report rate when it is used to detect fault in the nonlinear process,a new algorithm which is a combination of kernel function and T-PLS (KT-PLS) is proposed in this paper. Firstly,the low dimension process data is projected to the high dimension feature space through kernel function,which can resolve the nonlinear problem.Secondly,the feature space is divided into four subspaces by the guiding of quality variable.These four subspaces include the quality directly correlated subspace,the quality orthogonal subspace,the quality uncorrelated subspace and the residual subspace. Finally,statistics D and Q are used to detect fault respectively.Applying this algorithm to the Tennessee Eastman process (TEP),the simulation results of various fault modes show that KT-PLS is better than T-PLS in monitoring strong nonlinear system.

To solve the temperature control problem of the catalyst baking furnace,a heuristic dynamical programming (HDP) control method is proposed.The optimal control policy in HDP scheme is approximated gradually by implementing policy evaluation and policy improvement repeatedly.The system dynamics and the critic model are established by artificial neural networks.The learning algorithm of the HDP controller is clarified based on the gradient-decent principle.The proposed controller is tested on a baking furnace in a certain catalyst company.The experimental results indicate that the HDP controller has stronger ability to accommodate different work conditions than PID controller.In comparison with PID controller,the control precision in HDP controller increases about 70%,and the average electric current in HDP controller decreases about 5%.

The fault detection and diagnosis is very important for a security of control systems.By analyzing deeply on the relationship between changes associated with process variables and between the device failure,a diagnosis technique is presented for process fault based on the correlation coefficient in process variables.The process diagnosis function is defined for interrelated variables based on the correlation coefficient (correlation coefficient,multi-correlation coefficient,partial correlation coefficient) constraints,whether the involved devices and instruments are fault is estimated according to the correlation coefficient change variety.If a fault in device or system occurs,the correlation coefficient and the diagnosis function changes will be caused,the fault location and the failure equipments are determined by logic analysis for diagnosis function value.Finally,the validity of the method is verified with a distillation column practical engineering case.

Non-dominated sorting evolution strategy (NSES) is used to solve the multi-objective optimization problem of coal gasifier.The solving of the two classical test function with NSES indicates that NSES is effective comparing to NSGA-2.Aspen Plus is applied to the simulation of the coal gasifier process.On this basis,with the ratio of oxygen to coal,ratio of water to coal and gasification pressure as optimization variables,and taking the yield of effective gases and cold gas efficiency as goals,the sensitivity analysis is completed.The results show that the three variables have more or less influence on gasification indexes.NSES is used to solve that multi-objective optimization model.The two goals' compromise solution including the cold gas efficiency and yield of effective gases can be achieved based on the Pareto front.

The problem to seek further heat integration for the hydration of ethylene oxide by reactive distillation (RD) to produce ethylene glycol,which involves reactions with highly thermal effect,was investigated.An innovative process,internally heat-integrated RD column (R-HIDiC),was proposed to maximize energy efficiency of the system,in which the reactive section and stripping section was partitioned into two parts with different operating pressures so that the internal heat transfer between reactive section and stripping section was allowed.Steady-state simulations based on Aspen Plus software were performed to investigate the feasibility of internal heat integration,and an energy integration design without the need of reboiler was provided by optimizing the heat distribution in R-HIDiC.The results obtained demonstrated that,comparing with the conventional design,the proposed R-HIDiC design did not need a reboiler,thus,the total energy input can be supplied by the heat of reaction of the hydration of ethylene oxide and the electricity power of compressor,which leads to a reduction of the operating and total cost by 47.2% and 39.1%,respectively.

Owning to the large amounts of factors for causing filamentous bulking sludge and many troubles in expressing its mechanistic model of causing filamentous sludge bulking,a simplified mechanistic model is proposed for sludge volume index (SVI) from the filamentous bacteria growth kinetics in this paper.First,the appearance of filamentous sludge bulking is studied.And then the relationships among the SVI and the impact factors are obtained by analyzing the filamentous sludge bulking process.Secondly,the main factors for SVI are ensured.The simplified mechanistic model of SVI is designed.Meanwhile,the parameters of this proposed model are adjusted by using the method of data analysis and statistics.Finally,this simplified mechanism model is used for a real wastewater treatment process.The results show that this proposed model can simulate the mechanistic of filamentous sludge bulking and predict SVI accurately.

In water bloom prevention and control,water bloom prediction is always a difficult problem. This paper proposes a new water bloom prediction method based on multiple characteristic factors time series analysis to take into account the integrated effect of multiple characteristic factors along with the periodicity and random effect of environmental variables,to solve the problem that existing bloom prediction is not accurate enough,and to analyze the correlation between influential factors and water bloom.A multidimensional hidden periodic-auto regression (MHPAR) model is put forward based on the characteristic factors time series.A water bloom prediction method and an influential factors analysis method are put forward by using multidimensional period stationary time series analysis.Comparing the proposed model with other traditional time series models,such as auto regression (AR) model,hidden periodic-auto regression (HPAR) model and multidimensional auto regression (MAR) model,it has been found that multidimensional hidden periodic-auto regression model is useful and accurate for establishing multiple characteristic factors time series of water bloom.

According to the measurement,control and signal processing requirements of chemical machinery and equipment,and taking two PCs connected with Ethernet as hardware platform,a multi-function virtual signal spectrum analyzer was developed using virtual instrument graphical development software package LabVIEW as software development tool.The arbitrary signal generator virtual instrument (VI) module was developed on a PC,and the spectrum analyzer VI module was designed on the other.The arbitrary signal generator VI module could generate two channels waveform signal with additive white noise.On the other hand,the spectrum analyzer VI module collected the waveform signal adding with noise generated by the arbitrary signal generator VI module through Ethernet,and carried out low pass filtering,window smoothing processing,amplitude spectrum and phase spectrum analyzing to the waveform signal.The test results on chemical machinery monitoring of a pilot plant showed that the virtual signal spectrum analyzer had good performance.

Recognition of gene promoter DNA sequence is difficult with the complex structure and the huge amount of data.In this paper,the positional densities of oligonucleotides are modeled by Gaussian mixture model.It can identify less frequent but important motifs,since the positional density is independent of the actual occurrence frequency of the oligonucleotide.These motifs generally correspond to the consensus sequences of transcription factor binding site.GMM is used as eukaryotic promoter LS-SVM kernel,which simplifies the LS-SVM as LS model.The algorithm is simplified and the computational complexity is decreased.The simulation results show the accuracy is improved compared with Bayesian classifier,and is same to LS-SVM with RBF kernel,moreover the model building time is shorter.

A problem of mode mixing occurred in implementation process of local mean decomposition (LMD) method,a fault diagnosis approach for rolling bearing based on ensemble local mean decomposition (ELMD) and least squares support vector machine (LS-SVM) was proposed.Firstly,by using ELMD method,the vibrational signal of rolling bearing was decomposed a series of product function (PF) components,and then the PF components which contain main fault information were selected,and the kurtosis coefficients and energy characteristic parameters extracted from selected PF components were regarded as fault feature which was served as input parameters of LS-SVM to identify the working status and fault types of rolling bearing.The analytic results of fault-free,inner-race fault and outer-race fault of rolling bearing indicate that the working status and fault types of rolling bearing can be identified accurately and effectively by using the approach based on ELMD and LS-SVM.

An adaptive control algorithm is proposed for continuous stirred tank reactor (CSTR) with unknown functions based on the approximation property of the neural networks.Because the considered reactor contains the nonlinear property and the unknown functions are included in the subsystem,it is a completed system and is very difficult to be controlled.In order to avoid the difficulties,a novel recursive procedure is given to remove the interconnection term and special approximated functions are defined to be approximated by using the neural networks.Using the Lyapunov method,the algorithm ensures that all the signals in the closed-loop are bounded and the output can converge to a neighborhood of zero.A simulation example is given to show effectiveness of the algorithm.