Mobility of bed is the important factor for jigging separation process.A soft sensing modeling method based on the least squares-support vector machine(LS-SVM)is developed to deal with the problem that mobility cannot be measured directly online.In full consideration of highly nonlinear and strong coupling characteristic of separation process,an LS-SVM multi-model method based on affinity propagation(AP)clustering is presented and applied to avoid bad accuracy of single model expressing multiple working positions.In the presented method,AP-clustering algorithm is used to cluster training samples.Then,the sub-models are trained by LS-SVM.Finally,the predicted values of the testing samples are estimated by the sub-models after it is classified by switchover.Simulation results show that a better prediction for mobility of jig bed is obtained by the LS-SVM multi-model method based on AP-clustering algorithm.

Considering the unreacted reactant p-xylene and by-product methyl acetate as feed impurities in the industrial acetic acid dehydration process using n-propyl acetate as entrainer,HOC and UNIQUAC models were chosen respectively to compute the strong nonideality of vapor and liquid phase of the quinary system.The binary parameters in the UNIQUAC model of methyl acetate-p-xylene and n-propyl acetate-p-xylene systems were obtained by regressing the phase equilibrium data from experiment.Combining with binary parameters of the other binary systems built-in simulation software,the mechanism model of heterogeneous azeotropic distillation was developed and the simulation of the industrial HAc dehydration process including dehydration column,PX pure column and entrainer recovered column was conducted with Aspen Plus.The simulation results of key parameters agree well with the process data with the errors within ?6%.Thus,the model can describe the HAc dehydration process accurately,and it lays a solid foundation for the further study of industrial acetic acid dehydration process.

In actual sintering process,ratios of coal are widely determined by experience.Meanwhile,the consumption models of coal built from many sintering experiments and statistics are too complicated to apply in sintering process.This paper proposed a minimal coke consumption calculating method.Firstly,a BPNN temperature field model of different BTP(burning through point)and thickness is worked out according to isotherm data from actual sintering process.Secondly,the samples of the relationship between BTP,thickness and energy consumption for temperature rising are found and LS-SVM algorithm is applied to build the model due to these characteristics.The energy consumptions of carbonate decomposition and water evaporation are calculated by analyses of physical and chemical changes.Finally,the minimal coke consumption is worked out by coke combustion release process analysis.The simulation and actual sintering data show that the result of this paper reflects the actual need of minimal coke consumption in sintering process and this calculating method can guide the actual production process.

As for the problem that the estimation precision of soft sensor model is not enough on line in chemical processing,a method of multi-model soft sensor is proposed based on improved clustering and weighted bagging.It improves clustering result by reducing error dividing probability with K-neighbors based on traditional fuzzy C-means clustering,and the training sample set is grouped into several feature sets with correlation analysis.At last,a multi-model is constructed by support vector machines adaptively according to weighted bagging algorithm of ensemble learning.The simulation results show that every feature model is assigned with weight reasonably,and the estimated accuracy of model is improved,and the generalization ability is better.

Dynamic rigorous model including kinetics and thermodynamics is established for the ethylene slurry polymerization plant.Thermodynamic properties are estimated by Kriging technique with the maximum relative error less than 2%.Based on the equation oriented model,control and state variables are synchronously discretized using orthogonal collocation finite elements(OCFE) method to develop simultaneous dynamic simulation for grade transition operations.The results are validated by comparing steady state data of five different grades obtained by Aspen Plus.The dynamic grade transition based on average molecular weight is also presented,which is close to that in Aspen Dynamic.

Because of the large scale feature of the rigorous model,process optimization on multi-stage columns could be difficult to converge.In this paper,a reduced-order model based on the group method is proposed for high purity distillation columns which correlates the component recoveries with absorption factors and number of theoretical stages.Considering the strongly nonlinear characteristics of absorption factor along the column of high purity distillation column,a piecewise linear fitting method is proposed to obtain the component recoveries.Simulations are conducted under different loads to compare the performance of the reduced-order model and the rigorous model.The results show that the model scale is dramatically decreased from 2377 variables in the rigorous model to 34 variables in the proposed reduced-order model,the convergence time is also greatly reduced from 6136 s to that around 1 s,but the model accuracy is not damaged with very small errors between these two models.

This paper introduces a design of the heat duty control strategy for methanol distillation column bottom.Based on the technical background of the processes and the analysis of the working conditions of the practical applications,taking into consideration that the dual heat exchanger is used as a reboiler in bottom of the distillation column,the heat duty control strategy is selected in the three-column methanol distillation control design.Simulations are given by using Aspen.In contrast to the regular temperature control scheme of methanol distillation,the method proposed in this paper has advantage of reducing the impurity content of the glycol ether.Moreover,the proposed methods can deal with the disturbance of flow rate and components variations.The quality of products could satisfy the industrial standard and requirement,which indicates that the proposed heat duty control scheme has definite significance and economic benefit in practice.

In order to measure the color of fabric in the batch dyeing using multi-dyes,a soft-sensing technique based on particle filter is proposed.It uses the particle filtering method to predict the concentration of dyes through the absorbance of dye liquor in the jet dyeing machine and then develops a soft-sensing model applied to predict the color of fabric.The feasibility and validity of the proposed technique in this research is verified by its application to real examples.The results show that the soft-sensing system has better measuring precision and their color differences are all less then 1.0 CIELAB,which meets the technical requirements of dyeing industry.

Purified terephthalic acid is one of the most important chemical raw materials.p-Xylene oxidation is a solid-liquid phased catalytic reaction process under high temperature and high pressure.The reactor is the core unit of purified terephthalic acid plant,which affects the product yield and quality noticeably.Current traditional control strategy can meet the demands of the industrial operation,but the variation of the key quality indexes-the concentration of 4-carboxybenzaldehyde in crude terephthalic acid and the oxygen content in the exhaust are big,which influence the further optimization of the process.In this paper,in order to solve this problem,a multivariable predictive control strategy fulfilled by the FRONT-Suite software is put forward.After the application,the control performance on the 4-carboxybenzaldehyde concentration in crude terephthalic acid and oxygen content in the exhaust of p-xylene oxidation process are improved obviously.

In order to improve the performance of chemical process alarm systems,it is imperative to optimize assignments of process alarm thresholds.In response to limitations of traditional threshold assignment methods,based on historical data,this paper firstly invokes kernel density estimation methods to identify process alarm states before an objective associated with alarm threshold optimization in terms of minimizing the probabilities of false and missed alarms is established along with enabling numerical solvers.Simulation results on TE process demonstrate that the proposed approaches can effectively reduce the number of false alarms as well as limit that of missed alarms.

The hydrogen system has become an important component of the refinery with the increasing demand of hydrogen.It is significant for the optimization of the hydrogen network to realize the minimization of economic objectives and stable operation under all the operation conditions.A flexible optimization strategy is proposed based on the varying operation conditions and demand of hydrogen system in refinery.A mixed integer nonlinear programming(MINLP)model is formulated to optimize the hydrogen network under multi-scenarios.Then the MINLP model is linearized and solved with Lingo software.A case study based on the data from a refinery is showed to demonstrate the effectiveness and feasibility of the presented flexible optimization strategy.The comparison between the proposed strategy and the original hydrogen network shows that the presented flexible optimization strategy has better flexibility,and significant savings and stable operation of hydrogen system is realized.The proposed strategy plays an important role in guiding the management of hydrogen system in refinery.

The iterative learning control under periodic operation is used to control molecular weight distribution(MWD)in semi-batch polymerization process in this work.Firstly,the effect of periodic control on the MWD of the semi-batch polymerization reactor is analyzed.Simulation results show that the broadening of molecular weight distribution is achieved through periodic operation,therefore the performance of polymerization is improved.Then the optimal control strategy based on particle swarm optimization(PSO)with an improved performance index is used to control the shape of MWD,in which the control variable is duty cycle of the periodic feeding.In simulation experiment,the control inputs can reach to the optimal values from batch to batch in the presence of model plant mismatches,and the MWD of the polymer approximate the desired MWD gradually,thus the effectiveness of the periodic operation controlling method on MWD is verified.

To enhance the robustness of the scheduling in refinery with operational uncertainty,a novel rescheduling method based on a two-level simulation system is proposed in this paper.The lower level model with uncertainty in the tank farm is used to simulate the deviation from the original plan.And the heuristic rule is further applied to the upper level simulation model according to the disruptions from the lower level to get the conservative plan which satisfies both material and components constrains simultaneously.Subsequently,by fixing the value of binary decision variables and penalizing the simulated deviation in the MILP model,a sub-optimal solution is obtained with less computational effort.The effectiveness of the proposed rescheduling framework is illustrated through a tank farm operation case in real world refinery.

A predictive control algorithm based on local modeling is presented for a class of non-linear systems.In order to obtain reasonable training data,both the Euclidean distance between the sample data and angle of vector are considered.And the algorithm is applied in the simulation control of aromatics isomerization process.Firstly,the local-model is obtained as predictive model by least squares support vector machine(LS-SVM)algorithm.And then,the control values are obtained by the rolling particle swarm optimization(PSO)algorithm.Finally,the presented algorithm is applied in the simulation control of aromatics isomerization process,and the results show the effectiveness and robustness of the algorithm.

In order to improve the productivity and energy saving level of styrene in the dehydrogenation of ethylbenzene,optimization is an effective technological mean.The application of improved non-dominated sorting genetic algorithm is studied in optimization for operating conditions of dehydrogenation of ethylbenzene.Conversion and selectivity of the process of dehydrogenation of ethylbenzene to styrene are considered as the two objectives,and the kinetic model and process conditions are the constraints of the problems.NSGA-Ⅱ(non-dominated sorting genetic algorithm)is used to solve the optimization question of above dehydrogenation of ethylbenzene process.According to the obtained Pareto optimal solution set,the influence of operating conditions on conversion and selectivity of dehydrogenation of ethylbenzene is analyzed.Fuzzy comprehensive evaluation method is studied to satisfy specified demanding,supplying referenced optimal operating conditions.The results demonstrated good performance of NSGA-Ⅱ for achieving global optimal.With this algorithm,a satisfactory solution in different operating constraints can be obtained.

For the problem of adding cold charges operation in the process of copper converter,a optimization method based on improving genetic algorithm is proposed.The GA was improved by population and individual dissimilarity degree and dissimilarity interval in order to enhanced global optimization and accelerate the convergence rate.Then,on the basic of predictive value of residual heat to the copper matte converting process,the IGA is used for the global optimization problem of adding cold charges operation.The simulation results and production application demonstrate the effectiveness of the scheme which is beneficial to the operation optimization of copper converter,increasing throughput of cold charges and copper output.

Dry desulphurization system of slurry fluidized bed boiler has wide range of applications for its low cost and high desulfurization efficiency.Since fluidized bed boiler operating conditions have important influence on the desulfurization efficiency,and the increased desulfurizer can improve the desulfurization efficiency while reduce the thermal efficiency of boiler,there exists optimal operating conditions.In this paper,the SO₂ concentration prediction model was firstly built with LSSVM,and then the accuracy and fast performance of the model was verified.Secondly,a comprehensive and accurate model of boiler thermal efficiency affected by desulfurizer was established through mechanism analysis. Finally,according to minimizing the enterprise’s overall operating costs,a total objective function for operation optimization was built.Based on SO₂ forecasting model and boiler thermal efficiency model,and by solving the overall optimization problem,the best boiler operating conditions under different SO₂ concentration limits and different SO₂ emission costs were obtained.Through the operation optimization,the cost of the desulfurization operation was significantly reduced,and the desulfurization efficiency can be significantly improved.What has done in this paper has important guiding significance on the optimal operation for the dry desulphurization system of fluidized bed boiler.

Water quality is related to the national economy and a major event,water turbidity after filtering in the water supply system is one of the key indicators of water quality.Against to large inertia,large delay,model parameters time varying and much disturbance of water turbidity control in city supply system,a new control algorithm is proposed to obtain better control results.The specific idea of the algorithm is the double controller:the track controller and the disturbance controller in view of different task.The former adopts a nonlinear compensation PI controller to track the set point,the latter uses a traditional PI controller to restrain the disturbance from the process.The gain K of the nonlinear compensation part is very important to the track controller and adopts expert rules according to the system response.The simulation comparison of the new algorithm and the traditional PID feedback control algorithm,Smith predictive control and dual PI control algorithm has been done in this paper.Finally the new control algorithm is applied to a pilot base.The results of simulation and experiment show that the new control algorithm can overcome disturbance in time,is not sensitive to changes of model parameters and has a strong robustness and a good tracking ability to ensure the stability of water quality.

A novel swarm intelligence optimization AC_ICPSO(ant colony and immune clone particle swarm optimization)algorithm is proposed.It combines ACO(ant colony optimization)and PSO(particle swarm optimization)to conduct dynamic swarm query.According to introducing crossover and mutation operator,encoding repeatedly,iterative choice,etc.,it leads to widen data range,improves search precision and convergence efficiency,avoids premature convergence,and reduces complexity of the conventional ACO or PSO algorithm.Then AC_ ICPSO is used to optimize the parameters of LSSVM(least square support vector machines)to predict the melt index of polypropylene,so the best model AC_ICPSO_LSSVM is obtained.The detailed researches on the optimized model are carried out based on the data from a real plant,and the result shows that the proposed approach has great prediction accuracy and effectiveness.

For coal gasification operation optimization problem,a multi-population competitive co-evolutionary cultural differential evolution(MCCDE)algorithm is proposed.In MCCDE,a competitive co-evolutionary strategy based on differential evolution and a fitness value evaluation method are designed.And some ideas from cultural algorithm are also introduced into MCCDE.Meanwhile,an operation optimization model is constructed.MCCDE algorithm is used to solve the problem of operation optimization model.Simulations with a Texaco gasification unit for example testified that optimized operation variables can be found and the effective gas rate can be increased by the optimization model and algorithm.Finally,a coal gasification operation optimization system software is designed and developed,modeling,control and optimization technologies can be used in practical production to gain more economic benefits through the use of the coal gasification operation optimization system software.

To solve optimal control problem,the control vector parameterization(CVP)approach is often used and the time grid nodes are usually fixed.In each time intervals,the controls are approximated by a function of time,which is determined by parameters.The partition of time grid will take effects on the accuracy and performance of the numerical optimization algorithm used to solve the original optimal control problem.A CVP method with variable time nodes is presented to optimize both of the control parameters and time nodes in the grid partition.The time node parameter is introduced by sigmoid function which approximate the switch procedure in the piecewise constant parameterization method.The derivatives of the performance to time nodes are derived and the time nodes constraints are handled by a given algorithm.An example of optimal control problem with two controls is solved by the proposed method.The different time grid partitions of two controls are obtained which show a higher approximation of the optimal control trajectories.

A neighbourhood-clonal selection learning algorithm(N-Clonalg)is proposed in this paper,which is combined with the idea of biological immune clonal selection system and multi-agent technology. Different from other artificial immune algorithms,N-Clonalg is based on the grid environment,and contains three main search operators,i.e.,N-clonal selection,N-competition and self-learning operators.Combining global and local searching operations,N-Clonalg overcomes the phenomena of precocious and slow convergence,and can better achieve the global optimal solutions effectively in the individual space,which is proved in multi-modal benchmarks.Distillation resources optimization shows that it has better search performance.

The raw material purchasing and storage cost are the capital bottleneck of iron and steel enterprises.According to the purchasing and consuming characteristics of sinter material plant,the iron ore raw material inventory optimization model committed to minimum cost was established.Furthermore,an optimation method which combine the particle swam optimation(PSO)and genetic algorithm(GA)was developed to solve the question.To prove the effectiveness of this method,it was validated by the actual running data of a 360 m² sintering production line in an iron and steel enterprise.The simulation results showed that the model could feedback the real condition of raw material inventory in sinter plant,and the optimal solution could be obtained by the GA-PSO method,which could give a reliable support for the material purchasing decision.

The mechanical model of fed-batch penicillin fermentation has been well studied for the last decades,but most of them can hardly be implemented to control fermentation process.Due to this problem,this paper uses Birol’s unstructured model,adjusts temperature and pH dynamics of the fermentation and deduces a simplified penicillin unstructured model.To increase production of penicillin,substrate feedrate which is critical in fermentation process is optimized.Because of the nonlinearity and constraints of the mechanical model,sequential quadratic programming(SQP)algorithm is used to the whole process,in which the feedrate trajectory is divided equally into several intervals to enhance optimization efficiency.Optimization results show that the penicillin concentration and yield are increased compared to the normal input.

The combined orthogonal polynomial neural network has been used to model the molecular weight distribution(MWD)of polymers,therefore MWD is decomposed into a function of its moments.In this paper,tracking of a desired MWD is realized through choosing moment vector as directive control goal.In order to obtain the control solution for the non-affine nonlinear multivariate system,a model architecture utilizing an improved nonlinear autoregressive orthogonal polynomial neural network is proposed,and the linear relationship between the model outputs and current control strategy U(k)is realized.Taking the higher-dimensional controlled variable into consideration,the paper shows the equivalent of independent variables in moment vector and parameters of distribution function,thus the criterion of moment vector dimensionality reduction are proposed to transforming the higher-dimensional output control problem into low-dimensional.Based on the improved neural network model,the output feedback control method is used for the moment control of MWD,and then the tracking of desired MWD is realized.The control method is tested on styrene polymerization reacted in CSTR,and simulation results proved the effectiveness of the method.This paper proposed a new solution for modeling and control of nonlinear multivariate system.

In this paper,a new adaptive fuzzy sliding mode control is used to deal with complex chemical process with multivariable nonlinear and non-minimum phase systems.This method for sliding mode control has good robustness,but existing the chattering problems of controller output.The fuzzy control is used to soften the control signal and simple the fuzzy control,with the combination of sliding mode control can make full use of the system information.By adopting a new adaptive fuzzy factor to adjust proportional variable domain,the control signal is softened and reduces the output chattering of sliding mode controller.Meanwhile,the fuzzy sliding mode control algorithms and stability analysis are given; simplified general fuzzy rule is got.To make the control system has strong robustness,better adaptive ability and high control precision by using the scaling factor to adjust input amount of domain and rules of membership functions on line.Finally,the results of simulation experiment on nonlinear SISO and MIMO chemical models show that even big changes or heavy interference is happened in its operation points; the control system still has good disturbance and strong robustness.

Agglomeration fault of fluidized bed reactor not only influences product quality,but also influences production seriously.In order to monitor fluidized bed reactor agglomeration fault,fault monitoring method based on low-frequency piezoelectric acoustic sensor and voiceprint extraction is proposed.Acoustic signals created by the polymer impacting on the inner bed wall of the fluidized bed reactor were monitored with pasting piezoelectric ceramic sensors on the outside wall of fluidized bed reactor,transmitting the charge with long shielded cables and audio sampling.The acoustic signals waveform in time-domain,their power spectrums and voiceprints of polymer under the conditions of normal and agglomeration fault have been analyzed and compared.The stability and distinction of voiceprints under normal and fault conditions have been especially compared.With voiceprint extraction and neural network model,agglomeration fault were monitored.The monitoring model has also been verified with voiceprints extracted from other acoustic signals sampled from the sensor mounted at different position(1 meter apart),the same diagnose results are achieved.It shows that the method proposed has high time-space domain robustness.The original signals were re-sampled with different signal extraction rates.Voiceprint extraction,training and verification of monitoring model respect to re-sampled data with different extraction rates have been implemented,results show that agglomeration fault monitoring result will not be influenced with appropriate reduced signal sampling rates.This provided a new system architecture and implementation method for fault monitoring of fluidized bed reactor agglomeration.

The fault detection time of many methods based on kernel independent component analysis(KICA)is easily affected by the order of independent components(ICs)and the number of dominant ICs chosen empirically.Aiming at this problem,a method based on KICA and Gaussian mixture model(GMM)is proposed.KICA is used to extract ICs from the dataset measured under the normal condition,GMM is adopted to fit the probability density function of each IC,and then a monitoring statistic and corresponding control limit are constructed based on the GMM.The average value of each IC’s monitoring statistic is calculated and applied to judge its non-Gaussian degree.Each strong non-Gaussian IC is monitored independently by the built monitoring statistic,and the weak non-Gaussian part is monitored by principal component analysis.The simulation results on the Tennessee Eastman process illustrate that,in contrast to the fault detection methods based on KICA,the proposed method needn’t sort the ICs and select the number of dominant ICs avoiding their effects on fault detection time,make effective use of process information,and shorten the fault detection latency.

Many fault detection methods based on kernel independent component analysis(KICA)only consider the extraction of non-Gaussian information,but the preservation of local neighborhood structure is usually ignored.Aiming at this problem,a fault detection method based on modified KICA is proposed.The criterion of negentropy maximization in KICA considering only the extraction of non-Gaussian information is converted to the criterion of entropy minimization.The criterion of entropy minimization is then combined with the criterion of similar local neighborhood structure in locality preserving projections(LPP),and a new objective function taking both the extraction of non-Gaussian information and the preservation of local neighborhood structure into account is constructed.The particle swarm optimization algorithm(PSO)is utilized to optimize the objective function globally,and the monitoring statistics are built to monitor the process.The simulation results on Tennessee Eastman process illustrate that,in contrast to fault detection methods based on KICA,the proposed method can extract the non-Gaussian information of dataset while keeping the local neighborhood structure,shorten the fault detection latency and improve the fault detection rate effectively.

Multivariable statistical process control(MSPC)is developed in order to extract useful information from process data and utilize them for process monitoring.But when the change in process feed load or product composition happens,the conventional MSPC method does not function well for the process with multiple operation modes.In order to solve these problems,a novel process monitoring method is proposed based on multiple model external analysis and Greedy-KP1M.First,based on the traditional external analysis,multiple models modeling method is introduced to have a better performance. The multiple operation modes of process are eliminated by multiple models external analysis,and the residual error is got for monitoring.Then,to monitor the residual error,the method called Kernel Possibilistic one-Mean clustering(KP1M)is proposed.KP1M has a good ability to monitor nonlinear process.Its performance is similar with support vector data description(SVDD).But the computation complexity of KP1M is far less than SVDD’s.Moreover,to reduce the computation complexity furthermore,Greedy method is adopted to extract the feature samples for KP1M modeling.In the end,the proposed method is applied to monitor the TE(Tennessee Eastman)process and the ethylene cracking furnace to show its efficiency.

In the multi-classifier system,how to determine the weights of individual classifier in order to get more accurate results becomes a question that need to be solved.An optimal weight learning method is presented in this paper.First,the training samples are respectively input into the multi-classifier system based on Dempster-Shafer theory in order to obtain the output vector.Then the error is calculated by means of figuring up the distance between the output vector and class vector of corresponding training sample,and the objective function is defined as mean-square error of all the training samples.The optimal weight vector is obtained by means of minimizing the objective function.Finally,new samples are classified according to the optimal weight vector.The effectiveness of this method is illustrated by the UCI standard data set and electric actuator fault diagnostic experiment.

In the Texaco gasification process,the concentrations of syngas components are the key parameters to evaluate the gasification efficiency.A prediction model of the syngas components is designed for application in a real-world fertilizer plant.The model is a three-layer feedforward neural network,which adopts a chaotic differential evolution with generalized differentials(ChaoDEGD)as the learning algorithm.In ChaoDEGD,the generalized differential information between individuals is introduced into the mutation operation.Furthermore,chaotic mapping is brought on different individuals according to fitness ranking at each evolution phase,which preserves the population diversity so as to escape from the local minima.The experimental results indicate that ChaoDEGD is a competitive optimization algorithm and ChaoDEGD-NN based prediction model performs well in estimating the concentrations of CO,H₂,CO₂ in Texaco syngas.This would provide valuable instructions for the safety and stability of the Texaco gasification.

An online learning soft sensor method based on PLS kernel algorithm is presented for industrial processes with time-varying characteristics.By recursively learning representative samples,this method,utilizing PLS kernel algorithm,could improve soft sensor model’s ability of adaptation,which is more computationally efficient than NIPALS.And deleting one or more redundant samples according to a similarity-based criteria that takes both input and output information into consideration may build more effective training sample set.The foregoing scheme is applied to build an industrial polypropylene unit’s melt index model.The results indicate that the proposed method can efficiently track the change of melt index during grades transition.

In the flotation process,the concentrate grade and the tailing grade are crucial technical indices which can not be measured online continuously.They can hardly be described using accurate mathematical model for strong nonlinearities and uncertainties among technical indices and operating variables,mainly measured off-line by artificial laboratory.The long cycle of artificial laboratory is difficult to meet the control requirement of grade indices,so study of grade indices soft measurement method attracts more attention.By analyzing the relations between the technical indices and such boundary variables,a soft sensor model of technical indices based kernel principal component analysis(KPCA)and extreme learning machine(ELM)was proposed innovatively to estimate the concentrate grade and the tailing grade.To solve the outliers,missing data points of the outliers and deviation from normal values are detected.KPCA is applied to compress the input data,and select the nonlinear principle component.ELM is used to process regression modeling.The proposed model is successfully applied to the flotation process of a hematite ore processing plant in China.Industrial application results show that the soft sensor model has high accuracy and guidance to real production.

The conversion rate of the crude methanol is the primary indicator of methanol production,but also the key factor to influence the economic target.Based on the previous work,two random learning factor particle swarm optimizations with chaos are proposed.In the algorithms,the ergodicity of chaos is introduced respectively at early and late stage of evolution.The simulation of test functions evaluates the effectiveness of RLFPSOC.Finally,the proposed RLFPSOC,which is employed to optimize the parameter of neural network,is integrated with neural network to measure the methanol conversion rate.The results indicate that RLFPSOC-based neural network model can predict the methanol conversion rate well,which further verifies the global convergence of RLFPSOC.

Due to the lack of suitable on-line polymer property measurements,the control of multi-grade product quality in industrial polymerization reactors is difficult.In this article,a predictive model of polymer properties is deduced for industrial polyethylene process by combining the first principle model and the feature modeling scheme.Combining the extended Kalman filtering,a method of design the particle filtering joint estimation is proposed to update the estimation of polymer properties based on the off-line lab analysis data in this article.The application results of the proposed method to an industrial gas-phase polyethylene plant have verified its effectiveness and feasibility.With the proposed method,multi-grade polymer properties of industrial gas-phase polyethylene process can be on-line estimated and make it possible for achieving the advanced on-line multi-grade product quality control.

Some important parameters testing have certain error which brings some difficulty to ensure monitoring the production process and the real-time control of some important quality parameters.Because of the error data may be contained in the independent variables and dependent variables of the sample data,which may affect the accuracy of the model,so in this article we use the weighted least-square algorithm to give the punishment of square-errors of each sample different weights to overcome the abnormal influence of the training samples.Using the simulation data from the Pensim simulation platform to establish the weighted least squares vector machine(WLS-SVM)model in the Penicillin Fermentation by using particle swarm optimization(PSO)on the weighted least squares vector machine parameters optimization algorithm,through the simulation experiments show that the algorithm is used for the effectiveness of penicillin fermentation process modeling.

The extreme learning machine usually exist the problems on high-dimensional data modeling in chemical process.Aiming at solving these problems,the auto-associative neural network is combined,in which the auto-associative neural network is constructed to filter redundant information and extract characteristic components,and these characteristic components are trained by extreme learning machine.Thus,a data feature extraction based auto-associative neural network-extreme learning machine(AANN-ELM)is formed.Meanwhile,the effectiveness of this network is verified by the UCI standard data sets and the purified terephthalic acid(PTA)solvent system.The result indicates that AANN-ELM has the characteristics of fast learning speed,stable network output,and high model precision in handling with high-dimensional data,which will provide a new way to apply the neural network in complex chemical production.

The requirement of infrasound calibration is increasing,but the traditional reciprocity calibration method is easily influenced by gas leakage in low frequency and its calibration result is not good. The new method called laser pistonphone calibration is more suitable for infrasonic microphone calibration.This paper designs a laser pistonphone calibration system based on virtual instrument.The system employs ultra-low frequency technology.The distortion of the piston displacement is very small. The SPL(sound pressure level)in the cavity is stable and large-scale.The software design of the system is flexible and convenient.After sound pressure feedback is introduced,a sound source with stable SPL can be achieved.

In order to effectively evaluate ethylene industry’s energy efficiency level,and search for an opportunity to improve the energy efficiency,an ethylene industry’s energy efficiency evaluation method based on energy performance index(EPI)is proposed in this paper,which is according to data’s characteristic of the ethylene industry,and considers the activity effect,structure effect,intensity effect of energy consumption.Then the EPI and ethylene yield with data envelopment analysis(DEA)are analyzed.Finally,an example based on the energy data of several ethylene plants is presented.The results verify the accuracy and validity of the method and provide a new way to evaluate the efficiency of ethylene plants.

Internal thermally coupled distillation column(ITCDIC)is the most promising distillation energy-saving technology.It can save more than 40% energy compared with traditional distillation process,but so far it has not been widely used.The application of the ITCDIC technology in air separation column can bring a good energy saving effect.Based on the characteristic of cryogenic air separation process as well as three-component distillation,a new optimization model of internal thermally coupled air separation columns(ITCASC)is presented.Analysis of energy consumption is carried out detailed.Compared with the conventional air separation columns(CASC),compressor energy consumption decreases by 20.75%,the product value increases by 17.46% and energy consumption per unit of output decreases by 32.53%.The extraction rate and energy consumption of ITCASC are superior to conventional air separation columns.

A Logistic regression and Autoregressive Moving Average(ARMA)model-based approach to process alarm event prognosis is explicitly introduced in this paper.A sequence of process alarm events which includes states and duration of the alarm events can be extracted from historical data before establishing corresponding Logistic regression and ARMA models,thereby well predicting the process alarm events.A numerical example as well as industrial process data is employed to validate the effectiveness of the proposed methods.

Conventional PCA method is susceptible to intractable difficulties to monitor a process experiencing mode transitions.In this context,this paper presents a modified recursive PCA approach which combines on-line process singular value recognition algorithms and contributes to mode transition process monitoring.Specifically,an enabling algorithm of process singular value recognition is proposed,which accounts for accurately identifying ongoing mode transition of a process before launching recursive PCA algorithms to monitor the transition phases.TE process is employed as a case study,which demonstrates the benefits of the contribution.

For long determination period and inapplicable for the multi-peak absorbance curve of conventional simultaneous determination,an improved simultaneous determination method is proposed to determine individual dye concentrations in the mixed reactive dye liquors.Firstly,using the Gaussian approximation method fitting the model of fixed concentration of a single dye absorbance and wavelength. Secondly,the model of single dye absorbance,wavelength and concentration were derived according to the Lambert-Beer Law.The specific steps of the algorithm is provided to determine individual dye concentrations in the mixed reactive dye Liquors.At last,the individual dye concentrations determined results show that the effectiveness and efficiency of the proposed scheme.

This paper detailed analyses the evaluation architecture of key performance indicators(KPI)for manufacturing execution system(MES)in process industry,and a collection of KPI for manufacturing equipment is established.According to the dynamic analysis part of this evaluation architecture,a method based on steady-state optimization is proposed to determine the benchmark for KPI of manufacturing equipment.Finally,the calculation and benchmark determination for the produced ratio of a distillation column are achieved as an application case.

Aiming at improving the computational performance of bio-inspired algorithm based on membrane computing(BIAMC)in solving complex optimization problems in process manufacturing,a modified bio-inspired algorithm based on membrane computing(MBIAMC)is proposed.In MBIAMC,a new indeterministic abstraction rule is applied which substitutes the abstraction rule of BIAMC,and the algorithmic framework and other rules are inherited from BIAMC.For solving constrained optimization problems,the quadratic penalty function method is introduced in MBIAMC.Four constrained benchmark functions are used to evaluate computational performance of MBIAMC.The results and comparison with other two algorithms handling constraints problems reveal that MBIAMC is efficiency and superiority to BIAMC in accuracy and robustness.As a case study,MBIAMC is used to solve gasoline blending optimization problem,the better recipes and its lower computational cost validate its higher efficiency and more practicability.

An approach to solving large scale unit-commitment(UC)problem based on integer-coded genetic algorithm(GA)with novel repairing mechanism(r-ICGA)is presented.The GA chromosome consists of integer string,which has shorter length than binary string.Using the proposed repairing mechanism,new chromosomes produced in evolution process are repaired to comply with all constraints.As the alternative to penalty function method,the repairing mechanism turns solutions to feasible ones,and avoid coping with economic load dispatch(ELD)sub-problem for infeasible solutions.The algorithm is tested and validated in 6 cases with different scale up to 100 units.The solutions obtained by r-ICGA have lower operating costs,and the algorithm has approximate linear execution time versus unit number.These simulation results indicate that r-ICGA is more appropriate to large scale unit-commitment problem.

The automatically thickness measurement method for pipeline corrosion detection of ultrasound, which has the significant features as large amount of data and much interference, is required high adaptability.However, many algorithms for pipeline corrosion detection of ultrasound have many defects, such as high misdiagnosis rate, low accuracy or more time-consumed.For solving this problem, the two-step method which is based on FFT is proposed in this paper.Firstly, twice FFT algorithm is researched for solving high misdiagnosis rate problem.Secondly, the improved twice FFT method is presented to solve the low precision of traditional twice FFT method.The theoretical analysis and experiments shows that the presented method has high precision and low misdiagnosis rate which is the same as FFT and twice FFT respectively.The automatically thickness measurement for pipeline corrosion detection of ultrasound can be realized by this method.

In industrial practice, multi-tube heat exchangers,which involve both countercurrent and co-current flows are commonly used.Pinch technology is proposed for heat exchanger networks (HEN) with countercurrent flow, but cannot be applied to multi-tube HEN directly.Thus, it is necessary to analyze the relationship among minimum temperature difference, pinch characteristics and total cost quantitatively for multi-tube HEN.Considering the selection of tubes number and total cost of HEN, minimum temperature difference was optimized to minimize total annual cost.Then, it was used to synthesize multi-tube HENs based on pinch technology.Meanwhile, the correction factor for log mean temperature difference (LMTD) was calculated and the results were analyzed.Two case studies demonstrated the effectiveness and application prospect of the presented method.

Lysine fermentation process,which is characterized by nonlinear,large time-delay,multivariable dynamic coupling,is very difficult to realize crucial biochemical parameters on-line measurement.A soft-sensing modeling method was proposed based on PSO-SVM inversion theory.Firstly,the reversibility of the system was analyzed and the inverse extension model was built by introducing the feature information and abandoning the secondary information,and the inverse extension model was built offline with SVM which had good fitting capability and was adjusted online by PSO based on the error information between actual fermentation process input and model output.Finally,the inverse extension model was applied to the fed-batch L-lysine fermentation process and prediction with on-line soft-sensor of the directly immeasurable crucial biochemical parameters was realized.The simulations showed that the crucial biochemical parameters could be predicted.The results also showed that the proposed method was more accurate compared with the alternative modeling methods.