For integral processes with large lag, strong dynamic disturbance, the predictive control algorithm based on first-order state correction was proposed. Considering the integral effect of deviation between predictive and actual state, it exerts state correction on the first-order dimension. Thus the controller's robustness is improved significantly. By simulation experiments, the effectiveness of the state correction algorithm was confirmed. And the algorithm has been adopted for liquid level control of flash tank in a refinery. The practical application results showed that the control effect is outstanding.

The composite system of multiple liquid levels and furnace is a kind of complex industrial process with strong coupling, large delay and nonlinear characteristics. For the control of multiple liquid levels and furnace pass temperature in such systems, on the level of decoupling and balance, a variable period prediction control approach based on the total capacity is proposed. So multi-level control and pass temperature balance control is decoupled on the control structure, and coupling analysis shows the feasibility of the method. Then the overall coordination control can be realized. Simulation results on HYSYS Flow sheeting demonstrate the effectiveness of this approach.

For the combustion process of industrial circulating fluidized bed boiler (CFBB) with nonlinearity, time-variation, strong delay and multivariable coupling, considering how to satisfy the economic target with decreasing power consumption, an two level multivariable predictive control (MPC) method was proposed. The method based on the constrained control of main steam pressure, bed temperature and flue gas oxygen content is constructed to optimize the economic target. The industrial application result demonstrates that the proposed method can decrease the coal consumption of unit steam while assuring the steady of the main process parameters of the boiler combustion and bring high economical profit.

With the improvement of overall system performance requirements, control problems in distribution process has already become the current research focus. The output of the object of microscopic distribution process are controlled by macro way, not can arbitrary shaped curves track with offset-free. This paper propose an offset-free model predictive control algorithm of controlling the distribution process. First, the method describe the distribution process with B-splines. Second, using composite trapezoidal rule for discretizing proposition of rolling optimization. Finally, using the quadratic programming method to calculate the offset-free target to achieve offset-free control. It proved well performed at simulation experiment.

Control performance monitoring and assessment (CPM/A) plays great role in ensuring process working safely and efficiently. However, for a control system, the importance of the performance indicators is with different degree but in conflict, e.g. maintaining the operating point on the borderline is contradict with the aim of safety. In this work, a priority strategy for CPM/A system is presented, which monitoring and accessing the control system rank by rank so as to monitor the key variances relevant with safety in particular. A prioritized CPM/A system in multi-variable Model Predictive Control (MPC) system is developed for FCCU reactor-regenerator system and shows its advantages.

Gaseous emissions associated with the utility systems increasingly arouse people's attention in recent years. As a result of the severe environment situation, the design criteria for a modern utility system should not only include economic factors but also consider environmental requirements. A multi-objective mixed-integer linear programming (MOMILP) model for the operational planning of steam power system is demonstrated, which taking in account the different units and fuel selection and also the environmental concerns. The improved augmented e-constraint method is adopted to solve the multi-objective optimization problem of minimization of economic cost and minimization of environmental effect as its capability of producing the exact Pareto set. One motivation example is introduced to analyze the different environmental charge standards and corresponding optimization schemes of the same refinery. In addition, by the comparison, it's obvious to find the effectiveness of the proposed MOMILP model.

Industrial process plants are often characterized with problems of variable drifts,nonlinearity and time-variant. The time difference (TD) model was proposed by researchers to handle the drifting problems. However, the global model used under TD model cannot describe the data characteristic like the time-variant and high nonlinearity well. Moreover, the prediction accuracy will greatly decrease when change of process state occurs. In this paper, the time difference model and locally weighted partial least squares (LWPLS) are synthesized to enhance the adaptability of soft sensor models. In the TD-LWPLS based soft sensor framework, TD is used to reduce the effect of process drifts. Moreover, as a just-in-time (JITL) method, LWPLS is utilized to tackle nonlinearity and change of process state. A numerical example and an industrial application example have been carried out to test the effectiveness and feasibility of the proposed method. The results demonstrate that the TD technique with the LWPLS model can achieve the best prediction accuracy in both cases compared to two other methods.

To handle the multiscale and dynamic characteristics of blast furnace ironmaking process, a soft sensor model based on empirical mode decomposition (EMD) and Elman neural network is proposed. First, the original silicon content dataset is decomposed into a finite collection of intrinsic mode functions (IMFs) and a residue by EMD, obtaining relatively stationary sub-series from original data set. Second, each IMF and the residue are utilized to establish the corresponding Elman neural network model. To further improve the accuracy of prediction, the result of each sub-series is multiplied by a weight and then summed up to obtain the final silicon content. Here, all the weights are optimized by particle swarm optimization (PSO). The model was applied to the prediction of silicon content of blast furnace in a steel mill, and the result proved the effectiveness of the proposed method.

The receiver is an important part for photothermal conversion in the solar tower power station. Thus, modeling and simulation of the receiver is extremely significant for the safety and steady-operation of the plant. The sectional lumped parameter model of the molten salt receiver, whose prototype is Solar Two located in USA, is established by the space discretization of the heat tubes according to the law of conservation of energy and mass. The validity of the model is verified through the comparisons between the test results of the receiver in Solar Two plant and the simulation results. The heat transfer characteristics of the receiver in the cases where the inputs change and the received energy distribution isn't uniform are analyzed by the transient simulation, which can provide the foundation for the investigation of the aiming points strategy and safe operation of the station.

An iterative learning control (ILC) approach with input trajectory parameterization is proposed for batch processes. In the new approach, the main characteristics of the optimal input profile are obtained to parameterize the whole input trajectory with a few scholar decision variables. The proposed ILC method maintains the simplicity of the algorithm, while improving the optimizing control performance from batch to batch under uncertainties. A batch reactor is simulated to demonstrate the effectiveness of proposed ILC method.

Establishing effective model of scheduling for batch processes is the hot spot of the batch processes scheduling research. Continuous-time models are evolved as a promising tool for optimizing problems related to short-term scheduling of batch plants. Short-term scheduling of batch operations is an important part of scheduling problems. Based on the concept of unit-specific event, an improved mixed integer linear programming model for scheduling of batch plants is proposed. Several new variables and constraints are introduced to make stock transfer between units more flexible. The result shows that the new model can deal with scheduling problems of unlimited intermediate storage effectively and faster with less number of events.

Hydrogen is one of the most important by-products in catalytic reforming process, a hydrogen purity soft sensor will contribute to guiding production. However, the working condition of catalytic reforming process is complex and changeable, a single model soft sensor is hard to ensure the prediction accuracy. Aiming at this problem, this paper present a combined soft sensor model based on modified fast search clustering algorithm and Gaussian processes regression (GPR). The history sample are classified by the novel clustering algorithm and then each sub-model is built through GPR with the classified sub sample. Meanwhile the class identification model has been built by GPR as well. Finally, the combined model soft sensor is established in a switcher form. The combined is applied to a catalytic reformer and the result indicates that the proposed method has a good result and has certain practical value.

The biogeography-based optimization (BBO) is a new swarm intelligence algorithm. To improve the global searching ability, solve the prematurity of BBO, a heuristic mutation operator is designed, which based on the differential information among the population individuals. It makes up the lack of the heuristic information on Gauss, Cauchy mutation operators. And the nonlinear migration model was introduced to the BBO considering to the natural environment. Tests are carried out through four standard test functions on the standard BBO, GMBBO, CMBBO and HMBBO independently, the results shows that HMBBO has a preferable convergence rate and search accuracy. The application of gasoline blending scheduling shows that HMBBO is effective.

Methanation process in coal to synthetic natural gas (CTG) produces a large amount of waste heat. It will cause a huge loss of economic value and energy efficiency with this part of heat emitted into the atmosphere directly. LiBr absorption refrigeration and NH₃ absorption refrigeration cascade refrigeration technology (CRT) is driven by waste heat from methanation process. CRT can produce-40℃ ammonia used in rectisol which can replace a part of compression refrigeration. Thus, it can reduce power consumption significantly and increase energy utilization efficiency. For example, CRT is integrated with methanation applied in a 4 billion m³·a^-1 SNG plant. As a result, 16.2% compression refrigeration load is substituted, equivalent to saving 18000 tons standard coal per year. The dynamic payback period is about 1.7 years.

Alarm floods is a prevalent and difficult problem in alarm management of process industry. Alarm cluster analysis is helpful for alarm root cause analysis and alarm prediction. Aiming at the deficiencies of the current similarity measurement methods for alarm flood sequences, such as limitation of length of alarm sequences, computational complexity, depending on parameters, the discrete Fourier transform (DFT)-based method is employed to analysis on similarity among alarm flood sequences in the frequency domain. The Euclidean distance of the DFT power spectra of alarm flood sequences is proposed as a similarity distance metric for alarm floods, similarity distances of different alarm floods are evaluated. Dendrograms of alarm flood sequences by Unweighted Pair Group Method with Arithmetic mean (UPGMA) is obtained, according to similarity distance, determine the pattern of alarm floods and help operators identify the root cause of the abnormal for a rapid response. An application case of TE simulation process under different disturbances demonstrates validation and accuracy of the proposed method.

Signal-to-Ratio (SNR) of leak signal is the main factor influencing on missing alarm and false alarm in pipeline leak detection. Traditional de-noising methods, such as wavelet transform and EMD decomposition, are hard to promise stable and effective signal enhancement because of the selection of wavelet base, decomposition scale or reconstruction components. Through analysis of the transfer function of dynamic pressure transducer, it is found that the integral part in the output signal of dynamic pressure transducer, which can reflect the low frequency response characteristics of acoustic signal, is missing. Therefore, a gain-adjustable acoustic signal enhancement method based on time-integration is proposed. The test results with field data show that the proposed method has better performance for the signal enhancement with precise location and no complex parameter optimization, and it also provides an effective technical support for small leak detection and decreasing of the missing and false alarm rate.

The traditional functional link artificial neural network (FLANN) is inefficient in the high-dimensional data modeling of the chemical process, where the data has characteristics of multi-dimensional, strongly coupled and noisy. In order to dealing with this problem, the FLANN based on analytical hierarchy process (AHP-FLANN) is proposed. The analytical hierarchy process (AHP) is constructed to filter redundant information and extract characteristic components. And then these characteristic components are trained by the FLANN. Meanwhile, the proposed AHP-FLANN method is applied to analyze the ethylene production data in the chemical industry. Compared with the BP network and the FLANN, the AHP-FLANN has the advantages of fast convergence speed with high modeling accuracy and strong network stability. The experimental result shows that the proposed method can guide the ethylene production conditions and improve the efficiency of energy utilization during ethylene production process. It has the practical value in practice.

For the conventional radial basis function (RBF) neural network, there are many problems like uncertain nodes in the hidden layer, sensitivity to initial centers and slow convergence speed, etc. This paper proposes an RBF neural network model based on the fuzzy C-means method (FCM-RBF), with each cluster center obtained by fuzzy C-means clustering. And weights between the hidden layer and the output layer are trained by the gradient descent method based on error back-propagation (BP). The proposed method overcomes the sensitivity of the data center for traditional RBF model, determines optimally the number of nodes in the hidden layer of RBF neural network, and improves the network training speed and precision. Finally, the proposed method is applied in the production capacity forecast of the ethylene plants. The production statuses of ethylene plants of different technologies or different scales are analyzed and predicted to guide the ethylene production and improve energy efficiency. The empirical results demonstrate the effectiveness and practicability of the proposed algorithm.

Process modeling, optimization and control methods based on data-driven attract attention to both academic community and business circles in terms of its research domains and applications. Even in Big Data era, small sample problems cannot be ignored. In view of the difficulty of obtaining high learning accuracy with small-sample-set using traditional modeling methods, such as artificial neural networks (ANNs), extreme learning machine (ELMs), etc., a novel technology of multi-distribution mega-trend-diffusion (MD-MTD) is proposed to improve the learning accuracy of small-sample-set. The mega-trend-diffusion (MTD) is employed to estimate the acceptable range of the attribution of small sample. The uniform distribution and triangular distribution are added based on MTD to describe data characteristics, which are used to generate virtual samples and fill information gaps among observations in small sample. A benchmarking function is utilized to generate benchmarking samples under the orthogonal test and inhomogeneous sample test in order to verify the reasonability and effectiveness of the MD-MTD, and two industrial real-world datasets include MLCC and PTA are used to further confirm the practicability of the MD-MTD. The results of the validation tests manifest that the proposed MD-MTD can improve the learning accuracy of more than 8% for small sample.

Slow feature analysis (SFA) is an unsupervised liner learning algorithm and lacks the ability to consider class label information and data nonlinearity. In order to solve this problem, a novel nonlinear process fault detection method is proposed based on kernel slow feature discriminant analysis and support vector data description (KSFDA-SVDD). Firstly, process data is mapped from the original space into a high dimension feature space via kernel trick. Then, the discriminant matrix that maximizes the temporal variation of between-class pseudo-time series and minimizes the temporal variation of within-class pseudo-time series simultaneously is calculated. Finally, SVDD is applied to describe the distribution region of normal operation data which is projected to the discriminant matrix and one monitoring index is constructed to indicate the occurrence of the abnormal event. Simulation results on the continuous stirring tank reactor (CSTR) process show that the proposed method is more effective than the traditional KPCA method in terms of detecting faults.

In order to deal with the problem that locality preserving projections (LPP) does not take into account the global structure and dynamic characteristic of process data, a new fault detection method based on dynamic sparse locality preserving projections (DSLPP) is proposed. In the study, the original data matrix is firstly extended to a time-delay augmented matrix. Then, a sparse coefficient matrix which can represent global sparse reconstructive relationship of data is gotten by solving an optimal problem of sparse representation (SR). The sparse coefficient matrix combines with the objective function of LPP to form a new objective function for dimensionality reduction. The new dimensionality reduction algorithm can not only preserve the local neighbor structure of the original data space, but also have better effect in preserving the global sparse reconstructive relationship. At last, DSLPP-based T² and Q statistics are constructed respectively in the feature space and residual space for fault detection.The simulation results of Tennessee Eastman process demonstrate that the proposed method detects faults more quickly and achieves lower fault missing alarm rate than the LPP method.

In switch process of varying duty, it is difficult to determine the switching point and the switching criterion is imperfect. Utilizing the "full-voltage, zero-braking, static-regulation" control strategy, a "predictive Petri-net" is proposed in the paper. A predictor is used in the Petri-net which provides judgment elements for the transition of Petri-net. Based on the idea of "prediction-decision-prediction-decision", the judgment condition for switching process is added in this paper, and the self-optimization process of the Petri-net is realized. The temperature near the switching point is more stable. The test results on the laboratory furnace device show that when the switching point is uncertain, the predictive Petri-net can find the optimal switching point online and improve the response speed and stability of the system.

The performance of model-predictive control (MPC) is affected by many factors. Considering that current model quality index does not take the influence of disturbance into consideration, two indices are applied to realize real-time monitoring of system performance: historical performance index based on MPC objective function and covariance index based on model predictive residuals. The former monitors the whole performance and the latter reflects the influence of the model mismatch and the disturbance. They respond differently to different factors. Combining the re-identified result of the disturbance innovations, we can get preliminary diagnosis why the system performance decreases and narrow the scope of the source of performance degradation. Finally, the experimental validation on Wood-berry column demonstrates the effectiveness of this method.

In practice, model-plant mismatch(MPM) is a key factor that results in performance deterioration in model predictive control(MPC). Traditional correlation analysis between the prediction residual and the manipulated variable of a channel is usually affected by other manipulated variables and disturbance. The result of this process is unreliable, thus unable to locate the MPM accurately. Based on the above problems, partial correlation analysis is used to calculate the correlation between the prediction residual and the manipulated variable of each channel, under the premise of removing the effect of other manipulated variables and disturbance. The MPM problem is converted to a distribution problem of partial correlation coefficients in a certain interval. Whether a channel is mismatched is judged by observing the distribution graph of partial correlation coefficients. The experimental validation on the Shell tower demonstrates the effectiveness of this method.

In order to overcome the nonlinear, time-varying and multivariate of actual industrial processes, a kind of data-driven adaptive subspace predictive control method with forgetting factor was proposed . This method combined model predictive control with online subspace identification, the adaptive updating of variable forgetting factor was designed on the distance value of desired output and actual output at the same time, then the past and future forms of Hankel matrices were designed with the current forgetting factor, thus the online updated of predictive model was realized and the identification sensitivity and adaptability of controller for nonlinear and time-varying characteristics was improved. Finally, a simulating example with the quadruple tank was given to verify the validity of this method.

There will be some abnormal working states in the process of chemical production, so the controller of the system often needs to be reconfigured in order to keep the system running normally. In view of this situation, a control strategy based on piecewise affine (PWA) model is proposed. In the strategy, a system is modeled in PWA form and an explicit model predictive control algorithm is applied to control the system. When abnormal operating points are detected, formal verification is first used to decide whether the controller needs to be reconfigured or not. By this way, it not only saves production time, but also improves the control efficiency. An application case study on CSTR system is given to illustrate the effectiveness of this strategy.

Monitoring of five distance of hazardous chemicals storage warehouse stacking (distance, pile wall distance, zenith distance, spacing and channel spacing) is an important topic in dangerous goods warehouse safety supervision. Studying high precision positioning technology of dangerous goods stacking, is the premise of automately monitoring stacking safety. In this study, the main factors affecting the accuracy of UWB positioning of dangerous chemicals storage are analyzed, and an UWB four reference vectors compensation method applied on dangerous chemicals warehouse stacking positioning is proposed. Firstly, a reference system is established, and the monitoring area is divided into rectangular grids, each vertex of grids is named as a reference point, then the error vectors UWB positioning values at the reference points are obtained, and are took as the reference vector to calibrate the target points; Secondly, UWB tags are attached to dangerous chemical goods stacking, searching the grids around the tags, and calibrating the position value of the tags by the UWB four reference vectors compensation method. Finally, the corrected coordinates are considered as the final position of stacking. The experiment showed that this method can effectively improve the positioning precision of the dangerous chemicals warehouse stacking, it is suitable for monitoring the five distance of hazardous chemicals warehouse stacking.

Currently, domestic hazardous chemicals tend to be stacked in the warehouse without effective methods of monitoring stacking distance, distance from the wall, zenith distance and column spacing. Compared with other positioning technology, UWB technology can get a relatively accurate position of goods, but not accurate enough to meet the existing regulatory requirements. Combining with storage characteristics of hazardous chemicals, the research analyses positioning principles and the main factors that affects positioning accuracy of UWB (ultra-wide band) technology and suggests to monitor positioning of hazardous chemicals in the warehouse by UWB technology. The Thiessen polygon vectors compensation method is proposed to improve the positioning accuracy of hazardous chemicals in the warehouse. It is proved by experiments that this method can effectively improve the positioning accuracy and is suitable for the positioning and monitoring of hazardous chemicals.

Sintering process is a complex physical and chemical reaction process which possesses the characteristics such as large delay, strong nonlinearity and uncertainty, multi-objective and multi-constrained. Burning through point is a key factor during sintering, which affects heat status of sintering process. Bunker-level is related to the sintering safety directly. To optimize these two key parameters of the iron ore sintering process, a coordinated optimal control method is presented based on priority. First, an intelligent control model of the burning through points is established by using neural network predictive and fuzzy control. In addition, based on the analysis on the factors affecting the stability of bunker level, a bunker-level control model is set up according to the experience of the local expert, which can maintain the bunker level and insure the safety of the sintering process. Then, the coordination control model based on priority which takes the soft switch control is designed to combine two different controllers, so that it results in coordinated optimal control of ratio and trolley speed. Finally, an experiment platform system of sinter is designed to realize this coordinated optimal control method.

In the tube falling film evaporator, liquid level directly influences the pressure and the temperature in the evaporator and therefore influences the heating steam consumption. So, it is an important parameter for optimization of evaporator operation. But in the actual production, the liquid level is usually set empirically in a large range, so that the process cannot run in optimal situation. A liquid level optimal-setting method based on exergy analysis is proposed. By deeply analyzing the liquid level's effect on other evaporation parameters and using the actual running data of the sodium aluminate solution evaporation process for alumina production, the correlations between liquid level and other parameters are obtained. An optimization model based on the maximum exergy efficiency is then established based on the material balance of the evaporator and the exergy analysis method. The optimization model is then solved under a certain condition to get the exergy efficiency-liquid level curve. Finally the optimal level under different operating conditions is calculated, which provides guidance for the optimization of the actual production operation.

It is difficult to measure the process parameters online in the bauxite froth flotation process because the slurry deposits quickly. Especially, frequent change of the characteristics of the ore makes the process parameters change from time to time. So that, the static soft sensing models, such as the neural network model, which was obtained by a fixed set of training samples, may not track the dynamic characteristics of the process caused by change of the ore resource. And, thus, model mismatch problem occurs. In this paper, for model mismatch problem under various ore sources, dynamic RBF neural network modeling method based on the hidden layer node dynamic allocation and model parameters dynamic correction strategy is proposed. And the model is used for online measurement of the pH of the slurry in the flotation process, simulation results show that the dynamic model can solve the model mismatch problem well.

For the missing information problem caused by faulted temperature sensor in temperature detecting system on blast furnace (BF) wall, a soft-sensing method based on dynamic neural network is proposed. Firstly, the temperature sensor position model and regional temperature measurement model are built based on the structure of BF. Then, according to heat transfer mechanism, the correlation between temperature sensors in regional temperature measurement model is quantitatively calculated by using maximal information coefficient (MIC) method. Finally, the soft-sensing model for missing temperature information is proposed by using Elman neural network to identify the structure of the model. The effectiveness and feasibility of the proposed method is proved by the simulation results of the real-time producing data of blast furnace which satisfies the field detection accuracy requirement.

For recovery process of methyl acetate hydrolysis, this paper simulated process and designed integrated optimization by using Aspen Plus software. Firstly, analyses of thermodynamic and kinetic properties of the hydrolysis process were given. Secondly, the paper established steady-state process simulation structure of hydrolysis process and verified the accuracy of steady-state results. Then it used sequential optimization method to optimize process structure and operation parameters at the aim of achieving minimum production cost. The results showed that operating condition was the best and production cost was the lowest after optimization.

The distillation column with side reactors (SRC) process allowing a more flexible way of coupling reaction and separation leads to the increase in the freedom of the system and the number of design variables, which consequently increases the difficulty in the process simulation, optimization and dynamic control. For smooth operation and automatic control of the SRC process, with the SRC integration technique in the production of methyl acetate production process as an example, the steady-state optimization, dynamic control and simulation are discussed in the paper. To obtain the optimum integrated structure and the steady-state simulation of SRC for methyl acetate production, the systematic design approach based on the concept of independent reaction amount is applied to the process of SRC for methyl acetate production. Then multivariable control schemes based on product quality control with variable ratio control are designed and SRC of methyl acetate dynamic simulation based on process simulation software Aspen is used to verify the effectiveness of the control scheme. The simulation results show good control precision, robustness and dynamic follow performance of the control scheme.

In the process of chemical engineering, multiple manifold structures has different optimal number of nearest neighborhood under various operating modes. Locally linear embedding (LLE) algorithm based on clustering to select the nearest neighborhood is proposed for nonlinear monitoring. LLE algorithm was performed for dimensionality reduction and extract the available information in high-dimensional data. The mapping matrix from data space to feature space was obtained by local linear regression. The Silhouette index was selected as the clustering validity index to estimate the similarity between the embedded sample information, and further determine the optimal number of neighbors. Process monitoring statistics and its control limits were built based on the mapping matrix. Finally, the feasibility and efficiency of the proposed method were illustrated through the Tennessee Eastman process.

A linear parameter varying (LPV) method can develop the model of the multi-stage, non-linear process through the identification of linear multiple models. In recent years it has been of great concern. This paper investigates the identification of the LPV model with incomplete measurements. First, to indicate whether the output measurement is missing at each sampling instant, a binary variable is defined through which the availability of the output measurement is denoted. The key variable(s) can be taken as the scheduling variable(s) and the main operating point is determined based on the actual production process. Then, local models are constructed around each operating point, and EM algorithm is introduced to estimate their parameters. Both the missing data and the sampling data belonging are treated as the hidden variables. Finally, local models are combined according to an exponential weighting function. A simulated numerical example as well as the continuous stirred tank reactor (CSTR) are employed to demonstrate the effectiveness of the proposed method.

In many industrial processes, in addition to the online measurements with delay-free and low inaccuracy, there exist delayed measurements accurately obtained by laboratory analysis. The augmented state Kalman filter is employed to estimate the state by incorporating both the delayed and the delay-free measurements. To overcome the model-plant mismatch of the online soft-sensor model built by the delay-free measurements, the model deviation is employed to update the soft-sensor model. To follow the model drift, the model deviation is treated as a state, and it will be estimated when the offline measurements arrive. In the end the proposed method is used to estimate the tray compositions in the linearized nonlinear binary distillation column model and obtains good results.

Due to the difficulty in establishing the energy consumption model for the wastewater treatment process, an adaptive regressive kernel function is developed in this paper. Based on the characteristic analysis of the wastewater treatment process, the relationship between the energy consumption and the process variables is obtained. And an energy consumption model is established with the process variables. Meanwhile, the model parameters are adjusted adaptively by using the gradient descent algorithm to improve its accuracy. Finally, this proposed energy consumption model is applied to benchmark simulation platform and a real wastewater treatment process. The results show that this energy consumption model is able to display the operate cost of wastewater treatment process online according to the process variables, with good adaptivity and accuracy.

Due to the nonlinear and highly time-varying issues of wastewater treatment processes, a kind of multi-variable control method based on the recurrent fuzzy neural network (RFNN) is proposed. The proposed RFNN can obtain self-adaptive control accuracy of operating variables. The controller uses the learning rate on the basis of conventional BP learning algorithm on adaptive learning algorithm and the introduction of momentum to train network parameters, can avoid falling into local optimum network, which improved network control of the system accuracy. Finally, based on the benchmark simulation model (BSM1), experiments validate the effectiveness of the method that control the dissolved oxygen concentration in the fifth partition and nitrate nitrogen concentration in the second partition. Compared to PID, forward neural network and conventional recurrent neural network, the experimental results show that this control method can improve the adaptive control precision of the system.

It is difficult to control the dissolved oxygen concentration of the wastewater treatment, a novel approach of control method based on the self-organizing T-S fuzzy neural network (SOTSFNN) is proposed. The essence of the approach according to the actual environment adjust the neuron self-adaptation in time, based on the activity intensity comparisons of the fuzzy rules layer, and construct the appropriate control structure, thus increase the accuracy of control effect. Meanwhile, the parameters of the controller are adjusted on line using error back propagation algorithm. Finally, the controller is applied to Benchmark Simulation Model No.1. The results indicate that the proposed SOTSFNN controller can achieve better control effect for dissolved oxygen concentration with good adaptability.

Along with such dangerous sources as big fire, explosion and toxic matter leak in the chemical plants, the visual tracking technology is a simple yet effective solution. As an effective real-time visual target tracking algorithm, the tracking-learning-detection (TLD) has drawn wide attention around the world. In this paper, we propose a prediction-tracking-learning-detection (PTLD) based visual target tracking algorithm, which is obtained by making several improvements based on the original TLD algorithm. The improvements include employing Kalman filter in the detector of TLD for estimating the location of the target to reduce the scanning region of the detector and improve the speed of the detector; adding Markov model based target moving direction predictor in the detector of TLD to increase the discretion for target with similar appearance. In addition to ascending in the tracking speed by increasing the position and speed prediction, we use the spatiotemporal analysis that also greatly improves the tracking precision. Experimental results show that the proposed PTLD algorithm provides a means for robust real-time visual tracking.

It is difficult to rapidly and accurately detect the component content in the praseodymium/neodymium (Pr/Nd) extraction process. This paper proposes a multi-RBF model and its adaptive correction method of component content. Extracting the HSI image features of Pr/Nd mixed solution, the first moments of H and S components are selected as input variables. Using the subtractive clustering algorithm, the sample data are divided into several categories and the corresponding sub-models are obtained based on RBF neural network. To further realize the high-accuracy prediction of the element component content, a parameters adjustment strategy is designed to automatically adjust the network structure and parameters of sub-models when the change of operating environment or the object characteristics results in the accuracy of the prediction model doesn't meet control requirements. The comparison experiments on actual production data from Pr/Nd extraction process show that the proposed method can meet the high-accuracy and rapid requirements of element component content detection in rare earth extraction process.

Rare earth (RE) is a national major strategic resource, but there are some problems existed in the RE cascade extraction industry, such as poor levels of automation, large control error and low efficiency of manual adjustment. In this paper a non-linear generalized predictive control (GPC) method based on adaptive neural fuzzy inference system (ANFIS) is proposed to counter these problems. First, in consideration of the nonlinearity and dynamic characteristic of the extraction process, the ANFIS algorithm is employed to describe the process. Then, on the premise of high-precision of component content prediction, the GPC method is exploited to adjust the flows accurately and automatically. Finally, simulation experiments are carried out based on the dynamic data of Pr/Nd cascade extraction process. By the contrast with the conventional PID method, it is validated that the proposed approach is effective.

An adaptive control method using the two-dimensional proportional-integral-derivative (2D-PID) iterative learning control (ILC) is proposed for batch processes with time-varying parameters. First, the particle swarm optimization method is utilized to initialize the parameters of 2D-PID. Then, an auto-tuning neuron PID (ANPID) controller is adopted to adaptively tune the process within the batch operation. Moreover, considering the repetitive nature of batch processes, the PID-type ILC is further used to capture the useful information in historical batches. Consequently, the controller performance can be gradually improved batch to batch. The effect of the proposed controller is verified through a simulated batch fermentation process.

Batch process is an typical nolinear production process and can be simulated by a neuro-fuzzy model (NFM). In the previous research, a new model training method called PDF technology was proposed to successfully conquer the weak generalization ability which caused by the MSE rule based model training. But the density function is hard to estimate and the trained model are not stable when the target PDF can not given. To solve these problems, a new window width estimation method is introduced and also a contraction strategy with a PDF predictor is proposed when the target can not be given. Simulation results demonstrate that the proposed methods can get a more accurate density estimation and a more excellent model prediction ability.

Hydrogen utility consumption can be reduced significantly when purification unit is introduced in a hydrogen network. In general, design and targeting of the hydrogen network involving purification are more difficult compared to that of the network involving reuse only. In this paper, a simple method is presented for targeting the single contaminant hydrogen networks involving purification unit with fixed purified concentration model, based on the characteristics of the hydrogen networks. In the targeting procedure, the initial flow rate of the purified stream is first assumed as sufficient, and then the initial purification pinch is identified. Finally, the purification target can be obtained with a few relationships developed from material balances. If the initial purification pinch is estimated correctly, the purification target can be obtained by only one step simple calculation. If the initial purification pinch is not estimated correctly, the correct pinch point can be determined based on the results of the first step calculation. The target can be obtained by an additional step of simple calculation. The results of a few literature examples show that the purification stream consumption target of single-contaminant hydrogen networks involving purification unit can be obtained easily with the proposed method, which provides an effective tool for the targeting of hydrogen networks involving purification unit.

One of the important essences of distributed wastewater treatment system integration is to reduce the total treatment flow rate by lowering unnecessary stream mixing amount as much as possible. Based on this insight, a heuristic method is presented for design of distributed wastewater treatment network (WTN) of single contaminant with maximum inlet concentration constraints. Firstly, an initial network is developed based on the precedence order of treatment units and the allocation of streams to treatment units obtained by using the heuristic rules proposed. The final design can then be obtained with mass and flow rate balances. In the design procedure, the following factors will be used simultaneously: heuristic rules proposed, mass and flow rate balances, determination of the pinch point and maximum inlet concentration constraints. The results of a few literature examples show that the designs obtained with the proposed method are comparable to that obtained with mathematical programming approach. However, the proposed method is simple and of clear engineering insight.

The diversity of internal physical form of refrigeration system and the deep coupling between the system parameters make the system more intricate and the detection and diagnosis more complicated. Seven typical degrading faults of a refrigeration system, including system-level and component-level, were explored. The principal component analysis (PCA) was applied to extract the principal characters and reduce the dimension of faults samples. The probabilistic neural network (PNN) was used for fault diagnosis. The PCA could decompose the original 62 parameters into independent principal components and select a certain amount of principal components according to the cumulative contributions. Import these principal components as input data into PNN for fault diagnosis. Results indicate that the PNN combined with PCA is not sensitive to the spread value within a certain range. The combination also increased the correct rate and saved the elapsed time of diagnosis. Obviously, the use of PCA could effectively optimize the diagnosis performance of PNN.

Based on A/O nitrogen removal process and the kinetic model, Benchmark Simulation Model No.1, two significant factors, nitrate nitrogen and dissolved oxygen, affecting nitrogen removal in wastewater treatment process are considered. Linear active disturbance rejection control (LADRC) is adopted to realize the removal of nitrogen. The control laws designed for nitrate nitrogen and dissolved oxygen do not rely on faithful mathematical models, which make the effluent quality of wastewater be robust enough to uncertainties and fluctuant factors, and nice performance is obtained. Dynamic simulation results show that LADRC has better performance. It can improve the efficiency of nitrogen removal and the effluent quality of wastewater.

Primary steelmaking is one of the most energy intensive industrial processes in the world and many researches have been done to reduce production cost and CO₂ emissions of blast furnace. This paper formulates the above task as a multi-objective optimization problem, the main purpose is to optimize the production cost and CO₂ emissions in the process of blast furnace production and ingredients based on the nondominated sorting-based multi-objective genetic algorithm Ⅱ (NSGA-Ⅱ). It is important to find the Pareto-optimal frontier (PF) and Pareto-optimal solutions (PS) for the multi-objective optimization problem of blast furnace, because different state of operator can be selected in PS to largely reduce the emissions and still keep the steelmaking economically feasible. Furthermore, simulation results verify the effectiveness of the proposed method for the multi-objective optimization model in the process of blast furnace production and ingredients. After optimization, the cost was reduced by about 144 CNY, and CO₂ emissions were reduced by 67 kg.

The generalized predictive control is proposed that based on the control strategy of microbial fuel cell, combined with the characteristics of microbial fuel cells, to investigate the problems of unstable power output in the initial operation stage and the long adjustment time during the operation of an MFC, compared with the MFC system of PID control method joined, joined generalized predictive control of MFC system output is able to avoid the response greatly jitter and fast response, good robustness and dynamic adjustment to ensure that the dynamic output curve fast and accurate tracking system settings. The model identification is carried by the least squares method with forgetting factor to get linear model as a predictive model. Then, generalized predictive control (GPC) can adjust effectively the output response of an MFC under random influent flow at constant external resistance and acetate concentration. The simulation results show that GPC can achieve a good control effect and system robustness adjustment process has also been greatly improved in terms of speed control response. Effective implementation of the optimization of dynamic performance and robust performance of microbial fuel cell system to verify the proposed algorithm is effective and feasible.

Batch processes data is three dimensional data composed of lots of the batches, variables and time. The data contains abundant useful global and local structure information for process monitoring. The key of fault diagnosis is to fully extract the feature information of batch process. The traditional methods unfold three-dimensional data to two-dimensional data. The process inevitably leads to destructing internal structure of the data. The traditional methods usually only consider the global information data or local information data, then the useful process information can not be fully extracted, which leads to poor diagnosis. Aiming to above problems, a tensor global-local neighborhood preserving embedding (TGNPE) algorithm is proposed in this paper. First tensor factorization is used to deal with three-dimensional data directly which effectively save the internal structure of the data. Then the neighborhood preserving embedding algorithm is used to extract the local structure of the data information, at the same time considering the global information of the data. The data information can be fully extracted under keeping internal data structure. The contribution plot method is used to diagnose fault variables after detecting faults. The simulation results of penicillin fermentation process verified the effectiveness of the proposed algorithm.

A chemical process fault monitoring algorithms based on kernel entropy component analysis (KECA) is presented for the complexity and nonlinear of industrial chemical process data. The number of principal components selected by the KECA algorism is much less than the KPCA algorism, which can effectively reduce computational complexity. This is achieved by selections onto eigenvalue and eigenvector based on the value of Renyi entropy. Research shows that KECA reveals angular structure relating to the Renyi entropy of the input space data set. A new statistic—Cauchy-Schwarz divergence measure, namely the cosine value between vectors in kernel space, is proposed, which describes the similarity between different PDFs (probability density functions). It is shown that KECA has great advantages in detection latency and fault detection rate in comparing to KPCA by applying them to TE (Tennessee Eastman) process respectively.

Packed towers have been widely used in industrial productions. It is important to accurately predict the flooding velocity of packed towers. In industrial practice, there are many kinds of packings which can show different characteristics. Only using a single global model is still difficult to achieve satisfied prediction results. To overcome the problem, a new local modeling method is proposed to predict the flooding velocity. First, a recursive algorithm of ridge extreme learning machine with nodes growing is formulated, which can update the online model in an efficient manner. Moreover, using the just-in-time learning manner, the local recursive ridge parameter extreme learning machine (LRRELM)-based online modeling method is proposed. The experimental results show that the LRRELM model can explore more related information among data and thus to obtain better and more reliable prediction performance, compared with the related global models.