Considering the stability of production and the balance of load the process requires that the materiel quantity should change slowly or be kept unchanged. It can be solved by controlling the liquid level of buffer container. However, the classical PID method cannot control the level of the horizontal container and the processes with phase change and disturbances. Based on practical experience a zone control method combined with fuzzy control and classical control was proposed. This method took the high limit, low limit and changing value as the basis of the input fuzzy values. It changed the classical fuzzy methods and used a new strategy to transform the fuzzy outputs and the control outputs which transformed the fuzzy outputs to the expected level values. A tuning output term was added to enhance the control accuracy when considering large deviation between practical outputs and expected outputs. This control method allowed the liquid level to fluctuate within the set range, and guaranteed a slow and stable change of materiel quantity. When the liquid level went beyond the set fluctuation range or the variable liquid level exceeded the set point, it was adjusted to ensure the stable load of the downstream unit. A practical application proved the effectiveness of the zone control method.

In the pipeline transportation of liquid，the flow parameters are essential for management and operation. Based on the physical model of pipeline，flow parameters prediction was studied by means of simulation. And several kinds of boundary conditions were also investigated. The results from both theoretical analysis and simulation showed that the boundary condition would determine the waveform of the model outputs and influence the dynamics of the simulation. It can be concluded that using the pressure measurements at both upstream and downstream would have a relatively shorter setting time for the online simulation which would ensure quicker response when leak detection system based on real-time simulation was adopted. This can be used as an aid for the selection of boundary condition of the leak detection method based on real-time simulation.

Due to the lack of on-line analyzer, resin quality such as melt index and density can not be on-line measured and controlled in the industrial polyethylene process. In this paper, a predictive model of resin quality was deduced for industrial polyethylene process based on the kinetics of ethylene polymerization. According to the approximation error upper bound of the deduced predictive model, based on the multiple fading extended Kalman filter a method of design the suboptimal strong tracking filter was proposed to update the estimation of resin quality based on the off-line lab analytical data. The application of the proposed method to an industrial Unipol licensed linear low-density polyethylene process verified its feasibility and effectiveness. With the proposed method, resin quality of industrial polyethylene process can be on-line estimated and on-line resin quality advanced control can be achieved.

This study provides insights into the optimal operation of one of the most important refinery units, namely, the crude distillation unit (CDU）. A steady state model was developed to simulate an industrial CDU by using a process simulator, HYSYS. The elitist non-dominated sorting genetic algorithm (NSGA-Ⅱ) was used to solve a meaningful multi-objective optimization problem. It was observed that the economic benefit of the CDU could be increased while keeping the product quality within acceptable limits. This procedure used was quite general and could be applied to other CDUs.

A moving horizon pseudo-linear regression（MHPLR）method is proposed which is more robust than the original PLR in noisy environment，and requires no on-line optimization. Utilizing the proposed MHPLR，a time series model was recursively estimated on-line to describe the dynamics of the unmeasured disturbances in DMC control system. And the original DMC algorithm was modified in terms of the disturbance prediction. Applied to the PTA solvent dehydration tower，the modified DMC showed effectiveness and efficiency in improving the control performance and reducing the cost.

For a class of single-input and single-output discrete-time nonlinear systems，a nonlinear adaptive proportional-integral-differential（PID）control method was proposed by using neural networks and multiple models. Such control method was composed of a linear adaptive PID controller，a neural-based nonlinear adaptive PID controller and a switching mechanism. The linear adaptive PID controller was used to guarantee the boundedness of all signals in the closed-loop system，while the neural-based nonlinear adaptive PID controller was employed to improve the performance of the closed-loop system. By introducing a reasonable switching mechanism，the stability of the closed-loop could be guaranteed，while the control performance was improved. Theoretical analysis illustrated that the proposed control method could guarantee the boundedness of all signals in the closed-loop system，while the tracking error would convergent to any given compact set if the structure and parameters of the neural networks were properly chosen. Then the proposed control method was applied to a continuous stirred tank reactor（CSTR）. Simulation result of CSTR demonstrated the effectiveness of the proposed control method. Since the proposed control method was based on the incremental digital PID controller，it had a bright application prospect in industrial process control.

Soft sensing technique is an effective method to estimate variables which are difficult to be measured on-line in industrial processes, and the core problem of soft sensing technique is construction of an appropriate mathematical model. Support vector machine (SVM) algorithm is a machine learning method based on statistical theory. Least squares support vector machine (LSSVM) is a development of the SVM, and has a faster velocity than the standard SVM. Similar to SVM, LSSVM also has the problem of parameter selection. The differential evolution (DE) method was proposed to select hyper-parameter of LSSVM. At last DE-LSSVM was presented for soft sensor modeling on testing the content of 4-carboxybenzaldehyde (4-CBA) in terephthalic acid, and the result was satisfied.

The electro-fused magnesia furnace is one of the main equipment used to produce electro-fused magnesia. Aimed at multiple variables, strong nonlinearity and coupling among variables, as well as strong random disturbance of the fused magnesia production process, an intelligent optimal control strategy based on the integration of case-based reasoning and rule-base reasoning was proposed. First, the technical process of fused magnesia production was introduced. Next, the intelligent optimal control strategy including the optimal set model based on rule-based reasoning and the optimal set compensation based on case-based reasoning was discussed in detail. Finally，the intelligent optimal control system was developed and successfully applied to a real fused magnesia production process. The proposed intelligent optimal control strategy demonstrated reliable, accurate and timely control performance.

Pneumatic control valves stiction is very common in process industry. A big fluctuation of process will be caused by stiction and productivity will also be lowered. Among lots of research results, the method of describing stiction of pneumatic control valves and detecting stiction developed by Kano is simple and effective. However, some defects in the Kano’s method are pointed out by the authors. The model structure of pneumatic control valve of Kano was improved to overcome the existing defects so that the input and output were consistent with the physical process. To deal with multiple parameters in the statistical detection method, guide rules were presented through the analysis of the factors that affected the detection results. Finally, the stiction phenomenon was compiled to a module by using the S-function technology in Matlab Simulink. Simulation was made under the Matlab Simulink framework and the statistical detection method under disturbance was discussed.

Based on some features in lead-zinc sintering process，such as time-delay，nonlinearity and uncertainty，a fuzzy predictive control method for burning through point was proposed. First，to deal with the uncertainty of vertical burning velocity，a fuzzy T-S（Takagi-Sugeno）model for burning through point was established based on the piecewise linearization characteristics between burning through point and sinter strand velocity. Then，the model predictive controller was designed based on the global model obtained by the dynamic linearization method. The simulation results illustrated the effectiveness of the method proposed. The precision values of the fuzzy T-S model were respectively 89.0% and 95.0% while the allowed relative errors were within 2% and 3%. The fuzzy T-S model was much more precise than the neural network model for burning through point based on process parameters. The rise time and regulating time of the fuzzy predictive control method were respectively 9 min and 18 min which were both much shorter than those by using the fuzzy control method for burning through point in lead-zinc sintering process.

Particle swarm optimization (PSO) is an effective evolutionary method which is used to search the function extreme. It is simple and has fast convergence, but the convergence accuracy of this algorithm is not high, and it can easily fall into the local extreme points. The effect of inertia weight in PSO was analyzed. Motivated by the idea of power function, a new non-linear strategy for decreasing inertia weight (DIW) was proposed based on the existing linear DIW . Then a novel hierarchical multisensor data fusion algorithm adopting this strategy was presented and the weight factor of the data fusion was estimated. The distinctive feature of this algorithm was its capability of fusing data in a near optimal manner when no information about the reliability of the information sources, the degree of redundancy/complementarities of the information sources and the hierarchy structure is available. It obtained the effective information from the fusion data, removed the noise disturbance successfully and got the favorable results.

Hybrid particle swarm optimization was presented to improve the optimizing efficiency of the particle swarm by changing the optimizing strategy of the global best particle. Aimed at the problem of optimization with a limit on computing time, such as the state prediction of a typical equipment in process industry, hybrid particle swarm optimization took the global best position found by the particle swarm as a special particle, which performed the gradient descending optimization. By adding the individual gradient descending optimization of the global best particle to the optimization iterations, the global search and local search were combined in hybrid particle swarm optimization. The hybridism of this new particle swarm optimization improved the optimizing efficiency of the particle swarm, and reduced the time of optimization computing. In the test of a real application, hybrid particle swarm optimization was applied to the state prediction of the continuous stirred tank reactor (CSTR), which is a typical equipment of the process industry. In the test training of neural network that was used in the prediction of the concentration of the CSTR product, hybrid particle swarm optimization took less optimizing iterations than the traditional particle swarm optimization, and took less optimization computing time, which showed that hybrid particle swarm optimization could reduce the computing time of optimization as the original intent of this research.

A modified differential evolution algorithm with dynamic scaling factor and adaptive mutation operator was proposed in this paper. Dynamic scaling factor which was determined according to population information during the evolutionary process, was introduced for handling the constraint optimization problem; adaptive mutation operator which mutate partial individuals based on population diversity was adapted to overcome premature evolution and enhance the probability of finding global optimum. Polypropylene grade transition optimization model was established and the modified differential evolution algorithm was applied to solve the problem. Experiment results demonstrated that both search ability and efficient were superior to those of the original differential evolution algorithm.

The key technique for neural network based pipeline leak detection is features extraction. In this paper，piezoelectric dynamic pressure based leak signal transient was chosen as research object，the differences between leak signals of upstream and downstream under different operational conditions were analyzed，and leak signal enhancement with wavelet decomposition was introduced. Positive and negative interval divisions of dynamic pressure signal was proposed. The differences of weighted signal sums，signal mean values，signal peaks of every two successive intervals were selected as the features of leak signal，and their calculations and relative scaling transformations were presented. The feasibility criteria for lengthwise and breadthwise evaluation of leak signal features were presented，and evaluation was done with features extracted from field data and the feasibility was verified. At last，a neural network based leak diagnose model with both features from upstream and downstream and its training，testing results were given. Long term and real time monitoring of pipeline leak showed that the features extraction methodology proposed here had reasonable operational conditions adaptability，and provided an encouraging technical support for robust diagnosis of pipeline leakage.

As an important procedure of the manufacturing process in refinery, oil blending could be abstracted as a complex non-linear optimization problem (NLP) with many constraints. It is difficult to obtain satisfying optimum solution by traditional methods. According to the oil blending and scheduling problem, a two particle swarms optimization algorithm with a mutation operator was presented. The proposed algorithm constructed two swarms of particles with different velocity restrictions, introduced a communication mechanism and a special mutation operator, and sequentially elevated the swarms to higher ability and velocity of global convergence. The new method was illustrated with a recipe optimization problem of gasoline blending, and the feasibility and effectiveness of the proposed algorithm was experimentally confirmed by the simulation results.

PET solid-state polycondensation process involves spatial mass transfer of both solid reaction particles and reactor bed, which is a multi-dimensional and multi-phase objective. Aiming at reaction process control, a simplified one-dimensional dynamic model of PET solid-state polycondensation reactor based on existing reaction kinetics models was developed. The model emphasized not only reversible reactions, but also the diffusion of small molecular products inside and outside the particles. It approximated the concentration gradient of small molecular products due to diffusion with an effective coefficient, which borrowed the idea of pseudo-homogeneous fixed bed reactor model. Finally with the experimental data reported in the literature the paper numerically solved the distributed parameter model, which were partial differential equations, and analyzed the dynamic reaction process to get concentration profiles and product quality indices values, which were compared with the data in the literature to show the effectiveness of the developed model.

A new nonlinear model predictive control algorithm using velocity-based linearization was presented. The model used in the derivation of the control algorithm was obtained by using a velocity-based linearization method，and had a linear structure with variable parameters. The model parameters were then determined by the operation conditions of the system. It was shown that the linearized model approximated well the original nonlinear one. The model predictive control algorithm presented in this paper was based on the Levenberg-Marquardt algorithm，which is efficient in computation and provides a general framework for model predictive control design. A simulation study on a nonlinear continuous stirred tank reactor（CSTR）showed that the proposed control algorithm was effective and applicable to many nonlinear industrial systems.

A new decoupling internal model control（IMC）method was proposed for the multi-variable non-square systems with multiple time delays by introducing the conception of generalized inverse. The traditional IMC based on the inverses of nonsingular matrices was confined to square systems，so the generalized inverse was imported to design the internal model controllers by calculation in frequency domain. Time delays were approximated by the Taylor expansion diagrams. To guarantee the stability and regularity of system，special filters were designed to counteract unstable poles brought forth by the Taylor approximation. The results showed that when the model did not mis-match badly，the output curves had less than 20% overshoot and almost zero deviation from steady state. The output also had good performance about dynamic decoupling，and multiple time delays control. But the control system was sensitive to the change of time delays. Based on IMC system，the response would be better if the model matched more satisfactorily.

Ball mill coal pulverizing system of pelletizing plant is a complex nonlinear multivariable process with strongly coupling and time-delay，and its operation often varies significantly. The automatic control of such a system is a research focus in the process control area. A multivariable adaptive PID artificial neural network（ANN）controller was introduced，which was based on the characteristics of particle swarm optimization（PSO）algorithm searching the parameter space concurrently and efficiently，and the self-regulation and adaptability of PID artificial neuron networks. Decoupling control technology based on the PID-ANN was used to eliminate the coupling between loops. Particle swarm optimization algorithm was also adopted to optimize the weights of neural networks. Simulation results showed that controller method proposed had better control quality，adaptive decoupling ability and self-learning function. The new control strategy could overcome nonlinear and strong coupling features of the system in a wide range and is expected to have great potential for engineering application.

To deal with the problem of the strong non-linearity and large time delay in the coking process, the neural network prediction model for coke quality and coke yield，energy consumption of coke oven and the optimal control model with coke quality as constraint, coke yield and energy consumption as objective function were established based on principal components analysis and grey relational analysis of the process parameters. An integrated optimal control method, which combined fuzzy C-means clustering to realize coarse optimization and combined differential evolution to realize fine optimization, was proposed to optimize the process parameters and provide guidance for operation optimization. The simulation results showed that the method was efficient in restricting the fluctuation of operating conditions to achieve the production target of high coke yield and good coke quality at low energy consumption. It provided a new idea for the modeling and optimization control of complex industrial processes.

A kinetics model involving 18 lumped components and 31 reactions was selected and a mechanism model was built based on HYSYS simulation software. To establish the process optimization model, the objective was to maximize the yield of aromatics and the decision variables were inlet temperatures of four reactors. The optimization problem was solved by differential evolution and the constraints were handled by feasibility-based rule. The simulation results demonstrated the increase of aromatics yield and confirmed the effectiveness of differential evolution.

The carbonation column used in manufacture of soda ash is not only a multiphase reactor but also a multistage crystallizer. The complex processes include chemical absorption，reactive precipitation，multiphase flow，heat and mass transfer，and other physico-chemical processes taking place in the column. Industrial application of a commercial software of model predictive control（MPC）for five groups of twenty-five carbonation columns in a soda plant was introduced in this paper. An MPC system，which was composed of nearly one hundred controlled variables，manipulated variables and disturbance variables，was developed to deal with the constrained multivariable control problem on-line of the carbonation columns. Industrial application results showed that the MPC software could ensure the best operation for a long time and realize the ultimate highest operation potential of the carbonation systems by reducing the consumption of materials，improving product quality，and minimizing operating cost.

Due to its synthetic and complex characteristics，the combustion process in the hematite ore-filled shaft furnace is noted for complex mechanism and frequent change of operating conditions，which results in frequent occurrence of faults and unsteady production. In order to reduce the faults ratio during the combustion process，an intelligent faults prediction approach was developed based on the combination of case-based reasoning（CBR）with soft-sensing. The soft-sensing model could estimate the key technical parameters which were difficult to measure online，and provide some information about the faults. Then，the fault prediction model based on case retrieval and reuse was adopted to make a thorough analysis on the combustion process. The model could provide the occurring probability of some typical faults，followed by corresponding operation instructions. The proposed fault prediction system was applied to the practical combustion process in a shaft furnace，and evidently eliminated the fault ratio.

Kernel principal component analysis (KPCA) is suitable for nonlinear process monitoring, but it suffers many limitations, such as great calculation load, and poor real time performance. A monitoring method based on sparse KPCA (SKPCA) was proposed to decrease calculation load and improve real time monitoring. SKPCA was firstly used to weight the normal modeling data, and the minority of data with high weight could basically represent the information of the whole data, so modeling data could be largely reduced. Following this, KPCA model and the monitoring indices were built based on the sparse modeling data. In the end, taking a chemical separation process for example, KPCA and SKPCA were compared in terms of monitoring result and real time performance, and the superiority of the proposed SKPCA method was demonstrated.

Process transition during start-up，shut-down or product changeover is frequently encountered in chemical industry. Processes are more prone to various malfunctions and unknown disturbances during transitions. Fault detection and diagnosis during process transitions is critical to ensure process safety and production capacity. A novel modeling method，two-dimensional dynamic principal component analysis（2DDPCA），was developed for monitoring batch process transition in author’s previous work. To follow up，a fault diagnosis method was proposed in this paper. Process characteristics changed by faults were decomposed into “within-batch” and “batch-to-batch” information. Based on this extracted information，contribution plot，associated with the change of fault variables correlation in the optimal region of support，can then be used to isolate and diagnose the abnormal process variables. Simulation results showed the feasibility and validity of the proposed method.

For several complex industry processes，the original fault sources are difficult to identify by using kernel principal component analysis（kernel PCA）methods. And during the modeling and online dynamic monitoring process，the calculation of the kernel matrix K is a bottleneck problem for a large data set. An integrated fault diagnosis method based on feature sample extracting and kernel PCA was developed. Firstly，a feature extraction method was adopted to pre-process the modeling data set for solving the calculation problem of the kernel matrix K. Secondly，Hotelling statistics，T² and SPE of kernel PCA were adopted to detect system fault. Once fault was detected，the gradient algorithm of kernel function was used to define two new statistics，C_T² and C_SPE，which represented the contribution of each variable to Hotelling T² and SPE respectively. According to the degree of contribution，the fault variables might be identified from these correlative variables. To demonstrate the performance，the proposed method was applied to the Tennessee Eastman（TE）process. The simulation results showed that the proposed method could effectively identify various types of fault sources.

The monitoring and fault diagnosis of large-scale power plant processes that exhibit non-stationary and/or time-varying behavior were discussed. The work considered statistically-based monitoring technique, which was related to the multivariate statistical process control (MSPC) framework. Particular focus is on principal component analysis (PCA)，as this technique allows distinguishing between cause and effect variables consequently. To demonstrate the difficulties of monitoring processes with non-stationary and time-varying behavior, the use of conventional PCA was compared with its recursive and fast moving-window counterparts. A recently proposed recursive moving window technique was employed because of its ability in adapting to process changes and its computational efficiency. The advance in fault detection was demonstrated by comparing fast moving-window PCA (MWPCA) with the conventional PCA. In addition, this paper proposes to plot the scaled variables in conjunction with fast MWPCA for fault diagnosis, which proves to be effective in power utility process application.

By combining the sliding mode control for mismatched uncertainties with the linear matrix inequality（LMI）approach，a novel robust sliding mode observer for a class of uncertain dynamic system with mismatched uncertainties was designed. The necessary and sufficient condition of stabilizing the sliding mode observer and the LMI existence theorem were presented and strict verification was done to guarantee robustness to uncertainties of systems and disturbances. The Lyapunov function was used as the judgement condition for stabilizing observers，and the convergence rate between observer and system was changed by regulating the sliding-mode strategy so as to attain the desired performance of state estimation. Simultaneously，without the canonical transformation of the dynamic systems，the linear feedback matrix and nonlinear feedback matrix of the robust sliding mode observer were solved by LMI that has advantages in computation. The problems of actuator fault detection and reconstruction for a class of uncertain dynamic system with mismatched uncertainties were discussed. By applying the equivalence output error concept and LMI approach，the optimizing strategy of reconstructing actuator fault can be designed，and a new actuator fault detection and reconstruction design method based on the designed sliding mode observer was proposed to obtain the fault information on-line. The design procedure was described and nonlinear simulation results were presented to demonstrate the approach.

Considering the fact that feature extraction of infra-red image for burden surface temperature profile in blast furnace is a hard problem，an optimized strategy of threshold selection based on empirical knowledge and information entropy was proposed. Firstly，an image was segmented respectively by the two-peaked histogram method，the Otsu method and the consistency criteria method. Then a criterion to value the image quality based on information entropy was made by rules，which was related to the image sizes of objects. The appropriate threshold value，by which image segmentation was implemented to judge the burden temperature distribution most efficiently，was obtained. Actual application showed that the results of image segmentation were consistent with the expected image features，which satisfied the demands of industrial application and could help operators to judge the blast furnace conditions.

One of the critical problems in the operation of chemical process is occurrence of abnormal events. Therefore a process monitoring system that can detect and diagnose abnormal events is important for effective and stable operation of chemical process. A new type of timed fuzzy Petri net（tFPN）approach to prognostication and diagnosis of abnormal events was proposed in this paper. In tFPN，a timing factor was associated with the transition and the degree of truth of rule，with which the reliabilities and appearing times of abnormal events could be inferred automatically. The procedures of abnormal events monitoring based on tFPN models were presented in detail. Finally，the proposed techniques and solutions were demonstrated through a polypropylene reactor case study，which showed promising results by prognostication of abnormal events and diagnosis of root causes.

Ultrasonic detection is one of the important ways to inspect the wall-loss defects and cracks in-line for oil pipeline. Because of the complicated condition in pipeline the recognition of ultrasonic detection echoes is difficult. This is a high-dimensional classification problem. An effective method based on support vector machine（SVM），which is suitable for small-sample，non-linear and high-dimensional recognition for classification and recognition of pipeline corrosion defects was presented. The ultrasonic A-scan time-series were considered characteristic vectors. By unifying the characteristics extraction and pattern recognition of pipeline corrosion defects the classified decision function of ultrasonic detection echo signals was established. Experiments showed that the classified recognition of break interfaces of pipelines was accurate and clear and the method was suitable for in-line detection of pipeline corrosion defects.

An intelligent modeling approach was developed for batch reactor to solve modeling difficulties of non-linear characteristics，dynamic process and with partially unmeasurable states. Base on separability of reaction，the topology of model was established via structure approaching hybrid neural networks（SAHNN）. Compared with normal neural networks，SAHNN has optimized structure and more nodes which represent actual states. This modeling approach utilized virtual supervisor-artificial immune algorithm（VS-AIA）to solve problems of training neural network with unmeasurable states. It firstly initialized information of unmeasurable states during reaction with only a little mechanism information（ascending or descending curve）to make them act as virtual supervisors. Then it improved the precision of modeling by optimizing virtual supervisor and training network weights at the same time. During training，immune population was used to explore new better solutions and avoid wrong direction. Detailed process of modeling and training neural networks were described in dynamical modeling of batch condensation reaction of producing promotor for vulcanizing rubber. The simulation result proved that the approach was effective.

Internal thermally coupled distillation column（ITCDIC）is a frontier of high-efficient energy-saving distillation technology. A novel APC strategy based on generic model control（GMC）was therefore proposed to overcome the increasing nonlinearity with increasing product purity,which was difficult to deal with by the traditional linear control strategy. The basic algorithm of GMC and the application to ITCDIC were presented. Also the parameter preference method was given. The benzene-toluene system was adopted as the illustration example. The detailed comparative researches were carried out between GMC and traditional PID controller. The control performance of both servo control under step disturbance in feed composition and feed flow rate and fix-value control under step changes in set-point proves the efficiency of this proposed APC strategy for high purity ITCDIC process.

Fault prediction for a class of unknown-model multivariate continuous processes with a hidden fault was studied，and a solution was given based on statistical process monitoring（SPM）approach. A principle component analysis（PCA）model using sample data under normal state was built,then the characteristic value for fault prediction was constructed,and time series analysis and prediction were applied to the characteristic value to predict the remaining useful life（RUL）of the system. Aiming at the linear time invariant system，a characteristic value was proposed and the prediction error of RUL was analyzed under some assumptions for system structure and hidden fault. A case study on a CSTR showed the efficiency of the proposed approach.

By considering the characteristics of information generation of chemical industry product configuration model，a new method of constructing chemical equipment configuration model based on axiomatic design theory was presented. In this method，functional domain mapping into physical domain was introduced，functional requirements with constraints were defined as function constraint integrated component（FCIC），design parameters with feature were defined as structure feature integrated component（SFIC），and the configuration model based on SFIC was constructed by establishing database and the characteristic relationship of SFIC features. Axiomatic design made configuration modeling more scientific，and helped the designer decide more quickly and accurately. Finally，a case study of configuration design for air separation equipment was presented.

In the rotary kiln alumina production process, because of the complexity and variability of rotary kiln burning zone conditions, some important quality index related process parameters can not be detected continuously on-line. Detecting the different burning zone conditions on-line is a key factor for the whole process automation of alumina industry. The current method depends on flame observation by naked eye. In order to realize automated recognition of burning zone conditions, a method which learned experience and knowledge from naked eye observation was proposed to recognize burning zone conditions by utilizing the image processing technique and pattern classification method. At first, features were extracted from flame images of rotary kiln burning zone and were combined with some important process parameters to constitute a hybrid feature vector. Then a model with a binary tree based SVM (support vector machine) was constructed. At last, a flame image recognition system was developed. The system was successfully applied to a domestic alumina plant, and good economic benefit was realized.

Based on the scope of change of parameters，multiple sliding mode variable structure controllers were designed for a kind of nonlinear system with uncertain parameters. A switching controller was formed with a given index switching function. Under the guarantee of Lyapunov stability，the controller of system would be switched among multiple sliding mode variable structure controllers according to the switching condition. The mechanism of switching could improve the transient response greatly. A robotic arm was studied as a nonlinear system. Multiple sliding mode variable structure controllers were set up according to the dynamic equation of the robotic arm. An index switching function based on output error was given for the design of switching controller of the robotic arm. To test the effectiveness of the switching controller，four simulation examples according to different scopes of parameter change were investigated. From the simulation，it was concluded that the switching controller could enhance the control performance greatly.

A complex exothermic batch reactor model was developed by using structure approaching hybrid neural networks（SAHNN）. The optimal reactor temperature profiles were obtained via the PSO-SQP algorithm solving maximum product concentration problem based on recurrent neural network（RNN）.Considering model-plant mismatches and unmeasured disturbances，a novel extended integral square error index（EISE）was proposed，which introduced mismatches of model-plant into the optimal control profile. The approach used a feedback channel for the control and therefore dramatically enhanced the robustness and anti-disturbance performance. The stability analysis of the one-step-ahead control strategy through SAHNN-based model was described based on Lyapunov theory in detail. The result fully demonstrated the effectiveness of the proposed optimal control profile.

Aiming at an industrial DMF recovery plant, a kind of region expert coordination control system (RECCS) based on expert inference mechanism was proposed. With an S7-300 PLC based distributed control architecture, the expert control system can identify the operation states of regions from coupling data of the plant. The regional PID controllers were coordinated with the expert coordinating rules to realize plant-wide automation of the DMF recovery process. Simulation results on distillation models illustrated the effectiveness of the coordination control system. Easily integrated with DCS，the proposed expert region coordination control technique could be widely used in industrial applications.

The benchmark is an essential issue of the performance assessment problem of the control systems. In this paper，the authors extend the theories of fundamental design limitations to model predictive control（MPC）systems to establish a performance benchmark. Considering constraints on the frequency domain explicitly，an output feedback robust model predictive controller was developed for the multi-input/multi-output systems with disturbances. The feedback controller developed could be explicitly set up，without online optimization. With this controller the closed-loop systems reached the limit of control performance，so that it could be served as the benchmark controller to assess the other MPC’s using the process routine data. The performance index was established. A performance assessment procedure was proposed for evaluating the other MPC systems. Numerical example well demonstrated the proposed performance assessment procedure.

At present, the modeling method used in the batch process by signed directed graph (SDG) is oriented to the process in normal running state, which cannot describe the operating logic in abnormal running state, and it is inadequate to cover all faults and hazards. When the SDG model used for computer-aided hazard and operability study (HAZOP) is built for the complex chemical process including the batch process, the model of the batch process cannot always be associated to that of the other continuous systems. In order to meet the demand of computer-aided HAZOP, the paper presents a dynamic SDG modeling method based on the present SDG-HAZOP modeling method to deal with the batch process. The idea and procedures of dynamic modeling were elaborated. A real case about the fly ash removal in a complex process of coal-to-oil was modeled and analyzed by using the proposed method, which can resolve two problems at the same time: the sequential model about the SDG modeling of the batch process and the maximal covering of the faults and hazards.

Water-bloom prediction is important to environmental protection, and the mechanism model of water-bloom is the basis of water-bloom prediction. A model for algae growth was set up based on the analysis and research about the mechanism of water-bloom engendering. The model described the relationship of chlorophyll-a and temperature, solar illumination, overall phosphor, and was applicable to the lakes whose nutrition factor is restricted by phosphor in nutrition circle, without considering the influence of hydrodynamics. And an optimal method of estimating optimal parameters in a nonlinear minimum problem with constraints was proposed. The parameters in the model, such as maximum growth rate of alga, half - saturation coefficient of light, half - saturation coefficient of phosphor, maximum death rate of alga, net loss velocity of alga, and so on, were evaluated by the multi-parameter simultaneous estimation method and validated by experimental data. Through simulation and experiments, the result showed that the model could better describe the change of water-bloom from engendering to burst-out and provide reference for water-bloom prediction and control.