Xiaofeng Lin

Optimal control for a class of nonlinear systems with state delay based on Adaptive Dynamic Programming with ε-error bound

In this paper, a finite-horizon ε-optimal control for a class of nonlinear systems with state delay is proposed by Adaptive Dynamic Programming (ADP) algorithm. First of all, the performance index function is defined and the Hamilton-Jacobi-Bellman (HJB) equation is obtained for the problem, the convergence of the iterative algorithm is also presented. Then, ADP algorithm for finite-horizon optimal control is introduced with an ε-error bound so as to get the ε-optimal control, and BP neural network is used to implement ADP algorithm. At last, an example is given to demonstrate the effectiveness of the proposed algorithm.

Modeling of vertical mill raw meal grinding process and optimal setting of operating parameters based on wavelet neural network

The stability of vertical mill raw meal grinding process affect the yield and quality of cement clinker. Due to the nonlinear of grinding process, random variation of working conditions, and large lag of the offline index test, it is difficult to establish an accurate mathematics model, thus cannot collect the optimizing operating parameters of vertical mill in time. In this paper, based on the principal component analysis (PCA) for the related variables, a production index prediction model of vertical mill raw meal grinding process was established using wavelet neural network (WNN) and compared with the BP network model, and the validity of the novel model was verified. Then, based on the prediction model and related constraint conditions, the parametric optimization model was established, wherein, the optimal operating setting value under typical working conditions was obtained by using particle swarm optimization algorithm, and an optimal case base was established; through the case inquiry and revision, the optimal set points under the current conditions was obtained. The simulation results showed that, the novel wavelet neural network model and the parameter optimizing setting method could adapt to the changing of process indicators, and could provide optimal parameter value to make the production performance meet expectations, meanwhile achieved the optimizing goal.

An improved method of DHP for optimal control in the clarifying process of sugar cane juice

Clarifying process of sugar cane juice is a dynamic nonlinear system which has the characteristics of strong non-linearity, multi-constraint, large time-delay, multi-input and other characteristics of complex nonlinear systems. In this paper, Elman neural network is applied to the model of the clarifying process of sugar cane juice. An improved method of dual heuristic programming (DHP) affiliated to the approximate dynamic programming (ADP) family is employed to the optional control of neutralized pH value and purified juice pH value in clarifying process of sugar cane juice. The main advantage of this method is to add an “approximate” neural network, which takes use of the states and actions during delay-time of the system to calculate partial derivative of state with respect to action, consequently it overcomes the difficulty of using the system model to calculate the partial derivative of state with respect to action because of the large lagging time. As a result the DHP algorithm is more suitable for real-time process control and on-line application. Finally, the results of the simulation indicate that the improved DHP method has good control effect.

Study on GA-based Training Algorithm for Extreme Learning Machine

In view of the prediction accuracy of Extreme Learning Machines (ELM) is affected by its input weights and hidden layer neurons thresholds, an improved training method for ELM with Genetic Algorithms (GA-ELM) is proposed in this paper. In GA-ELM, after selection, crossover and mutation of Genetic Algorithm (GA), we will get the optimal weights and thresholds, in initial which are randomly obtained by ELM, then to enhance the generalization performance of ELM. The simulation results show that, compared with other algorithms, the GA-ELM has better prediction accuracy.

Adaptive dynamic programming-based optimal tracking control for nonlinear systems using general value iteration

For the optimal tracking control problem of affine nonlinear systems, a general value iteration algorithm based on adaptive dynamic programming is proposed in this paper. By system transformation, the optimal tracking problem is converted into the optimal regulating problem for the tracking error dynamics. Then, general value iteration algorithm is developed to obtain the optimal control with convergence analysis. Considering the advantages of echo state network, we use three echo state networks with levenberg-Marquardt (LM) adjusting algorithm to approximate the system, the cost function and the control law. A simulation example is given to demonstrate the effectiveness of the presented scheme.

Predictive model of production index for sugar clarification process by GDFNN

Clarification process is significant to the cane sugar product, because its production index have direct effect on the output and quality of refined sugar. To maintain the index always in the range of expected value through adjusting operation parameters, an index predictive model is need. In this paper, the principle component analysis(PCA) and other statistical method were employed to deal with massive field data first, then built the generalized dynamic fuzzy neural network(GDFNN) predictive model, and finally the new model was compared with the back propagation(BP) network one on various performances.

Research on water level optimal control of boiler drum based on dual heuristic dynamic programming

Boiler drum system is an important component of a thermal power plant or industrial production, and the water level is a critical parameter of boiler drum control system. Because of non-linear, strong coupling and large disturbance, it is difficult to reach a suitable working state of drum system by using traditional control methods. It is necessary to explore new methods to realize optimal control of drum water level. The back propagation (BP) neural network model of boiler drum system is built in this paper firstly, then the optimal control of the drum system by the dual heuristic dynamic programming (DHP) algorithm is realized, and compared with the heuristic dynamic programming (HDP) algorithm at last. The result shows that the DHP optimization algorithm has good performance in control precision and rejecting process disturbances.

Optimal control for industrial sucrose crystallization with action dependent heuristic dynamic programming

This paper applies a neural-network-based approximate dynamic programming (ADP) method, namely, the action dependent heuristic dynamic programming (ADHDP), to an industrial sucrose crystallization optimal control problem. The industrial sucrose crystallization is a nonlinear and slow time-varying process. It is quite difficult to establish a precise mechanism model of the crystallization because of complex internal mechanism and interacting variables. We developed a neural network model of the crystallization based on the data from the actual sugar boiling process of sugar factory. The ADHDP is a learning- and approximation-based approach which can solve the optimization control problem of nonlinear system. The paper covers the basic principle of this learning scheme and the design of neural network controller based on the approach. The result of simulation shows the controller based on action dependent heuristic dynamic programming approach can optimize industrial sucrose crystallization.

Neuro-controller of cement rotary kiln temperature with adaptive critic designs

The production process of the cement rotary kiln is a typical engineering thermodynamics with large inertia, lagging and nonlinearity. So it is very difficult to control this process accurately using traditional control theory. In order to guarantee the process to be stable, and to produce the high-grade cement clinker, it is important to make the temperature of the sintering zone stable. Artificial neural networks offer a solution to this problem due to their advantages, such as self-organization, self-adaptivity and fault tolerance. This paper introduces a novel nonlinear optimal neuro-controller which is based on adaptive critic design and uses the structure of action-dependant heuristic dynamic programming (ADHDP). The principle of ADHDP is presented. An action network and a critic network are set up in such a way that they basically learn from interactions based on local measurement to optimize the neuro-controller. The ADHDP neuro-controller has a simple frame-work and is independent from the system model. A simulation of the cement rotary kiln is carried out using Matlab/Simulink. The simulation results show that using the ADHDP neuro-controller it is possible to keep the temperature of sintering zone stable in a certain range, and the temperature can meet the requirements of cement clinker production. Simulation results also are presented to show that the neuro-controller with the ACD has the potential to control the cement rotary kiln.

Modelling of the vertical raw cement mill grinding process based on the echo state network

It is known that the variable is strong coupling, nonlinear, multivariable and large time-delay dynamic characteristics in the raw cement vertical mill grinding process. Against the problem which is difficult to establish accurate mathematical model, this paper establishes a production index prediction model of vertical mill raw meal grinding process by using echo state network and Kalman Filter. And its to use a cement plant raw cement vertical mill grinding process parameters to training and testing data of the model. After comparing with the BP neural network model, the experimental results show that the proposed modeling method is effective, and the raw cement vertical mill grinding process operation process stability has increased.