Research Into Multi-dimensional Taylor Network Integrated Optimal Control Of Cement Firing System

In recent years,more and more enterprises have paid attention to the control and optimization of the process industry system as the direction of reform for the management level of the cement production process industry.However,due to the strong coupling,large time lag and nonlinear characteristics of cement systems,traditional control algorithms cannot achieve good control effects.Aiming at these problems,the intelligent optimization control idea is applied to the cement production system and the integration of two-layer optimization and control scheme based on the Recurrent Multi-dimensional Taylor Network is designed.In the upper optimization layer,the simple structure of the Recurrent Multi-dimensional Taylor Network is utilized to obtain the optimal reference curve via the direct method.Next,the model predictive control scheme is designed to track the reference curve,such that the cement production system can guarantee the best economic benefits on the basis of system stability.Firstly,a modeling scheme based on Recurrent Multi-dimensional Taylor Network is proposed to identify the cement firing system model.Since the construction of system model can affect the subsequent optimization control process,it is an urgent problem to establish a highprecision system model.In order to solve this problem,a Recurrent Multi-dimensional Taylor Network is designed by virtue of the advantages of static Multi-dimensional Taylor Network,which effectively characterizes the dynamic characteristics of the system and obtains the concise explicit expression of the system.The model parameters are trained by back-propagation through time algorithm and the gradient updating formula is obtained.The optimal parameters of Recurrent Multi-dimensional Taylor Network can be obtained after iterative computation.In order to further improve the training efficiency,the weight of the previous time is taken as the initial value of the current time,that is,hot start.Based on the historical data of the industrial site,the economic model of the cement firing system,the calciner system model and the rotary kiln subsystem model are identified,and the effectiveness of the identification scheme is verified.Secondly,a nonlinear model predictive control scheme is designed for the calciner and rotary kiln subsystems of the firing system,so as to realize the control of the nonlinear,timedelay,multivariable and constrained complex system.Since the economic benefit is based on the system stability,it is particularly important to design a controller with fast response and stability.Firstly,for the discrete nonlinear system with input delay,a positive definite function is constructed,which is composed of the quadratic terms of the predicted output and reference trajectory errors and the control signal,and the corresponding cost function is designed to solve the tracking control problem.On the basis of proving the feasibility of recurrent optimization of the model predictive control scheme,the optimal cost function is used as the Lyapunov function of the system to prove the stability of the closed-loop system.The tracking control and antiinterference experiments of the calciner and rotary kiln systems are carried out respectively to verify the dynamic characteristics and robust stability of the controller.Finally,a two-layer optimization control integration scheme is proposed,in which the upper layer is the economic real-time optimization problem and the lower layer is the real-time control problem.In the upper optimization layer,a dynamic real-time optimization scheme is proposed by introducing the recurrent multi-dimensional Taylor network economic model,which effectively solves the problem that the frequent dynamic changes of the system and the changes of the optimization index are not considered in the traditional steady-state real-time optimization problem.In addition,because it is very difficult to calculate the analytical solution of the nonlinear system optimization problem by using the minimum principle,the direct method is used to solve the economic dynamic real-time optimization problem numerically,and the corresponding optimal control sequence is finally obtained.The economic dynamic real-time optimization problem is solved by experiments,and the economic indicators with different weight ratios are compared,which verifies the effectiveness of the optimization scheme and effectively makes up for the shortcomings of the steady-state real-time optimization scheme.Finally,the simulation experiment of the optimization and control of the sintering system considering the economic benefits verifies the feasibility of the integrated scheme of the two-level optimization control.