Steady-State Optimization Control Study Of Complicated Industry System

It has become the focus of people to make the complicated industrial engineering operated steadily, safely, high-quality and high-effectively, along with the fact that industry process run increasingly to large-scale and automation. It's also brought forward higher request to procedure of production. Steady-state optimization can find out the equipment parameters or handicraft variable that can maximize or minimize the function index or target function of the system according to the characterstic of the process system. It's an advanced control technique of little devotion and quick effect, and it's also one of the difficulties to apply control theory to actual fields. Steady-state optimization is mainly composed of modeling and optimization. The corresponding improved algorithm has been applied and verified by simulation in this paper, based on analyzing domestic and foreign's situation of the modeling and optimization, every method's principle, realizing method, shortcoming and their limitation.In the modeling part, focus on the characteristics of multivariate, nonlinearity, strong coupling, time-varying, time-lag and uncertainty in modern complicated industry process, it is difficult to build strict system model by using traditional modeling method.Using modern intelligent modeling method-artificial neural network to build system model is approved in this paper.It not only can approach any nonlinearity, but also has large-scaled parallel process, knowledge distributed store, strong self-learning and fault-tolerance well and so on. Radial Rasis Function Neural Network is a typical local neural network and can approaches optimal very well. In this paper, we use nearest neighbor- clustering algorithm as the algorithm of number links in the hidden layer and weight of the Radial Rasis Function Neural Network, and we verified it by nonlinear example.In optimization part, in order to exert the advantages of the group algorithm, such as, succinct, easy to realize, little parameters to adjust and needn't information of gradient, we consider to improve the particle group optimizing algorithm, and to combine it with other optimizing algorithms. Simulated Annealing algorithm has great ability of global optimization, to avoid the particle group optimizing algorithm trapped into local optimization and improve rate of convergence. We integrated the method of simulated annealing algorithm into basic particle group algorithm and improved linear decreasing strategy of inertia weight and integrated it. The simulation of the standard testing function verified the validity.Finally, decomposing cumene hydroperoxide tooken as optimization object, we carried out large amount of data collection and analysis via study of the productive technology. We determined the controllable parameters and the optimization target. We got satisfactory effect, using nearest neighbor-clustering radial rasis function neural network and improved particle group optimizing algorithm to modeling and optimizing.