基于神经网络的随机动态系统的优化新方法

An Approach for Modelling and Optimization of Stochastic Dynamic Systems Based on Neural Networks

基于神经网络方法研究了一类随机动态系统的建模和最优化问题。首先将采样样本和相关的时间区域段上的平均值样本,用来训练一个多层反传网络,以得到该随机过程的神经网络模型;然后,基于一给定的随机目标函数,该网络模型用来寻优以获得系统操作变量的最优均值设定值。这一网络建模和优化方法用于解决一尿素合成塔的优化问题。仿真结果表明了这一方法的有效性。

The sample patterns and the corresponding mean patterns are used to train a multilayer back propagation network to derive a model for stochastic dynamic system. Then, the model is used to synthesize the maen of the inputs which minimize a given stochastic objective function. Finally, the above results are applied to solve the optimization problem of a practical urea reactor. Simulation results show that this approach is effective.