神经网络非线性预测优化控制及仿真研究

Artificial Neural Network-Based Nonlinear Predictive Optimal Control and Simulation

针对暖通空调等一类时变多输入多输出非线性过程控制系统,采用神经网络作为优化反馈控制器求解优化反馈解,并利用预测控制滚动优化能够克服干扰和不确定性影响的优势,采用基于Hamilton-Lagrange 方法和预测滚动优化算法训练多层前向神经网络,同时对系统中某些不能直接测量且受到多种因素影响、计算复杂的时变参数也利用神经网络进行预测,以实现对象特性的实时预测。利用该控制方法对某变风量暖通空调模型进行了仿真,优化指标取舒适性指标和耗能量之和。仿真结果表明,采用此方法,在模型不确定和存在外在干扰的情况下可以得到较好的控制效果。

A multi-layer forward neural network acted as the optimal feedback controller, which was trained with optimization algorithm based on the Hamilton-Lagrange method as well as the predictive optimization to compensate for disturbances and uncertain plant nonlinearities. The controller can approximate the optimal feedback solution of nonlinear-time-varying systems without the complexities of computation and storage problems of the classical optimal methods. Additional neural networks were used to predict some time-varying parameters to achieve the real-time predication of the dynamic behavior. An optimal control system was designed to control a Variable-Air-Volume (VAV) system, which aimed at optimizing thermal comfort and energy consumption. Simulation results illustrate the effectiveness of this technique.