基于遗传算法学习的复合神经网络自适应温度控制系统

Composite neural networks adaptive control system of temperaturebased on GA learning

针对一类温度控制系统中存在的非线性和参数不确定等问题,提出一种复合神经网络自适应控制结构.在控制系统中构造了神经网络正模型来再现被控对象的动态特性,用神经网络控制器实现优化控制律的非线性映射.文中选用了被控对象80组历史数据作为样本集,并利用遗传算法的全局搜索能力及高效率来训练多层前向神经网络的权系数.最后用升降温工艺曲线作为输入对温度控制系统进行仿真.仿真结果表明,应用遗传算法能够提高神经网络的学习效率.保证神经网络全局快速收敛,从而克服了传统的误差反传学习算法的一些缺点.证明了采用这种神经网络自适应控制结构.使神经网络控制器的输出可以适应对象参数和环境的变化.使温度控制系统具有很好的学习和自适应控制能力,取得了良好的控制效果.

A kind of composite neural network adaptive control structure is proposed in this paper with the purpose of tackling problems such as nonlinearity and uncertainty in a temperature control system. The neural network positive model is constructed to represent the dynamic characteristics of the controlled object, and the neural network controller is used to realize the nonlinear mapping of optimal control rules. By taking eighty troops of history data to serve as samples the genetic algorithm with its searching ability and high efficiency is successfully used to train the weights of the multi-layer forward neural network. Then, the raising and falling temperature technics curve is acted as the input to simulate the temperature control system. The simulation results indicate the capability of genetic algorithm in fast learning of neural networks, guaranteeing a rapid global convergence and overcoming some shortcomings of the traditional error back propagation algorithms. It is shown that if this neural network adaptive control structure is applied to the temperature control system, the output of the neural network controller can adapt to changes of the object parameters and environment, and hence the temperature control system will have a nice learning and self-adaptive capability and lead to a good control result.