遗传算法在BP网络PID控制中的应用仿真

摘要

为了有效地克服BP网络存在的局限性,即控制效果严重依赖于权值矩阵初始值,对空调系统模型进行了多次仿真,结果权值矩阵初始值的优化点分布与取值区间具有相关性,即在特定优化区间内,满意的权值矩阵初始值数量较多.引用遗传算法全局搜索能力和过程流程图,并引入到BP网络PID控制之中,既利用了遗传算法的全局搜索能力进行权值矩阵区间优化,同时利用BP网络的局部搜索能力与实时处理能力,有效地解决了常规BP网络控制的局限性.仿真结果表明.混合算法优于常规BP网络整定PID控制方法,并可推广到其它BP网络应用领域.%For overcoming effectively the limitation of BP neural network, namely the control performance is dependent mostly on the initialization of weight matrix, simulation of air - conditioning system is carried out. Results of simulation show that the distribution of satisfactory weight matrix is related to the interval range, that is to say, within the optimized interval range the number of satisfactory initialization of weight matrix is multiple. And the global searching ability and flowchart of genetic algorithm are introduced and then a genetic algorithm is introduced into PID control based BP neural network, so the method possesses both the global searching ability of genetic algorithm and the good local searching ability of neural network, thus overcoming the limitation of BP network. Results of simulation show that the hybrid algorithm is superior to conventional PID control based BP neural network, and can be popularized in other fields of BP neural network.