基于改进的PSO算法的球磨机PID神经网络控制系统

PID Neural Network Control System of Ball Mill Based on Modified PSO Algorithm

球磨机制粉系统是一个复杂的多变量系统,具有强耦合、非线性、大迟延、慢时变等特点,很难建立精确的数学模型,采用常规的控制策略难以获得满意的控制效果。针对上述问题,在对球磨机制粉系统动态特性进行分析的基础上,提出了一种不依赖于被控对象数学模型的多变量PID神经网络解耦控制策略;为进一步提高控制器性能,利用一种改进的PSO算法对PID神经网络的权值初值进行离线优化训练,然后采用BP算法对权值进行在线调整,避免网络陷入局部极小值,保证了系统不会出现大的超调和震荡。仿真结果表明,该策略可以保证球磨机控制系统有大范围的鲁棒性和适应性,能较好地解决球磨机制粉系统的耦合性、时变性等问题,具有优良的解耦机制和控制品质。

Ball mill system is a complex multivariable system, which has characteristics of strong coupling, nonlinearity, large delay and slow time-varying, so it is difficult to build its precise mathematical model and achieve satisfying control effect with conventional control strategy. In view of the problem, the paper proposed a multivariable PID neural network control strategy which is independent of the mathematical model of controlled object based on analysis of dynamic characteristics of the ball mill system. In order to improve the performance of controller further, a modified PSO algorithm was used to off-line optimize training of initial value of weights of PID neural network, and the weights were adjusted by BP algorithm on-line, so as to avoid that network falls into the local minimum, and ensure the system cannot overshoot and shake greatly. The simulation results showed that the strategy can guarantee robustness and adaptability of the control system of ball mill in a large range, and solve problems of coupling and time-varying, which has good coupling mechanism and control quality.