基于BP神经网络PID的液压支架初撑力自适应控制

液压支架初撑力对顶板的控制具有重要作用，采用三位四通手动操纵阀的开环控制或两位三通电磁换向阀的先导控制，很难使初撑力达到设定值并保持稳定，即使达到设定值也存在压力降和波动现象。基于此，建立了立柱电液力控制系统数学模型，利用MATLAB分析了系统的稳定性，得到系统的Pole-Zero图右半S平面不存在开环零点和极点，系统为最小相位系统；Nyquist图逆时针绕（-1，j0）的圈数为0，系统相位裕度为941°，幅值裕度为107 dB，系统稳定；阶跃响应115 s趋于稳定，脉冲响应90 s趋于稳定。提出了基于BP神经网络的PID初撑力自适应控制方法，并建立了三层神经网络控制模型，误差控制采用二次型性能指标；采用有监督的Hebb学习规则和梯度下降法对输出层和隐含层的权值系数进行更新，经训练得到PID控制器的三个控制参数。仿真结果表明:期望输入为阶跃信号时，立柱达到初撑力并稳定需要约885 s，期望输入为方波信号时，立柱达到初撑力并稳定需要约91 s，相比没有采用BP神经网络PID控制，其响应时间提高了约13倍。

Adaptive control of setting load of hydraulic support based on BP neural network PID

The setting load of hydraulic support plays an important role in the roof controlThere are two methods to control the setting load of hydraulic support，one is open loop control by manual control valve of three position four port，the other is pilot control by solenoid directional control valve of two position three portHowever，these two methods can hardly make the setting load reach the expected value and remain stableEven when the expected value is reached，pressure drop and fluctuation generally existBased on this，a mathematical model of electrohydraulic force control system is established，then the stability of the system is analyzed by using MATLABIt is obtained that there are no open loop zeros and poles in the right half S plane of the Pole Zero diagram of the system，so the system is a minimum phase system；the number of cycles of counter clockwise winding（-1，j0） from the Nyquist diagram is 0，and the system phase margin is 941° and the amplitude margin is 107 dB，so the system is stable；the step response is stable for 115 s，the impulse response is stable for 90 sAn adaptive PID control method based on BP Neural Network is proposed，and a three layer neural network control model is establishedQuadratic performance index is used to control errorThe weight coefficients of the output and hidden layers are updated by using supervised Hebb learning rules and gradient descent algorithmThen three control parameters of the PID controller are obtained by trainingThe simulation results show that:it takes about 885 s for the setting load to reach the expected value and maintain stability when the expected input is step signal，and 91 s when the expected input is the square wave signalCompared with no BP neural network PID control，the response time is increased by about 13 times.