Neural Network Identification And Control On Quadruped Robot Hydraulic Driving Unit

In recent years, the legged robot which as a new type of high-tech product plays an increasingly important role in the field of military action, disaster relief and polar exploration, and its research work get wide attention at home and abroad. With the development of science and technology, the control performance of quadruped robot has higher requirements. However, hydraulic unit has the characteristics of hysteresis, time-varying and nonlinearity. It can increase the difficulty of identification and control. Therefore, it has value of engineering application to improve the identification precision and research the new intelligent control algorithm of the electro-hydraulic servo system.Quadruped robot hydraulic driving unit is studied in this thesis. The linear differential equations model on valve-controlled cylinder of the position servo system and single degree of freedom driving force system are established according to hydraulic driving unit the hardware-in-loop simulation test bench. The model structure (order) is verified by the model response curves obtained based on MATLAB simulation, which provides the theoretical basis for the subsequent neural network identification.Secondly, The BP network and Elman network are used to identify model of the quadruped robot hydraulic driving position/force unit instead of the mathematical model of linear differential equations that can not represent the actual system. Quasi Newton algorithm, LM algorithm and adaptive learning rate algorithm are used to modify the weight function in order to improve the convergence speed, the accuracy and the generalization ability. The mean square error and the normalized mean square error are used to modify the error function. The identification accuracy is verified by comparing the experimental data with identification curves, which lays a foundation for further research of the control theory.Finally, the neural network PID controller is designed in order to improve the static characteristics of the quadruped robot hydraulic driving position/force unit. On-line adjustment PID parameters are realized based on the gradient descent BP neural network to solve the slow convergence rate and linear and discontinuous of the neuron transfer function. The experimental results show that the controller is effective on the force/position servo system under different load cases, which provides a reliable basis for the neural network control of quadruped robot hydraulic driving position/force unit.The hydraulic drive unit of quadruped robot is recognised by static and dynamic neural network in this thesis. And hydraulic driving position/force unit is controlled. The study provides a reliable basis for the neural network control of quadruped robot hydraulic driving position/force unit.