粒子群小波神经网络在交流伺服系统中的应用

Application of Particle Swarm Optimization Wavelet Neural Network on AC Servo System

针对大功率交流伺服系统存在的非线性特性以及不确定扰动而难以建立其精确数学模型的问题,提出了一种基于粒子群小波神经网络的系统辨识方法。由于粒子群算法具有避免局部极小所带来的系统不稳定、收敛速度快等优点,将小波神经网络的各连接权值和各阈值作为粒子群算法里粒子的位置向量,并且将该网络的权值和阈值按照粒子群算法寻求最优值,取代了传统的梯度下降法。将该算法与传统的小波神经网络的辨识结果进行了比较,表明基于粒子群小波神经网络优化算法的函数逼近误差能力、网络性能方面均比传统的小波神经网络算法有着显著的提高,并且有效的解决了局部极小值问题。

As a result of the non-linear characteristics and the uncertain disturbances in high-power AC servo system, it is difficult to construct an accurate mathematical model. In order to solve this problem, a system identification method based on particle swarm optimization wavelet neural network was proposed. Due to the advantages of the particle swarm optimization including avoidance of the local minimum of the unstable system and fast convergence rate, the particle swarm optimization was used to optimize the parameter of wavelet neural network. In the particle’s positions vector, the connection weights and the threshold value as a particle swarm optimization algorithm. Moreover, according to the optimal value in the particle swarm algorithm, the network weights and threshold value was replaced the traditional gradient descent method. Compared with traditional wavelet neural network, the identification result shows that the particle group of wave neural network algorithm has reduced function approximation error and improved ability of the network performance significantly. And to a certain extent, it also solves the local minimum value problem.