高斯变异粒子群优化径向基神经网络铣削力预测

Gaussian Variational Particle Swarm Optimized Radial Basis Neural NetworkMilling Force Prediction

针对传统电流-切削力神经网络预测模型精度不高且神经网络参数难以确定的问题,将主轴电流与驱动轴电流同时考虑作为输入样本,提出帐篷映射下高斯变异粒子群优化径向基神经网络算法。算法在改进收缩粒子群径向基神经网络(改进CFA PSO-RBF)的基础上,对粒子位置初始化采用帐篷映射(Tent Map),同时提出粒子动态高斯变异。该算法能够均匀化粒子初始位置,控制变异过程,并有效避免算法陷入局部最优的早熟问题。基于此方法进行算法对比分析实验,结果表明:同时考虑主轴与驱动轴电流,较单一考虑主轴电流,铣削力预测精度更高;在该算法下,随机15次训练结果平均均方根误差低于BP、RBF、改进CFA PSO-RBF神经网络,能够有效提高铣削力预测精度。

Aiming at the problems of the traditional current-cutting force neural network prediction model with low accuracy and difficult to determine the parameters of neural network,considering both spindle current and drive shaft current as input samples,the Gaussian variational particle swarm optimized radial basis neural network algorithm under Tent mapping was proposed.The algorithm was based on the improved contraction particle swarm radial basis neural network(improved CFA PSO-RBF),Tent map was adopted for particle position initialization,and particle dynamic Gaussian variation was also proposed.Using the algorithm,the initial position of particles could be homogenized,the mutation process could be controlled,and the algorithm falling into the local optimal premature problem could be effectively avoided.Based on this method,the algorithm comparison and analysis experiments were carried out.The results show that:the milling force prediction accuracy is higher when considering the spindle and drive shaft currents at the same time than considering the spindle current alone;under this algorithm,the average root-mean-square error of the randomized 15 training results is lower than that of the BP,RBF,and the improved CFA PSO-RBF neural network,by which the accuracy of the milling force prediction can be effectively improved.