样本数量对切削力的神经网络预测精度的影响

Effect of Number of Training Samples on ANN Prediction Accuracy for Cutting Force

为用尽量少的训练样本达到预测目的,通过不同数量训练样本训练网络的对比试验,分析了训练样本数量对基于列文伯格-马夸尔特算法的切削力的神经网络预测精度的影响.用统计学平均幅值和均方差作为误差的评价指标,探讨了训练样本数量与预测精度的关系.研究结果表明:用40～50组样本训练网络,就可以实现特定切削用量范围内切削力的准确预测.

In order to predict cutting force using training samples as few as possible, tramlng samples with different numbers were selected to train an artificial neural network （ANN） respectively, and the effect of the number of training samples on ANN prediction accuracy for cutting force based on the LM （Lenvenberg-Marquardt） algorithm was analyzed by contrast experiments. Statistic mean amplitude and mean square error were taken as the evaluation indexes for forecast results, and the relationship between the prediction accuracy for cutting force and the number of training samples was investigated. The research result indicates that 40 to 50 groups of training samples may be sufficient to obtain accurate cutting force within the certain range of cutting parameters.