基于GA-LM-BP神经网络的拖拉机可靠性预测

Tractor Reliability Prediction Based on GA-LM-BP Neural Network

针对经典BP神经网络在可靠性预测中存在收敛速度慢、易陷入局部极小值等问题,利用遗传算法对BP神经网络的权值和阈值进行初步寻优并将其作为网络训练的初始值。通过LM算法对BP神经网络进行局部优化,采用现场跟踪实验得到的拖拉机故障数据,建立GA-LM-BP神经网络拖拉机可靠性预测模型。选取MAE、RMSE、NRMSE作为网络模型的评价指标,与经典BP网络模型、LM-BP网络模型进行对比,结果表明:采用GA-LM-BP神经网络模型预测时的效果最优,与经典BP网络模型相比三项指标分别降低了5.47%、6.30%和11.14%,与LM-BP网络模型相比三项指标分别降低了1.86%、5.31%和6.27%,表明GA-LM-BP神经网络模型具有更好的预测效果。最后采用皮尔逊相关系数法进一步证实了上述模型的精确性。

Aiming at the problems of slow convergence speed and easy to fall into local minimum in reliability prediction of classical BP neural network,the paper used a genetic algorithm to preliminarily optimize the weights and thresholds of BP neural network and took them as the initial values of network training.First,the BP neural network was locally optimized through LM algorithm.Then,using the tractor fault data obtained from field tracking experiment,the GA-LM-BP neural network tractor reliability prediction model was established.MAE,RMSE and NRMSE were selected as the evaluation indexes of the network model.Compared with the classical BP network model and LM-BP network model,the GA-LM-BP neural network mode has the best prediction effect;Compared with the classical BP network model,the three indexes are reduced by 5.47%,6.30% and 11.14%;And compared with the LM-BP network model,the three indexes are reduced by 1.86%,5.31% and 6.27%,It shows that GA-LM-BP neural network model has better prediction effect.Finally,the accuracy of the model was further confirmed by Pearson correlation coefficient method.