基于BP人工神经网络的发动机生产作业环境综合评价模型

Comprehensive evaluation model of engine production environment based on BP artificial neural network

针对输入与输出之间高度非线性映射的发动机生产作业环境综合评价问题,文章应用误差反向传播(error back propagation,BP)人工神经网络构建综合评价模型。通过分析发动机生产作业环境的特点、主要影响因素及其危害,建立发动机生产作业环境评价指标体系,并确定每个单项指标的分级标准;将温度、湿度、气流速度、油雾、噪声以及照度6个指标作为模型输入,舒适度等级作为模型输出,建立3层BP神经网络模型;并应用贝叶斯正则化和动量梯度下降法较好地解决了传统BP人工神经网络训练高精度和预测低精度的过拟合现象。实验结果表明,基于该模型的评价结果符合实际情况,对作业环境改善具有指导意义。

A comprehensive evaluation model is built on the basis of the error back propagation（BP） ar- tificial neural network, focusing on the engine production environment with high nonlinear mapping between its inputs and outputs. Firstly, the evaluation system of engine production environment is built by analyzing the characteristics, affecting factors and hazards of engine production environment, and the classification criterion of each single index is identified. Then a three layer BP neural network model is established and the temperature, humidity, air velocity, oil mist, noise and illumination are taken as model inputs, and the comfort degree is taken as model output. The over-fitting problem of high training precision and low forecasting precision of conventional BP artificial neural network is well solved by utilizing Bayesian regularization with momentum gradient descending method. The ex- perimental results indicate that the proposed model is well coincident with the practical condition, dis- playing good guiding significance for the improvement of engine production environment.