基于BP神经网络的油冷器压降及换热量预测

油冷器作为发动机散热部件之一,压降和换热量是评估其性能的重要指标,但油冷器中传热与流动规律错综复杂,所以对其压降和换热量进行预测存在一定难度。本研究提出了一种基于BP神经网络和特征工程的预测方法。该方法通过实验获得不同结构类型下冷油器数据,对样本数据进行插值和增强等方法解决样本量分布不均的问题,并根据相关性计算Shah-Focke关联式、Gray and Web关联式、A.R.Wieting关联式等相关经验公式与本文实验结果相关性,并筛选出相关性最高的关联式来构造新特征,最后利用BP神经网络模型进行预测。结果表明,Shah-Focke关联式与本文实验结果相关性最高,且该经验公式特征的引入对模型有积极影响,预测精度提升50%,令压降预测误差为6%,换热量预测误差为4%。

Prediction of pressure drop and heat exchange of oil cooler based on BP neural network

The oil cooler is one of the engine heat dissipation components. Pressure drop and heat exchange are im-portant indicators to evaluate its performance. However, the heat transfer and flow laws in the oil cooler are complicated, so it is difficult to predict the pressure drop and heat exchange. . This study proposes a prediction method based on BP neural network and feature engineering. This method obtains the data of the oil cooler under different structure types through experiments, interpolates and enhances the sample data to solve the problem of uneven sample size distribution, and calculates the Shah-Focke correlation, Gray and Web correlation, AR according to the correlation Wieting correlation and other related empirical formulas are correlated with the experimental results in this paper, and the correlation with the highest correlation is selected to construct new features, and finally the BP neural network model is used for pre-diction. The results show that the Shah-Focke correlation is the most relevant to the experimental results in this paper, and the introduction of this empirical formula feature has a positive effect on the model. The prediction accuracy is improved by 50%, the pressure drop prediction error is 6%, and the heat exchange amount prediction error is 4%.