BP神经网络在双燃料发动机排放预测中的应用

Emission Prediction of a Dual-Fuel Engine Based on Back Propagation Neural Network

运用BP神经网络(Back Propagation Network)的自学习以及非线性逼近能力,对双燃料发动机排气中CO、HC、NO_x和碳烟的浓度进行拟合和预测。搭建神经网络模型,通过采集双燃料发动机排气浓度数据对神经网络模型进行训练和验证。当BP神经网络训练过程中样本和模型计算值的线性相关系数R大于0.9,且用于验证的数据和模型运算值误差在可忽略范围内,则所建的神经网络模型能够有效预测双燃料发动机的排气浓度。训练结果显示,CO、HC、NO_x和碳烟浓度的模型计算值和实测值线性相关系数R都大于0.9,说明神经网络具有较强的拟合能力;验证结果显示,预测值和实测值的相对平均误差都小于10%,能够满足实际需求。结果表明,运用神经网络模型能够有效预测双燃料发动机的排放。

The emission characteristics of a marine LNG/diesel dual-fuel engine, namely the concentrations of CO,HC,NO_x and soot, were fitted and predicted applying the self-learning and nonlinear approximation abilities of BP network(Back Propagation Network). Firstly, a set of emission concentration of the dual-fuel engine was obtained in bench test to train the BP network. Secondly, the BP network model was trained by analyzing and matching the measured values of CO,HC,NO_x and soot concentration. The model could be considered to predict the emission performance accurately if the linear correlation coefficient(R)of the input data's train result exceeded 0.9 and the errors between output data and measured value could be ignored. As showed in the results, the R values of CO,HC,NO_x and soot concentration model and the correlation coefficient(R)of the measured value were both above 0.9. This result showed that the network had strong fitting capacity. In addition,the average error of the predict value and the measured value was less than 10% from the result. To summarize, it was indicated that this model could predict the emission performance of dual-fuel engine effectively.