基于PSO-BP神经网络的摩托车排放预测

Motorcycle Emission Prediction Based on PSO-BP Neural Network

为了预测摩托车的排放特性,依据台架试验中的某摩托车速度、加速度、λ值与排放数据,利用MATLAB搭建了基于PSO-BP神经网络的摩托车排放预测模型。排放预测模型以双隐含层BP神经网络为基础,使用粒子群算法优化神经网络的权值与阈值,选取速度、加速度与λ值作为输入数据,CO、NO_x和THC的排放浓度作为输出数据。预测结果表明:CO、NO_x、THC排放因子的最大相对误差为8.32%,预测值与实测值相关性强。该模型可为摩托车排放研究提供依据,具有一定的实用价值。

Based on the experimental datameasured in bench test, in terms of speed, acceleration,and emissions of a motorcycle,a motorcycle emission prediction model is constructed using PSO-BP neural network in MATLAB. On the foundation of the double hidden layered BP neural network, this emission prediction model introduces the particle swarm optimization algorithm to optimise the weights and thresholds of the neural network. Besides, it selects vehicle speed, acceleration and as the input data and the PPM of CO, NOx and THC as the output data. The result shows that the maximum relative error of the emission factor of CO, NOx and THC is 8.32%, thus demonstrating a strong correlation between the predicted value and the raw value. The model is able to provide a fundamental guideline for investigating the motorcycle′s emission.