| 摘要:以玉柴YC6K420LN-C31型柴油机为研究对象，基于RBF(radial basis function)神经网络算法建立发动机数据模型，采用PSO(particle swarm optimization)算法进行基于模型的多目标优化研究。研究表明:RBF神经网络建立的NOx、总碳氢化合物(THC)、CO和燃油消耗率(brake specific fuel consumption,BSFC)数据模型的决定系数R2分别为0.93、0.98、0.96和0.95，模型的预测准确度均大于90%，拟合优度和预测能力满足多目标优化的需求；采用PSO算法对发动机进行多目标优化，将适应度目标NOx、THC、CO和BSFC的权重最终均设置为0.25，生成控制图谱并进行台架验证，在推进特性工况下总排放量和油耗相比于原机平均降低了22.9%与5.3%。 |
| 关键词: 柴油机 DOE RBF神经网络 PSO算法 多目标优化 适应度函数 |
|
| Multi-Objective Optimization of Diesel Engine PerformanceBased on PSO Algorithm |
| Wang Shuang,Song Enzhe,Zhao Guofeng,Yao Chong,Dong Quan |
| School of Power and Energy Engineering, Harbin Engineering University,Heilongjiang Harbin 150001 |
| Abstract:Taking YC6K420LN-C31 diesel engine as the research object, the engine data model was established based on the RBF neural network algorithm, and the model-based multi-objective optimization research was carried out using the PSO algorithm. The results show that the decision coefficients of NOx, THC, CO and BSFC data models based on RBF neural network are 0.93,0.98,0.96 and 0.95 respectively, the prediction accuracy of the models is greater than 90%, and the goodness of fit and prediction ability meet the needs of multi-objective optimization. The PSO algorithm was used to optimize the engine for multiple goals, and the weights of the fitness goals NOx, THC, CO, and BSFC were finally all set to 0.25. The control map was generated and the bench verification was performed, the total emissions and fuel consumption were reduced by 22.9% and 5.3% on average compared to the original engine under the propulsion mode. |
| Key words: diesel engine DOE RBF neural network PSO algorithm multi-objective optimization fitness function |
