基于DOE和神经网络的增压柴油机实时模型建模方法研究

Development of Real-Time Capable Turbocharged Diesel Engine Model Based on DOE and Neural Network

基于一维仿真物理模型,研究了发动机实时模型的建模理论和方法。将发动机划分为进排气管路、中冷器、涡轮增压器和气缸四个子系统,分别探讨了各个子系统的建模方法,并重点研究了利用DOE和人工神经网络构建发动机气缸模型的方法。在此基础上,以道依茨BFM1015增压中冷柴油机为对象,对该方法进行了验证。研究结果表明:与传统的面向控制的平均值模型相比,运用新方法建立的面向控制的发动机实时模型误差小于5%,具有精度高、对试验数据依赖低等特点。

Abstract ： An approach of real-time modeling for turbocharged diesel engine based on 1-D physical model was presented. The whole diesel engine system was divided into four subsystems, intake and exhaust pipe system, intercooler, turbocharger and cylinder, and modeling methods of each subsystem was analyzed respectively. Using DOE and neural network, the engine cylinder model was established. Based on the approach, a real-model of DEUTZ BFM1015 turbocharged and intercooled diesel engine was established and applied to a transient simulation. Results show that different from traditional control-oriented mean value model, the real-time engine model has advantages of higher accuracy and less reliance on test data,maximum error is less than 5 ~.