Research On Modeling Of Emission Characteristics Of Internal Combustion Engine And Application Based On BP Neural Network

Research on the emission and its control of internal combustion engine, whose application enlarging and quantity increasing quickly (automobile is a representative), are becoming a hotspot, because air pollution control becomes stricter. Considering the research trouble such as costly and complicated testing equipments, lots of affected factor in mathematic modeling and with no clear law to be followed etc, a model based on neural network theory with general testing equipments, less testing data and normal complicated mathematic calculation is presented in this paper, which predicted emission characteristics of stationary or automobile operating conditions and control effect of inlet ingredient to emission result. The main work is as follows:1) For establishment of the stationary emission predication model,a complete modeling method based on BP neural network is brought forward, which is a result of understand of neural network and internal combustion engine. According to this, a stationary predication model is set up with training and checking samples got from 6135ZG diesel tests and selected by a new method-variable edge orthonormal planning method put forward by author . The model applies to emission prediction of the stationary and thirteen operating conditions successfully.2) Based on the study of the differences and influence factors of transient and stationary condition emission a double model method predicting the emission in automobile operating condition based on engine stationary testing was put forward. First, transfer the test cycles from automobile condition to engine condition; second, according to the work and emission characters of nonturbocharged for low pressure turbocharged diesel, assume that transition process is quasi stationary one and diesel temperature rises linearly in cold starting period. Finally, the neural cells of input layer are increasedand study samples indicating the thermodynamic status of engine are added to the stationary model , and a emission predication model during warming processing after cold starting is found. This model combined with the stationary model above, can predict emission law of automobile run according to ECE-15 per second.3) With the neural network theory introduced into emission control research, a modeling method to predict multi-factors comprehensive influence based on single factor test is set up. With this model, a research how the inlet ingredients control diesel emission is completed and the range of each composition, which got a good emission result while characteristics of power and economy were affected little is obtained for ROBIN DY41D diesel.The basis of this predication method is the test data, its predicating result not depend on the diesel n\mathematic model. When structural parameters and emission formation mechanism are not deeply understood ,this method can be used with less general equipments and get a good result with less funds and shorter research period. Additionally, limited by test condition, some emission characteristics can not be obtained but can be predicated with the capability of high nonlinear mapping and spreading characteristics of neural network. So this is a real new, quick and accurate method.Only depending on testing data and not related to fuel and structure, this model can applies to different nonturbocharged(or low turbocharged) engines, such as diesel, gasoline engine, CNG and LPG, and is easy to spread.So, this new method with lower requirement can facilitate execution of emission rules, own higher theory meaning and wider engineering practical value.