Zero-D predictable combustion model based on neural network and modeling

Zero-D predictable combustion model on the basis of neural network was put forward, which is appropriate to the combustion prediction for both steady and dynamic engine simulation. Main procedures for building a predictable model were introduced, including calculation for the rate of heat release(RoHR), parameterization for RoHR, establishing and training the neural network. Firstly, the in-cylinder pressure curve was smoothed using average method and the RoHR was obtained by thermodynamics. Then, mathematical algorithms were adopted to fit the RoHR in tri-Wiebe function. To solve the multiple solutions in fitting, some constraints were put forward by analysis of parameters meanings. Lastly the radial basis function(RBF)neutral network was established and trained to complete the zero-D predictable combustion model. The accuracy of the model was validated by the training error analysis and comparison between predicted results and experimental data.