Accurate and comprehensive reconstruction of in-cylinder combustion process is essential for timely monitoring of engine combustion state. This article developed a method based on the zero-dimensional (0-D) physical model integrated with big data. The traditional 0-D prediction model based on cumulative fuel mass is improved, the factor of in-cylinder temperature is introduced to adjust the heat release rate, which solves the problem of difficulty in calibrating the heat release rate. Then, convolutional neural network-gated recurrent unit (CNN-GRU), as a deep neural network, including a special convolutional layer and a gated recurrent unit (GRU) neural network is designed for the parameters to be calibrated in the model. The 0-D predictive combustion model is constructed by combining the physical model with CNN-GRU, the combustion process is simplified and reconstructed. The fitting results show that the 0-D physical model based on improved cumulative fuel mass approach is an effective method to reflect the heat release law. Under non-calibration conditions, the root mean square error (RMSE) value of peak firing pressure (PFP) based on CNN-GRU prediction model is 0.5862. The prediction model is a promising method to realize online fitting and optimization of combustion process.
Keywords: 0-D model; Deep learning; Diesel engine; Prediction model.
© 2023 College of Power and Energy Engineering Harbin Engineeing University. Published by Elsevier Ltd.