Embedding-Graph-Neural-Network for Transient NOx Emissions Prediction

Recently, Acritical Intelligent (AI) methodologies such as Long and Short-term Memory (LSTM) have been widely considered promising tools for engine performance calibration, especially emission prediction optimization, Transformer is also gradually applied to sequence prediction. To carry out high-precision control predicting long time step sequences required. However, LSTM has the problem of gradient disappearance on too input output sequences, cannot reflect dynamic features historic information which derives from cycle-by-cycle combustion events, leads low accuracy weak algorithm adaptability due inherent limitations encoder-decoder structure. In this paper, considering highly nonlinear relation between multi-dimensional operating parameters data outputs, an Embedding-Graph-Neural-Network (EGNN) model was developed combined with self-attention mechanism adaptive graph generation part GNN capture relationship improve ability reduce number simplify network Then, a sensor embedding method adopted make adapt characteristics different sensors, so impact experimental hardware accuracy. The results show that under condition long-time forecasting, error our decreased by 31.04% average compared five other baseline models, demonstrates EGNN can potentially be used in future calibration procedures.