Machine Learning Prediction Based on Demand Power of FPLG

This paper presents a method for predicting the operating parameters of an FPLG based on the demanded power. First, a 1D FPLG model was developed in AMESim, based on established structural principles and a characterization of stable operation. The model was validated at specific operating points using an experimental prototype. Due to the limited number of available operating points in the prototype, the model boundaries were explored, and the influence of input variables was analyzed. Ultimately, injected mass, spark timing, and injection timing were selected as the primary control parameters. Further analysis examined how variations in these parameters affect the system's steady-state operation, and the relationship between input parameters, output efficiency, and power was established. Based on this relationship, two rules - optimal efficiency and stable operation - were proposed. These rules were integrated with a three-layer coupled machine learning model to form an FPLG-specific prediction and control strategy. Finally, the effectiveness of the machine learning predictions was validated using a demand power curve based on the FTP75 test driving cycle.FPLGMachine