Research on Online Learning Architecture for Engine Combustion Prediction on Multi-Platform Simulation

To address the limitations of conventional offline data-driven models for engine parameter prediction in HIL testing, including poor generalization and inefficient use of supplementary data, this study develops an innovative cross-platform online learning architecture that integrates a pre-trained Python-based Wiebe parameter prediction model with high-fidelity MATLAB/Simulink engine simulation. The proposed framework incorporates five key functional modules (real-time data processing, online regression prediction, performance evaluation, incremental learning optimization, and engine simulation) to enable dynamic adaptation to varying engine conditions through seamless integration of Python’s incremental learning algorithms with Simulink’s simulation environment. By implementing a kth order polynomial decay learning rate strategy, the architecture significantly improves model convergence under limited training conditions while enhancing real-time performance and reliability in HIL testing scenarios. Experimental results demonstrate a 15% improvement in prediction accuracy compared to traditional offline methods, confirming the technical advantages of this MATLAB/Simulink/Python-based online learning approach for engine parameter prediction in industrial testing applications.