Investigations on multi-scale chaotic characteristics and hybrid prediction model for combustion system in a premixed lean-burn natural gas engine

• Published in: Fuel (Guildford) 2025-02, Vol.381, p.133393, Article 133393
• Main Authors: Ding, Hao, He, Shuai-Feng, Ding, Shun-Liang, Ke, Yun, Yao, Chong, Song, En-Zhe
• Format: Article
• Language: English
• Subjects: Combustion system; Excess air coefficient; Hybrid prediction; Multi-scale chaotic characteristics; Natural gas engine
• ISSN: 0016-2361
• DOI: 10.1016/j.fuel.2024.133393

•The multi-scale fluctuation characteristics of engine combustion system are studied.•The multi-scale chaotic characteristics of engine combustion system are studied.•The IWNN is proposed with adaptive error compensation performance.•The hybrid prediction model for the engine combustion system is established. This work investigates the multi-scale chaotic characteristics and prediction model of combustion system in an electronically controlled natural gas engine with varying excess air coefficients (λ). The indicated mean effective pressure (IMEP) is calculated based on the measured in-cylinder pressure. The wavelet decomposition method is applied to process the calculated IMEP time series, yielding a set of sub-sequence signals with varying frequency components. Phase space reconstruction and the largest Lyapunov Exponent (LLE) are employed to analyze these sub-sequence signals under different λ conditions, enabling both qualitative and quantitative identification of multi-scale chaotic features in the combustion system. Results indicate that combustion instability in the natural gas engine evolves across multiple time scales. The high-frequency signal D1 makes the most significant contribution to the fluctuations in the IMEP time series near the lean-burn limit. The engine combustion system exhibits notable multi-scale chaotic characteristics akin to those of typical chaotic systems. The attractor trajectories of signal D1 are disorderly distributed and show distinct chaotic features, with randomness being the predominant factor. Signal D2 exhibits the strongest chaotic characteristics, followed by signal D3 and S3. Based on these results, a hybrid prediction model for the engine combustion system is developed by integrating a chaotic prediction algorithm, an improved wavelet neural network (IWNN), and a genetic algorithm (GA). Results demonstrate that the established hybrid prediction model can accurately forecast abnormal combustion behaviors in the subsequent engine cycle. The average relative error (MRE) of the predictions is only 0.43 % when λ = 1.6, but accuracy declines when λ > 1.6. Compared to other prediction models, the hybrid prediction model developed in this paper achieves the highest accuracy for the combustion system across different λ conditions. The theoretical analysis and development of this model support the adoption of active control strategies to enhance the combustion stability, efficiency, power, and emissions of natural gas engi