Probabilistic Guaranteed Gradient Learning-Based Spark Advance Self-Optimizing Control for Spark-Ignited Engines.

Area of Science:

• Internal Combustion Engines
• Control Systems Engineering
• Combustion Science

Background:

• Spark advance (SA) critically influences efficiency in spark-ignited (SI) engines.
• Online self-optimizing control (SOC) for SA faces stochastic optimization challenges due to combustion variability.
• Existing gradient-based methods using dithers introduce undesirable noise in SA control decisions.

Purpose of the Study:

• To develop a dither-free SOC scheme for SA control in SI engines.
• To maximize indicated fuel conversion efficiency (IFCE) while ensuring stable control sequences.
• To address the stochastic nature of combustion for robust engine optimization.

Main Methods:

• Implementation of a gradient descent-based SOC scheme utilizing probabilistic guaranteed gradient learning (PGGL).
• PGGL approximates gradients using statistical distributions of past samples with adaptive sample size adjustment.
• Analysis of convergence performance based on probability distribution.

Main Results:

• The proposed PGGL-based SOC scheme guarantees gradient learning accuracy and adaptively adjusts sample size.
• It achieves a balance between rapid response and stable decision sequences.
• Experimental validation on an SI engine test bench confirmed successful operation near optimal IFCE.

Conclusions:

• The PGGL-based SOC scheme effectively optimizes SA for maximum IFCE in SI engines.
• The method provides a fast response and stable SA behavior under various operating conditions.
• This approach offers a robust solution for online engine efficiency optimization.