The core of this patent lies in proposing a method for training endogenized models of multi-agent systems. By utilizing sampling tasks, extracting Agent embedding representations and meta-feature ...