Behavioral intelligence optimization methods in higher education based on multi-agent collaborative decision-making model

Abstract

In response to the demand for optimizing teaching behaviors in the smart campus scenario, this paper introduces computer technologies such as multi-agent reinforcement learning, deep neural network time series modeling, and attention mechanism, and constructs an intelligent optimization method for higher education teaching behaviors based on a multi-agent collaborative decision-making model. Through dual-stream time series encoding and cross-agent attention to obtain the classroom state representation, combined with a centralized value network and multi-agent deep reinforcement learning framework, and trained under the "offline pre-training + online incremental update" process. Experimental results show that compared with the rule-based strategy, the classroom participation rate and the completion rate of tasks on time have increased by approximately 15.2% and 12.7% respectively, and compared with the single-agent method, they have increased by approximately 7.9% and 6.3% respectively. The comprehensive score reaches 86.5 points, and the resource consumption increase is controlled within 3.4%. This verifies the effectiveness of computer technology-driven multi-agent collaborative decision-making in teaching behavior optimization. The research results provide a technical path that can be implemented in engineering for the transformation of higher education teaching management from experience-based decision-making to data and intelligent decision-making.