Integrated energy particle swarm optimization method with multi-head self-attention

To address the issues of slow convergence and susceptibility to local optima in traditional particle swarm optimization (PSO) algorithms within the energy domain, this paper proposes an Attention-PSO optimization method based on a multi-head self-attention mechanism. Unlike conventional approaches that modify particle update parameters, the proposed method represents particles—carrying trend information as vectors, and employs a self-attention mechanism to rapidly focus on high-quality particles. This enhances the global search capability of the swarm and improves convergence speed, accuracy, and stability under multiple constraints. Using a combined cooling, heating, and power (CCHP) system—including gas turbines, boilers, and electric chillers—as the experimental case, comparative results demonstrate that the improved algorithm achieves better coordination among multiple objective indicators, significantly accelerates convergence, and optimizes overall system performance, thereby providing a reliable basis for subsequent energy planning.