As a kind of distributed energy storage system, ice-storage air conditioning (IAC) system is an effective way for transferring cooling load. A parallel adaptive critic design method is constructed for solving the optimal control problem of IAC systems. Adaptive critic design is also called adaptive dynamic programming (ADP). The initial value function of the parallel ADP method is obtained by using particle swarm optimization (PSO) method to pre-train the weights of the neural networks. Another PSO algorithm is utilized in each iteration to achieve the target control, which avoids solving a nonlinear equation in traditional ADP algorithms. Finally, the efficiency of the parallel ADP algorithm will be illustrated through the experiment results.