An artificial bee colony-assisted reinforcement learning approach for improvement in optimizing performance of interconnected systems

This paper focuses on the decentralized optimal control issue for networked nonlinear interconnected systems by the intelligent optimization algorithm with reinforcement learning (RL). Different from existing studies employing the gradient descent method, the artificial bee colony (ABC) strategy is employed to tune the weight vector, which has the main advantage of conducting global and local searches in each iteration to avoid sinking into local optimal solutions. Furthermore, to achieve privacy-preserving and improve the security of network establishing integer data through quantization technology, a Paillier encryption and decryption mechanism is introduced to map the transmitted information to ciphertext space. Meanwhile, within the framework of critic designs, the Paillier encryption and decryption-based local Hamilton-Jacobi-Bellman equation (HJBE) is solved by utilizing a single critic neural network. An ABC-assisted RL algorithm is developed to solve the HJBE and then tackle the optimal control problem. Based on this, the stability of dynamics in the sense of uniformly ultimately bounded is conducted by envisioned optimal control. Eventually, simulation results demonstrate the effectiveness and availability of the proposed algorithm.