The mutation operations and related control parameters play important roles in the performance of the differential evolution algorithm. Learning optimal policies for these strategies and parameters through reinforcement learning is a hot topic. However, most of the current studies focus on either mutation strategy selection or the control parameters alone while the others keep fixed or self-adaptive, resulting in deteriorated performances. To address this gap, this paper proposes a framework for the joint adaptation of mutation strategies and related control parameters based on deep reinforcement learning. In this method, the distributed proximal policy optimization algorithm is employed to train the agents to dynamically select the optimal combination of mutation strategies and control parameters. To enhance the agent’s learning of the optimal policy, information derived from fitness landscape analysis is incorporated into the state representations. The training is conducted on the black-box optimization benchmark test problems, which are capable of generating large-scale test instances. Numerical results on the new problems from CEC2013 and CEC2017 test suites, and the real-world application of rover trajectory planning demonstrate that the proposed approach achieves competitive performance compared to state-of-the-art methods. The adaptation behavior and the contribution of learning are also thoroughly analyzed.

Most current studies formulate the cloud workflow scheduling as a single-objective or multi-objective optimization problem with at most three objectives, which is unable to fully meet practical scenarios′ needs. To address the limitations above, many-objective cloud workflow scheduling model was established, where many indicators such as time, cost, reliability, resource consumption, load balancing, were taken into account. Then, an improved co-evolutionary algorithm was introduced to address this problem, where dual-stage search strategy and multi-indicator cooperation mechanism were employed to effectively balance the convergence and diversity of solution set, so as to enhance the search capability of algorithm. Experiments on seven types of real life workflow instances reveal that our proposal outperforms the existing peers and can find better scheduling schemes in most cases.

To address the issue of the lack of generalization capability of deep reinforcement learning in flexible job shop scheduling problems, a method combining curriculum learning and deep reinforcement learning was proposed. The training instance difficulty was dynamically adjusted, with an emphasis on enhancing the training of the most difficult instances, to adapt to different data distributions and avoid the forgetting problem during the learning process. Simulation test results demonstrate that the algorithm maintained decent performance on large-scale untrained problems and benchmark datasets. It achieves better performance on four large-scale untrained problems with two artificial distributions. Compared to exact methods and metaheuristic methods, for problem instances with larger computational complexity, it could rapidly obtain solutions of decent quality. Moreover, the algorithm can adapt to flexible job shop scheduling problems with different data distributions, exhibiting a relatively fast convergence speed and good generalization capability.

This paper considers the impact of setup time in production scheduling and proposes energy-aware distributed hybrid flow shop scheduling problem with sequence-dependent setup time (EADHFSP-ST) that simultaneously optimizes the makespan and the energy consumption. We develop a mixed integer linear programming model to describe this problem and present a two-stage adaptive memetic algorithm (TAMA) with a surprisingly popular mechanism. First, a hybrid initialization strategy is designed based on the two optimization objectives to ensure the convergence and diversity of solutions. Second, multiple population co-evolutionary approaches are proposed for global search to escape from traditional cross-randomization and to balance exploration and exploitation. Third, considering that the memetic algorithm (MA) framework is less efficient due to the randomness in the selection of local search operators, TAMA is proposed to balance the local and global searches. The first stage accumulates more experience for updating the surprisingly popular algorithm (SPA) model to guide the second stage operator selection and ensures population convergence. The second stage gets rid of local optimization and designs an elite archive to ensure population diversity. Fourth, five problem-specific operators are designed, and non-critical path deceleration and right-shift strategies are designed for energy efficiency. Finally, to evaluate the performance of the proposed algorithm, multiple experiments are performed on a benchmark with 45 instances. The experimental results show that the proposed TAMA can solve the problem effectively.