Research On The Distributed Green Shop Scheduling Problem Under Time-of-Use Electricity Tariffs And Its Optimization Algorithm

More and more manufacturing enterprises carry out production and manufacturing activities from the perspective of resource conservation and environmental protection under the dual pressure of climate change and limited resources.The time-of-use electricity traffic,which encourages manufacturers to shift high-energy-consuming production tasks to low-demand periods to reduce peak-demand pressure,is implemented in many regions.Electric energy is one of the main energy consumed in the manufacturing process of the industry.Renewable energy is introduced to reduce electricity consumption and further reduce the total electricity charges,which has become one of the green development goals of manufacturing enterprises.The distributed manufacturing model has become an inevitable trend in the development of manufacturing enterprises with global economic integration and market competition intensified.The distributed shop scheduling problem has become an important combinatorial optimization problem in modern manufacturing systems.The present research shows that this problem is an NP-hard problem that is solved efficiently in a limited time by the intelligent optimization algorithm of reasonable operation.The moth-flame optimization algorithm(MFO),which is inspired by the special navigation mechanism of the moth in nature called transverse orientation,is an intelligent optimization algorithm.MFO algorithm has the advantages of simple structure and easy implementation.In recent years,it has received great attention from experts in various fields.The MFO algorithm is deeply studied and the problem properties are combined in this paper.The corresponding algorithms are designed to solve the distributed green shop scheduling problem under time-of-use electricity traffic.The main research contents are summarized as follows.(1)A comprehensive learning moth-flame optimization with low discrepancy sequence(CLMFOLDS)is designed aiming at the disadvantages of the random initialization method,such as the unbalanced initial population distribution and the weak global and local search ability of the original MFO algorithm.The low discrepancy sequence is employed to generate an initial population with uniform distribution in the search space.The comprehensive learning strategy,which utilizes the information from the entire population to update the position of the moth,is introduced to improve the global search ability of the algorithm.The external storage is designed to preserve the suboptimal solution during the iteration.An elimination mechanism,which eliminates bad solutions from the population,is adopted to further enhance the local search ability of the algorithm.Experimental results show that the CLMFOLDS algorithm performs well in solving complex optimization problems by comparing the algorithm including the original MFO algorithm.(2)Reinforcement learning driven moth-flame optimization algorithm(RLMFO)is proposed to overcome the shortcomings of the original MFO algorithm,such as poor population diversity and prematurity convergence.Opposition-based learning is used to generate the initial population to improve the diversity of the population.Reinforcement learning is introduced to balance the local and global search of the algorithm to solve the prematurity convergence problem of the original MFO.A strategy pool containing Gaussian Mutation,Cauchy Mutation,Lévy Mutation,and elite strategy is designed to store strategies with different functions.The RLMFO algorithm is tested with four comparison algorithms on the CEC 2017 benchmark test suite.The statistical analysis results show that the proposed algorithm is significantly superior to the contrast algorithm in solving constrained realvalue optimization problems.(3)A moth-flame optimization algorithm based on inverse reinforcement learning and Q-Learning mechanism(IRLMFO)is devised to balance the local and global search capabilities of MFO and improve the reliability of the reward function.Inverse reinforcement learning is employed to obtain the reward function,which is more reliable than a manual setup.The reward function is utilized in the Q-Learning mechanism to select the optimal search operation in each generation to balance the global and local search capabilities of the algorithm.The strategy pool with different operations and the competition mechanism are designed,respectively.The performance of the IRLMFO algorithm is verified in the CEC 2017 benchmark test suite.The experimental results show that the IRLMFO algorithm has good performance in solving large-scale real-value optimization problems.(4)Distributed heterogeneous no-wait flow shop scheduling problem under time-of-use electricity tariffs(DHNWFSP-TOU)is studied.The maximum completion time and the total electricity charges are considered as two objectives from the time and economic benefit of the enterprise.A cooperative optimization framework based on inverse reinforcement learning(IRLCOF)is proposed to solve this problem.Two construction heuristic algorithms are designed to cooperatively create the initial population and six search operations combined with the properties of the problem are devised.Inverse reinforcement learning is employed to generate the reward function used in the Q-Learning mechanism.Reinforcement learning is utilized to select the optimal search operation in each generation.The 720 benchmark instances are used to verify the validity of IRLCOF.Experimental results show that IRLCOF performs better than the comparison algorithm in solving the DHNWFSP-TOU problem.(5)Distributed heterogeneous no-wait flow shop scheduling problem with renewable-based microgrids(DHNWFSP-RM)is studied based on DHNWFSPTOU.Renewable energy is introduced in the existing problem model based on the current green development of new needs.The key idea of the RLMFO algorithm is applied to solve this problem.Reinforcement learning driven cooperative optimization algorithm(RLCOA)is proposed and the problem property is transformed into knowledge.The improved Nawaz-Enscore-Ham heuristic and the earliest completion time rule are employed to cooperatively generate the initial population.A local intensification mechanism based on reinforcement learning is designed to exploit the potential solution sufficiently.Reducing electricity charges based on the knowledge transformed by the problem property is designed to reduce the total electricity charges of the factory.The Taillard problem is used to demonstrate the effectiveness of RLCOA.Experimental results show that RLCOA is superior to the contrast algorithm in solving DHNWFSP-RM.