Research On Production Optimization Scheduling Based On Evolutionary Algorithm

The complex and volatile market environment requires the workshop to produce multi-batch and multi-variety products in a short period of time.In order to obtain the above objectives,the workshop scheduling plan should be optimized so that the production materials are properly allocated to maximize resource utilization.The research on the optimal scheduling of production in the workshop is an important guarantee for realizing green,economical and efficient manufacturing.The above research has application value and practical significance for improving resource utilization and the intelligent level of manufacturing.This paper studies the optimal scheduling of production and related algorithms in static and dynamic environments.The study are listed as follows:1.When using the existing evolutionary algorithm to solve the flexible job-shop scheduling problem(FJSP)in the static environment.Because of the lack of selection strategy,the size of the selected population is greatly reduced,or the individuals in the population are crowded in a narrow area.This study adds limited information to the stable matching and proposes a limited stable stable matching(LSTM)strategy to solve the above problems.This strategy limits the preference of sub-problems to individuals and balances the diversity and convergence of selected populations.In turn,a variety of scheduling programs with good performance are obtained to guide production.The LSTM was validated by benchmark instances(15 instances)and real instance.The experimental results show that the algorithm with LSTM as the selection strategy has the best convergence in 11 benchmark instances and real instance,and the best diversity in 13 benchmark instances and real instance.2.In this paper,the dual-selection-strategy and the dual-information-based evolution strategy are used to solve the problems of “excessive convergence”,“elite solution loss” and“slow convergence” of existing evolutionary algorithms.Furthermore,this paper proposes dual-information-based evolution and dual-selection strategy in evolutionary multiobjective optimization(MOEA/D-DIDS).The algorithm uses both individual historical information and neighborhood information to guide evolution.It increases the amount of evolution and speeds up the convergence.In the selection operation,the adaptive limited stable matching strategy is used to select the parent population,and the convergence and diversity of the parent population are balanced to overcome the excessive convergence.The stable matching strategy is used to select the offspring population and send it to the next generation,so that the individuals in the population are selected again to retain the elite solution.The numerical simulations on multiobjective(2-,3-,5-,8-,10-,15-objective)benchmark instances verify that MOEA/D-DIDS can obtain convergence and diversity populations at a faster speed of convergence.3.In solving the dynamic flexible job-shop scheduling problem(DFJSP),the robustness of the schedule must be taken as the optimization objective.However,the existing robust computing methods have the disadvantages of “large computation” and “low precision”.This paper propose a robustness measure based on Extreme Learning Machine(ELM)to solve the above disadvantages.Based on machine breakdown,operation and slack-time information,the input information of ELM is generated,and the robustness of the schedule is calculated through the prediction function of ELM.Subsequently,the robustness measure and MOEA/D-DIDS are used to solve the DFJSP in a cooperative manner.This study uses a benchmark instances with different machine breakdown for simulation verification.The experimental results show that the robustness measure has the best performance on the 93.3%instances.Compared with the existing algorithms,MOEA/D-DIDS has better performance when solving DFJSP.