Research On Multi-Objective Optimization Algorithms Of Prefabricated Concrete Manufacturing Scheduling Based On Reinforcement Learning

From the Guiding Opinions of the General Office of the State Council on Vigorously Developing Prefabricated Building,developing prefabricated building is an eseential evolution in the way of construction.Precast product(PC)is the important part of the prefabricated building,whose manufacturing process is an optimization problem with multi-factor,multi-constraint,and multi-objective,being the key factor in construction process,producing cost,and energy saving.Currently,most researches model the PC scheduling as flow-shop scheduling problem,where most time is wasted in machine idle and the complex characteristics in manufacturing cannot be completely described.Based on the above-mentioned problems,this thesis fully considers the complex constraints and objective characteristics,including distributed schediling,time-of-use electricity price,machine processing speed,crane transportation,group schedulingm,and the integration of production and transportation.And the mixed integer linear programming(MILP)models are built based on the PC manufacturing as a flexible job shop scheduling problem(FJSP).Based on the problem properties,numerous global and local search operators are designed.To different manufacturing circumstances,the deep reinforcement learning algorithms with diversified architectures are proposed,where the reinforcement learning and evolutionary algorithms were integrated,constructing an efficient simulation platform of PC production and transportation.The main work of this thesis can be concluded as follows.(1)For the MILP modeling of PC production scheduling in FJSP,the constraints of overlapping at steam curing operation and unbalanced processing time are considered,where the objectives are makespan and total energy consumption.In addition,the distributed PC manufacturing scheduling problems at multiple factories are also considered.In this thesis,the hybrid algorithm of the estimation of distribution algorithm and variable neighborhood search is designed.The numerical experiment results show that the hybrid algorithm can not only improve exploration ability but enhance exploitation ability in optimization.(2)For the complex scheduling problem in real PC production,time-of-use electricity price and machine process speed are happened in the real PC manufacturing;therefore,this thesis considers the constraints and machine setup and idle conditions.A deep Q-network model is proposed,in which the state features depict characteristics in all scheduling processes.The actions are designed according to problem properties,enhancing the optimization of objectives.(3)For the crane tranpostation in PC manufacturing,the time and energy consumption of crane transportation should be considered.In addition,the setup,idle,and processing stages of machines are also considered.The whole PC scheduling problem is complex,where makespan and total energy consumption are minimized simultaneously.The hybrid algorithm of the estimation of distribution algorithm and deep Q-network is designed to improve the optimization performance.(4)For PC manufacturing in multiple factories,the grouping constraint is also considered for physical distribution factors.In group scheduling constraint,the PCs in the same group should be performed in the same factory,and the setup time is not needed;while the setup time of the PCs of the same groups should be considered.To solve this distributed PC group scheduling problem,this thesis designed muti-DQN construction,where the strategies of single factory and multiple factories are separately managed by two DQNs,and another DQN is designed to control the selection of the two DQNs.Ultimately,the solution refinement strategies are proposed to further optimize objectives.(5)For the whole process from PC production to building site in distributed scheduling,the corrsponding mathematic model is built,this thesis utilizes improved DQN model to solve the integrated problem,which has significant meaning in solving real-world manufacturing and transportation problems.According to the comparison experiments with mathematical model and competitive algorithms,the proposed DQN model can satisfy the production and transportation of PC in multiple factories.