Cooperative Co-Evolution Algorithms For Complex Flexible Scheduling

Flexible scheduling is one of the core technologies to achieve intelligent manufacturing.Under the condition that flexible scheduling satisfies the constraints of manufacturing systems,flexible scheduling breaks through the restriction of resource uniqueness in traditional schedul-ing and processes the ordered production operations via allocating different resources to achieve the maximum production efficiency.However,increasing market competition and customer demand result in the increasing scale of flexible scheduling.Meanwhile,there exist unavoid-able uncertainties in the actual manufacturing systems.Uncertainties include the uncertainties with prior knowledge such as the periodic aging of equipment and the uncertainties with unex?pected incidents such as the failure of processing resources.Therefore,how to allocate resources flexibly and maximize efficiency in large-scale and uncertain environments are the keys to the promotion and application of flexible scheduling.Cooperative co-evolution algorithms improve the performance by constructing multiple sub-populations.Multiple sub-populations cooperate to accommodate the large-scale and uncertain evolutionary environments of complex systems.However,due to the interdependence between operation sequence and resource allocation of flexible scheduling,the existing cooperative co-evolution algorithms cannot be applied to com-plex flexible scheduling directly.Therefore,this dissertation proposes the corresponding coop-erative co-evolution algorithms for large-scale flexible scheduling,uncertain flexible scheduling with prior knowledge and uncertain flexible scheduling with unexpected incidents to minimize Makespan.The main research contents and innovations are listed as follows:1.Distributed cooperative coevolution algorithm.For the low probability of the asso?dated operations decomposed into the same group and the performance degradation of local search caused by the increasing problem size of large-scale flexible scheduling,this dissertation proposes a distributed cooperative co-evolution algorithm.The proposed algorithm includes a multiple-time random re-decomposition strategy which can increase the probability of the as-sociated operations decomposed into the same group,and an improved local search strategy.The valid encoding and decoding strategies are designed to filter a large number of infeasible solutions.The local search strategy is proposed to enhance the ability of local search via recon-structing critical path by moving critical operations.The cooperative co-evolution mechanism is developed to apply the proposed algorithm to the open source distributed computing frame?work.The probability of the associated operations decomposed into the same group is increased by dynamically adjusting the decomposition to improve the efficiency.The proposed algorithm achieves a lower error rate of Makespan on benchmark and generated datasets.2.Hybrid cooperative co-evolution algorithm.For uncertain flexible scheduling with prior knowledge,the processing time of each operation is an interval value modeled by one triangular fuzzy number.For the unbalance between exploration and exploitation in decision space and the performance degradation of the fixed parameters,this dissertation proposes a hybrid coopera-tive co-evolution algorithm.The proposed algorithm includes the conversion mechanism used for combing particle swarm optimization with genetic algorithm while considering exploration and exploitation,and a parameter self-adaptive strategy used for improving the effectiveness of parameters.The sorting mechanism which considers multiple attributes is designed to im?prove sorting efficiency.The conversion mechanism between real number encoding and integer number encoding is proposed for combining genetic algorithm with particle swarm optimization algorithm to balance the exploration and exploitation in the decision space.The parameter self-adaptive strategy based on contribution is proposed to improve the effectiveness of parameters.The proposed algorithm achieves the effective optimization of fuzzy Makespan.3.Learning-based cooperative co-evolution algorithm.For uncertain flexible scheduling with unexpected incidents,the processing time of each operation is a random value modeled by a probability distribution.For the influence on the flexible scheduling optimization caused by the dynamic relevant relationship and dependent relationship among operations in the evolution-ary process,this dissertation proposes two learning-based cooperative co-evolution algorithms.Cooperative co-evolution algorithm with Markov random field based decomposition strategy excavates the relevant relationship by constructing Markov random filed in the state of pre?processing.Cooperative co-evolution algorithm with Bayesian optimization algorithm based decomposition strategy excavates the dependent relationship by constructing the Bayesian net-work in the stage of the evolutionary process.The influence on flexible scheduling optimiza-tion is reduced through considering the excavated relationships.Both two proposed algorithms achieve the lower expectation of Makespan on the benchmarks with uniform distribution,Gaus-sian distribution and exponential distribution.