Research On Multi-project Scheduling Of Customized Manufacturing Enterprise Group

Customized manufacturing enterprise group exists universally in equipment manufacturing industry. The production of customized manufacturing enterprise group is usually driven by customers’orders, and possesses the characteristics of engeneering to order (ETO), single and mini-batch products, several subcompanies cooperating across regin, and multiple projects paralleling, etc. So the multi-project scheduling and planning of such enterprise group not only needs to consider the complex of ordinary project scheduling of the independent enterprise, but also take into account the hierarchical desision and coordination, independent planning of each subcompany, in addition to multi-project selection and allocation, resource sharing, and centralized purchasing. But the current study of multi-project scheduling primarily focuses on the centralized decision way of independent enterprise or decentralized decision way of virtual enterprise alliance, so that the solving model and the optimization methods derived from the above decision way are not suitable for the multi-project scheduling in terms of hierarchical controlling in the enterprise group. Hence, it is very necessay to, under the background of enterprise collectivization, analyze the new issues of multi-project scheduling, and study the solving model and algorithms based on the hierarchical decision and coordination; thus, it can enhance the resource utilization rate, strenthen the coordination among different subcompanies, shorten the makespan of projets, and reduce the total cost of projects.In this thesis, research work maily includes the following aspects:(1) After analyzing the hierarchical controlling mode, a simplified two-layer hierarchical coordination architecture and its corresponding bilevel programming mathematical model were proposed, and then, according to these models, the specified problems solved in group layer and subcompay layer repectively and the corresponding solving algorithms were proposed, too.(2) The project scheduling in a subcompay comes down to a resource-constrained project or multi-project scheduling problem (RCPSP/RCMPSP), but RCMPSP can turn into RCPSP by merging the single-projects. Therefore) we proposed a dynamic diversity evolution strategy (DDES) to solve the RCPSP. DDES can balance the strength between global exploration and the local exploitation by dynamically controlling the diversity of the population and restarting the diversity evolution. Besides, for the problem-specific characteristics of the PCPSP, three new operators are incorporated in the DDES, namely, a max resource utilization rate two-point-crossover operator to recombine the individuals, an enhanced insert-based mutation operator to overcome the premature phenomena, and a diversity-based elitism selection operator to select the next generation individuals. Simulated experiment showed that DDES is better than basic evolution strategy (BES), and, comparing with other state-of-the-art heuristics, can balance the solution quality and efficiency while producing a relative small number of schedules.(3) Under the circumstance of enterprise group, the headquarters instead of the subcompanies receives the orders, and then decides which ones to choose and how to allocate them among the subcompanies according to the resource available quantities and unit costs; finally, each subcompay schedules its allocated projects subject to limited renewable resources. The goal of this problem is to pursuit max total profit of the enterprise group. We presented a mathematic model to describle this problem and then proposed two heuristics, fixed priority based algorithm (FPA) and variable priority based algorithm (VPA), and a hybrid intelligent algorithm, hybrid genetic algorithm and particle swarm optimization (GA-PSO), to solve it. Moreover, we also added a search space compression and a serarch time cut down strategies into the GA-PSO. The simulated experiment showed that GA-PSO outperforms FPA and VPA in any problem scale, and the space and time compression strategies also can greatly reduce the search time on the premise of guaranteeing solution quality.(4) For taking into account the resource sharing situation in the enterprise group, we built up the mathematic model for this multi-project scheduling problem with an objective of minimizing the projects’total cost. We first proposed a DDES-based centralized method (DDES-CM) to solve it, and then proposed a discrete particle swarm optimization algorithm integrated with path relingking (DPSO-PR) after analyzing the deficiency of the DDES-CM. DPSO-PR develops a new particle position updating method for AllDifferent problem, because defining the orders of shared resources is actually an AllDifferent problem. In the phase of resource initialization, a fast allocation approach based on the priority of max resource utilization cost is used, and in the phase of resource reallocation, a serial allocation method is used. The simulated experiment showed that in the most cases, especially in the medium and large scales, the total project cost obtained by DPSO-PR is less than that got by DDES-CM, and the decrement rate of cost increases while the problem scale increases; moreover, comparing with the continuous AllDifferent particle swarm optimization (PSO-CA), it also indicated that the particle postion updating method proposed in DPSO-PR can more easily find a better solution.(5) In allusion to the centralized purchasing in the enterprise group, we built up the mathematic model for this multi-project scheduling problem with the objective of minimizing the projects’total cost. After defining some related concepts of the problem, we pointed out that this problem actually is a grouping problem that has uncertain number of groups. Then a traditional genetic grouping based algorithm for centralized purchasing (GGA-CP) and a grouping discrete particle swarm optimization with tabu list (GDPSO-TL) are developed to deal with it. In GDPSO-TL, applying normalization of particle decoding to solve the redundancy, and tabu list to avoid researching the existing groups. The simulated experiment demonstrated that the total project cost computed by centralized purchasing method which used both in GDPSO-TL and GGA-CP is significantly smaller than that got by decentralized mothod; moreove, under the most circanstances, GDPSO-TL can find better solutions than GGA-CP, and in small and medium scales, the solution efficiency of GDPSO-TL is better than GGA-CP.(6) In the real-life use, the enterprise group is usually hard to think about the multi-project selection and allocation, resource sharing and centralized purchasing factors at the same time from the beginning, so we propose a staging solving procedure. Then we apply it, in addition to the former methodologies, to solve the multi-project scheduling problem in a large mould enterprise group and develop a multi-project scheduling management system. Finally, a real multi-project scheduling case in this enterprise group is used to verify the effectiveness of the proposed research methods again.