Rolling horizon lot-sizing planning of wood-based panel based on an improved particle swarm optimization

ObjectiveIn general, lot-sizing planning (LSP) is the basis of master production schedule and material requirements planning. Obviously, optimization of LSP is the key to save cost and materials. At present, bulk-production mode is widely adopted in wood-based panel industries. For arranging the productive task, the detail planning needs to be made in advance. However, the scheduled planning usually needs to be re-planned to cope with the change of market requirement dynamically. For this reason, an improved particle swarm optimization (PSO) and a kind of rolling horizon planning (RHP) method are proposed to solve the LSP problem of wood-based panels in this paper.

MethodFirstly, according to the characters of production process, a series of decision variables and constrains were determined, and then, a mathematical mixed integer decision method was built for LSP problem. Secondly, an improved PSO algorithm was designed by reducing the dimension of the solution space. At last, with RHP method, the improved PSO algorithm was used to solve a series of sub-LSP problems repeatedly. A simulation example was designed to verify the performance of the LSP model and improve PSO algorithm.

ResultAccording to the simulation results, to a wood-based panel production line with an annual output of 100 000 m³, when the planning horizon and total planning period were set to 3 months and 12 months respectively, 1.8% production cost could be saved through RHP method and the improved PSO approach. In addition, to the same model with production capacity constrains, the production cost could be reduced by 0.9%.

ConclusionRHP is an efficient way to make lot-sizing planning of wood-based panel. The important advantage of RHP is that the production schedule could be adjusted with the changing market requirement dynamically. Due to the NP-hard of RHP, the intelligent swarm optimizations could solve these problems with an acceptable period of time. In addition, future research will focus on the model of LSP with actual constrains and the algorithms of intelligent swarm optimization.