Study On Multi-period Batch Plant Scheduling Under Demand Uncertainty

At a time of economic crisis, chemical production enterprises would like to stay competitive with other companies and pursue profits maximization, which need to make and adjust the production schedule including machines and materials according to the market changes. But there are many uncertain factors in the ever-changing market, such as the demand or the price of the products, the price of the raw materials and etc. The process of making the production schedule is an optimization problem. The uncertain parameters are treated as their expectation by the traditional deterministic method, which may cause the losses of the interests and the even greater risk. Owing to the uncertainty, the stochastic optimization plays an important role in the decision-making process.The scheduling decision-making process for multi-period batch plant under demand uncertainty is studied in this dissertation. The main content of this paper is listed as follows:(1) The realization of the demand uncertainty is studied firstly. The scenario tree is employed to describe the evolution of the uncertain parameter over the time horizon. The adjacency matrix and incidence matrix are used for the mathematical description of the scenario tree. An improved mathematical model is proposed in this paper, which makes the scenario tree and the mathematical model connection close together.(2) The problem of the multi-period batch plant scheduling under demand uncertainty is a multi-stage stochastic programming problem, which is hard to solve. To overcome the great difficulty of the calculation, the hybrid approximation solution strategy based on the decomposition of the scenario tree is proposed, which both employs the shrinking-horizon strategy and recourse strategy. The original large-scale MILP problem is decomposed into several small-scale MILP sub-problems, which largely reduce the computation load and complexity.(3) From the case studies, the feasibility, effectiveness and efficiency of the proposed hybrid approximation solution strategy is verified, which provides consistently superior performance in terms of solution quality without a significant increase in the computational effort.