Agent Based Production Scheduling Optimization System For A Refinery Plant

Production scheduling is a very important mean to improve economic efficiency and market competitiveness for refineries. Refineries have a continuous production process, some complex production equipments, and many unit equipments. So, there are lots of variables in scheduling models, and the dimensions of constraints are very big. These are not conducive to solving the model.As the initial population of widely used optimization algorithms (such as GA and PSO) is randomly selected, the optimization results have a certain degree of randomness. As usual, we can hardly get the best optimization results just by optimizing one time. In order to gain the best optimization results, we must compute repeatedly, and select the best one from all the results.According to the feature of the typical production scheduling model of a refinery plant and the disadvantages of common optimization algorithms, this thesis designs a Multi-Agent based production scheduling system for a refinery plant, and implements it on the JADE platform. The main works are as follows:First, considering the status and features of the refinery plant, local scheduling model for each equipment are presented. There are five processing units in the refinery plant, including two atmospheric vacuum distillation units, a catalytic cracker, a solvent oil production unit and a gas separation equipment. For each unit, the objective function of scheduling is to minimize the cost of purchasing raw materials and storage products, and constrains are mass balance, equipment processing, capacity quality demand, tank capacity and inventory constraint. Then two common optimization algorithms, GA and PSO, are tested and verified with two examples. And the advantages and shortcomings of two algorithms are summarized. Then, according to the work in the above, the scheduling system based on multi-Agent technology of the refinery plant is established. The system consists of a Management Agent, five local scheduling Agents, three algorithm Agents and a knowledge base Agent. Local scheduling Agents are divided by five processing units and algorithm Agents includes three algorithms, such as GA, PSO and Simplex Algorithm. In this thesis, the task of each Agent is described, and the collaborative interaction between Agents is described as well. In order to implement the Multi-Agent system by computer, the thesis also studies the related technologies of JADE framework, and develops the production scheduling optimization multi-Agent system on JADE platform.Finally, the designed multi-agent system is verified by a real refinery scheduling problem. The Multi-Agent system gives more stable scheduling program, and the feasibility of the system is proved. The interactions between Agents show that Agent-based optimization system reduces the difficulty of scheduling optimization, overcomes the randomness of the optimization results, and improves the stability of optimization.