Research On The Batching Decision And Batch Scheduling Problems In The Petrochemical Production Process

Currently, when the competition of global economy is getting more intense, it is most important for the survival and development of petrochemical enterprises to enhance the production efficiency and meet the various market demands which have been a primary aspect for the enterprises to occupy market share in the fierce competition. The scientific production management is one of the most efficient ways for petrochemical enterprises to promote their core competitiveness while the production planning and scheduling are the key factors in the production management. Only the reasonablely making and executing the production planning and scheduling can shorten production cycle, cut down material and energy consumption, decrease production cost, and improve product quality.This dissertation which is based on the research background of batch processes in the petrochemical industry focuses on the modeling techniques and approximate solution methods for the production planning and scheduling problems as follows. The continuous time modeling method is studied for the parallel machine scheduling problem. Moreover, the continous and discrete particle swarm optimization (PSO) algorithms are proposed to solve the batching decision problem and the campaign decision problem, respectively. As for the scheduling with batching decision problems, a hybrid approach based on PSO and artificial immune algorithm (AIA) is presented to solve the single machine scheduling problem integrating batching decision. Besides, a hybrid algorithm of PSO and differential evolution (DE) is investigated to deal with the parallel machine scheduling with batching decision problem. In terms of multi-stage lot streaming problem, a discrete PSO algorithm based on the ordinal optimization (00) framework is proposed to solve it. The major research results of this dissertation are as follows:1) The continous time modeling method is studied for the uniform parallel machine scheduling problem. The main idea of this modeling method lies in two aspects. On the one hand, binary variables are adopted to assign tasks to the time-slots of units when the numbers of time-slots on different units have been given in prior. On the other hand, continous variables are introduced in the model to define the time-tables of time-slots. Moreover, based on index definations of assignment variables and different positions of empty time-slots on time axes,3-index and2-index MILP models are presented. To improve the efficiency further, two heuristic algorithms are proposed to reduce the numbers of time-slots and two types of heuristic models are presented. Finally, both the superiority and disadvantage of all the formulations presented are analyzed by some experiments.2) The petrochemical batching decision problem is studied where the storage facilities and processing units of a batch plant are linked by divergent, convergent, cyclic, multi-inputs and multi-outputs material flows. The batching decision problem is to determine the size and number of batches for different tasks with objective of optimization some production target while satisfying the recipe constraints. The difficulties of this problem lie in the flexible input and output proportions of a task, limited storage capacity of intermediate products, some perishable intermediate products and complex material flows. Based on some preliminary properties, an improved continuous PSO algorithm is developed to solve it. To deal with the mass constraints from complicated recipes, a novel constraint handling mechanism including a forward repair strategy and a constraint fitness-based method is incorporated into the proposed PSO algorithm to speed up the convergence of particles toward the feasible area. Besides, a local search strategy for the global best particle found so far improves the performance of PSO further. The performance of the proposed PSO algorithm is evaluated by comparing the experimental results with the optimal solutions and the lower bounds obtained by CPLEX solver. The proposed PSO obtained the optimal solutions for all the test instances in the small-scale instance set and the maximal deviation from the lower bound is no more than5%in the large-scale instance set.3) The petrochemical campaign decision planning problem with fixed batch size is studied. The campaign decision problem is to determine the number and length of campaigns for different products over a planning horizon such that some production target is optimized where a campaign is defined as the production amount of a specific product type of one contiguous production run. This problem is characterized by the variable capacity of facilities in different periods, product-dependent setup times and costs, and the campaign restriction that no two successive campaigns may produce the same product. An improved discrete PSO algorithm is presented to solve it. In the proposed PSO, a "product-to-period" binary representation for the discrete particle is designed in terms of some properties analyzed in prior. Moreover, the velocity of particles is redefined and a new strategy is developed to move a particle to the new position. To escape from local minima, a disturbance strategy for the position and velocity of a particle is also introduced during the iteration process of the PSO. The performance on the proposed PSO algorithm is evaluated by comparing the experimental results with the commercial optimization software on mass instances with different scales.4) The single machine scheduling problem integrating batching decision for batch processes in the petrochemical industry is investigated. Some special features are considered in this problem, such as variable batch size, identical batch processing times and sequence-independent setup costs. This problem is formulated as a mixed-integer nonlinear programming model with the objective of minimization of the total weighted tardiness costs and set-up costs. A hybrid approach based on PSO and AIA is proposed to solve it where a novel particle solution representation is designed for representing a batching scheme and the schedule of batches is determined by a heuristic method. In addition, a repair procedure is adopted to rectify those infeasible particles during the iteration process of the PSO and the AIA mechanism is incorporated to enhance the performance of PSO. The experimental results on randomly generated instances with different structures show that the poposed PSO is superior to the standard PSO and GA.5) The parallel unit scheduling with batching decision problem is studied. In this problem, the variable batch size and batch processing time, the limited capacity of units and alternative processing units are considered. A mixed-integer linear programming model is developed for the problem with the objective of makespan minimization. To solve this problem, a PSO algorithm is presented where the encoding and decoding methods are designed based on some analysis about the characteristics of this problem. A scale-based repair procedure is also applied to amend the infeasible particles in the swarm. To improve the quality of solutions further, DE algorithm is merged into the proposed PSO. The computational results on randomly generated test instances show that the proposed PSO algorithm outperforms the standard PSO method.6) In multi-stage assembly lines production environment, multi-job lot streaming problem as a class of scheduling problems with batching decisions is investigated. To improve the computational efficiency, a discrete PSO based on the OO framework is proposed to solve it where the OO algorithm is act as the main framework and the PSO is embedded into this framework as a sub-procedure. The number of fitness evaluation in the proposed algorithm is reduced effectively since the searching objective is relaxed to find a satisfying solution instead of an optimal solution. A novel encoding scheme based on continuous values is designed, which converts continuous position values of particles into job sequences. Given a job sequence in a particle, the single-job lot streaming problem is solved for each job in the sequence one at a time by a heuristic algorithm. Moreover, the population is initialized by using two types of heuristic including a constructive and a random one to enhance both the quality and the diversity of solutions in the swarm simultaneously. Computational results and comparisons with the results reported in the literature demonstrate a1.42%decrease on the solution gap and the maximal deviation from the lower bound is no more than5.43%.