Research On Multi-objective Swarm Intelligence Algorithm For Hot Rolling Production Planning And Load Distribution

In recent years, with the increasing competition in the steel industry, most of the ironand steel enterprises want to reduce production costs and improve product quality. Hotrolling process is one of the key processes in the steel production, and it has two importantoptimization problems: hot rolling production planning (HRPP) and load distribution. TheHRPP problem has an important impact on production efficiency and production costs,while load distribution is an important way to improve product quality. Therefore, to studythe optimization methods for the HRPP and load distribution problems has a practicalsignificance.The HRPP and load distribution essentially belong to the multi-objective optimizationproblems. Most of the previous methods belong to single objective optimization based onthe weighted-sum approach. However, it is difficult to determine the weight coefficients inpractice, especially when the objectives have different orders of magnitude. Therefore, weadapt multi-objective swarm intelligence algorithm (MOSIA) to optimize the HRPP andload distribution problems. The MOSIA can not only avoid the selection of weightcoefficients, but also generate more than one Pareto-optimal solution in one run, whichprovides more decision-making flexibility for decision-makers.As for the HRPP and load distribution problems, a MOSIA framework based on theMaximin fitness function is proposed to improve the performance and decrease thecomputational complexity. In the MOSIA, fitness assignment, elitism preservation anddiversity preservation are three most important mechanisms. This framework assignsfitness, preserves elitism and diversity based on the modified Maximin fitness function. Inorder to improve the diversity performance, a hybrid diversity preservation strategy is alsoproposed. Meanwhile, by introducing a two-dimensional array to store the min fitnesses,the computational complexity of the framework decreases to O(mN2), which reflects the idea of trading space for time.In this dissertation, the HRPP problem is formulated as a multi-objective prizecollecting vehicle routing problem (PCVRP) model, which considers the selection processof the candidate slabs. Moreover, a Pareto max-min ant system algorithm (P-MMAS) isproposed to solve this model. On the basis of MMAS, P-MMAS modifies the statetransition rule, the pheromone updating rule, the local search rule and pheromone trailsmoothing mechanism, as well as employs the MOSIA framework based on the Maximinfitness function to preserve elitism and diversity. Then, a multi-objective optimizationmethod for the HRPP problem is proposed. Firstly, P-MMAS is used to minimize thepenalties caused by jumps between adjacent slabs, and maximize the prizes collectedsimultaneously. Then a multi-objective decision-making approach based on TOPSIS isused to select the final rolling batch from the Pareto-optimal solutions.In this dissertation, a multi-objective load distribution model that takes into accountthe rolling force margin balance, roll wear ratio and strip shape control, is presented. Then,a local search based multi-objective particle swarm optimization algorithm (LS-MOPSO)is used to solve this model. It introduces a local search strategy into the multi-objectiveoptimization, and adapts the mathematical programming method to approach thePareto-optimal solutions quickly. Moreover, it also employs the MOSIA framework basedon the Maximin fitness function to preserve elitism and diversity. Meanwhile, a Gaussianmutation operator is introduced to avoid LS-MOPSO premature convergence. Finally, anefficient constraint handling method is proposed to handle the constraints in the loaddistribution model. The simulation results based on practical production data indicate thatLS-MOPSO can not only achieve a better performance in comparison with the empiricalsolution, but also find the conflicting relationship between different objectives, whichreflects the advantage of multi-objective load distribution optimization and theeffectiveness of LS-MOPSO.