Multi-objective Optimization Algorithms Base On Differential Evolution And Its Application In Aluminum Hot Rolling Schedule

With the rapid development of domestic aluminum industry, the aluminum alloy is widely used in the paper, automobile, shipbuilding, high-speed train, electronic industry, etc. The overall level of domestic hot rolling equipment can’t catch the world’s advanced level, especially for the high-end products, which are mostly rely on import. This dissertation conducts a practical research for “1+4” aluminum hot tandem rolling, aims at contributing reasonable rolling schedules to improve the products quality.The precise model of rolling processing and the efficient method of multi-objective optimization are two major basis of rolling schedule. The model of rolling process is the mathematical expression of rolling technology, which highly effect the forecast precision. As a mathematical tool, the multi-objective optimization algorithm play an important role in the modeling and scheduling.Aim to solve the large-scale multi-objective optimization problem in practice, a multi-objective differential evolution algorithm(DE) based on the physical programming have been proposed. Physical programming maps the objective function to preference function, quantifies the designer’s preference, and make the design process naturally. DE is a newly-develop evolutionary calculation technology, which is suitable for the problem can’t be solved by the traditional mathematical method. Combining the advantages of two methods, a multi-objective optimization algorithm aiming to apply in the practice is proposed. The simulation shows the result of proposed algorithm converge to the special position of Pareto front based the preference.Aim to solve the constrained multi-objective optimization problem, this dissertation puts forward a new selection strategy based on environment pareto dominated, which provides a judgment, especially between the feasible solution and non-feasible solution. On this basis, the improved mutation strategy, the scalar factor adaptive strategy, the crossover probability adaptive strategy compose a differential evolution algorithm based on environment pareto dominated selection strategy in constrained multi-objective optimization problem. The test functions of classical CTP have been run 30 times by the proposed algorithm and the other 3 algorithm, respectively. The simulation shows that the proposed algorithm processes higher stability and solving accuracy.Focus on solving the low forecast accuracy of the rolling model, a new model of friction coefficient is proposed based on the research and analysis of a large amount of field data. On the basis of the former model, the new model considers the influence of the rolling temperature and improves the forecast accuracy greatly. To make a further improvement on the accuracy of forecast model, a dynamic self-learning method based on the data similarity is applied. The self-learning coefficient updates based on specification of rolling products and information of the rolling mills before and after rolling. The simulation result shows that this method improve the forecast accuracy of the model effectively.These proposed algorithms have been applied to optimize aluminum hot rolling schedule, respectively. The simulation shows the proposed algorithms improve the performance compared with the original rolling schedule. To verify the algorithm effectiveness, the experiments have been conduct in the field. The result shows the proposed methods meet the requirement in the field basically and have instructive significance to practical production.