DE算法的改进及其在系统辨识中的应用

对于一类高维、非光滑及非线性的约束优化问题,传统的搜索方法不能很好地求得全局最优解,而DE算法可以处理这类问题.为了提高DE算法收敛到全局最优的概率和精度,在基本DE算法的基础上,运用变步长梯度法和记忆库,得到改进的DE算法,并将改进的DE算法应用于实际水槽的模型参数辨识.经过测试对象、采集数据、选择模型结构、辨识参数和验证模型,结果表明,改进的DE算法使辨识系统参数收敛到全局最优的能力增强,收敛概率和精度得到提高,模型偏差平方和更小.

An improved DE algorithm and the application in system identification

For a class of high-dimensional,non-smooth,nonlinear,constrained optimization problems,traditional search method is unable to find the global optimum solutions,whereas the DE algorithm can solve this problem.To improve convergence and accuracy of the DE algorithm,the basic DE algorithm is improved through combing variable step-size gradient method and memory bank.The improved algorithm is applied to identify the parameters in the tank model.After testing object,collecting data,choosing model structure,identifying parameters,and validating model,the testing result shows that the improved DE algorithm is able to improve the convergence probability and precision in the global optimum.The error sum of square is smaller too.