高斯骨架差分进化算法与回声状态网络的结合应用

Combined application of Gaussian skeletal differential evolution algorithm and echo state network

针对回声状态网络(ESN)对于不同时间序列的学习上无法有效地确定储备池参数的问题,提出一种新型预测模型。利用改进的高斯骨架差分进化算法(DE)来优化回声状态网络。在DE算法中引入了变异策略选择因子,并将选择因子随个体共同参与进化,使每个个体执行当前最适合的变异策略。改善了原始DE算法进化过程中的盲目性,同时选择因子的动态自适应特性保持了骨架算法近似无参数的优点,最后为避免算法早熟加入停滞扰动策略改善算法的寻优性能。为验证模型的有效性,对Mackey-Glass时间序列、赣州月平均气温数据集进行仿真实验。由实验结果可知,该模型可以提高时间序列的预测精度,且具有良好的泛化能力及实际应用价值。

A new prediction model was proposed to address the problem that the Echo state networks(ESN)cannot identify reservoir parameters effectively when learning different time series,using modified Gaussian Skeleton Differential Evolution(DE)algorithm to optimize the ESN.The mutation strategy selection factor was introduced in it,and the selection factor was involved in the evolution with individuals,so each individual executed the best mutation strategy.The blindness in the evolution of the original DE algorithm was improved and the dynamic adaptive characteristics of the selection factor kept the advantage approximately parameter-free as the skeleton algorithm.Finally,to avoid the algorithm Premature,stagnation disturbance strategy was added to improve the optimization performance of the algorithm.To verify the effectiveness of the method,simulation experiments were carried out on the Mackey-Glass time series and the data sets of Ganzhou monthly average temperature.The experimental results showed that the model proposed in this paper can improve the prediction accuracy of time series with good generalization ability and practicality.