改进的快速跟踪回声状态网络及PM2.5预测

针对递归最小二乘回声状态网络在噪声环境中预测精度不高的问题,提出了一种改进的快速跟踪回声状态网络。首先在递归最小二乘回声状态网络结构的基础上,将自适应调节的可变遗忘因子加入其代价函数中,用改进的递归最小二乘法对网络输出权值进行训练,得到快速跟踪回声状态网络;然后利用经典Lorenz混沌系统验证快速跟踪回声状态网络的有效性;最后利用灰关联法分析各相关变量与PM2.5的关联度,建立PM2.5浓度值辅助变量集合,将辅助变量集合输入到快速跟踪回声状态网络进行PM2.5浓度值预测。实验表明,与传统回声状态网络、递归最小二乘回声状态网络预测效果相比,快速跟踪回声状态网络的预测方法精度佳,抗噪声能力强。

Improved Fast Tracking Echo State Network and PM2.5 Prediction

Aiming at the problem that the recursive least squares echo state network has low prediction accuracy in noisy environments, an improved fast tracking echo state network was proposed. Firstly, the variable forgetting factor which can be adaptively adjusted was added into the cost function of network, and the output weights of the network were trained by the improved recursive least squares method to obtain a fast tracking echo state network. Then, the effectiveness of the proposed network was verified by classical Lorenz chaotic system. Finally, the grey correlation method was used to analyze the correlation between the relevant variables and PM2.5, and the auxiliary variable set of PM2.5 concentration value was established. The auxiliary variable set was input into the fast tracking echo state network to predict the PM2.5 concentration value. Experimental results show that compared with the traditional echo state network and the recursive least squares echo state network, the improved state echo network has better prediction accuracy and stronger anti-noise ability.