基于自组织递归模糊神经网络的PM2.5浓度预测

针对PM2.5浓度非线性动态变化的特点，提出了一种自组织递归模糊神经网络(self-organizing recurrent fuzzy neural network，SORFNN)方法预测PM2.5小时浓度。首先，通过分析影响PM2.5浓度的多种因素，利用主成分分析法(principal component analysis，PCA)筛选出与PM2.5浓度相关性较强的特征变量作为神经网络的输入变量。然后，根据ε准则和偏最小二乘算法(partial least squares，PLS)进行规则化层神经元的增删，实现递归模糊神经网络结构的自动调整，并采用学习率自适应的梯度下降算法调整模型中心、宽度和权值等参数，建立PM2.5预测模型。最后，利用典型非线性系统辨识和实际PM2.5浓度预测实验进行验证。实验结果表明，所设计的自组织递归模糊神经网络结构精简且预测精度高，较好地满足了PM2.5实时预测的要求。
To address the nonlinear dynamic variation in the concentration of fine particulate matter (PM2.5), in this paper, we propose a novel self-organizing recurrent fuzzy neural network (SORFNN) for predicting the hourly PM2.5 concentration. First, we analyzed the factors affecting PM2.5 concentration by principal component analysis to identify the characteristic variables and used them as input variables in the neural network. Next, we added or deleted a nerve cell to the regularized layer, based on the ε criterion and partial least squares algorithm, to automatically adjust the recurrent fuzzy neural network. In addition, we applied the adaptive gradient descent algorithm to adjust parameters such as the centers, widths and weights to establish a PM2.5 model. Lastly, to verify the results, we conducted experiments in typical nonlinear system identification and actual PM2.5 concentration prediction. The experimental results show that the proposed SORFNN is compact in structure, has high prediction accuracy, and can satisfy the real-time prediction requirements of PM2.5 concentration