Severe geomagnetic storms caused by the solar wind disturbances have harmful influences on the operation of modern equipment and systems. The modeling and forecasting of AE index are extremely useful to understand the geomagnetic substorms. This study presents a novel cloud‐nonlinear autoregressive with exogenous input (NARX) model to predict AE index 1 hr ahead. The cloud‐NARX model provides AE index forecasting results, with a correlation coefficient of 0.87 on the data of whole year 2015. The benchmarks on the data of the two interested periods of 17–21 March 2015 and 22–26 June 2015 are presented. The presented model uses uncertainty “cloud” model and cloud transformation to quantify the uncertainty throughout the structure detection, parameter estimation, and model prediction. The new predicted band can be generated to forecast AE index with confidence interval. The proposed method provides a new way to evaluate the model based on uncertainty analysis, revealing the reliability of model, and visualize the bias of model prediction.
Air pollution causes significant negative impacts on climate, environment and human health. Monitoring and forecasting the PM2.5, one of the most dangerous pollutant, are crucial. However, the strong prediction uncertainty in peak periods can potentially increase the prediction error and decrease the model reliability. To overcome this problem, prediction intervals are needed to quantify the uncertainty and provide information of the confidence in the prediction. In this article, an improved cloud-NARX estimation algorithm is developed to quantify the uncertainty and produce prediction intervals. The proposed method integrates a new recursive estimation procedure and two new criteria, which significantly improve the training speed and prediction interval accuracy. The proposed method is applied to predict PM2.5 for one hour ahead. From our results, the proposed method achieves higher accuracies of both average predictions and prediction intervals than other methods. This study provides a novel framework for quantifying the uncertainty of time series prediction, and to improve the model robustness.