An Extreme Learning Machine and Gene Expression Programming-Based Hybrid Model for Daily Precipitation Prediction

Accurate daily precipitation prediction is crucially important. However, it is difficult to predict the precipitation accurately due to inherently complex meteorological factors and dynamic behavior of weather. Recently, considerable attention has been devoted in soft computing-based prediction approaches. This work presents a scheme to reduce the risk of Extreme Learning Machine (ELM) modeling error using Gene Expression Programming (GEP) to improve the prediction performance, and develops an ELM-GEP hybrid model for regional daily quantitative precipitation prediction. In this study, firstly, we use ELM for modeling the data sample of daily rainfall to construct a main model. Secondly, we use GEP for modeling the error of the main model as a compensation of the main model to reduce the prediction error. We conducted eight experiments of two different types of daily precipitation prediction problems using five metrics to evaluate our proposed model performance. Experimental results show that our model is comparable or even superior to five state-of-the-art models with high reliability in terms of all metrics on all datasets. It indicates that the proposed method is a promising alternative prediction tool for higher accuracy and credibility of regional daily precipitation prediction.