Precipitation Modeling and Prediction Based on Fuzzy-Control Multi-cellular Gene Expression Programming and Wavelet Transform

Abstract

Accurate and timely precipitation prediction is very important to development and management of regional water resources, flood disaster prevention/control and people’s daily activities and production plans. However, the prediction accuracy is greatly affected by nonlinear and non-stationary features of precipitation data and noise. Many researches show that Multi-cellular Gene Expression Programming (MC-GEP) algorithm has strong function mining ability. This paper designed a Fuzzy-control Multi-cellular Gene Expression Programming algorithm (FMC-GEP) based on Multi-cellular Gene Expression Programming, which used fuzzy control theory to dynamically adjust the probability of genetic manipulation. Then we coupled Fuzzy-control Multi-cellular Gene Expression Programming algorithm with wavelet transform to develop a novel algorithm for precipitation prediction (abbreviated as WT_FMC-GEP). To verify the prediction performance, we conducted experiments using RMSE and MAE as evaluation metrics on the real precipitation data sets in three regions of different continents where climate vary widely. The results show that the WT_FMC-GEP algorithm outperforms other existing prediction algorithms including BP neural network, support vector regression and Gene Expression Programming. It also outperforms such algorithm based on MC-GEP and wavelet transform, thus having a good application prospect.