Water Level Prediction Based on Polyelement Correlation Analysis and Improved BP Neural Network

To discover the useful information in a large amount of hydrological data set becomes a big challenge in hydrological data mining. Water level Prediction has great significance for the state flood control. However, current approaches has low accuracy and bad adaptability. In this paper, we put forward a new forecasting approach based on polyelem-ent correlation analysis and improved BP neural network. First, we used correlation analysis technique to obtain the most relatively influential factors of water level, rainfall and temperature. These two factors and water level were put together to train the improved double-hidden neural network model, then we used LMDP optimization algorithm to optimize the model. The data of experiment is the daily water level, rainfall and temperature data from Zhari Namco observation station. The experimental results based on the five evaluation criteria demonstrate that the proposed method has high accuracy, low error and be superior to the traditional prediction model.