Study On Runoff Prediction In Jingou River Basin Based On Artificial Neural Network

As a large-scale complex system, water resource system which is comprehensively influenced by climate, physical geography, social development, human survival needs, and many other factors has complex features such as multiple time scale, randomness, mutagenicity, etc. Meanwhile, water resource is one of the most basic elements which supports human life, economic growth and its sustainable development. The change of runoff factors lead the change of the whole water resource. Therefore, the establishment of runoff prediction model could coordinate various kinds of water contradiction during comprehensive utilization of water resources at the maximum limit, and ensure the sustainable development of economy, society, ecology and environment.This thesis takes the monthly runoff of47years during the year of1957to2003monitoried by Bajiahu hydrologic station which is located in the middle reaches of Jingou river basin as the research object. The earlier39-year monthly runoff datum from1957to1995are taken as the training samples, and the latter8-year monthly runoff datum from1996to2003are taken as the test samples. The neural network prediction model is established according to the training and test samples.First of all, routine statistical method is used to analyzed the monthly runoff time series of Jingou river basin to get to know the change trend and variation period preliminarily. Then, wavelet analysis which has a good ability to characterize local features of signal in both time and scale (frequency) fields is used to decompose the original time series to low and high frequency to know the change law furthermore. At last, the prediction of runoff time series is taken based on the change law.The artificial neural network method is used to forecast the runoff time series. The programming is taken in the MATLAB environment. The most fast grads descent methodology of BP net work is adopted to forecast each detail signal decomposed by wavelet transform. Then the prediction results are remodeled to get the final prediction results. And the predictive value errors of different months are analyzed. The results showed that:this neural network prediction model has a good prediction effect to the results from june to october, the errors are all below4%, which can play the guiding practice. But the prediction effect of the other months is not ideal, the maximum error reached53.1%.The LM-BP and RBF neural network prediction models are establish to forecast monthly runoff of the other months aiming at the condition of the too large error. The prediction results distribution to each month according to monthly runoff per year to get the monthly runoff prediction results. The results showed that: The prediction effect of LM-BP network model is better than that of RBF network model according to the results from november to may next year, the average errors are19.4%and22.9%respectively. Both the LM-BP and RBF network models are better than the BP network model.According to the error analysis of this prediction model, it is found that the establishment of different neural network prediction models for different months has good reference function in both theory and practice in practical application of monthly runoff prediction in Jingou river basin.