The Research On Medium-long Term Hydrological Forecast At Yangtze Three Gorges

Medium-long term hydrological forecast plays an important role in flood control and drought relief, reservoir operation, water resources planning and management, etc. At present medium-long term hydrological forecast is a field falling behind relatively in hydrology, still in the stage being exploring and developing. The Yangtze Three Gorges is a key project harnessing and developing the Yangtze River, utilizing the water resource of Yangtze River comprehensively. Demanding for satisfying flood protection, electric power generation, shipping, etc, the reservoir operation must put optimization controller into practice. The discharge forecast at dam site of Three Gorges rate is the main basis of optimizing regulation. In the current, the medium-long term hydrological forecast at Yangtze Three Gorges exist the problem that the accuracy is low, and have difficult to guide to production practice effectively.In this paper, the predictors such as rainfall and atmospheric circulation which affect the hydrological phenomenon in earlier stage at Yichang station which is the represented station of Yangtze Three Gorges are selected by combining cause-and-effect and statistical analysis with fuzzy analysis. Based on this, the medium-long term hydrological forecasting model of fuzzy pattern recognition and artificial neural network (ANN) are set up to forecast the annual maximum peak discharge and monthly discharge. The main conclusions of the research are shown as follows:(1) The present paper poses the predictors selection method of fuzzy pattern recognition based on the theory and model of fuzzy pattern recognition, proceed from single influencing factor, taken max r as the objective function, aimed at the optimal relativity to the features of classifications and objects combined by many influencing factors. Practice has shown that predictors selected in earlier stage are reasonable and the method is worth popularization and application.(2) When forecasting with the fuzzy pattern recognition model, takes the absolute value of relativity to eigenvalue of objects and predictors as the initial weights of predictors conform to real hydrological phenomenon.(3) LMBP algorithm shows low accuracy and high risk when single forecasting, but efficient and stabilization when training. Self-adaptive BP algorithm introduces momental factor, so the probability of concussion reduces, and has relatively high accuracy and rapid training speed. So we recommend LMBP algorithm when confirming network architecture and self-adaptive BP algorithm when forecasting.(4) When simulating error is about 6%-8%, the ANN model works out satisfactorily and can confirm training accuracy for applied based on it.(5) Predictors optimized from fuzzy method can apply to ANN model, and the forecast result is better than fuzzy pattern recognition model.(6)The models set up in the research are reasonable and showing good results. The achievements play great roles in operating and managing Yangtze Three Gorges hydro-power station.