Intelligent Algorithms For Long-term Forecasting And Operation Of Hydropower System

Forecast and operation is core of management and scheduling of hydropower system. Nearly 10 years, it is one of front issues in this research area to apply intelligent algorithms to the forecast and optimization modeling analysis. Intelligent algorithms expand the traditional calculation mode, not requiring the establishment of a precise model. They are well suited for those very hard to be solved effectively and even unable to be solved using traditional method, for the difficulty of establishing an effective formalization model. The intelligent algorithms and their hybrid models provide a new idea and approach to perform the long-term forecast and operation of hydropower station(s). Based on the project background of hydropower station(s) in Yunnan power grid's Lanchangjiang River basin and Northeast power grid's Hunjiang River basin, long-tem forecast and operation modeling methods are studied in-depth using intelligent algorithms such as neural network, fuzzy system, support vector machine, SCE-UA algorithm, ant colony optimization algorithm. And a decision support system for long-term forecast and operation of hydropower station(s) is developed to apply the intelligent models to relevant project. The main contents are as follow:(1) A long-term discharge forecast model is developed using an adaptive-network-based fuzzy inference system (ANFIS) based on subtractive clustering. Due to the grid partition's curse of dimensionality, subtractive clustering is used to divide fuzzy space and generate initial fuzzy rules from given input data. Secondly, ANFIS tunes parameters by a hybrid learning algorithm combines the backpropagation gradient descent and the least squares estimate method. Utilizing the long-term observations of discharges of monthly river flow discharges during 1953-2003 in Manwan hydropower, the model performance is compared with the ARMA model and ANN model. It has been demonstrated that the ANFIS model based on subtractive clustering is an effective and accurate method for long-term discharge prediction.(2) Support vector machine (SVM) model, which parameters are estimated by SCE-UA algorithm, is presented as a promising method for long-term discharge prediction. The Radial basis function is introduced in the establishment of the model describing the run-off hydrograph, and the SCE-UA algorithm is applied to identify the parameters of SVM. The exponential transformation is used to help quickly and precisely search the optimal parameters. The SVM prediction model is tested by the long-term observations of monthly...