- 學位論文
基於粒子群演算法之風速及發電量預測
A study of wind speed and power generation prediction based on particle swarm opt
摘要
風速的不穩定變化將迫使風力發電機組發電量輸出產出不穩定。若可準確預測風速及風機發電量輸出，便可藉其評估風力發電之經濟效益。傳統預測方法者眾，為提昇預測準確率，本文應用線性時序模組、雙指數平滑法及廣義回歸神經網路三種預測法，加上上述三種方法分別結合K-means群聚分類、最鄰近群聚分類及相關性群聚分類而成之九種預測法，共十二種預測模組於風速及發電量預測。並利用粒子群演算法找出一組最適合此十二種預測法之權重集合，使各預測模組在預測中充分發揮效用。 本研究首先利用台北機場氣象站之風速資料，並以風力發電模擬平台以實驗方式取得風力發電機之發電量輸出。其次，利用十二種傳統預測模組的預測結果，並利用粒子群演算法找出一組最合適此十二種預測模組之權重集合。最後，以合併預測法合併十二種預測結果，並加上粒子群演算法求得的最佳權重組合進行預測。實驗結果顯示，利用粒子群演算法求得合併預測法的最佳權重無論在發電量或風速預測上，預測準確率明顯高於其餘十二種傳統預測法。
並列摘要
The output power generation of a wind turbine will fluctuate according to the unstable wind speed. If we can predict the wind power generation and its speed precisely, the economic benefits and the cost of a wind turbine can be evaluated. Therefore, to improve the prediction accuracy, this paper employs three prediction methods, including linear time-series based model, double exponential smoothing and general regression neural network and their combinations with K-means clustering, nearest neighbor clustering and correlation clustering individually to predict the wind speed and power generation. Then, particle swarm optimization is used to find out the optimal weight set of the twelve prediction methods which help them to perform well. First, this paper executes an experiment using a simulation platform and the wind speed data from the weather station at Taipei airport to obtain the actual output power generation of the wind turbine. Second, particle swarm optimization is used to find out the optimal weight set of the twelve various consequences by twelve prediction methods. Finally, combination forecasting combines the twelve consequences together with the optimal weight set for prediction. The simulation result indicates that the combination forecasting with the optimal weight set has superior accuracy to other prediction methods, whether on prediction of power generation or wind speed.