基於RSCMAC的澎湖中屯風力發電機組功率預測

Power Forecast of Penghu Zhongtun Wind turbine group based on RSCMAC

王振宇(Zhen-yu Wang)

指導教授 ： 江青瓚

健行科技大學/電資學院/電機工程系碩士班/碩士(2021年)

摘要

本論文之目的為採用小腦模型控制器結合風力發電機發電與氣象監測站之資料，預測中屯風機群的發電狀況。本論文風機地點選用中屯風機公園中8支風機來進行三種不同資料，結合回歸型最簡類化型小腦模型控制器(Recurrent simple addressing structure for Cerebellar Controller with General Basis Function ,RSCMAC)為基礎之預測模型建立。一為僅採用風機群監測資料的模型訓練與測試，二為氣象監測站的模型訓練與測試，三為風機群與氣象監測站的模型訓練與測試。在訓練時會將相關資料分散存於RSCMAC之各個記憶體中，之後於預測時再整合輸出。在學習時輸出值會與目標值比對，將誤差值回傳，進行各記憶體之修正，重複循環學習直到學習完成，在單使用風機群資料結合RSCMAC訓練後與目標值相比之絕對誤差在4%以內，而使用風機群使用經過訓練之模型測試後與目標值相比之絕對誤差在5%以內，使用氣象監測站結合RSCMAC訓練後與目標值相比之絕對誤差在7.5%以內，而使用氣象監測站使用經過訓練之模型測試後與目標值相比之絕對誤差在9.5%以內，而風機群與氣象監測站結合RSCMAC訓練後與目標值相比之絕對誤差在4%以內，而風機群與氣象監測站使用經過訓練之模型測試後與目標值相比之絕對誤差在10.5%以內。
以本計劃所得到的結果來看，風機群的方法可以用在更加廣域的風場上，這更可以大幅提升，電網上的電源調度，也間接的提升風力發電的穩定性與發展潛力。

關鍵字

預測 ； 風力發電機 ； 回歸型最簡類化型小腦模型控制器

並列摘要

The purpose of this paper is to combine the cerebellum model controller with the wind turbine power generation and weather monitoring station data to predict the power generation status of the Zhongtun wind turbine group. In this paper, eight fans in Zhongtun Fan Park are selected as the fan location in this paper to establish a prediction model based on three different data combined with Recurrent simple addressing structure for Cerebellar Controller with General Basis Function ,(RSCMAC). One is model training and testing using only wind turbine monitoring data, the other is model training and testing at weather monitoring stations, and the third is model training and testing wind turbines and meteorological monitoring stations. During training, relevant data will be scattered in each memory of RSCMA, and then output will be integrated during prediction. During learning, the output value will be compared with the target value, the error value will be returned, and each memory will be corrected, and the cyclic learning will be repeated until the learning is completed. The absolute value compared with the target value after using the fan group data combined with RSCMAC training The error is within 4%, and the absolute error compared with the target value after using the trained model test with the wind turbine group is within 5%, and the absolute error compared with the target value after using the weather monitoring station combined with RSCMAC training is within 7.5% , And the absolute error compared with the target value after using the weather monitoring station and the trained model test is within 9.5%, and the absolute error compared with the target value after the wind turbine group and the weather monitoring station combined with RSCMAC training is within 4%, The absolute error of the wind turbine group and the weather monitoring station compared with the target value after testing with the trained model is within 10.5%.
Judging from the results obtained in this plan, the wind turbine method can be used in a wider area of wind farms, which can greatly improve the power dispatch on the grid, and indirectly improve the stability and development potential of wind power.

並列關鍵字

Prediction ； Wind Turbine ； RSCMAC

參考文獻

[1] C. Tian, D. Lei, S. Gao and X. Liao, “A Study of Grey Theory Used in Prediction of Annual Wind Power Generation,” Electric Information and Control Engineering (ICEICE), 2011 International Conference, 15-17 April, 2011. Pages 1952 - 1955.

[2] Yuan Li, Qian Sun, Brad Lehman, Siyuan Lu, Hendrik F. Hamann, Joseph Simmons, Jon Black, “A Machine-Learning Approach for Regional Photovoltaic Power Forecasting,” IEEE Trans. Power Forecasting, 2016 IEEE Power and Energy Society General Meeting (PESGM).

[3] Haixiang Zang, Zhi Liang, Mian Guo, Zeyu Qian, Zhinong Wei, Guoqiang Sun, “Short-term Wind Speed Forecasting Based on anEEMD-CAPSO-RVM Model,” IEEE Trans. forecasts, 2016 IEEE PES Asia-Pacific Power and Energy Engineering Conference (APPEEC).

[4] Yi Zhang, He xu Sun, Ying jun Guo, Zhi ping Lu, “Improved grey model with rolling method for wind power prediction,” IEEE Trans. forecasts, 2016 35th Chinese Control Conference (CCC).

[5] Ming-Jian Cui, De-Ping Ke, Yuan-Zhang Sun, Di Gan, Jie Zhang, Bri-Mathias Hodge, “A scenario generation method for wind power ramp events forecasting,” IEEE Trans. forecasts, 2015 IEEE Power & Energy Society General Meeting.