基於機器學習的台灣風電場發電功率預測研究

Abstract

全球不可再生能源短缺，風能作為一種環境友好型能源，成為替代化石能源的重要選擇。但風具有間歇性、波動性、隨機性等特點。風力發電給電力系統的穩定性帶來了一定的挑戰。而對風場的風功率進行預測是解決這一問題的重要途徑。風速預測是風功率預測的重要組成部分。 在此背景下，本文針對風場的短期功率預測，重點完成以下幾方面的工作： 根據歷史記錄資料，使用時間序列法建立模型，根據前3小時的資料預測下一小時的風速，並建立風速與功率曲線。 針對Back Propagation（缩写 BP）神經網路，通過對比相同輸入的不同結構之神經網路，確定誤差最小的神經網路架構；對風機的輸入資料進行預處理，利用神經網路確定對功率輸出相關度最大的特徵。 對於支援向量機，收斂速度快,學習能力強,泛化能力好,即使風速劇烈變化時,也能有效的預測序列的變化趨勢。 最後建立了組合預測模型，有效的提升了預測精度，減小了預測誤差，降低了不穩定性。

Because of the shortage of non-renewable energy, wind energy, which an environmentally friendly energy, has become an important alternative to fossil fuels. But the wind has the characteristics of intermittence, volatility and randomness. Wind power has brought some challenges to the stability of power system. The prediction of wind power is an important way to solve this problem. Wind speed forecasting is an important part of wind power prediction. Under this background, this paper focuses on the following aspects of the short-term wind power prediction: Use the time series method to build a model based on historical data, and predicts wind speed with three-hour-ahead wind speed, and establish a wind speed and power curve. For the BP neural network, the neural network architecture with the smallest error is determined by comparing the neural networks of different structures with the same input; the input data of the wind turbine is preprocessed, and the neural network is used to determine the variable with the highest correlation with the power output. For the support vector machine(SVM), the convergence speed is fast, the learning ability is strong, and the generalization ability is good. Even when the wind speed changes drastically, the trend of the sequence can be effectively predicted. Finally, a combined prediction model is established, which effectively improves the prediction accuracy, and reduces the prediction error, and reduces the instability.