類神經網路應用於太陽能光電系統發電量預測研究

The Power Generation Forecast of Photovoltaic System by Artificial Neural Network

温修培(Wen, Xiu-Pei) ； 盧信忠(Lu, Hsin-Chung)

《弘光學報》 87期 (2021/03) Pp. 17-31

https://doi.org/10.6615/HAR.202103_(87).0002

摘要

因應石化燃料的縮減與其造成的健康危害，再生能源已逐漸受到重視，目前太陽能發電量在全球呈現正成長的趨勢，故如能預測其發電量，則對太陽能發電量的評估具有相當大的價值。在本研究中擬利用多重線性回歸（Multiple linear regression：MLR）與類神經網路中多重認知器（Multilayer Perceptron：MLP）網路模式進行太陽光電發電量的模擬，輸入資料為氣象資料（日射量、日照小時、溫度、濕度、降雨量、雲遮量等）與環保署PM_(10)，PM_(2.5)資料，並以兩種統計指標評（R^2，RMSE）評估模式的優劣，太陽能發電量選取台電公司的澎湖、台中與興達發電廠的發電資料，選取的時間為2017～2018的日發電量，以2017年的資料當為訓練組，並將2017年資料取得的模式用以預測2018年的發電量。結果顯示類神經網路MLP模式於澎湖、台中與興達發電廠的預測結果皆優於MLR模式，其2018年預測結果的R^2皆大於0.7以上，顯示MLP可成功預測發電量與氣象條件的關係，預測結果可當電廠設置及成本收益預估的參考。

關鍵字

多重線性回歸 ； 類神經網路 ； 多重認知器 ； 太陽能

並列摘要

The renewable power was received a highly attention recently because of the reduction use of fossil fuel and their harm to health. Additionally, the current solar power generation shows a positive growth trend around the world. Therefore, if the power generation can be predicted, the predicted results will be a valued reference for the setup of solar power generation station. In this study, the Multilayer Perceptron (MLP) network mode and the Multiple linear regression (MLR) were used to simulate the solar photovoltaic power generation. The input data is the meteorological data (radiation, sunshine hours, temperature, humidity, rainfall, cloud cover, etc.) and the PM_(10), PM_(2.5) data from monitoring stations of Environmental Protection Agency. Two statistical indicators (R^2, RMSE) were taken to evaluate the fitness of two predicted methods. The daily power generation from 2017 to 2018 were selected from Taipower's Penghu, Taichung and Xingda Power Plant. The 2017 data was used as the training data, and the trained model obtained from 2017 data was used to predict the power generation in 2018. The results show that the predicted results of the neural network (MLP model) in Penghu, Taichung and Xingda power plants are better than the MLR model. The R^2 values are all greater than 0.7, which shows that MLP can successfully predict the relationship between power generation and weather conditions. The forecast result can be used as a reference for power plant settings and cost-benefit estimation.

並列關鍵字

Multiple linear regression ； Artificial neural networks ； Multilayer Perceptron ； Solar photovoltaic power

參考文獻

J. H. Yousif, H. A. Kazem, and J. Boland, “Predictive models for photovoltaic electricity production in hot weather conditions,＂ Energy, vol. 10, pp. 971-889, 2017. ; https://doi.org/10.3390/en10070971.

EnergyTrend, 太陽能話題持續延燒，2025年全球發電量預計將達到969GW, 2017, https://www.energytrend.com.tw/news/20171225-14308712.html.

太陽能 2 年推動計畫，2016，http://mrpv.org.tw/page/f80b9aa7.

M. W. Gardner, and S. R. Dorling,“Meteorological adjusted trends in UK daily maximum surface ozone concentrations,＂ Atmospheric Environment, vol. 34, pp. 171-176, 2000.

M. Kolehmainen, H. Martikainen, and J. Ruuskanen, “Neural networks and periodic components used in air quality forecasting,＂ Atmospheric Environment, vol. 35, pp. 815-825, 2001.