短時(shí)風(fēng)速預(yù)測的智能方法研究
發(fā)布時(shí)間:2021-04-12 12:17
經(jīng)濟(jì)的發(fā)展使得人們的生活質(zhì)量越來越高,但隨之而來的環(huán)境問題也愈來愈嚴(yán)重,譬如大氣污染、海洋污染、森林火災(zāi)、病毒肆虐、石油資源的日漸匱乏等問題,因此,充分利用風(fēng)能等可再生的清潔能源是極其重要的。精確的風(fēng)速預(yù)測能夠幫助我們有效地利用風(fēng)能資源,但由于風(fēng)速的隨機(jī)性以及非穩(wěn)定性使得風(fēng)速的預(yù)測極具挑戰(zhàn)性,故需要尋找合適的預(yù)測模型才能精確地預(yù)測風(fēng)速。針對風(fēng)速預(yù)測,本文提出了一種基于權(quán)重的組合模型,文章主要從三個(gè)方面對風(fēng)速的預(yù)測展開了探討:(1)數(shù)據(jù)降噪方法的選擇;(2)單一模型的預(yù)測研究;(3)組合模型的預(yù)測研究。由于原始風(fēng)速數(shù)據(jù)具有非平穩(wěn)性等特征,直接使用原始風(fēng)速數(shù)據(jù)進(jìn)行預(yù)測研究會使得預(yù)測效果不理想,在預(yù)測之前需要對原始數(shù)據(jù)進(jìn)行數(shù)據(jù)重構(gòu)等降噪處理,本文選擇了經(jīng)驗(yàn)?zāi)B(tài)分解(EMD)、集成的經(jīng)驗(yàn)?zāi)B(tài)分解(EEMD)、小波變換(WD)、高斯濾波器和均值濾波器五種方法分別對六個(gè)站點(diǎn)的原始風(fēng)速數(shù)據(jù)進(jìn)行降噪處理,將處理后的數(shù)據(jù)基于長短時(shí)記憶神經(jīng)網(wǎng)絡(luò)LSTM分別做預(yù)測實(shí)驗(yàn),得出EMD和WD方法預(yù)測效果明顯優(yōu)于其他幾種。單一模型的預(yù)測選擇支持向量回歸SVR、多層感知機(jī)MLP、循環(huán)神經(jīng)網(wǎng)絡(luò)RNN、長短時(shí)記憶神經(jīng)網(wǎng)絡(luò)...
【文章來源】:蘭州大學(xué)甘肅省 211工程院校 985工程院校 教育部直屬院校
【文章頁數(shù)】:57 頁
【學(xué)位級別】:碩士
【部分圖文】:
六個(gè)站點(diǎn)原始風(fēng)速數(shù)據(jù)來源
蘭州大學(xué)碩士學(xué)位論文短時(shí)風(fēng)速預(yù)測的智能方法研究21表3.1不同降噪方法的預(yù)測結(jié)果站點(diǎn)模型MAE(m/s)RMSE(m/s)MAPE(%)SMAPE(%)站點(diǎn)1EMD_LSTM0.36640.44676.73306.9387EEMD_LSTM0.35800.46266.48996.7140WD_LSTM0.26000.337974.79804.8445MEAN_LSTM0.39660.52377.35617.3821GAUSS_LSTM0.61360.800611.316611.3366ORGINAL_LSTM0.37210.48006.97816.8643站點(diǎn)3EMD_LSTM0.27340.30374.63934.7249EEMD_LSTM0.38560.34155.41325.3472WD_LSTM0.29200.29044.44524.5300MEAN_LSTM0.27520.26124.05683.9482GAUSS_LSTM0.70660.680410.457610.4477ORGINAL_LSTM0.38670.45347.13826.9047站點(diǎn)5EMD_LSTM0.33380.42126.41056.1204EEMD_LSTM0.32890.40386.24956.035WD_LSTM0.21820.27564.11194.0885MEAN_LSTM0.37270.47196.84146.9616GAUSS_LSTM0.56200.690110.795310.2028ORGINAL_LSTM0.34990.45136.61376.3943站點(diǎn)7EMD_LSTM0.30060.36844.58934.6457EEMD_LSTM0.33690.40965.10825.1548WD_LSTM0.24700.31683.81113.8064MEAN_LSTM0.44050.57346.9736.6957GAUSS_LSTM0.59790.74469.3299.0072OTGINAL_LSTM0.37240.48235.69285.7135站點(diǎn)8EMD_LSTM0.27250.36494.88024.8238EEMD_LSTM0.33420.43245.87515.9058WD_LSTM0.24600.32874.37454.3341MEAN_LSTM0.43340.57977.87227.4964GAUSS_LSTM0.62340.807411.299910.6695ORGINAL_LSTM0.37850.50786.66126.5885站點(diǎn)9EMD_LSTM0.30540.38037.00796.6899EEMD_LSTM0.24450.31725.44625.4220WD_LSTM0.22270.28834.97594.9103MEAN_LSTM0.35200.43618.02937.7262GAUSS_LSTM0.39330.50258.83698.6773ORGINAL_LSTM0.31070.40376.93496.8361站點(diǎn)平均EMD_L
蘭州大學(xué)碩士學(xué)位論文短時(shí)風(fēng)速預(yù)測的智能方法研究25第四章單一模型預(yù)測本章為單一模型的預(yù)測,首先選擇EMD和WD兩種數(shù)據(jù)降噪方法對六個(gè)站點(diǎn)的數(shù)據(jù)分別做重構(gòu)處理,得到降噪后的數(shù)據(jù)。仿真實(shí)驗(yàn)分為EMD重構(gòu)數(shù)據(jù)和WD重構(gòu)數(shù)據(jù)兩組,選擇SVR、MLP、RNN、LSTM、GRU五種模型為單一模型,分別進(jìn)行實(shí)驗(yàn)分析。4.1基于EMD數(shù)據(jù)重構(gòu)的單模型風(fēng)速預(yù)測選擇經(jīng)驗(yàn)?zāi)B(tài)分解EMD為數(shù)據(jù)降噪方法,分別對六個(gè)站點(diǎn)原始風(fēng)速數(shù)據(jù)做降噪處理,將處理后的數(shù)據(jù)作為SVR、MLP、RNN、LSTM、GRU五種單一模型的輸入值,分別進(jìn)行風(fēng)速預(yù)測。圖4.1給出了站點(diǎn)3和站點(diǎn)8各單一模型基于EMD的預(yù)測趨勢圖。圖4.1部分站點(diǎn)基于EMD的單一模型風(fēng)速預(yù)測趨勢圖
【參考文獻(xiàn)】:
期刊論文
[1]基于PSO優(yōu)化LSSVM的短期風(fēng)速預(yù)測[J]. 孫斌,姚海濤. 電力系統(tǒng)保護(hù)與控制. 2012(05)
本文編號:3133282
【文章來源】:蘭州大學(xué)甘肅省 211工程院校 985工程院校 教育部直屬院校
【文章頁數(shù)】:57 頁
【學(xué)位級別】:碩士
【部分圖文】:
六個(gè)站點(diǎn)原始風(fēng)速數(shù)據(jù)來源
蘭州大學(xué)碩士學(xué)位論文短時(shí)風(fēng)速預(yù)測的智能方法研究21表3.1不同降噪方法的預(yù)測結(jié)果站點(diǎn)模型MAE(m/s)RMSE(m/s)MAPE(%)SMAPE(%)站點(diǎn)1EMD_LSTM0.36640.44676.73306.9387EEMD_LSTM0.35800.46266.48996.7140WD_LSTM0.26000.337974.79804.8445MEAN_LSTM0.39660.52377.35617.3821GAUSS_LSTM0.61360.800611.316611.3366ORGINAL_LSTM0.37210.48006.97816.8643站點(diǎn)3EMD_LSTM0.27340.30374.63934.7249EEMD_LSTM0.38560.34155.41325.3472WD_LSTM0.29200.29044.44524.5300MEAN_LSTM0.27520.26124.05683.9482GAUSS_LSTM0.70660.680410.457610.4477ORGINAL_LSTM0.38670.45347.13826.9047站點(diǎn)5EMD_LSTM0.33380.42126.41056.1204EEMD_LSTM0.32890.40386.24956.035WD_LSTM0.21820.27564.11194.0885MEAN_LSTM0.37270.47196.84146.9616GAUSS_LSTM0.56200.690110.795310.2028ORGINAL_LSTM0.34990.45136.61376.3943站點(diǎn)7EMD_LSTM0.30060.36844.58934.6457EEMD_LSTM0.33690.40965.10825.1548WD_LSTM0.24700.31683.81113.8064MEAN_LSTM0.44050.57346.9736.6957GAUSS_LSTM0.59790.74469.3299.0072OTGINAL_LSTM0.37240.48235.69285.7135站點(diǎn)8EMD_LSTM0.27250.36494.88024.8238EEMD_LSTM0.33420.43245.87515.9058WD_LSTM0.24600.32874.37454.3341MEAN_LSTM0.43340.57977.87227.4964GAUSS_LSTM0.62340.807411.299910.6695ORGINAL_LSTM0.37850.50786.66126.5885站點(diǎn)9EMD_LSTM0.30540.38037.00796.6899EEMD_LSTM0.24450.31725.44625.4220WD_LSTM0.22270.28834.97594.9103MEAN_LSTM0.35200.43618.02937.7262GAUSS_LSTM0.39330.50258.83698.6773ORGINAL_LSTM0.31070.40376.93496.8361站點(diǎn)平均EMD_L
蘭州大學(xué)碩士學(xué)位論文短時(shí)風(fēng)速預(yù)測的智能方法研究25第四章單一模型預(yù)測本章為單一模型的預(yù)測,首先選擇EMD和WD兩種數(shù)據(jù)降噪方法對六個(gè)站點(diǎn)的數(shù)據(jù)分別做重構(gòu)處理,得到降噪后的數(shù)據(jù)。仿真實(shí)驗(yàn)分為EMD重構(gòu)數(shù)據(jù)和WD重構(gòu)數(shù)據(jù)兩組,選擇SVR、MLP、RNN、LSTM、GRU五種模型為單一模型,分別進(jìn)行實(shí)驗(yàn)分析。4.1基于EMD數(shù)據(jù)重構(gòu)的單模型風(fēng)速預(yù)測選擇經(jīng)驗(yàn)?zāi)B(tài)分解EMD為數(shù)據(jù)降噪方法,分別對六個(gè)站點(diǎn)原始風(fēng)速數(shù)據(jù)做降噪處理,將處理后的數(shù)據(jù)作為SVR、MLP、RNN、LSTM、GRU五種單一模型的輸入值,分別進(jìn)行風(fēng)速預(yù)測。圖4.1給出了站點(diǎn)3和站點(diǎn)8各單一模型基于EMD的預(yù)測趨勢圖。圖4.1部分站點(diǎn)基于EMD的單一模型風(fēng)速預(yù)測趨勢圖
【參考文獻(xiàn)】:
期刊論文
[1]基于PSO優(yōu)化LSSVM的短期風(fēng)速預(yù)測[J]. 孫斌,姚海濤. 電力系統(tǒng)保護(hù)與控制. 2012(05)
本文編號:3133282
本文鏈接:http://sikaile.net/kejilunwen/dianlidianqilunwen/3133282.html
最近更新
教材專著