基于CEEMDAN-VMD和優(yōu)化LSTM的電力短期負(fù)荷預(yù)測(cè)
馬藝銘
(國(guó)網(wǎng)遼寧省電力有限公司大連供電公司,遼寧 大連 116001)
摘 要 :電力系統(tǒng)負(fù)荷具有波動(dòng)性高、隨機(jī)性強(qiáng)、不確定性及復(fù)雜度高的特點(diǎn),為進(jìn)一步提高電力 短期負(fù)荷預(yù)測(cè)精度,需要深層次挖掘數(shù)據(jù)間的非線性關(guān)系。提出了一種基于自適應(yīng)噪聲完備集合經(jīng)驗(yàn)?zāi)?態(tài)分解 (CEEMDAN) 和變分模態(tài)分解 (VMD) 二次模態(tài)分解的長(zhǎng)短期記憶 (LSTM) 網(wǎng)絡(luò)電力短期負(fù)荷預(yù)測(cè)模型。在利用CEEMDAN對(duì)原始數(shù)據(jù)序列進(jìn)行初次模態(tài)分解得到序分量后,采用K-means手段將序分量樣本熵 (SampEn/SE) 聚類為三部分,對(duì)其中的強(qiáng)非平穩(wěn)序列進(jìn)行VMD技術(shù)的二次分解以減弱其非平穩(wěn)性,將二次分解后得到的序分量與初次模態(tài)分解得到的中低頻序分量構(gòu)建為新的組合后分別通過(guò)粒子群優(yōu)化算法 (PSO) 得到最優(yōu)超參數(shù),代入?yún)?shù)訓(xùn)練后得到各分量最優(yōu) LSTM 模型,并融合各模型預(yù)測(cè)結(jié)果得到最終負(fù)荷預(yù)測(cè)值。通過(guò)實(shí)驗(yàn)表明,相較于其他模型,所提方法在實(shí)際預(yù)測(cè)中具備更好的模型性能和更高的 預(yù)測(cè)精度。
關(guān)鍵詞: 短期負(fù)荷預(yù)測(cè);二次模態(tài)分解;自適應(yīng)噪聲完備集合經(jīng)驗(yàn)?zāi)B(tài)分解;變分模態(tài)分解;樣本熵; 粒子群優(yōu)化 ;長(zhǎng)短期記憶網(wǎng)絡(luò)
中圖分類號(hào) :TM715 文獻(xiàn)標(biāo)識(shí)碼 :A 文章編號(hào) :1007-3175(2025)11-0041-07
Short-Time Power Load Forecasting Based on CEEMDANVMD and Optimazed LSTM
MA Yi-ming
(State Grid Liaoning Electric Power Co., Ltd. Dalian Power Supply Company, Dalian 116001, China)
Abstract: Power system load is characterized by high volatility, strong randomness, high uncertainty, and high complexity. To further improve the accuracy of short-term power load forecasting, it is necessary to deeply explore the nonlinear relationships between data. A short-term power load forecasting model based on long short-term memory (LSTM) network with secondary modal decomposition combining complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and variational mode decomposition (VMD) is proposed. After the initial modal decomposition of the original data sequence using CEEMDAN to obtain sequential components, the sample entropy (SampEn/SE) of the sequential components is clustered into three parts by K-means method. The strongly non-stationary sequences among them are subjected to secondary decomposition using VMD technology to reduce their non-stationarity. The sequential components obtained from the secondary decomposition and the medium-low frequency sequential components from the initial modal decomposition are constructed into new combinations, and the optimal hyperparameters are obtained for each combination through the particle swarm optimization (PSO) algorithm. After parameter training, the optimal LSTM model for each component is obtained, and the final load forecasting value is derived by fusing the prediction results of each model. Experimental results show that compared with other models, the proposed method exhibits better model performance and higher forecasting accuracy in practical predictions.
Key words: short-term load forecasting; secondary modal decomposition; complete ensemble empirical mode decomposition with adaptive noise ; variational mode decomposition; sample entropy; particle swarm optimization; long short-term memory network
參考文獻(xiàn)
[1] 李建林,丁子洋,游洪灝,等 . 構(gòu)網(wǎng)型儲(chǔ)能支撐新型電力系統(tǒng)穩(wěn)定運(yùn)行研究[J]. 高壓電器,2023, 59(7):1-11.
[2] 李剛,方鴻,劉云鵬,等 . 新型電力系統(tǒng)中的大模型驅(qū)動(dòng)技術(shù):現(xiàn)狀、機(jī)遇與挑戰(zhàn)[J]. 高電壓技術(shù), 2024,50(7):2864-2878.
[3] 楊海柱,田馥銘,張鵬,等 . 基于CEEMD-FE和AOALSSVM的短期電力負(fù)荷預(yù)測(cè)[J]. 電力系統(tǒng)保護(hù)與控制, 2022,50(13):126-133.
[4] 戴浩男,張辰灝,甄釗,等 . 基于時(shí)空特征聚類和雙層動(dòng)態(tài)圖卷積網(wǎng)絡(luò)建模的短期凈負(fù)荷預(yù)測(cè) [J]. 高電壓 技術(shù),2024,50(9):3914-3923.
[5] 解振學(xué),林帆,王若谷,等 . 基于時(shí)序動(dòng)態(tài)回歸的超短期光伏發(fā)電功率預(yù)測(cè)方法[J]. 智慧電力,2022, 50(7):45-51.
[6] FUMO N, RAFE BISWAS M A.Regression analysis for prediction of residential energy consumption[J]. Renewable and Sustainable Energy Reviews,2015, 47 :332-343.
[7] AMBER K P, ASLAM M W, HUSSAIN S K.Electricity consumption forecasting models for administration buildings of the UK higher education sector[J]. Energy and Buildings,2015,90 :127-136.
[8] 孔祥玉,馬玉瑩,艾芊,等 . 新型電力系統(tǒng)多元用戶的用電特征建模與用電負(fù)荷預(yù)測(cè)綜述 [J]. 電力系統(tǒng)自 動(dòng)化,2023,47(13):2-17.
[9] 彭曙蓉,彭家宜,楊云皓,等 . 基于時(shí)變深度前饋神經(jīng)網(wǎng)絡(luò)的風(fēng)電功率概率密度預(yù)測(cè) [J]. 電力科學(xué)與技術(shù) 學(xué)報(bào),2023,38(3):84-93.
[10] 申洪濤,李飛,史輪,等 . 基于氣象數(shù)據(jù)降維與混合深度學(xué)習(xí)的短期電力負(fù)荷預(yù)測(cè) [J]. 電力建設(shè),2024, 45(1) :13-21.
[11] 冉啟武,張宇航 . 基于模態(tài)分解及GRU-XGBoost短期 電力負(fù)荷預(yù)測(cè) [J]. 電網(wǎng)與清潔能源,2024,40(4) : 18-27.
[12] 趙倩,鄭貴林 . 基于WD-LSSVM-LSTM模型的短期電力 負(fù)荷預(yù)測(cè) [J]. 電測(cè)與儀表,2023,60(1) :23-28.
[13] 李焱,賈雅君,李磊,等 . 基于隨機(jī)森林算法的短期電力負(fù)荷預(yù)測(cè) [J]. 電力系統(tǒng)保護(hù)與控制,2020, 48(21):117- 124.
[14] 張雲(yún)欽,程起澤,蔣文杰,等 . 基于EMD-PCA-LSTM的光伏功率預(yù)測(cè)模型 [J]. 太陽(yáng)能學(xué)報(bào),2021,42(9): 62-69.
[15] WU Z, HUANG N E.Ensemble empirical mode decomposition:A noise-assisted data analysis method[J].Advances in Adapt Data Analysis, 2009,1(1) :1-41.
[16] TORRES M E, COLOMINAS M A, SCHLOTTHAUER G, et al.A complete ensemble empirical mode decomposition with adaptive noise[C]//2011 IEEE International Conference on Acoustics, Speech and Signal Processing(ICASSP),2011 :4144-4147.
[17] 楊德州,劉嘉明,宋汶秦,等 . 基于改進(jìn)型自適應(yīng)白噪聲完備集成經(jīng)驗(yàn)?zāi)B(tài)分解的工業(yè)用戶負(fù)荷預(yù)測(cè)方法[J]. 電力系統(tǒng)保護(hù)與控制,2022,50(4):36-43.
[18] 陳錦鵬,胡志堅(jiān),陳緯楠,等 . 二次模態(tài)分解組合DBiLSTM-MLR 的綜合能源系統(tǒng)負(fù)荷預(yù)測(cè) [J]. 電力系統(tǒng) 自動(dòng)化,2021,45(13):85-94.
[19] 趙星宇,吳泉軍,朱威 . 基于CEEMDAN和TCN-LSTM模型的短期電力負(fù)荷預(yù)測(cè)[J]. 科學(xué)技術(shù)與工程,2023, 23(4):1557-1564.
[20] RICHMAN J S, MOORMAN J R.Physiological timeseries analysis using approximate entropy and sample entropy[J].American Journal of Physiology Heart and Circulatory Physiology,2000,278(6): H2039-H2049.