基于長短時記憶網(wǎng)絡的電力系統(tǒng)負荷預測方法研究
王鑫琪,李闖,焦晗,李焱飛
(南京工程學院 電力工程學院,江蘇 南京 211167)
摘 要:準確的負荷預測對保持電網(wǎng)的穩(wěn)定性和提高當?shù)亟?jīng)濟效益、節(jié)約成本有重大幫助??紤]到負荷數(shù)據(jù)帶有時序性,以及智能電網(wǎng)的發(fā)展所帶來的數(shù)據(jù)量的增大,建立了長短時記憶網(wǎng)絡(LSTM)模型來對未來用電量進行短期負荷預測。針對Adam訓練算法可能存在的收斂問題,對其進行了改進,并通過MATLAB軟件對LSTM網(wǎng)絡進行建模,通過與BP神經(jīng)網(wǎng)絡進行對比,結果表明,LSTM模型具有更高的精確度以及實用性。
關鍵詞:短期負荷預測;BP神經(jīng)網(wǎng)絡;長短時記憶網(wǎng)絡;Adam算法
中圖分類號:TM715 文獻標識碼:A 文章編號:1007-3175(2019)11-0017-04
Research on Power System Load Forecasting Method Based on Long-Term and Short-Term Memory Network
WANG Xin-qi, LI Chuang, JIAO Han, LI Yan-fei
(School of Electric Power Engineering, Nanjing Institute of Technology, Nanjing 2111 67, China)
Abstract: Load forecasting is an important part of power system dispatching. Accurate load forecasting is of great help to maintain grid stability and improve local economic benefits and cost. Considering the sequential nature of load data and the increase in data volume brought about by the development of smart grids, a long-term and short-term memory network (LSTM) model was established to make shortterm predictions of future electricity consumption. On this basis, the convergence problem of Adam training algorithm may be improved. It is simulated by MATLAB software and compared with BP neural network. The results show that the LSTM model has higher accuracy and practicability.
Key words: short-term load forecast; BP neural network; long-term and short-term memory network; Adam algorithm
參考文獻
[1] 徐堯強,方樂恒,趙冬華,王凱. 基于LSTM神經(jīng)網(wǎng)絡的用電量預測[J]. 電力大數(shù)據(jù),2017,20(8):25-29.
[2] 張宇航,邱才明,賀興,凌澤南,石鑫. 一種基于LSTM神經(jīng)網(wǎng)絡的短期用電負荷預測方法[J]. 電力信息與通信技術,2017,15(9):19-25.
[3] 陳卓,孫龍祥. 基于深度學習LSTM網(wǎng)絡的短期電力負荷預測方法[J]. 電子技術,2018,47(1):39-41.
[4] 劉進波,周克男,任明會. 基于相關因子的BP神經(jīng)網(wǎng)絡電力負荷短期預測[J]. 信息記錄材料,2018,19(1):211-214.
[5] 陳英慧,商淼,李婷,高紅艷. 基于GA-BP算法的呼和浩特地區(qū)短期電力負荷預測[J]. 山東工業(yè)技術,2018(5):170.
[6] 王德文,孫志偉. 電力用戶側大數(shù)據(jù)分析與并行負荷預測[J]. 中國電機工程學報,2015,35(3):527-537.
[7] 劉文博. 基于神經(jīng)網(wǎng)絡的短期負荷預測方法研究[D]. 杭州:浙江大學,2017.
[8] 程宇也. 基于人工神經(jīng)網(wǎng)絡的短期電力負荷預測研究[D]. 杭州:浙江大學,2017.
[9] 朱振偉. 氣象因素對電網(wǎng)負荷特性影響的研究[D].杭州:浙江大學,2008.
[10] 李萌,程浩忠,楊宗麟,韓新陽,楊鏡非. 采用分形插值的典型日負荷曲線改進預測方法[J]. 電力系統(tǒng)及其自動化學報,2015,27(3):36-41.
[11] 陳鯉江,景程,吳姚鑫,鄭水泉. 數(shù)學表達式的歸一化方法研究[J]. 浙江工業(yè)大學學報,2012,40(2):229-232.