基于改进LSTM的火电厂负荷预测方法研究

打开文本图片集

中图分类号：TM621 文献标志码：A 文章编号：2095-2945（2025）26-0139-04

Abstract：Heatingloadforecastingisaprerequisiteforguiding heatingoperationmanagementandscheduling.Heatingload forecastingisatimeseriesforecastingproblemthatrequiresustouseavailablehistoricalrecordsandweatherinformatioto predictthereal-timeheatingloadforthenext24hours.Inthispaper，ashort-termheatingloadforecastingmodelbasedona carefullydesignedtandemlongshort-termmemory（LSTM）recurentneuralnetworkwasproposed.Wedemonstratedtheprocessof datapreprocesinganddesignthelossfunction toimprovetheperformanceofthemodel.Wealsocombinedtheensemble strategywiththeLSTMmodeltoenhanceitsgeneralzationabilityandrobustness.Ontheofline（historical）testdata，the proposedmodelisabletomakesatisfactorypredictions tomeetheneedsofthelocalpowerplantInaditiontotheoflinetest， weapliedthemodeltotheonlinesystemofapowerplantin ShandongProvince.Duringtheheatingseasonof2O18，themodel continuouslymadepreditionswithouthumaninterventionforfourmonths.Theperformanceofthemodelintheonlinetestis comparable to the ofline experimental results using historical data，achieving satisfactory test results.

Keywords：deep learning; load forecasting;recurrent neural network; time series; LSTM

负荷预测对大型能源系统的规划和优化至关重要，特别是在电力和供热领域。（剩余4912字）