High Precision LSTM Model for Short-Time Load Forecasting in Power Systems
Tomasz Ciechulski, S. Osowski
Abstract
The paper presents the application of recurrent LSTM neural networks for short-time load forecasting in the Polish Power System (PPS) and a small region of a power system in Central Poland. The objective of the present work was to develop an efficient and accurate method of forecasting the 24-h pattern of power load with a 1-h and 24-h horizon. LSTM showed effectiveness in predicting the irregular trends in time series. The final forecast is estimated using an ensemble consisted of five independent predictions. Numerical experiments proved the superiority of the ensemble above single predictor resulting in a reduction of the MAPE the RMSE error by more than 6% in both forecasting tasks.
Topics & Concepts
Electric power systemArtificial neural networkComputer scienceMean squared errorMean absolute percentage errorPower (physics)Ensemble forecastingTime seriesReduction (mathematics)Recurrent neural networkSeries (stratigraphy)Work (physics)Mean squared prediction errorArtificial intelligenceMachine learningData miningStatisticsMathematicsEngineeringMechanical engineeringBiologyPhysicsGeometryQuantum mechanicsPaleontologyEnergy Load and Power ForecastingStock Market Forecasting MethodsHydrological Forecasting Using AI