Research On Power Load Forecasting Method Based On Machine Learning

The amount of electricity load data acquired at the smart meter end is increasing as Io T artificial intelligence technology advances.Accurate and timely power load prediction findings are critical for power system stability,as well as decision support for medium and long-term grid scale construction and large-scale maintenance timing.With the continued development of deep learning and machine learning algorithms,it is vital to include machine learning deep learning approaches into power load forecasting in an effective and scientific manner in order to improve the quality of power grid operation.1)Given that medium-and long-term power load trends are susceptible to economic and meteorological factors,and traditional trend forecasting methods are difficult to predict accurately,this paper proposes the EMDIA medium-and long-term power load forecasting method,using Hong Kong’s power consumption from 2000 to 2021 as the research object,and taking economic and meteorological factors into account.Adaboost,isometric mapping(Isomap),and empirical modal decomposition(EMD)are all fully integrated within the EMDIA a lgorithm.The EMD approach is used to dissect the load and its affecting elements into numerous intrinsic modal functions(IMFs)and residual terms,taking into account the impact of meteorological and economic factors;the power load is split into fluctua ting and trending terms by correlation analysis.1)The meteorological and economic components are then dimensionally reduced using the Isomap method to decrease correlation while reducing redundant quantities in the components;Adaboost is then used for ind ividual forecasting;and the forecasting results are ultimately overlaid.EMDIA compares the applicability and reliability of the novel combined forecasting model suggested in this study against standard forecasting methods using all four assessment criteria.2)The short-term electric load fluctuates often,and the trajectory of the fluctuation shift is erratic.This research suggests an attention mechanism-enhanced CNN-GRU short-term electric load forecasting model(CGA).CNN are utilized in the model to feature extract historical meteorological data and load data.The model then uses the GRU model for prediction in the eventuality that significant information is lost due to the power load data being too long in a time series.The "decomposition-integration" concept is used to further enhance the forecasting capabilities of the model.The sparrow algorithm,which decomposes the electric load into many parts via VMD,addresses the parameter selection problem for VMD(PSVMD).The components are predicted separately and the superimposed components are then forecasted.in order to further improve the forecasting,the model’s capacity.The PSVMD-CGA model,when compared to the CGA model,uses the data decomposition concept to lessen the nonlinearity of the load and makes it simpler for the CGA to identify the pattern of fluctuation between the model input and output.The overall research results demonstrate that the PSVMD-CGA model is superior,as evidenced by its higher prediction accuracy and stronger prediction per formance.Accurate medium and long term prediction provides favorable reference for power grid and power company’s power supply system reconstruction and long-term planning.Short term prediction provides decision support for ensuring power system stabilit y and power dispatching.Figure [39] Table [13] Reference [82]...