Research And Application Of Short-term Power Load Probabilistic Forecasting Technology And Interpretability Analysis

In the current power system,power load forecasting has become a crucial and vi-tal responsibility for the growth of the power industry.The basis for the safe,reliable,and cost-effective operation of the electricity system is accurate load forecasting.With the growth of society and the occurrence of more harsh weather,people’s need for power continues to rise,and load forecasting has become increasingly crucial.With the rapid advancement of artificial intelligence,deep learning’s outstanding feature extraction ca-pabilities have led to its widespread application in load forecasting.However,inadequate data mining is one of the primary factors affecting the precision of load forecasting.In ad-dition,it is difficult for decision makers to have faith in deep models because they cannot reveal more dependable information.Consequently,this paper investigates the problem of power load forecasting,adopts a specific feature extraction method for different types of factors affecting power load,and predicts the overall probability distribution of the load by probabilistic forecasting,thereby providing a broader range of load uncertainty information.On this premise,an interpretability component is introduced,which allows the model to interpret the prediction findings to some extent and improves the model’s dependability.Finally,it is used to the smart city power load forecasting problem.The main work and contributions of this paper are as follows:1.A probabilistic short-term power load forecasting model(SPLPF)is presented to address the issue of insufficient information extraction in load forecasting.SPLPF replaces the load probability model’s single modeling method by modeling the tar-get sequence and covariates independently,filtering noisy information,and min-ing effective features.Quantifying the prediction uncertainty and providing more complete forecast information,a probabilistic forecasting method based on quantile regression is utilized to provide a broader range of load uncertainty information.2.Aiming at the problems of insufficient trustworthiness and interpretability of the existing load forecasting models and inability to provide reasonable explanations for the forecasting results,this paper studies the interpretability method of the deep model.In the target sequence decomposition network,the prior knowledge of time series decomposition is added to limit the decomposition content of the decomposi-tion block,so that the output results can be understood by researchers after visual-ization? The contribution degree is used to explain the prediction basis of the model,and an interpretable short-term power load probability prediction model SPLPF-I is formed.3.This paper investigates and implements the power load forecasting task for smart cities.The power load data is collected,compiled into a data collection,and ap-plied to SPLPF-I using Internet of Things technology.In addition,the findings of this paper are incorporated into the national key R&D project of the Ministry of Science and Technology entitled ”Key Technologies and Demonstrations of the In-ternet of Things and Smart City” in order to monitor the urban public infrastructure and provide a basis for subsequent risk prediction and decision-making.