Research On Urban Short-term Electricity Load Forecasting Based On Deep Learning

The imbalance between energy supply and demand has led to a global energy crisis,and energy transformation is imminent under the double carbon target.Domestic power demand is growing,and accurate multi-time scale power load forecasting can alleviate the contradiction between supply and demand in the power system and help the power system to carry out macro control and dispatch.Electricity load forecasting is a typical time series forecasting task.Electricity load data often exhibits different characteristics such as large scale,trend,periodicity and volatility.Weather factors,economic factors and human characteristics are key factors affecting urban electricity loads and load forecasting has received extensive attention from researchers since the emergence of the electricity industry.The access to a large number of Io T data collection devices,the explosive growth of multi-dimensional data and the increasingly demanding requirements for prediction accuracy have led to the difficulty of classical parametric models as well as traditional machine learning algorithms to meet the high accuracy requirements of long series prediction tasks.Since the rise of deep learning,various types of algorithms represented by CNNs,RNNs and Transformer models have achieved fruitful results in electricity load forecasting tasks.This paper uses the publicly available Panama Electricity Load dataset from Google’s Kaggle big data platform as the subject of this research,with the following main work:(1)Introduces the characteristics related to electric load data,analyzes the hourly load data given in the dataset from January 3,2015 to March 10,2020,as well as multidimensional influencing factors such as weather variables and holiday information,and performs pre-processing operations such as data cleaning and data scaling.(2)The structure and principles of the Informer algorithm are presented.The Informer algorithm,which has shown excellent performance in long series forecasting tasks,is used in the field of short-term electricity load forecasting.Based on the performance of the model tests,it is shown that the default MSE loss function used by Informer leads to errors in forecasting mainly due to differences in the shape of the peak and trough of the load curve.Further tests have shown that Informer-DILATE can produce prediction curves that better fit the true load values in terms of shape and time.(3)For the problems of many hyperparameters and complex model structure of deep learning networks,the population intelligence optimization algorithm is used to tune the model.The Pelican Optimisation Algorithm(POA)is a recent development of the Population Intelligent Optimisation algorithm with powerful optimisation discovery capabilities.the SPM chaos mapping and Levy flight strategies are used to further improve the optimisation accuracy and convergence speed of the Improved Pelican Optimisation algorithm.The Improved Pelican Optimisation Algorithm(IPOA)was used to tune the hyperparameters and optimisation model structure to make it more suitable for the electricity load forecasting task and to improve its forecasting performance and generalisation capabilities.An electricity load dataset from Panama was selected for short-term electricity load forecasting experiments on several time scales and better forecasting performance was achieved on several evaluation metrics.Among them,on the prediction experiments with 168 and 336 steps,the R~2of the model in this paper are 2.53%and 6.6%higher than those of the Informer model,which has better prediction accuracy.In summary,analyzing the characteristics of electric load data and combining the advantages of Informer model,DILATE loss function and IPOA,the prediction model of this paper is proposed.The proposed model is evaluated in short-term urban load forecasting,and experimental results demonstrate its superior forecasting accuracy.The research findings are both innovative and practical,contributing significantly to the advancement of the short-term urban electricity load forecasting field.