Research On Urban Residential Load Forecasting Based On Improved LSTM Network And Informer

Electric energy is one of the most important energy sources for future smart city development.In cities,there are different power consumption groups,and the proportion of residents’ electricity consumption in the city’s total electricity consumption is increasing.Therefore,accurate estimation of residential electricity consumption will help the power sector to formulate a scientific and rational power production scheduling plan,effectively reduce the consumption of power resources,and thus promote the normal operation of the power system,which is crucial to the development of smart cities in the future.It is also of great significance for the development of modern electricity and smart grids.With the expansion of urban areas,the diversification of household appliances,and the continuous development of society,the factors that affect the power load are gradually enriched.Traditional forecasting methods have been unable to achieve satisfactory results in increasingly complex load sequences.Therefore,how to accurately describe the relevant factors that affect the electricity consumption in urban residential areas,and use practical and effective methods to forecast are the main challenges facing the current residential load forecasting field.This paper first analyzes the historical residential load data to understand its interaction with external factors,and then fully mines the information contained in the residential load data itself through time series mining technology,and finally uses the multi-layer processed data to train the model.At present,the application of hybrid models and deep learning algorithms in the field of load forecasting is the main research direction of researchers.In this paper,two models are used to predict the electricity consumption of urban residential areas,one is the hybrid model LSTM-Ada Boost,and the other is an improved deep learning algorithm based on Transformer-Informer.The main contents of the paper include the following aspects:(1)Analysis of external influencing factors and preprocessing of residential load data.First,the external factors that affect the electricity demand in residential areas,such as temperature,humidity,holidays and other data,use mutual information(MI)to carry out correlation analysis,so as to find out the external factors that really affect the electricity demand;secondly,the residential load sequence data The characteristics of the data are analyzed to make more accurate predictions,such as periodic detection,to grasp the laws of residential electricity consumption behavior,etc.;finally,the data is standardized to eliminate errors caused by different data dimensions.(2)A short-term residential load forecasting model LSTM-Ada Boost is constructed that incorporates ensemble learning Long Short-Term Memory(LSTM)networks.The model uses the LSTM network with time-series memory function as the individual learner for ensemble learning;then uses the Ada Boost ensemble algorithm to train multiple individual learners in series and calculate the weight of each individual learner.Different individual learners have different According to the weight size,the individual learners with smaller prediction errors have larger weights;finally,all individual learners are linearly combined according to the weights of the individual learners to form a strong learner,and the final prediction result is output.(3)The long-term series prediction of the power load in urban residential areas can make the optimal allocation of power resources in advance and ensure the balance between power supply and demand.In order to improve the ability of long-sequence residential load forecasting,this paper applies the Informer model to the field of urban residential load forecasting for the first time.The Informer model proposed at the Association for the Advancement of Artificial Intelligence utilizes its unique encoderdecoder architecture and probabilistic sparse self-attention mechanism to accurately capture long-term correlations between long-sequence inputs and outputs.At the same time,in order to reduce the impact of abnormal points on the model prediction,this paper uses the Log-Cosh loss function to train the Informer model to further improve the accuracy of load prediction for urban residential areas.The experiment uses the MIT smart microgrid data,which records low-resolution historical electricity consumption data from 400 households and meteorological data from the surrounding areas of the household,which is very suitable for simulating microgrids.The experimental results confirm the feasibility of the proposed two methods.The results show that the LSTM-Ada Boost ensemble model has higher prediction accuracy than single prediction methods such as LSTM network,support vector machine(SVM)and CART decision tree.The method can improve the accuracy and robustness of residential load forecasting;compared with the LSTM-Ada Boost ensemble model,the Informer model does not show better results than the LSTM-Ada Boost ensemble model when predicting shorter time series.But as the prediction horizon increases,the Informer model is able to more efficiently and accurately capture long-term correlations between long sequences of inputs and outputs.At the same time,the improvement of the loss function further improves the accuracy of the model’s prediction of electricity consumption in residential areas.