Research On Power Load Forecasting Based On Neural Network

Power consumption has a huge impact on social progress and economic development.With social development and progress,global power consumption is increasing rapidly.Electricity consumption is divided into three sectors,namely industrial sector,commercial sector and residential sector,of which more than 65% is consumed in residential areas.The efficiency of power management system has a great influence on the utilization rate of electric energy.In order to minimize power outages and enhance user experience,the function and service ability of power management system must be improved.In China,power load forecasting plays an important role in promoting the security and stability of the power system to provide services to users and improve the service efficiency of the national power system.Short-term load forecasting is mainly based on daily load.Before the short-term load forecasting process,the correlation analysis method is used to analyze the potential law of load and meteorological information,and the influence of weather on the load curve is studied.It is related to the establishment of safe,reliable and economic operation strategy of power system,which plays a key role and has a great responsibility.Therefore,power grid load forecasting is faced with new challenges,and higher requirements are put forward for the accuracy of forecasting than before.At present,there are two problems in power load forecasting: 1.Because there are too many climate factors affecting short-term changes,too many feature inputs will increase the calculation time of short-term load forecasting model;2.Second,there are some problems in the processing of time series data in the existing models.Based on the above two problems,this paper proposes a short-term load forecasting model based on neural network.The main contents are as follows:(1)By analyzing the characteristics of the load in the target area,it is concluded that season and working day have a great influence on the short-term load,which can be used as the characteristics of the final model input.Weather data and load data are preprocessed before correlation analysis and forecasting models.Spearman algorithm was used to conduct correlation analysis,and it was concluded that among the numerous climate factors,the climate factor that was correlated with the daily load was the one that simplified the input characteristics for the subsequent model prediction,reduced the calculation time and strengthened the stability of the training model.(2)After analyzing the characteristics of correlation,ATT-LSTM,a model combined with LSTM of attention mechanism,is used for load prediction.This method uses attention mechanism to give different weights to the characteristics of input sequences,as well as the gate mechanism of long and short-term memory neural network,which has the advantage of long-term memory,so a combination model of both is proposed.It has been proved that compared with the traditional algorithm,the prediction accuracy is improved,the cost of power generation and power supply dispatching is saved,and the shortage or waste of power supply is avoided.(3)Although ATT-LSTM improves the prediction accuracy,it is not applicable to the case of small sample size.When the number of samples is limited,the accuracy effect of load prediction is general.Moreover,in the case of sufficient samples,the calculation time of the model with the two-layer LSTM structure is too long.If the single-layer LSTM is adopted,the structural accuracy is not high,so the prediction model based on LSTM is not practical.Aiming at the shortcomings of ATT-LSTM model,a method combining SENET with time domain convolutional network(TCN),STCN,is proposed.TCN has efficient calculation and stable gradient,which reduces calculation time and improves accuracy.SENET can further improve model performance.The actual data experiments show that this method has a higher prediction accuracy,and the calculation time is better than the existing prediction models.Experimental comparison with All-LSTM and TCN verifies the effectiveness of the improved method and its high practicability.Accurate and timely load forecasting can grasp the regional load variation law in time,and provide reliable technical support for power grid enterprises’ decision-making and power dispatching,so that power grid enterprises can provide more stable and reliable power service.