Research Of Short-Term Load Forecasting Based On Load Characteristic Analysis And Neural Network

Due to the growing problems of fuel supply constraints,global warming and the environment,there is an urgent need to improve energy efficiency and reduce carbon emissions,where electricity consumption has a significant impact on carbon emissions,energy use efficiency and industrial development decisions.In the development of modern power systems,reliable and accurate regional electricity load forecasting is essential for power system operation and planning,which provides basic information for economic operation of power systems,dispatching power and assessing system security.The accuracy of load forecasting can be influenced by load time series data on the one hand,and there will be a correlation between load values at the current moment and load values at the historical moment.On the other hand,load changes can be affected by external factors such as weather,holidays,and power consumption behavior.Considering multiple variable factor inputs can contain more reference information to build more accurate load forecasting models,however,input factors with low correlation may cause too much noise to affect load forecasting results.With the continuous improvement of power systems and the continuous development of smart grid,load characteristic analysis is the basic work for the power market to formulate demand response strategy,manage orderly power consumption and load forecasting.Due to the large amount of data,fine granularity of data,and affected by weather conditions,holidays,economic development and other factors,the trend of power load has the characteristics of timing,randomness and nonlinearity.Combined with the results of load characteristic analysis,this paper studies the short-term load forecasting method based on neural network.The main work is as follows:(1)In the short-term load forecasting problem with univariate electric load feature inputs,a bidirectional long short-term memory neural network(Bi-LSTM)model based on Attention mechanism(Attention)is constructed.The standard long short-term memory neural network(LSTM)can effectively alleviate the long-term dependence problem of long-range sequence inputs,but as the load data volume increases and the input sequence length increases,the LSTM model structure becomes more complex and the computational complexity increases substantially,limiting the ability to process information.In this paper,we combine the Attention mechanism and the Bi-LSTM model,and introduce the Attention mechanism into the time series features for adaptive weight assignment so that the Bi-LSTM layer can identify more important features.By using Bi-LSTM as the prediction model,the structure of bidirectional information input and the information ”memory” function in the LSTM network,the complexity of the prediction model is reduced,the network capacity is larger,and the generalization performance is better.(2)In the short-term load forecasting problem based on multiple feature inputs,a short-term load forecasting method combining convolutional neural network(CNN)and bidirectional long short-term memory neural network(Bi-LSTM)is designed.The method first extracts the nonlinear features between the influencing factors and the electric load through the CNN network,and the network layers,network structure,the use of activation function and the convolutional kernel in the CNN are designed.Then,the extracted features are used as the input of the Bi-LSTM model,and the Bi-LSTM model is trained and analyzed to output the predicted load data.The algorithm test results show that the combined model approach significantly outperforms other load prediction methods performed for comparison.