Research On Short-Term Load Forecasting Using Long Short Term Memory

The electrical energy consumption in the community has increased significantly from time to time.The demand and supply of electrical energy must be balanced to avoid damaging power suppliers and power consumers.Therefore,the accuracy of load forecasting plays a vital role in planning and operation.The electricity load is generally unstable,volatile,and varies so it can cause losses if not properly planned and operated.Poor or bad planning and operation will be very detrimental to electricity suppliers and can also harm consumers.Also,to predict the electric load accurately is challenging.In this study,a hybrid model that combines the Convolutional Neural Network(CNN)and Long Short Term Memory(LSTM)with added by Attention Mechanism(AM)and Bidirectional direction to help improve the accuracy of forecasting electric load.First,the data information brings to the CNN layer to extract its features to find out trends and get the pattern from the data.Next,the data information goes to the LSTM layer for prediction.In this LSTM layer,there is an attention layer whose function is to emphasize the more important information Besides that,in this LSTM layer uses bidirectional direction,which helps the LSTM to read not only the previous data information but also the future data information.In this study also uses dropout layer to deal with data overfitting problems.The proposed model has been tested using real data from the AMI.To prove the effectiveness of the proposed model,the experimental results of the proposed model are compared with several benchmark models that have been used in the previous research on short term load forecasting.As a result,the proposed model shows better results with enough small error compared to other models.