Short-Term Load Forecasting for Commercial Building Using Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) Network with Similar Day Selection Model

Load forecasting is essential in power systems for reliable and efficient energy planning and operation. Commercial buildings usually account for 20% of all energy used, with approximately 30% being wasted. Accurate load forecasting for commercial buildings can help improve operational efficiency. For accurate forecasting load, deep learning models have been used. Furthermore, the selection of input data has become important because the forecasting results can vary depending on which input data is trained. However, although various hybrid models have used historical sequential data as input data using the sliding window approach, they did not consider the hourly correlation between factors and load while selecting input data. In this paper, a hybrid convolutional neural network-long short-term memory network is used in combination with a similar day selection model to overcome these limitations by selecting the data of similar days as input data and by considering the hourly correlation with factors. The proposed method is found to be effective by comparing the performance of the traditional methods using convolutional neural or long short-term memory network.