Research on Short-Term Power Load Forecasting of CNN-LSTM Based on Attention Mechanism

Compared to the traditional power grid, the smart grid's abundance of big data poses limitations for traditional power load forecasting methods in terms of learning and mining data information. In this paper, we propose the construction of a combination model using deep learning algorithms, in order to achieve accurate and intelligent short-term prediction of power load. This approach is in response to existing prediction models' limitations. To achieve this, we introduce the AM-CNN-LSTM load forecasting combination model, based on the CNN-LSTM deep learning model. The CNN-LSTM model addresses the limitation of a single LSTM model, which only takes into account the past information of the time-series data and disregards future information. This model enables a more precise extraction of data features, enhancing the accuracy of the prediction model. In addition, the proposed model introduces an attention mechanism (AM) to consider the varying contributions of historical data to data points at different times. The model combines AM with CNN and LSTM, addressing the low accuracy and poor generalization issues compared to traditional load forecasting models.