Short-term power load forecasting based on TCN-LSTM model

Accurate electric load forecasting is crucial for maintaining the stability and efficiency of power operations. The significant volatility and randomness, influenced by various external factors, make precise load forecasting a considerable challenge. To address this issue, we propose a short-term electric load forecasting method utilizing a TCN-LSTM neural network model. Initially, the k-means algorithm is applied to perform cluster on the raw load-related data, identifying groups of similar days. Subsequently, grey relational analysis is employed to assess the similarity between these groups, highlighting the key factors affecting load fluctuations. These critical factors are then incorporated into the TCN-LSTM neural network model for training and forecasting. The TCN component captures long-term dependencies within the load data, while the LSTM segment addresses shorter-term dependencies. The model is then trained and validated using actual load data. Validation results demonstrate that, compared to traditional forecasting models, our method significantly improves accuracy and generalization, effectively predicting short-term changes in electric load. This advancement provides more precise decision support for the scheduling of power systems.