Enhancing accuracy in point-interval load forecasting: A new strategy based on data augmentation, customized deep learning, and weighted linear error correction

Accurate load forecasting plays a crucial role in ensuring and maintaining the stability and reliability of the power system. However, due to the diverse electricity consumption behaviors of users, the volatility and randomness of load data have been continuously increasing, making traditional point forecasting paradigm insufficient to provide enough information for Independent System Operators (ISOs). Furthermore, most studies neglecting the information contained in the error sequence, which makes it difficult to achieve more accurate forecasting results. To this end, this paper proposed a novel hybrid point-interval load forecasting framework. Specifically, we first obtain the trend data and denoised data by the improved Variational Mode Decomposition and extract peak data to get the 3-dimensional data after data augmentation, which allowing the model to improve the data utilization, especially for irregular load patterns and limited feature availability. Customized deep learning neural network integrates multi-layer Convolutional Neural Networks (CNNs), Bidirectional Gated Recurrent Unit (BiGRU), and multi-head self-attention mechanism, which can enhance feature extraction and capture long-sequence dependencies at the same time, and Quantile regression loss function (QRLoss) is introduced to achieve accurate point-interval forecasting. After getting forecasting results from the neural network, we employ an improved optimization algorithm to optimize the linear error correction function to attain more precise prediction outcomes. To validate the advanced nature of our model, we conducted experiments using four sets of load data from New South Wales and performed Diebold-Mariano test. The experimental results show that the Mean Absolute Percentage Error of our model is less than 0.7% across the four datasets, with the Prediction Interval Coverage Probability close to 1.00, which significantly outperforms baseline models, providing accurate and comprehensive forecasting information for power system management.