Multi-Load Short-term Forecasting for Integrated Energy Systems Based on Multi-features and BiLSTM-KAN

With the continuous deployment and development of integrated energy systems, accurate multi-load forecasting has become an important guarantee for the safe and reliable operation of integrated energy systems. To this end, a multi-load short-term forecasting method for integrated energy systems based on bidirectional long short-term memory and Kolmogorov-Arnold networks (BiLSTM-KAN) is proposed. Firstly, the convolutional neural networks and improved convolutional block attention module (CNN-ICBAM) are used to extract the coupling features among multiple loads in the time domain. The improved CBAM adopts the adaptive KAN to help the model better extract load features. At the same time, the Fourier transform is used to convert the load data to the frequency domain, and CNN is used to extract the coupling features in the frequency domain, thus achieving simultaneous extraction of time domain features and frequency domain features. Then, the long short-term memory (LSTM) network is used to extract periodic features in the load data, and finally, the KAN network is used to fit and obtain the prediction results. This paper uses the integrated energy system load data from the Tempe campus of Arizona State University for validation and compares it with several existing methods. The results show that the weighted average absolute percentage error of this method is minimal.