Multi-Sector User Air Conditioning Load Extraction Method Based on SFOA-Transformer-BiGRU-AKDE

Aiming at the challenges of the difficulty of dynamic separation of air conditioning loads, significant industry heterogeneity and lack of uncertainty quantification in the massive user shut-off load data, this paper proposes a highprecision air conditioning load mining model integrating SFOA-Transformer-BiGRU-AKDE. Based on the user load profile characteristics for industry label division, wind speed, temperature, humidity, and shutdown load characteristic data are constructed for each industry; secondly, the model learning parameters are optimised by SFOA, and the TransformerBiGRU cascade network realises the weather-load coupling relationship modelling and air-conditioning load dynamic decoupling.And Adaptive Kernel Density Estimation(AKDE) is introduced to perform adaptive bandwidth kernel density estimation on the prediction residuals to generate probability density distributions to quantify the prediction uncertainty.