A multi-dimensional time series modeling technology for internal resources of virtual power plants

To address the issue of insufficient modeling accuracy caused by the complex time - series characteristics and high - dimensional coupling of multi - industry resources in virtual power plants, this paper proposes a multi - dimensional time -series analysis model based on CEEMDAN_VMD_CNNbiLSTM. Three typical demand - side resources, namely blast furnace blowers (industrial), commercial building air - conditioning systems (commercial), and electric vehicles (residential), are selected to construct a multi - industry high - dimensional time -series dataset. The CEEMDAN - VMD hybrid modal decomposition is used to eliminate the modal aliasing of cross - industry signals, and the CNN - BiLSTM network is combined to achieve in - depth extraction of spatio - temporal features and bidirectional dependency modeling. Experiments show that: (1) The determination coefficient of the commercial user model is the best (97.10%), and the MAE and MAPE are reduced to 3.74 and 1.58% respectively; (2) The prediction error fluctuation of residential users is the smallest (MAE = 0.65); (3) The modeling accuracy of industrial users and electric vehicle resources is significantly improved, with determination coefficients reaching 92.10% and 87.65%. This verifies its ability to decouple the complex resource dynamics of virtual power plants and its adaptability across different scenarios, providing technical support for the precise regulation of virtual power plants with a high proportion of renewable energy integration.