A Deep Learning-Based Multi-Layer Dynamic Multivariate Prediction Model for Power System Time Series Data Prediction Model

Multivariate time series forecasting is an important but challenging problem in future power system. Existing fore-casting models struggle to handle the increasing amount of time series data due to the continuous upgrading of system structures, leading to a decrease in model performance and making it difficult to meet the accurate prediction and response speed requirements of the system. To address this issue, this paper proposes a multilayer dynamic multi-type prediction model based on deep learning, designed to adapt to the complex and diverse dynamic changes brought about by the high uncertainty of renewable energy sources in future power systems. Firstly, the model introduces a dynamic variable convolutional structure to address the different application scenarios of multiple regions and roles in real power systems. Secondly, a three-task prediction framework is proposed to adapt to the multi-scale features of temporal information in power systems. Then, an imbalanced prediction structure is designed to concentrate computing re-sources on the main tasks. Finally, to meet the fast response requirements of the prediction system, a combination of linear and nonlinear structures is proposed to construct a dynamic multivariate forecasting model, simplifying the network structure design and accelerating training speed while ensuring prediction performance. Experiments on three real datasets demonstrate the strong generalization performancestrong generalization performance and effectiveness of this model in complex multivariate time series forecasting.