Application of Deep Learning Models Based on Chaos Modeling in Power Internet of Things Forecasting Tasks

Deep learning has provided a reliable technological foundation for time series prediction tasks. In the field of time series prediction, deep learning models are widely used in the Power Internet of Things due to their powerful fitting capabilities. However, efficiently capturing the relationships between time data remains a challenge. To overcome this challenge, this paper proposes a time series prediction model called the CTN-former based on chaos theory and a self-attention mechanism. CTN-former can detect chaos in time series through a carefully designed chaotic theory network (CTN), derive the relationships between time data, and perform phase space reconstruction. Additionally, to improve efficiency, the model includes a master-slave self-attention module to reduce quadratic complexity. Finally, experiments were conducted on four large-scale public datasets from three different domains. The results demonstrate that compared to traditional time series prediction models, the CTN-former achieves a relative improvement rate of 12.7% in time series prediction tasks, and the CTN-former achieves a relative improvement rate of 12.7% in time series prediction tasks and provides a novel solution for time series prediction by recovering the motion trajectory between chaotic time series, thereby enhancing the data utilization efficiency.