A Koopman Theory-Based Time Series Prediction Model for Renewable Energy Power

With the increasing penetration of renewable energy in power systems, its inherent uncertainty has significantly impacted system operation states. Accurate prediction of renewable energy power represents a critical technical bottleneck for ensuring reliable grid integration and effective utilization. To address this challenge, this paper proposes a Koopman theory-based time series prediction model for renewable energy power. The output power of renewable energy such as wind power and photovoltaic generation is treated as non-stationary time series. Using a Koopman theory-based prediction method for non-stationary time-series, the time series is decomposed into timevarying and time-invariant components, which are then mapped to an infinite-dimensional linear space for analysis and prediction. A hierarchical approximation framework is constructed, employing a multi-layer residual fitting prediction iterative program to progressively enhance prediction accuracy. Compared with conventional methods such as LSTM and ARIMA, the proposed method demonstrates superior accuracy in short-term prediction for renewable energy power.