Adaptive Learning Based Load Forecasting Curve Data Extraction for Power Grid Operation Planning

Given that load forecasting data is often generated by different departments within the power system or obtained through predictive analysis tools, there are often discrepancies in format and time scale compared to the requirements of the Security-Constrained Unit Commitment (SCUC) algorithm. Typ-ically provided as load curves, this data necessitates conversion of format and refinement of time scales to meet the needs of SCUC input. This paper proposed a multi-time-scale data extraction method based on adaptive learning rates (ALR), aimed at addressing the mismatch between load forecasting data and SCUC algorithm inputs in power system operation planning. The proposed method dynamically adjusts the learning rate according to the local slope and variability of the curve, automatically adapting the ALR for data point extraction. This approach ensures high-precision data extraction across different time scales and complex curve scenarios, reducing the workload of manual data extraction and enhancing the intelligence and automation of the process.