A Model-Free Approach to Dynamic Operating Envelopes Calculation Considering Model Uncertainty

The increasing penetration of distributed energy resources poses challenges for distribution network safety, with simultaneous power injection potentially causing issues such as voltage overruns. To address this, Dynamic Operating Envelopes (DOEs) are essential for setting time-varying power injection limits for each producer to ensure safe network operation. Traditional model-based methods rely on detailed electrical models, which are time-consuming and costly to obtain. This paper proposes a model-free approach to DOEs computation using a deep learning model that combines CNN, LSTM, and ABKDE. The model predicts voltage magnitudes while ABKDE generates confidence intervals, accounting for model uncertainty. A heuristic algorithm is then used to calculate the DOEs iteratively. The proposed method is validated on the IEEE 33-bus distribution system, demonstrating strong performance and effectiveness.