Risk Prediction Method for Power Flow and Voltage Violations Based on Adaptively Weighted Multi‐Task Deep Learning

ABSTRACT

With the increasing integration of high‐penetration renewable energy into distribution networks, power flow and node voltage violations have emerged as critical threats to system stability. Considering the physical coupling between these two types of risks, this paper presents a risk prediction method for power flow and voltage violations using adaptively weighted multi‐task deep learning (AWMDL). First, based on the IEEE 33‐bus distribution network model, an active distribution network simulation model is developed in MATLAB to obtain time‐series data on power flow and node voltages. Using the generated simulation dataset, the XGBoost algorithm is introduced to perform dual selection of input time steps and feature variables. Considering the distinct characteristics of the two types of risks, a soft sharing‐based multi‐task deep learning model is constructed, employing BiGRU as the shared layer model, CNN‐1D as the model for the power flow violation classification task, and TCN as the model for the voltage violation risk regression task. Meanwhile, an adaptive weighting strategy based on the loss function is proposed to achieve joint optimization of the model. Case study results demonstrate that the proposed method outperforms conventional single‐task approaches in comprehensive prediction performance, and each core module is shown to significantly boost prediction effectiveness.