Distributed Photovoltaic Distribution Voltage Prediction Based on eXtreme Gradient Boosting and Time Convolutional Networks

Abstract

The current distribution voltage prediction methods have low accuracy and cannot realize more efficient system power allocation. To address this, the study proposes a distributed photovoltaic distribution voltage prediction model based on gradient boosting tree and time convolutional network. The model uses eXtreme gradient boosting for feature selection and time convolutional network and two-layer prediction strategy for voltage prediction. Then, the model is improved and optimized using residual module with bottle sea sheath algorithm. The outcomes indicated that the evaluation results of the model’s prediction intervals coverage probability, prediction interval normalized average width, R-squared, root mean square error, and running time indexes were 97.47%, 0.02, 97.48%, 0.15, and 2.11s, respectively. The research-designed model, when applied to voltage control, can effectively prevent voltage overrun situations. The research-designed voltage prediction model has important practical application value for the prediction and control of photovoltaic distribution voltage in power systems.

Keywords

• XGBoost
• TCN
• N-BEATS
• SSA
• voltage prediction
• distributed photovoltaics