Power System Energy Prediction Error Optimization Model Based on Self-Supervised Learning Temporal Convolutional Network

Read full text Summarise with VIRAT
Authors: Han Gong, Xiao Hu, Chenchen Qu, Xue Yang, Chao Lin, Wenyan Chen
Cited by: 0
References: 8

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Abstract Roughly capturing the general trend of load changes leads to large prediction errors. To address this, a power system power prediction error optimization model based on a self-supervised learning temporal convolutional network is proposed. A feature extraction unit containing modules such as self-attention is constructed to implicitly represent historical power interaction information through multi-dimensional transformation and calculation. Various operations are combined, and a mutual information maximization unit is constructed. After preprocessing and training, latent temporal features of power are mined. Based on a temporal convolutional network integrating multiple convolutions, an attention mechanism and autoregressive components are introduced, integrating linear and nonlinear components. The Dropout method is used to reduce overfitting, thus constructing a power prediction model. The nonparametric statistical kernel density estimation method is used to model the deviation probability distribution, calculate the deviation interval probability, analyze the effective deviation, and then adjust the model parameters accordingly, completing the construction of the power system power prediction error optimization model. Experiments show that the predicted curve of the proposed method has a high degree of fit with the actual curve. Key indicators such as MAE, RMSE, and MAPE are all controlled within 2%. Compared with the comparison methods, its residual curve fluctuates more closely around the zero value with small fluctuations, and the R 2 value is significantly higher, significantly reducing the prediction error.

Fields of Study & Subjects

Energy Load and Power Forecasting Stock Market Forecasting Methods Smart Grid and Power Systems

Publication details

Journal / source
Journal of Physics Conference Series
Volume
3198
Issue
1
Pages
012001–012001
Publication date
1 March 2026
Language
EN