GNN-Hybrid: A Dynamic Residual Optimization Hybrid Learning Framework for Long-Term Sequence Prediction

View at publisher Summarise with VIRAT
Authors: Yuxin Zhang, Zitong Pan
Cited by: 0

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

In this paper, a hybrid prediction framework based on neural networks is proposed for the in-depth study of the multi-scenario sequence prediction problem. First, a gray prediction model based on the equal dimensional complementary mechanism (IDCM) is used to construct the base prediction, which solves the problem of cumulative error common to traditional methods in long-term prediction. In order to further optimize the prediction accuracy, residual correction is performed in combination with a multilayer perceptron (MLP), which enhances the fitting ability to nonlinear fluctuations through a three-layer structure with the ReLU activation function. Next, a GM(1,1) model is used for trend prediction, and after processing the sequences using the cumulative generation operation, the residuals are quadratically optimized by the MLP, which significantly improves the prediction accuracy. Finally, a hybrid framework combining Random Forest and XGBoost is constructed, in which the macro variables are extrapolated through a linear trend model with an exponential smoothing method, Random Forest captures the nonlinear relationships among variables, and XGBoost corrects the residuals, thus enhancing the stability and robustness of the model. Through dynamic updating and residual correction, the framework effectively balances the prediction of long-term trends and nonlinear fluctuations, and improves the overall accuracy and stability. Future research will further extend the applicability of the framework to support multi-scenario applications such as edge computing, and introduce more complex deep learning structures to enhance the prediction performance and the deployment capability of the framework. The method is highly interpretable and can provide high-precision prediction support in multiple fields.

Fields of Study & Subjects

Grey System Theory Applications Forecasting Techniques and Applications Stock Market Forecasting Methods

Publication details

Journal / source
IEEE International Conference on Power, Electronics and Green Energy (ICPEGE)
Pages
1697–1703
Publication date
28 January 2026
Language
EN