Dynamic Matching of Photovoltaic Energy Supply and Demand on Highway Based on Transformer Method of Deep Learning

Abstract

In the era of rapid transportation electrification, the integration of power system automation and transportation system networking promotes the construction of comprehensive integration of the highway industry. Due to the huge project and the difficulty of construction, traditional highway construction cannot fully consider the supporting construction of electrification and networks and often lays monitoring and network facilities near the power facilities around the path. With the maturity of photovoltaic and other new energy technologies, and considering the increasing demand for modern highways, we put forward the design requirements and transformation standards of energy-supporting facilities. Based on the whole path, the power-supporting projects are realized to establish a new highway project with electrification, networking, and whole-process monitoring. This study addresses the issue of supply–demand mismatch in photovoltaic energy by proposing an integrated nonlinear regression framework based on the Transformer architecture to impute missing data. Considering the impact of energy storage conversion energy loss on the calculation effect of the model, and the input unit extracts the global degradation external characteristics, the long linear regression model is used to design the potential of recurrent neural network (RNN) shallow feature extraction. Multiple parallel RNN-transformer (RNN-T) are set to extract the external characteristics of power supply and demand balance without sharing the relevant parameters. The experimental process in the simulation platform built has verified that the model established in this study reduces the complexity and eases the accuracy error caused by data calculation. While in the matching ability and prediction accuracy, the test results are significantly better than the traditional method because its overall performance is improved by 25.2% and 29.7%, respectively, which supports that the improved model can learn the relationship between multilayer heterogeneous data simultaneously. Therefore, it improves the model’s deep learning ability in the photovoltaic power supply and demand relationship model. High-precision prediction effect is also achieved, which has great market value and practical significance for future road development.