SimVGNets: Similarity-Based Visibility Graph Networks for Carbon Price Forecasting
Shengzhong Mao, Xiao‐Jun Zeng
Abstract
In response to global warming, carbon trading market emerges to reduce carbon emissions. However, uncertain fluctuations and complicated price mechanisms in the market have changed the characteristics of carbon data. Existing methods cannot track underlying structures among carbon data streams, failing to explain high variations and predict carbon prices accurately. In this case, how to predict carbon price is a problem demanding new and prompt solutions. Unlike existing predictive mechanisms with fixed structure, in this work, we investigate carbon price from a novel perspective of visibility graph networks, developing a variant forecasting framework based on visible relationships. Firstly, carbon prices are mapped into networks where data are transformed into corresponding nodes utilizing visibility algorithm. Then a novel similarity index is proposed to measure similarity distribution of nodes converted from carbon data, which extracts past and recent price information rather than only considering recent input, resulting in the loss of past knowledge. The forecasting model is finally established based on mutual visibility machamism and similarity of constructed carbon network. To verify the proposed method, experiments are conducted on real carbon datasets, and the results demonstrate the method is superior to existing models, suggesting the predictive potentials of graph network science. It is hoped that this work can act as a bridge between carbon data and networks, motivating further research on carbon price forecasting from graph network views.