Carbon price prediction based on multidimensional association rules and optimized multi-factor LSTM model

Precise forecasting of carbon prices serves as a crucial reference for operational production and investment strategies, which assumes great implications for prompting the healthy evolution of the carbon market. This study puts forward a multi-factor hybrid prediction framework, MARM-WOA-LSTM, which amalgamates three frontier technologies to realize a dependable forecast of carbon prices in Hubei. First, multidimensional association rule mining (MARM) is exploited to quantitatively assess the association and influence degree among eight aspects of factors and carbon prices, after which the filtered high-correlation factors are fed into the forecasting model. Thereafter, the whale optimization algorithm (WOA) is implemented to autonomously search and optimize the hyperparameters of the long short-term memory (LSTM) network, thereby enhancing the forecasting effect further. The results illustrate strong correlations of diverse uncertainty indexes and agricultural prices with carbon prices, where geopolitical risks and corn prices weigh more heavily. Moreover, MARM not only effectively cuts down the count of variables but also markedly augments the predictive accuracy. Relative to the single-factor LSTM model, models without MARM and WOA, as well as other comparative models, the developed system exhibits remarkable performance, thus fully attesting to its broad prospect and practical value in the field of carbon price forecasting.