An Enhanced Particle Swarm Optimization Long Short-Term Memory Network Hybrid Model for Predicting Residential Daily CO2 Emissions

This study aims to establish an accurate hybrid model for predicting residential daily carbon dioxide (CO2) emissions, offering essential theoretical insights and data support for decision-makers in the construction industry. A hybrid model named CRLPSO-LSTM was proposed, which integrates an enhanced particle swarm optimization (CRLPSO) algorithm with a long short-term memory (LSTM) network. The CRLPSO algorithm enhances population quality, diversity, and global search efficiency by introducing improved circle chaotic mapping, optimizing worst mutations, and incorporating the Lévy flight strategy. The performance of the CRLPSO algorithm was rigorously evaluated using 23 internationally recognized standard test functions. Subsequently, the CRLPSO algorithm was employed to optimize the parameters of the LSTM model. Experimental validation was performed on three datasets from China, the United States, and Russia, each exhibiting distinct emissions characteristics: China with high emissions and high volatility, the United States with medium emissions and medium volatility, and Russia with low emissions and low volatility. The results indicate that the CRLPSO-LSTM hybrid model outperformed other hybrid models in predicting residential daily CO2 emissions, as demonstrated by superior R2, MAE, and MSE metrics. This study underscores the effectiveness and broad applicability of the CRLPSO-LSTM hybrid model, offering a robust theoretical foundation and data support for advancing the sustainable development goals.