Ultra-Accurate CO2 Emission Forecasting for the Cement Industry Using FbOA-Optimized Neural NODE Models
The cement sector is a linchpin of global infrastructure and is also one of the world’s most significant industrial sources of CO2 emissions, accounting for about 7-8% of anthropogenic emissions. The proper prediction of cementgenerated emissions is thus essential for designing mitigation strategies, planning industrial transitions, and evaluating progress toward carbon-neutrality goals. This paper proposes a new time-series forecasting model that combines Neural Ordinary Differential Equations (NODE) with the Football Optimization Algorithm (FbOA) to enable automated, data-driven hyperparameter optimization. The performance of NODE is compared with Seq2Seq and ConvLSTM models for global CO2 emis-sions from cement production in baseline settings, and subsequently metaheuristically optimized using FbOA, PSO, MVO, WOA, and GA. The baseline experiments demonstrate that NODE, with an MSE of 0.00745, RMSE of 0.0863, MAE of 0.0515, and high levels of agreement (NSE = 0.91, WI = 0.905), outperforms both Seq2Seq and ConvLSTM. Upon hyperparameter optimization, the FbOA + NODE combination achieves significant performance improvement, with MSE of 3.95×10−7 , RMSE of 6.28×10−3 , and MAE of 3.42 × 10−4 , r = 0.977, R2 = 0.973, NSE = 0.975 and WI = 0.98. Competing optimizers (PSO, MVO, WOA, GA) also improve NODE’s performance, and across all important metrics, they are consistently below FbOA. The findings indicate that integrating NODE and FbOA yields an accurate, stable, and computationally inexpensive model for predicting cement-associated CO2 emissions, offering a potential avenue for data-driven climate and industrial planning.
Volume & Issue
Vol. Volume 5 / Iss. Issue 1