Article: Economic growth and carbon neutrality balancing: an optimal decision-making framework for emission reduction policies driven by reinforcement learning Journal: International Journal of Innovation and Sustainable Development (IJISD) 2026 Vol.20 No.8 pp.65 - 83 Abstract: To tackle dynamic carbon emission modelling and policy optimisation challenges amid climate change, this paper introduces a novel integrated intelligent decision-making framework named Autoformer-DQN-deep forest framework (ADDF), which synergises Autoformer, deep Q-networks (DQN), and Deep Forest (DF). Autoformer enables precise longterm trend and periodic pattern capture via deep decomposition and forecasting of carbon emission and economic indicator time series. DQN realises dynamic policy optimisation based on real-time states and predictions, while DF conducts multi-granularity feature extraction and importance analysis for interpretable modelling and key factor identification. Experiments show the framework outperforms traditional models in forecasting accuracy and policy effectiveness, achieving a 6.7% annual average carbon emission intensity reduction with only 0.9% gross domestic product (GDP) loss. Parameter sensitivity analysis and ablation studies confirm each module's necessity and synergistic mechanisms. Inderscience Publishers - linking academia, business and industry through research

Title: Economic growth and carbon neutrality balancing: an optimal decision-making framework for emission reduction policies driven by reinforcement learning

Authors: Jing Zhang; Tao Zhang

Addresses: School of Economics and Management, Liaoning Institute of Science and Technology, Benxi, Liaoning, 117004, China ' School of Economics and Management, Liaoning Institute of Science and Technology, Benxi, Liaoning, 117004, China

Abstract: To tackle dynamic carbon emission modelling and policy optimisation challenges amid climate change, this paper introduces a novel integrated intelligent decision-making framework named Autoformer-DQN-deep forest framework (ADDF), which synergises Autoformer, deep Q-networks (DQN), and Deep Forest (DF). Autoformer enables precise longterm trend and periodic pattern capture via deep decomposition and forecasting of carbon emission and economic indicator time series. DQN realises dynamic policy optimisation based on real-time states and predictions, while DF conducts multi-granularity feature extraction and importance analysis for interpretable modelling and key factor identification. Experiments show the framework outperforms traditional models in forecasting accuracy and policy effectiveness, achieving a 6.7% annual average carbon emission intensity reduction with only 0.9% gross domestic product (GDP) loss. Parameter sensitivity analysis and ablation studies confirm each module's necessity and synergistic mechanisms.

Keywords: carbon emission reduction; time series forecasting; reinforcement learning; feature extraction; dynamic decision-making.

DOI: 10.1504/IJISD.2026.155845

International Journal of Innovation and Sustainable Development, 2026 Vol.20 No.8, pp.65 - 83

Received: 12 Dec 2025
Accepted: 08 May 2026
Published online: 24 Aug 2026 *

Title: Economic growth and carbon neutrality balancing: an optimal decision-making framework for emission reduction policies driven by reinforcement learning

Keep up-to-date