[論文レビュー] Stable Time Series Prediction of Enterprise Carbon Emissions Based on Causal Inference

Zitao Hong, Zhen Peng+1 more|arXiv (Cornell University)|Jan 31, 2026

Energy, Environment, Economic Growth被引用数 0

ひとこと要約

The paper presents Stable-CarbonNet, a cross-environmental temporal prediction framework that combines causal stable learning with adaptive temporal corrections to predict enterprise carbon emissions under distribution shifts, improving robustness across regions, industries, and policies.

ABSTRACT

Against the backdrop of ongoing carbon peaking and carbon neutrality goals, accurate prediction of enterprise carbon emission trends constitutes an essential foundation for energy structure optimization and low-carbon transformation decision-making. Nevertheless, significant heterogeneity persists across regions, industries and individual enterprises regarding energy structure, production scale, policy intensity and governance efficacy, resulting in pronounced distribution shifts and non-stationarity in carbon emission data across both temporal and spatial dimensions. Such cross-regional and cross-enterprise data drift not only compromises the accuracy of carbon emission reporting but substantially undermines the guidance value of predictive models for production planning and carbon quota trading decisions. To address this critical challenge, we integrate causal inference perspectives with stable learning methodologies and time-series modelling, proposing a stable temporal prediction mechanism tailored to distribution shift environments. This mechanism incorporates enterprise-level energy inputs, capital investment, labour deployment, carbon pricing, governmental interventions and policy implementation intensity, constructing a risk consistency-constrained stable learning framework that extracts causal stable features (robust against external perturbations yet demonstrating long-term stable effects on carbon dioxide emissions) from multi-environment samples across diverse policies, regions and industrial sectors. Furthermore, through adaptive normalization and sample reweighting strategies, the approach dynamically rectifies temporal non-stationarity induced by economic fluctuations and policy transitions, ultimately enhancing model generalization capability and explainability in complex environments.

研究の動機と目的

• Identify long-term, cross-environmental causal drivers of enterprise carbon emissions that remain stable across regions, industries, and policy regimes.
• Develop a stable learning framework that extracts invariant features and uses sample reweighting to reduce spurious correlations.
• Incorporate adaptive normalization and temporal weighting to handle non-stationarity and policy/policy-shock effects.
• Connect stable causal features with temporal dynamics to improve cross-environment generalization and interpretability for decision-making.

提案手法

• Formulate the prediction problem with environment-specific emissions Y(i,t) and features X(i,t) under environment e_t.
• Decompose the emission process into stable causal features f*(X^c) and environment-specific disturbances h^(e)(X^s).
• Use a shared representation Phi_theta(X) with linear head w to enforce cross-environmental stability, optimizing min_theta,w sum_e R_e(w^T Phi_theta(X)) + lambda S(theta,w).
• Introduce a gradient-consistency stability term S(theta,w) that minimizes the squared norm of gradients across environments.
• Apply temporal weighting omega_t and adaptive normalization N_e to address non-stationarity and cross-environment scale differences, yielding a final objective combining stability and temporal correction.
• Provide an end-to-end training scheme with backpropagation where lambda controls the trade-off between stability and fit.

実験結果

リサーチクエスチョン

• RQ1 What stable causal drivers of enterprise carbon emissions persist across cross-environment settings (regions, industries, policies)?
• RQ2 Can a stable learning framework combined with temporal adaptation achieve robust prediction under distribution shifts compared to standard ERM models?
• RQ3 How do adaptive normalization and sample reweighting influence cross-environment generalization and interpretability of the model?
• RQ4 What is the impact of the stability parameter lambda on predictive invariance versus fitting accuracy?
• RQ5 Do multi-environment data improve understanding of cross-regional and cross-industry emission dynamics for policy and trading decisions?

主な発見

• The proposed method yields superior robustness and accuracy in cross-regional and cross-industry prediction tasks.
• The approach achieves reductions in average prediction error of approximately 10% to 15% compared with conventional models.
• Stable features identified are robust against external perturbations and demonstrate long-term stable effects on emissions.
• Adaptive normalization and sample reweighting effectively correct for temporal non-stationarity induced by economic fluctuations and policy transitions.
• Cross-environment constraints connect causal invariance with temporal dynamics to enable more reliable carbon forecasting for enterprise decisions.