Carbon price fluctuation forecasting using an adaptive dual-channel residual attention neural network optimized with white shark optimizer and blockchain-based data provenance

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:Provides a novel hybrid model combining neural networks, metaheuristic optimization, and blockchain for carbon price prediction with rigorous empirical validation.

🏢実務担当者:Offers a high-accuracy forecasting tool usable by trading firms for risk management and arbitrage in regulated carbon markets.

🏛政策担当者:Highlights how reliable price forecasts and data provenance can support market monitoring and regulatory oversight.

炭素市場は排出削減と持続可能な経済移行において重要な役割を果たすが、炭素価格系列は高いボラティリティ、非線形ダイナミクス、複雑なクロス特徴量間の相互作用を示すため、正確な予測は困難である。現在の予測モデルは、時間依存性、クロス指標間の関係、最適化効率、およびデータ管理の信頼性を効果的に同時に扱うことができない。これらの欠点に対処するため、本研究論文はADRGPNN-WSOと名付けられた統合予測アプローチを提案する。このフレームワークは、ノイズ低減にFuzzy K-Top Matching Value (FKMV)、非線形の時間的およびクロス特徴量の挙動をモデル化するAdaptive Dual-Channel Residual Group Pulse-Coupled Neural Network (ADRGPNN)、重みを適応的に最適化するWhite Shark Optimizer (WSO)、そして安全で検証可能なデータ来歴を提供するDecentralized Chameleon Hash Functionsを備えた編集可能なBlockchain (RB-DCHF)を使用する。湖北、上海、深圳の各取引所における炭素取引データを用いた実験では、提案モデルが従来モデルおよびtransformerベースのモデルと比較して、MSE 0.0189、MAE 0.0987、RMSE 0.1375、MAPE 5.88、R² 0.942という優れた予測性能を示した。統計的有意性検定により、性能改善の強固さがさらに実証された。提案されたフレームワークは、規制された炭素市場における取引企業、リスク管理、および政策立案を支援するために、信頼性が高くガバナンスに配慮した予測に基づくソリューションを提供する。

AI 翻訳（deepseek-v4-flash）。 正確を期す場合は下の原文を参照してください。

Carbon markets play a critical role in emission reduction and sustainable economic transition; however, carbon price series exhibit high volatility, nonlinear dynamics, and complex cross-feature interactions, making accurate forecasting challenging. Current forecasting models cannot effectively address both time dependence, cross-indicator relationships, and optimization efficiency, as well as data control reliability. To address these shortcomings, this research paper proposes an integrated forecasting approach, named ADRGPNN-WSO. The framework uses Fuzzy K-Top Matching Value (FKMV) to reduce noise, an Adaptive Dual-Channel Residual Group Pulse-Coupled Neural Network (ADRGPNN) to model nonlinear temporal and cross-feature behaviors, and the White Shark Optimizer (WSO) to adaptively optimize weights, a redactable Blockchain with Decentralized Chameleon Hash Functions (RB-DCHF) to provide secure and verifiable data provenance. The experiment with carbon trading data at the Hubei, Shanghai, and Shenzhen exchanges shows that the proposed model has better forecasting performance, with an MSE of 0.0189, MAE of 0.0987, RMSE of 0.1375, MAPE of 5.88, and R 2 of 0.942, compared with the conventional and transformer-based models. Statistical significance tests further demonstrate the strength of the performance improvements. The suggested framework offers a predictive solution based on reliable, governance-conscious forecasting to aid trading companies, risk management, and policy-making in regulated carbon markets.

• semanticscholar https://doi.org/10.1038/s41598-026-43184-6first seen 2026-05-15 17:18:06 · last seen 2026-08-12 05:16:06