基于深宽网络模型的碳价预测——基于LSTM-BLS的湖北碳排放权交易价格预测
Based on the Deep and Broad Network Model for Carbon Price Forecasting—Prediction of Hubei Carbon Emission Rights Trading Price Based on LSTM-BLS

近年来，随着温室效应的愈发严重，极端天气事件频发，控制碳排放成为了当前全世界人们最为紧迫的共同任务。碳交易市场作为控制减碳的重要金融工具，其健康发展依赖于稳定的碳价。准确地预测碳价不仅有助于投资者进行投资，也有助于政策制定者制定合理的制度。本研究提出了一种结合长短期记忆网络(LSTM)和宽度学习(BLS)的LSTM-BLS预测模型，以提高碳价预测的准确性和效率。通过对我国8个地方碳交易中发展时间较长，交易量较多的湖北碳交易所的碳价数据进行实证分析，发现引入宽度学习之后的模型预测效果优于基线模型和单一长短期记忆神经网络模型。研究结果为碳市场参与者提供了更可靠的预测工具。然而，本研究也具有一定的局限性，未来工作中将进一步考虑影响碳价的多因素数据，不断对模型进行优化。
In recent years, as the greenhouse effect has intensified, extreme weather events have become more frequent, making carbon emission control the most urgent common task for people worldwide. The carbon trading market, as an important financial instrument for reducing carbon emissions, relies on stable carbon prices for its healthy development. Accurate carbon price forecasting not only helps investors in their investment decisions but also assists policymakers in establishing reasonable systems. This study proposes an LSTM-BLS prediction model that combines Long Short-Term Memory Networks (LSTM) with Broad Learning System (BLS) to improve the accuracy and efficiency of carbon price forecasting. Through empirical analysis of the carbon price data from the Hubei Carbon Exchange, one of the eight local carbon trading markets in China with a longer development history and higher trading volume, we found that the model’s predictive performance, after incorporating width learning, is superior to that of the baseline model and the single LSTM model. The research results provide carbon market participants with a more reliable forecasting tool. However, this study also has certain limitations, and future work will further consider multi-factor data that affects carbon prices, continuously optimizing the model.