基于多源信息融合的碳价格预测模型  

Carbon Price Prediction Model Based on Multi-source Information Fusion

作　　者：郑一鸣 连子炎 兰子梦 ZHENG Yi-ming;LIAN Zi-yan;LAN Zi-meng(School of Internet,Anhui University,Hefei 230039,China)

机构地区：[1]安徽大学互联网学院,合肥230039

出　　处：《价值工程》2022年第28期94-96,共3页Value Engineering

基　　金：国家自然科学基金(72001001,71871001,71701001,71771001)资助;安徽省自然科学基金(No.2008085QG334);安徽大学大学生创新创业项目(202110357254)。

摘　　要：搜索引擎为大数据时代的时间序列预测提供了多源信息。传统的预测方法通常仅基于历史交易数据或影响因素,没有同时考虑多源信息的特点。因此,本论文为提高预测精度,充分利用各种类型的数据信息,筛选9个能够反映民众关注热度的关键词,并降维处理,结合历史交易信息,分别构建带有不同输入层的LSTM和LSSVM预测模型。本文以碳价格预测为例,结果表明,同时考虑关注热度和历史交易信息的LSSVM模型在预测精度和稳定性方面表现突出。Search engines now give multi-source information for time series prediction in the era of"big data".Traditional forecasting approaches are typically focused on past transaction data or influential factors,and do not take into account the unique properties of multi-source data.As a result,this article selects 9 keywords that might reflect the public's interest and decreases dimensionality.In order to increase prediction accuracy and full utilization of various sorts of data information,LSTM and LSSVM prediction models with multiple input layers are built using past transaction data.The results suggest that the LSSVM model,which uses both the attention heat and previous transaction information,performs better in terms of forecast accuracy and stability when it comes to carbon price prediction.

关 键 词：多源信息 碳价格 时间序列预测 LSSVM模型 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论]