在“双碳”战略目标和“构建以新能源为核心的新型电力系统”的大背景下，能源结构加速调整，以天然气、电力等清洁能源发电为主的能源新消费在稳步增长，但煤炭资源作为我国最丰富的能源，对保障能源安全、推动经济发展具有至关重要的作用。准确预测煤炭价格有助于稳定市场运行、帮助企业提供科学决策和支持。以往煤价预测模型多集中于传统宏观经济指标，但由于政治地缘、气候变化和政府政策等不确定因素的作用和大量文本数据的生成，情感因素对煤炭价格预测的影响性逐渐突显。情感分析作为一种自然语言处理技术，能够有效挖掘文本数据中的情感信息，为煤价预测提供更准确及时的信息。因此，本文引入情感分析技术，融合煤炭非情感影响因素和煤炭新闻信息，构建煤炭价格组合预测模型，为煤炭企业规避市场风险提供参考依据。

本文以秦皇岛港山西优混5500K动力煤的日度价格为研究对象，首先，通过对煤炭价格影响因素已有文献的研究，总结了煤炭价格波动的非情感影响因素和情感影响因素，建立了系统的影响煤炭价格的指标体系。其次，使用 BERT 模型代替传统的 TF-IDF、Word2vec 模型表示词向量，提出基于BERT-BiGRU的情感分析模型，对煤炭市场新闻进行情感分类并量化得出情感指数序列。然后构建了基于单变量的EEMD-LSTM煤炭价格组合预测模型和融合煤炭市场新闻的EEMD-BERT-BiGRU-ATT-LSTM煤炭价格组合预测模型，并对其进行检验分析。研究结论如下：（1）本文提出的情感分析模型对煤炭市场的情感分类效果较好，煤炭市场的新闻文本数据为煤炭价格预测提供信息，提高了煤炭价格的预测精度；（2）基于单变量的煤价预测模型中通过时间序列分解方法，极大提高了煤炭价格预测的准确性，解决了原始序列非线性造成的预测误差；（3）融合多因素的煤价预测模型表现最好，不同时间步长对煤炭价格预测的准确率不同，其中，单步预测的效果最好，MAE，RMSE，MAPE以及R2四个评价指标表现最优。本文研究对预测煤炭价格提供了新的思路，加入新闻文本数据预测煤炭价格有意义，能更好地帮助企业决策和政府调整政策。

Under the background of the "dual-carbon" strategic goal and "building a new type of power system with new energy as the core", the energy structure is accelerating the adjustment, and the new consumption of energy is mainly based on natural gas and clean energy of electric power generation is growing steadily, but the coal resources, as the most basic and abundant energy source in China, play a vital role in guaranteeing energy security and promoting economic development. However, coal resources, as the most basic and abundant energy source in China, play a crucial role in guaranteeing energy security and promoting economic development. Accurate prediction of coal prices is important to stabilize market operations and provide scientific decision-making and support. In the past, prediction models mostly focused on traditional macroeconomic indicators, but due to the role of uncertain factors such as political geography, climate change and government policies, and the generation of a large amount of text data, the influence of sentiment indicators on coal price prediction has gradually come to the fore. Sentiment analysis, as a natural language processing technology, can effectively mine the sentiment information in text data to provide more accurate and timely information for coal price prediction. Therefore, this paper introduces sentiment analysis technology, non-emotional influencing factors and coal news information, and constructs a coal price combination prediction model to provide a reference basis for coal enterprises to avoid market risks.

This paper takes the daily price of Shanxi Premium Blend 5500K Power Coal in Qinhuangdao Port as the research object, firstly, through the research on the influence factors of coal price in the existing literature, it summarizes the non-sentiment influencing factors and sentiment influencing factors of coal price fluctuation, and establishes a systematic indicator system for influencing coal price. Secondly, the BERT model is used instead of the traditional TF-IDF and Word2vec models to represent word vectors, and a sentiment analysis model based on BERT-BiGRU is proposed to classify the sentiment of coal market news and quantify the sentiment index sequence. Then, the univariate EEMD-LSTM coal price combination prediction model and the EEMD-BERT-BiGRU-ATT-LSTM coal price combination prediction model based on coal market news were constructed and analyzed. The conclusions of the study are as follows: (1) the sentiment analysis technique proposed in this paper is effective in classifying the sentiment of the coal market, and the news text data of the coal market provides information for coal price prediction, which improves the prediction accuracy of the coal price; (2) In the univariate coal price prediction model, the time series decomposition method greatly improves the accuracy of coal price prediction and solves the prediction error caused by the nonlinearity of the original series; (3) Coal price forecasting models that integrate multiple factors perform best. The accuracy of coal price prediction varies with different time steps, among which, the single-step prediction is the best, and the four evaluation indexes of MAE, RMSE, MAPE and R2 are the best performance. The research in this paper provides new ideas for predicting coal prices, and it is meaningful to add news text data to predict coal prices, which can better help enterprises and the government to make decisions and adjust policies.