Abstract: The emotional information contained in news textual data affects the decision-making process of managers and investors. Hence, it can be used to improve the forecasting accuracy of carbon price interval. In this paper, a two-layer decomposition model driven by textual data is proposed. Relevant news texts associated with the carbon emission in network are climbed, of which the sentiment scores are calculated, at first. Secondly, the carbon price interval and sentiment data are decomposed in EMD-WTS decomposition method, and the results are reconstructed into high, low frequency and trend items corresponding to time series based on SE algorithm. Eventually, the LSTM network is used to predict the obtained sequences, and the final prediction result is gained after superposition and integration. The empirical and comparative experiments show that the proposed model can not only effectively utilize the multi-source information, but also fully mine the detailed features in the time series of complex fluctuation of carbon price range, thus achieving a more significant prediction effect.

Key words: carbon price forecast, text data, two-layer decomposition, interval prediction, LSTM