Abstract: Schedule control is a key task of hydropower project management; A timely summary of schedule management information helps formulate and adjust the project schedules. In hydropower project construction, progress information is often presented in semi-structured or unstructured text forms, thereby inducing a difficulty in information extraction. An urgent issue is how to realize automation and intelligent mining of the text information of hydropower project progress. This paper presents a new intelligent extraction method of hydropower project schedule information based on an improved LDA method for intelligent extraction of the key information from schedule management texts. This method, based on the Gibbs sampling mechanism of the traditional LDA model, takes full consideration of the association relationship between words, and improves semantic association between words, closeness between words, and accuracy in the description of topic words. It is applied to practical hydropower projects to analyze 221 weekly reports on construction supervision, and extracted the key words of 12 themes, with the main and secondary processes extracted through calculations. Results show that our improved LDA topic model is better than the traditional LDA and it helps improve word extraction efficiency and information mining efficiency for hydropower construction.

Key words: hydropower project, construction progress, keyword extraction, improved-LDA topic model, co-occurrence, text intelligence analysis