Data-driven regional characteristic analysis and partition prediction of support load in deep well and ultra-long working face
-
-
Abstract
Accurate prediction of hydraulic support load plays an important role in improving the adaptability of support and the stability of surrounding rock control, where high-quality and large-scale time series data and effective prediction methods are needed. However, the cyclic training and modeling of intercepting hundreds of sets of load data in the same time period consumes a lot of computing resources and takes a long time to train. And the heterogeneity of the stress environment of the overlying strata on the working face and the asynchronous caving step distance lead to different support loads in different areas of the working face. In view of the obvious different load in different areas of the working face caused by the long-term cyclic dynamic load and partition failure of the roof overlying rock of the fully mechanized mining face, and the problem that the load prediction of hydraulic support group under dynamic area cannot be realized, a novel predicting scheme of hydraulic support group load in working face with respect to regional characteristics is proposed. Specifically, meanshift adaptive clustering algorithm is used to realize the region division of fully mechanized mining face firstly then the regional characteristics of the working face are analyzed. Secondly, an attention-based LSTM algorithm combined with the production technology is proposed. Taking the regional support load as the input, a one-time multi-input and multi-output prediction model of regional support load is established, which verifies the prediction effectiveness of the proposed input-output feature engineering. Finally, based on the division results of the working face area, regionalized hydraulic support group load prediction models are established based on the proposed attention-based LSTM algorithm to achieve high-precision prediction of the hydraulic support load of the fully mechanized mining face. By considering the regional distribution characteristics of the working face and proposing a multi-input and multiple-output feature engineering, the hydraulic support group load prediction based on the dynamic update of the working face area can be realized, which can be used for the follow-up predicting analysis of the strata behaviors provides a basis for guiding the safe and efficient mining.
-
-