Real time prediction technology for load bearing effect of hydraulic support after initial support based on data-driven approach

In the actual production of coal working face, due to the roof conditions, mining influence, hydraulic support attitude and their mutual influence, the column pressure after the initial support of the hydraulic support may change, thus affecting the load bearing effect after the initial support of the support. The pressure failure of hydraulic supports after initial support may lead to problems such as coal wall lining, inter frame leakage, support leaning forward, and overturning. At present, most of the control strategies for the initial support force of hydraulic supports in intelligent mining working faces directly determine whether the column pressure reaches the rated initial support force when lifting the column. It lacks consideration for the load bearing effect caused by the change in column pressure after initial support. In order to solve the above problems, a real-time prediction method for the load bearing effect of hydraulic support after initial support based on column pressure data-driven approach is proposed. The method divides the historical data of column pressure within 3 minutes after initial support of hydraulic supports into 6 typical working conditions. The method classifies the 6 typical working conditions into effective load bearing or failure load bearing according to the different load bearing effects after initial support. Through correlation analysis, five feature factors affecting the load bearing effect of the support after initial support are identified. The method manually annotates the effective or failed load bearing samples of the column, and extracts features. The method inputs the feature values into four different algorithms: decision tree, random forest, support vector machine, and K-nearest neighbor (KNN) to establish classification models. After comparative analysis, the random forest model has the highest precision, reaching 95.60%, which basically meets the accuracy requirements of the model application. A real-time prediction model for the load bearing effect of hydraulic supports after initial support based on random forest is established. On this basis, a real-time prediction system for the load bearing effect of hydraulic supports after initial support is developed and deployed to coal mine sites. After continuous operation for 25 days, the system collects the column pressure within 3 minutes after the initial support of the hydraulic support and could output the load bearing effect of the hydraulic support within 5 seconds. The accuracy of the prediction results compared with the actual operation records is 82.48%, indicating that the system has high accuracy and stability in predicting the load bearing effect.