Abstract
The mining roadways of fully mechanized mining faces are susceptible to complex disturbances, leading to deformation of the surrounding rock, breaking the mechanical balance of the advanced support section of the mining roadway, and potential roof safety incidents and equipment damage. Accurate prediction of the spatial attitude of the advanced hydraulic support group can provide a foundation for dynamically adjusting the spatial attitude to accommodate the deformation of the roadway's surrounding rock. The selection of training parameters in the conventional Long Short-Term Memory (LSTM) network is often random and involves a significant amount of effort, which further limits prediction accuracy and real-time performance. Based on this, the WOA algorithm was utilized to search for the optimal number of neurons in the hidden layer and the learning rate. A spatial attitude prediction method for the advanced hydraulic support group based on WOA-LSTM was proposed. Relying on the experimental monitoring platform, the attitude parameters of key nodes during the movement of the support group were obtained, and comparative experiments were carried out. The results indicate that, with 3,200 training samples for the top beam pitch angle and 600 iterations, the Mean Absolute Error (MAE) of the WOA-LSTM prediction model is 0.18°, the Root Mean Squared Error (RMSE) is 0.23°, and the Mean Absolute Percentage Error (MAPE) is 1.3%. Compared to the traditional LSTM model, these three error metrics are reduced by 0.2°, 0.2°, and 1.6%, respectively. This improvement enhances the accuracy and parameter optimization efficiency of the advanced support attitude prediction model, thereby providing robust theoretical and technical support for the intelligent, safe, and efficient mining operations of the advanced coupling support system.