Search

Rockburst can cause significant damage to infrastructure and equipment, and pose a substantial risk to the safety of mine workers. Effective prediction of short-term rockburst damage can greatly assist in the implementation of preventive measures and risk mitigation strategies. Therefore, the development of accurate and reliable models for rockburst damage prediction is of great importance. This study has developed a novel hybrid model, based on eXtreme Gradient Boosting (XGBoost) and Sand Cat Swarm Optimization (SCSO), for predicting the scale of short-term rockburst damage. The model, which leveraged a data set of 254 rockburst damage cases collected from rockburst event reports across multiple mines in Australia and Canada, was comprehensively evaluated via non-parametric statistical tests and compared with other prevalent models. The results demonstrate that the SCSO–XGBoost model achieved a test accuracy of 88.46% for the rockburst damage scale, presenting the best comprehensive performance among the models in this study. Furthermore, the SCSO showed significant advantages in reducing overfitting, greatly enhancing the generalization capability of the XGBoost model. Lastly, various model explanation methods were employed to analyze the contribution of input variables. It was found that Peak Particle Velocity (PPV), Geological Structure (GS), Ground Support System Capacity (GSSC), and Stress Condition Factor (SCF) provide the most significant contributions to predicting rockburst damage. This study explains the decision-making rationale behind the SCSO–XGBoost model's identification of different damage scales from both a holistic and sample perspective. In addition, a graphical user interface has been developed and demonstrated for testing purposes. In conclusion, as a study integrating a new SCSO metaheuristic method with the XGBoost ensemble model for rockburst damage identification, the established SCSO–XGBoost model effectively predicts the types of short-term rockburst damage, offering superior predictive performance and explainability compared to previous works. Highlights: This study presents a novel hybrid ensemble model for short-term rockburst damage prediction, combining a new metaheuristic method with the XGBoost model. It provides a comprehensive and detailed discussion on the optimization process and performance evaluation of the hybrid model. The model demonstrates significant advantages in reducing overfitting and enhancing its generalization capability. It exhibits excellent predictive performance and explainability in forecasting the types of short-term rockburst damage. [ABSTRACT FROM AUTHOR]

In this paper, a modified backpropagation neural network is combined with a chaos-search genetic algorithm and simulated annealing algorithm for very short term electrical energy demand prediction in deregulated power industry. Multiple modifications are carried out on the conventional backpropagation algorithm such as improvements in the momentum factor and adaptive learning rate. In the hybrid scheme, the initial parameters of the modified neural network are optimized by using the global search ability of genetic algorithm, improved by cat chaotic mapping to enrich its optimization capability. The solution set provided by the optimized genetic algorithm is further improved by using the strong local search ability of simulated annealing algorithm. The real data of New South Wales, Australian grid, is used in the experimentation for 1-h-ahead forecast with an emphasis on data analysis and preprocessing framework. The correlation analysis is used for the identification and selection of the most influential input variables. The simulation results reveal that the proposed combination technique effectively enhanced the prediction accuracy as compared to the available conventional methods. The prediction of 1-h-ahead load demand is critically important for decision-making response of the modern smart grid system. The acceptable precision of the proposed model concludes that it can be applied in the smart grid to enhance its demand responsiveness and other intelligent features. [ABSTRACT FROM AUTHOR]