A prediction model for high-geothermal rockburst intensity using a genetic projection pursuit algorithm

High-temperature rockbursts pose a critical challenge in deep underground engineering and resource exploitation. Consequently, predicting high-geothermal rockbursts has become a key scientific objective. In this paper, a genetic projection pursuit algorithm (GPPA) is proposed for the prediction of high-geothermal rockbursts by introducing the coefficient K, and utilizing multiple empirical criteria (Wet index, σc/σt, σθ/σc, and σ1/σc). Four empirical criteria were statistically analyzed for 147 sets of rockburst cases, yielding accuracies of 40%, 39%, 46% and 29%, respectively. After the implantation of optimal segmentation, there was an enhancement in accuracy by 12%, 9%, 6%, and 19%, respectively. The σθ/σc criterion exhibited superior performance, with a baseline accuracy of 46%. The GPPA model was tested and validated using four characteristic parameters (Wet index, σc/σt, σθ/σc, and σ1/σc) as inputs, revealing that the error ranged between 0.07 and 0.41. Successful validation was performed in the Sangzhuling Tunnel (four slight rockbursts) and Qirehataer Diversion Tunnel (one moderate rockburst), which matched field observations. Consequently, the proposed model offers guidance for predicting high-geothermal rockburst hazards.