鉴于岩爆机理的复杂性以及岩爆发生前后信号提取困难的现状，对高应力区进行岩爆倾向性预测研究具有现实意义。为提高岩爆预测的准确性，基于岩爆预测多维非线性的特点，选取4个影响岩爆发生的核心指标作为判决依据，结合粒子群优化算法（PSO）与径向基神经网络（RBF）建立了PSO-RBF神经网络岩爆预测模型。采用试错法确定隐含层节点数后，进一步利用国内外典型工程数据对模型参数隐含层基函数中心 c_i，隐含层节点宽度 {\sigma }_i以及隐含层与输出层间权重因子w进行学习优化以获取最优参数，并将所建立的模型应用于实际工程的岩爆倾向性预测。结果表明：利用该模型预测的岩爆等级与实际岩爆情况基本相符，相对误差率为10%，精度较以往预测方法有显著提高。

Rockburst is one of the typical dynamic disasters in the field of underground engineering.The forecast of rockburst tendency in high stress area is of great practical significance.Due to the complexity of rockburst mechanism，the existing prediction models were difficult to reflect the multi-dimensional nonlinear characteristics of rockburst，which result in the low rockburst tendency prediction accuracy.In order to forecast rockburst tendency more accurately, a new rockburst tendency forecast model was proposed by combining particle swarm optimization (PSO) with radial basis function neural network (RBF).After determining the number of the hidden layer nodes by trial-by-error method，the parameters of RBF neural network including the center of basic function，width of the hidden layer node and the output weights formed a multi-dimensional vector，and were optimized as population particle of the PSO algorithm for the purpose of getting the optimal solution within the scope of global solvable space.Further，this paper referenced domestic and foreign related literature and choose four major rockburst tendency indicators，including the uniaxial compressive strength，the rock stress index，the rock brittleness index and the elastic energy index.25 typical practical rockburst engineering cases were took as the learning samples to train the PSO-RBF neural network model parameters.Finally，the established model of PSO-RBF was applied to rockburst tendency prediction of practical engineering.The results show it is approved that the prediction results of the proposed model in this paper are approximately consistent with the actual rockburst status.The relative error rate of PSO-RBF prediction model is 10%，and the accurate is significantly improved than prevenient prediction method.The PSO-RBF neural network rockburst tendency prediction model has a certain practicality and could provide effective guidance for similar projects.