为克服传统预测模型结构单一、预测精度及稳定性不足等缺陷，提出多元体系组合预测模型的建模思路。首先，基于支持向量机、BP神经网络及ARMA模型3种单一预测模型，构建铁路隧道变形预测体系； 再以均方根误差、误差平方和及平均绝对误差等为评价准则或指标，构建各预测结果的误差评价体系，求解各单项预测模型的权值贡献指数，得到最优组合权值； 然后利用后验差检验、残差检验和关联度检验构建预测精度校验体系，对组合预测结果进行检验，评价预测模型的有效性； 最后，结合工程实例，对多元体系组合预测模型在特大断面隧道中的变形预测效果进行检验。结果表明： 多元评价体系组合模型预测相对误差值均小于2%，具有较高的预测精度，且较单一预测模型具有更高的预测精度，也一致通过相关检验，验证了多元体系组合预测模型的有效性。

Composite model of multiple evaluation system is put forward so as to overcome the shortcomings of traditional prediction models, such as single structure, poor prediction accuracy and poor stability. Firstly, a prediction system of railway tunnel construction deformation is established based on support vector machine (SVM), back propagation (BP) neural network, autoregressive moving average (ARMA) model. Then, the root mean square error, error sum of squares and mean absolute error are taken as evaluation indicators to establish error evaluation system of each predicting result, to solve weight contribution indicator of single prediction model and obtain optimal combination weights. And then, the prediction accuracy validation system is established based on after test, residual test and correlation test to check and evaluate the validity of the prediction model. Finally, the abovementioned model is applied to a superlarge crosssection railway tunnel. The results show that the relative error of the composite model of multiple evaluation system is less than 2%; the prediction accuracy is higher than that of single prediction model, and can meet the requirements of relevant tests.