基于BBO-SVM的大坝变形预测模型与性能验证

针对大坝监测数据小样本、高维度和非线性的特点，引入支持向量机( SVM) 机器学习方法，采用生物地理学优化算法( BBO) 优化其惩罚因子 c 和核函数参数 g，建立了基于 BBO-SVM 的大坝变形预测模型。结合 2011—2016 年水口大坝 4 个测点共 900 组环境量与效应量监测数据，对模型预测性能进行了验证，并将预测结果与 SVM、PSO-SVM 和 ABC-SVM 大坝变形预测模型进行对比。结果表明: 文中提出的 BBO-SVM 模型不仅预测精度高，且稳定性更好，4 个测点的均方根误差分别达到了 0. 332 0、0. 473 5、0. 405 7、0. 222 8，拟合优度分别达到了 0. 910 4、0. 961 0、0. 962 4、0. 956 9。本研究可提高大坝安全监测成果利用，对于大坝健康状态预测评估具有一定的工程指导意义。

BBO-SVM-based dam deformation prediction model and its performance verification

Aiming at the characteristics of small sample，high dimensionality and non-linearity of dam monitoring data，a support vector machine ( SVM) learning method is introduced and its penalty factor c and kernel function parameter g are optimized with a biogeographic-based optimization ( BBO) algorithm，and then a BBO-SVM-based dam deformation prediction model is established． In combination with 900 sets of the monitoring data of both the environment factors and the effect factors from four gauging points of Shuikou Dam from 2011 to 2016，the performance of the model is verified，from which the prediction results are compared with those from SVM，PSO-SVM and ABC-SVM dam deformation prediction models． The study result shows that the BBOSVM-based model does not only has a high precision，but also has a better stability，for which the root mean square errors at the four gauging points reach 0. 3320，0. 473 5，0. 4057 and 0. 222 8 with the fitting goodnesses of 0. 919 8，0. 961 0，0. 962 4 and 0. 956 9 respectively． The study can improve the utilization of dam safety monitoring results and has a certain engineering referential value for the prediction and assessment on the healthy status of dam．