边坡变形与地质活动,气候变化以及人工切坡等众多因素相关,是一种典型的非线性、非平稳随机过程,传统边坡变形预测方法存在预测精度低和泛化能力弱等问题。提出一种基于改进水循环算法(improved water cycle algorithm,IWCA)优化支持向...边坡变形与地质活动,气候变化以及人工切坡等众多因素相关,是一种典型的非线性、非平稳随机过程,传统边坡变形预测方法存在预测精度低和泛化能力弱等问题。提出一种基于改进水循环算法(improved water cycle algorithm,IWCA)优化支持向量回归(suppor vector regression,SVR)的边坡变形预测方法(IWCA-SVR)。SVR利用核函数将低维空间中的非线性边坡位移量数据映射到高维空间进行建模分析,不仅能够获得较高的预测精度,同时具有较强的泛化能力。由于SVR核参数和惩罚因子对预测性能影响较大,提出IWCA算法对其进行全局寻优,提升预测性能。最后将所提IWCA-SVR方法与粒子群优化的SVR(PSO-SVR)和极限学习机(ELM)等方法进行对比,结果表明IWCA-SVR在平均相对误差和均方根误差两项指标方面分别提升超过55.8%和54.9%,并且具有更强的泛化能力。展开更多
[Objective] The research aimed to study forecast models for frozen and melted dates of the river water in Ningxia-Inner Mongolia section of the Yellow River based on SVR optimized by particle swarm optimization algori...[Objective] The research aimed to study forecast models for frozen and melted dates of the river water in Ningxia-Inner Mongolia section of the Yellow River based on SVR optimized by particle swarm optimization algorithm. [Method] Correlation analysis and cause analysis were used to select suitable forecast factor combination of the ice regime. Particle swarm optimization algorithm was used to determine the optimal parameter to construct forecast model. The model was used to forecast frozen and melted dates of the river water in Ningxia-Inner Mongolia section of the Yellow River. [Result] The model had high prediction accuracy and short running time. Average forecast error was 3.51 d, and average running time was 10.464 s. Its forecast effect was better than that of the support vector regression optimized by genetic algorithm (GA) and back propagation type neural network (BPNN). It could accurately forecast frozen and melted dates of the river water. [Conclusion] SVR based on particle swarm optimization algorithm could be used for ice regime forecast.展开更多
基金Supported by National Nature Science Fund Item,China (51009065)Key Science and Technology Research Plan Program in Henan Province,China(112102110033)
文摘[Objective] The research aimed to study forecast models for frozen and melted dates of the river water in Ningxia-Inner Mongolia section of the Yellow River based on SVR optimized by particle swarm optimization algorithm. [Method] Correlation analysis and cause analysis were used to select suitable forecast factor combination of the ice regime. Particle swarm optimization algorithm was used to determine the optimal parameter to construct forecast model. The model was used to forecast frozen and melted dates of the river water in Ningxia-Inner Mongolia section of the Yellow River. [Result] The model had high prediction accuracy and short running time. Average forecast error was 3.51 d, and average running time was 10.464 s. Its forecast effect was better than that of the support vector regression optimized by genetic algorithm (GA) and back propagation type neural network (BPNN). It could accurately forecast frozen and melted dates of the river water. [Conclusion] SVR based on particle swarm optimization algorithm could be used for ice regime forecast.
