【摘要】 在地震综合预测投影寻踪研究工作中,投影寻踪回归算法是其中应用最多的一种方法.但一般投影寻踪回归算法构造技术较为复杂,采用多次局部光滑回归,计算量较大,外推较为繁杂,容易陷于局部解.在综合考虑传统投影寻踪回归算法特点的基础上,针对投影寻踪回归计算中存在的一些不利因素,给出了一定的解决思路:采用粒子群优化算法代替高斯-牛顿算法优化投影方向;采用厄米多项式代替分段线性光滑回归来拟合岭函数,以简化优化过程;参数优化无需分组,获得全局优化的岭函数.利用数值仿真技术进行基于粒子群优化算法与厄米多项式构建的投影寻踪回归模型建模能力与计算精度的检验,再将其应用于多维地震时间序列和一般多维无序地震样本回归综合建模预测中.通过计算和分析表明,基于粒子群优化算法与厄米多项式构建的投影寻踪回归模型具有简单、快速、有效的特点,在实际地震综合预测建模中取得了满意的效果,可作为地震预测的一种综合分析方法.更多还原

【Abstract】 In synthetic earthquake prediction,the projection pursuit (PP) algorithm is widely used,but the PP algorithm using partial smooth regression is quite complicate and time consuming,and its complicate extrapolation is apt to be trapped in local solution.To avoid these shortcomings,this paper uses particle swarm optimization approach,instead of Gauss-Newtonian algorithm,to optimize projecting direction,and applies Hermite polynomial,instead of piecewise linear regression,to fit ridge function.In our approach,the optimizing process is simplified,parameter optimization needs not to be grouped,and globally optimized ridge function is obtained.The modeling ability and calculation accuracy of the PP algorithm is tested by numerical simulations based on particle swarm optimization and Hermite polynomial fitting.Then the improved PP algorithm is applied to synthetic prediction modeling of poly-dimensional seismicity time series and general disordered seismic data samples.The applied result shows that the new PP regression model based on particle swarm optimization and Hermite polynomial fitting is simple in use,fast in computation and effective in optimization.This algorithm may serve as a useful tool in synthetic earthquake prediction study.更多还原