摘要
选用目前非线性方法中的2 个研究热点———BP神经网络和模拟退火算法,用优化后的BP神经网络为主框架,结合位场反演的特点,在反演过程中引入模拟退火算法,这样既利用了BP神经网络具有指导学习的功能来提高局部搜索性能,也利用了模拟退火算法的概率突跳性来实现最终的全局收敛性,从而提高了反演的速度和精度。通过模型试验验证了其有效性之后,将该方法应用于珠江口盆地潮汕坳陷的重力资料反演,结果较好地反映了剖面的地质情况。
Two nonlinear methods, Back-Propagation Neural Network ( BPNN) and Simulated Annealing Algorithm (SA), which are current research hotspots, are chosed. A hybrid optimization inversion method is obtained in the course of potential inversion by introducing SA into BPNN according to the characters of potential inversion. This method takes advantages of both BPNN with features of local research and SA with features of global research, and improves the rate and precision of the inversion. The method is used in the gravity inversion of Chaoshan col after its validity is validated by model test.
出处
《热带海洋学报》
CAS
CSCD
北大核心
2005年第2期86-93,共8页
Journal of Tropical Oceanography
基金
国家重点基础研究发展规划项目(G20000467 02)
关键词
混合最优化
重力反演
潮汕坳陷
hybrid optimization
gravity inversion
Chaoshan col
