基于PSO优化的RBF神经网络在地震测井联合反演中的应用
|
摘要
目前,钻井地质特征参数的获得主要依赖于地震、测井资料,对待钻井而言,则只有地震信息。而若缺乏详细的地质信息,利用地震信息很难精确地推算各种地质参数。可首先利用已钻井地震信息和测井信息的映射关系,结合待钻井的地震信息,来预测待钻井的测井信息。采用PSO优化的RBF神经网络算法进行地震测井反演,并将该算法应用于准噶尔盆地永字号井。该算法与最小二乘RBF神经网络算法和梯度下降RBF神经网络算法相比,在平均绝对误差、平均相对误差、最大误差、相关系数、数据方差以及收敛速度等方面都是最优的。
To obtain geologic characteristic parameters mainly depended on seismic and logging data, while for well drilling, only seismic information was available. It was difficult to predict various geologic data without detail geologic information. In combination of seismic information with the data of well drilling, the mapping relation between the seismic information and logging information could be firstly used to predict logging information in drilling, PSO optimized RBF neural network algorithm was used for seismic inversion and the algorithm was used in the Wells Yong of Junggar Basin. Comparing with least square RBF and gradient descent RBF neural network methods, it is optimized in areas of absolute error, average relative error, maximum error, correlation coefficient, variance and convergence rate.
To obtain geologic characteristic parameters mainly depended on seismic and logging data, while for well drilling, only seismic information was available. It was difficult to predict various geologic data without detail geologic information. In combination of seismic information with the data of well drilling, the mapping relation between the seismic information and logging information could be firstly used to predict logging information in drilling, PSO optimized RBF neural network algorithm was used for seismic inversion and the algorithm was used in the Wells Yong of Junggar Basin. Comparing with least square RBF and gradient descent RBF neural network methods, it is optimized in areas of absolute error, average relative error, maximum error, correlation coefficient, variance and convergence rate.
引文
[1]刘争平.人工神经网络在测井-地震资料联合反演中的应用研究[M]·北京:科学出版社,2003·
[2]Kennedy J,Eberhart R C·Particle swarmopti mization[J]·Proceedings of the IEEE International Conference on Neural Networks,Perth WA Australia,1995·1942~1948·
[3]Shi Y,Eberhart R·A modified particle swarmopti mizer[J]·Proceedings of IEEE Congress on Evolutionary Computation,Anchor-age,Alaska,1998·69~73·
[4]Guerra F A,Coelho L S·Radial basis neural network learning based on particle swarm opti mization to multistep prediction of chaotic Lorenz's system[J]·Proceedings of the Fifth International Conference on Hybrid Intelligent Systems,Rio de Janciro Brazil,2005·521~523·
[5]刘鑫朝,颜宏文·一种改进的粒子群优化RBF网络学习算法[J]·计算机技术与发展,2006,16(2):185~187·
[6]樊玮.粒子群优化方法及其实现[J]·航空计算技术,2004,34(3):39~42·
[7]谢晓峰,张文俊,杨之廉.粒子群算法综述[J]·控制与决策,2003,18(2):129~134·
[8]王炜,吴耿锋,张博锋等.径向基函数(RBF)神经网络及其应用[J]·地震,2005,25(2):19~25·
[9]张小军,冯宏伟.基于径向基函数神经网络的车牌识别技术[J]·西北大学学报(自然科学网络版),2006,4(2):21~24·
[10]魏海坤.神经网络结构设计的理论与方法[M]·北京:国防工业出版社,2005·
[2]Kennedy J,Eberhart R C·Particle swarmopti mization[J]·Proceedings of the IEEE International Conference on Neural Networks,Perth WA Australia,1995·1942~1948·
[3]Shi Y,Eberhart R·A modified particle swarmopti mizer[J]·Proceedings of IEEE Congress on Evolutionary Computation,Anchor-age,Alaska,1998·69~73·
[4]Guerra F A,Coelho L S·Radial basis neural network learning based on particle swarm opti mization to multistep prediction of chaotic Lorenz's system[J]·Proceedings of the Fifth International Conference on Hybrid Intelligent Systems,Rio de Janciro Brazil,2005·521~523·
[5]刘鑫朝,颜宏文·一种改进的粒子群优化RBF网络学习算法[J]·计算机技术与发展,2006,16(2):185~187·
[6]樊玮.粒子群优化方法及其实现[J]·航空计算技术,2004,34(3):39~42·
[7]谢晓峰,张文俊,杨之廉.粒子群算法综述[J]·控制与决策,2003,18(2):129~134·
[8]王炜,吴耿锋,张博锋等.径向基函数(RBF)神经网络及其应用[J]·地震,2005,25(2):19~25·
[9]张小军,冯宏伟.基于径向基函数神经网络的车牌识别技术[J]·西北大学学报(自然科学网络版),2006,4(2):21~24·
[10]魏海坤.神经网络结构设计的理论与方法[M]·北京:国防工业出版社,2005·
版权所有：© 2023 中国地质图书馆 中国地质调查局地学文献中心