Study On Intelligent Optimization Algorithm With Application To Prestack AVO Nonlinear Inversion

In recently years, with the development and requirement of hydrocarbon exploration technique, especially in complex reservoirs, pre-stack seismic technique based on pre-stack data has become the focus of research in seismic exploration and development. P-wave and S-wave seismic velocities, density and other reservoir sensitive elastic parameters, as well as reservoir characteristic parameters can be obtained from pre-stack AVO inversion, which is based on pre-stack seismic data and makes seismic inversion interpretation develop from qualitative explanation to quantitative interpretation. Therefore, pre-stack AVO inversion has become the new force of seismic inversion. Since pre-stack AVO inversion is essentially a kind of nonlinear inversion, in this paper, the intelligent optimization algorithm with global optimization capability is developed and used in the research of pre-stack AVO nonlinear inversion.Firstly, this paper introduces geophysical inversion theory, pre-stack AVO nonlinear inversion theoretical basis and seismic inversion methods, etc. Pre-stack AVO inversion based on the Bayesian theory is developed. Objective equations of AVO inversion based on different probability density distribution functions are derived to obtain different inversion constraint equations. Based on this theory, a new constrained inversion method is proposed, which expresses prior probability distributing of model by the product of lateral error-probability density and longitudinal model parameter probability density. The background logging curves are added into the new objective equation to enhance low frequency information, it has remarkable effects to improve the accuracy and lateral continuity of the inversion results.The Intelligent optimization algorithm performed in this paper is mainly based on genetic algorithm (GA), so this paper introduces the basic theory, process steps and implementation process of simple genetic algorithm (SGA) in detail firstly, and improves GA based on adaptive mutation operation. GA is easily trapped into the local optimal solution and appeared premature convergence, it’s tough to solve these two problems effectively by improving a single algorithm. Therefore, in this paper, an improved adaptive genetic algorithm is presented combined with simulated annealing algorithm(SA), immune algorithm(IG) and particle swarm optimization algorithm(PSO). On the basis of these three mixed strategy of embedded, series hybrid and competition/cooperation, immune genetic algorithm(IGA), genetic simulated annealing algorithm(GASA) and GA-PSO co-evolutionary algorithm(CEA) are studied respectively, out of the three, IGA and GA-PSO CEA are critical in the study.Based on each optimized algorithm, inversion tests are performed on the layered homogeneous medium model and theoretical model w hich obtained from the actual maritime logging curves.In this paper, IGA with embedded hybrid features is proposed by adding the thought of IG into GA. Compared with GA, IG has vaccine extraction, vaccine inoculation and immune selection. Based on the antibody density adjusting mechanism, a simple and efficient probability formula of immune selection is constructed by simultaneously considering individual fitness and similarity. This formula can make the algorithm maintain the diversity of the population in the mid and later stage. Also it would effectively avoid the precocious phenomena and improve the global optimization capability of GA. Considering the special conditions of pre-stack AVO nonlinear inversion, this research applies an improved vaccine extract method, dynamic vaccine extraction, to add vaccine extraction into the construction of immune memory library. This method could ensure the optimal solution effectively and make the final output optimal solution be the history optimal solution.In the light of the idea of CEA, a collaborative GA-PSO CEA based on immune is presented. According to this algorithm, the population would be divided into two sub populations which perform GA and PSO evolution respectively, and merge into a new population after interior evolution. The new one gets further evolve under the immune operation and random reallocate individuals to perform the next iterative evolution. In the framework of co-evolutionary, the merits and demerits of GA, GA-PSO PSO and IG are considered comprehensively. The organic combination of these three algorithms has realized the advantage complementation and information exchange, enhance the global optimization capability of the algorithm. The superiorities of GA-PSO CEA show that this algorithm and CEA mixed strategy own great developing potential.On the basis of SGA, improved GA,GASA, IGA and GA-PSO CEA, inversion tests are performed on the layered homogeneous medium model and actual logging curves model respectively. And the inversion test on layered homogeneous medium model shows that the improved SGA could obtain a better inversion result of P-wave velocity, whereas the result of S-wave is difficult to distinguish the horizons which have little property differences. The accuracy of inversion results by using improved GA and three parameters inversion based on other three hybrid genetic algorithms are higher than SGA, and have a better anti noise performance. However, IGA and GA-PSO CEA have a higher resolution and anti noise performance than the improved GA in the actual logging curves model which has more inversion parameters, yet the GA-PSO CEA could effectively improve the inversion accuracy of density. As immune memory cells are applied in IGA and GA-PSO CEA based on immune thoughts, the algorithm could effectively protect the history optimal solution. Hence inversion result could well fit the variation trend of well logs and has evident advantages in the prediction of thin offshore oil-bearing reservoirs.In this paper, a pre-stack AVO nonlinear inversion based on GA-PSO CEA is performed in the fractured-vuggy carbonate reservoir of Tarim Basin, Xinjiang. Firstly, we reconstructed the well logs of SN1 well by neural network technique, and use the rock physical analysis to determine P-wave velocity and density as the reservoir sensitive elastic parameters. Then the actual pre-stack data is preprocessed to extract statistic wavelet with different angles. At the end of this paper, we carry out a pre-stack AVO nonlinear inversion based on GA-PSO CEA. The inversion result shows that low value anomalies at well location are in agreement with well interpretation, and we predict that the top and the internal of Yingshan formation are well developed, and point out the beneficial zones for the development of reservoirs. GA-PSO CEA applied in pre-stack AVO inversion for fractured-vuggy carbonate reservoir has certain practical applied value, but the accuracy and stability of density inversion based on actual seismic data have yet to be further improved.