Statistical analysis of fracture properties based on particle swarm optimization and Pearson correlation coefficient method

Prediction of reservoir fracture is the key to explore fracture-type reservoir. When a shear-wave propagates in anisotropic media containing fracture,it splits into two polarized shear waves: fast shear wave and slow shear wave. The polarization and time delay of the fast and slow shear wave can be used to predict the azimuth and density of fracture. The current identification method of fracture azimuth and fracture density is cross-correlation method. It is assumed that fast and slow shear waves were symmetrical wavelets after completely separating,and use the most similar characteristics of wavelets to identify fracture azimuth and density,but in the experiment the identification is poor in accuracy. Pearson correlation coefficient method is one of the methods for separating the fast wave and slow wave. This method is faster in calculating speed and better in noise immunity and resolution compared with the traditional cross-correlation method. Pearson correlation coefficient method is a non-linear problem,particle swarm optimization( PSO) is a good nonlinear global optimization method which converges fast and is easy to implement. In this study,PSO is combined with the Pearson correlation coefficient method to achieve identifying fracture property and improve the computational efficiency.

Cultural Binary Particle Swarm Optimization Algorithm and Its Application in Fault Diagnosis

Binary particle swarm optimization algorithm(BPSOA) has the excellent characters such as easy to implement and few set parameters.But it is tendentious to stick in the local optimal solutions and has slow convergence rate when the problem is complex.Cultural algorithm(CA) can exploit knowledge extracted during the search to improve the performance of an evolutionary algorithm and show higher intelligence in treating complicated problems.So it is proposed that integrating binary particle swarm algorithm into cultural algorithm frame to develop a more efficient cultural binary particle swarm algorithm (CBPSOA) for fault feature selection.In CBPSOA,BPSOA is used as the population space of CA;the evolution of belief space adopts crossover,mutation and selection operations;the designs of acceptance function and influence function are improved according to the evolution character of BPSOA.The tests of optimizing functions show the proposed algorithm is valid and effective.Finally,CBPSOA is applied for fault feature selection.The simulations on Tennessee Eastman process (TEP) show the CBPSOA can perform better and more quickly converge than initial BPSOA.And with fault feature selection,more satisfied performance of fault diagnosis is obtained.

RESEARCH ON OPTIMIZING THE MERGING RESULTS OF MULTIPLE INDEPENDENT RETRIEVAL SYSTEMS BY A DISCRETE PARTICLE SWARM OPTIMIZATION

The result merging for multiple Independent Resource Retrieval Systems (IRRSs), which is a key component in developing a meta-search engine, is a difficult problem that still not effectively solved. Most of the existing result merging methods, usually suffered a great influence from the usefulness weight of different IRRS results and overlap rate among them. In this paper, we proposed a scheme that being capable of coalescing and optimizing a group of existing multi-sources-retrieval merging results effectively by Discrete Particle Swarm Optimization (DPSO). The experimental results show that the DPSO, not only can overall outperform all the other result merging algorithms it employed, but also has better adaptability in application for unnecessarily taking into account different IRRS's usefulness weight and their overlap rate with respect to a concrete query. Compared to other result merging algorithms it employed, the DPSO's recognition precision can increase nearly 24.6%, while the precision standard deviation for different queries can decrease about 68.3%.

A Time/Space Separation Based 3D Fuzzy Modeling Approach for Nonlinear Spatially Distributed Systems

Spatially distributed systems （SDSs） are usually infinite-dimensional spatio-temporal systems with unknown nonlinearities. Therefore, to model such systems is difficult. In real applications, a low-dimensional model is required. In this paper, a time/space separation based 3D fuzzy modeling approach is proposed for unknown nonlinear SDSs using input-output data measurement. The main characteristics of this approach is that time/space separation and time/space reconstruction are fused into a novel 3D fuzzy system. The modeling methodology includes two stages. The first stage is 3D fuzzy structure modeling which is based on Mamdani fuzzy rules. The consequent sets of 3D fuzzy rules consist of spatial basis functions estimated by Karhunen-Love decomposition. The antecedent sets of 3D fuzzy rules are used to construct temporal coefficients. Going through 3D fuzzy rule inference, each rule realizes time/space synthesis. The second stage is parameter identification of 3D fuzzy system using particle swarm optimization algorithm. After an operation of defuzzification, the output of the 3D fuzzy system can reconstruct the spatio-temporal dynamics of the system. The model is suitable for the prediction and control design of the SDS since it is of low-dimension and simple nonlinear structure. The simulation and experiment are presented to show the effectiveness of the proposed modeling approach.