【作者】 焦健；

【导师】 林君；

【作者基本信息】 吉林大学 ， 测试计量技术及仪器， 2012， 博士

【摘要】 利用天然源面波信息的微动勘探技术具有简便、快捷、低成本、对观测环境无特殊要求的特点，在实际应用中越来越受到重视。微动勘探技术与其他物探手段相结合，对解决资源勘探领域的采区构造、寻找石油和其它矿产资源方面已显现出其潜在的优势，但由于是利用随机噪声，该方法的精度问题一直受人们关注。基于空间自相关法的微动勘探技术是利用小规模圆形地震仪排列的微动勘探方法。目前用微动勘探技术得到的估算结果精度要低于常规地震勘探方法。综上所述，完善微动勘探方法，探索该方法的优化改善路线，满足地震工程领域勘探工作的精度要求显的尤为重要。针对以上问题，本文展开了对基于空间自相关法的微动勘探技术的研究。根据微动面波波勘探的理论，分别研究了微动数据的预处理和分析，并根据实际信号进行了测试。在此基础上，研究分析了谱估计跟频散曲线计算精度的关系，提出了趋势外推法的谱估计优化方法。研究了层状介质中的微动面波及其频散曲线的正演算法（Knopoff算法），结合不同地层模型讨论分析了对频散曲线产生影响的因素。在正演研究的基础上，研究了微动面波的线性反演方法，对遗传算法和神经网络算法的相关理论进行了深入的研究，分析了算法存在的缺点，然后针对这两种方法的不足之处提出了优化改进策略，并根据正演算法模拟频散曲线的地层模型实现了微动面波反演的计算，提出了遗传算法结合神经网络算法的非线性全局优化方法。最后结合校园内的野外试验，进行了实地探测，验证了对微动勘探技术的研究并针对速度异常区进行了探测实验。更多还原

【Abstract】 Natural source surface wave information of micro-exploration technology issimple, fast, low-cost, no special requirements for the characteristics of theobserved environment, more and more attention in practical applications.Microcombination of exploration techniques and other geophysical instrumentsconstructed to solve the field of resource exploration and mining area, lookingfor oil and other mineral resources has been showing its potential advantages,but because of the use of random noise, the accuracy of the method has beenunderpeople’s attention.Space micro-exploration technology from the relevantlaw is based on the use of small-scale circular seismographs arranged inmicro-exploration.Micro exploration technology to get the estimates accuracy islower than the conventional seismic exploration methods.To sum up, improvemicro-exploration methods to explore the optimization of the method to improvethe route to meet the accuracy requirements of the exploration work in the fieldof earthquake engineering was particularly important.For the current space-based self-the Micro exploration of the relevant law in thedispersion curve calculation and stratigraphic inversion exists.Support of the keytechnologies of metal mine seismic exploration equipment "project"863"plansubject in the micro-dynamic wave propagation and frequency dispersion curveof linear inverse calculation method based on, based on the spatialautocorrelation methodmicro exploration technology related technology researchand analysis.The main research work has the following five parts:(1) Prediction theory based on least squares method and the outer derivativepush to optimize the spectral estimation method, a base trend extrapolation of the spectral estimation method.In this paper, the basic theory of spatialautocorrelation analysis of the data processing method, we can see fromspace-based self-correlation of micro-exploration technology IF the accuracyof dispersion curves depends on the accuracy of the micro-signal powerspectrum and cross-spectral projections.Traditional spectrum estimationmethods to the poor resolution and variance of performance is not good.Forsuch problems, the spectral estimation method based on trend extrapolation,the method uses the derivative of the least squares method and extrapolationprediction theory, through the push model to establish the basis of knowndata and then fitted by the least squares method of the derivativedata, andfinally the use of trend extrapolation of the signal spectrum estimation.Afterthe widening of the data length, the spectral resolution will beimproved.Experimental results show that the higher resolution spectrumestimation method proposed in this paper, the more pronounced peak andpeak frequency and peak significantly help improve the accuracy of thecalculation of dispersion curves, compared to the traditional fitting algorithmfor spectral estimation to infer the error.less than2.3%.(2) Forward theory of micro surface waves in layered media, according to thestructural model of the forward calculation.As the use of micro-surface waveson a layered half-space detection and inversion of the theoretical basis, thispaper has studied the micro-surface wave propagation in layered media andits dispersion characteristics, to study the micro-layered mediafixed surfacewave dispersion curves algorithm (Knopoff algorithms), and micro surfacewave forward modeling calculations based on constructed two stratigraphicmodel.Finally, experimental discussion of the factors that impact on thedispersion curve.(3) Rayleigh wave inversion for the genetic algorithm convergence is slow andpremature convergence problem, based on improving the original fitnessfunction and the introduction of the mutation operator of the optimizationstrategy, genetic algorithm optimization in micro surfacewave inversioncalculation.Genetic algorithm by virtue of its unique characteristics of thenonlinear global optimization method of inversion theory has been widelyused in the micro-exploration techniques based on spatial autocorrelationmethod for Inversion of velocity structure.But in the inversion of the genetic algorithm convergence is slow and prone to premature convergence.For suchproblems, the adaptive function, selection strategy, mutation strategy and theoptimization of the genetic operator strategy, increasing the fitness of thegenetic probability of selection to improve the convergence speed, theimproved algorithm optimizationprecocious situation has been significantlyimproved.The experimental results show that during the frequencydispersion curve inversion, the genetic algorithm optimized to improve theretrieval accuracy and convergence speed, compared to the traditionalgenetic algorithm has been dramatically improved, the number of iterationsis equal to only1of the traditional algorithm/computing speed, enhanced bytwo orders of magnitude, to meet the micro-exploration formation retrievalaccuracy. This method is applicable not only to the stratigraphic inversion,introduced optimization strategy, the same can be extended to other dataprocessing areas.(4) Micro surface wave inversion, slow convergence and hidden nodes select adifficult problem, a neural network based on batch learning and optimize thenetwork structure optimization strategy to achieve a neural networkalgorithm in the micro-surfacewave inversion calculation.Neural networkalgorithms have more problems in the actual calculation, for example, whenthe sample is too much will cause the weights to adjust the tone of thephenomenon, which makes network learning time lengthened, theconvergence slows down; In addition, the number of hidden layer nodesaffect the networkthe structure, resulting in the difficulty of the choice ofhidden layer nodes.In this paper, such problems raised by the introduction ofthe batch and optimize network structure optimization strategies to improvethe speed of network learning, reducing the iterative inversion times.Theexperimental results show the effectiveness of the method, the contrastanalysis showed that this method significantly reduces the iterative inversionof the number of neural network algorithm reduced the number of iterationscompared to the standard20%to improve the convergence speed, and hasgood resistance tointerference.(5) Characteristics of each combination of genetic algorithms and neuralnetwork algorithm, a nonlinear hybrid global optimization algorithm toachieve a hybrid optimization algorithm in the micro-surface wave inversion in the calculation.Genetic algorithm has the characteristics of the nonlinearglobal optimization, such as premature, poor local searching capability; theother hand, the neural network algorithm has strong characteristics of thelocal search ability, but there is slow convergence and the objective functionof the existence of localminimum of problems.Therefore, for these twomethods in the micro-surface wave inversion of less than existing methods,genetic algorithm artificial neural networks and hybrid optimizationmethod.First, the neural network training samples, the training result doesnot meet the call to the genetic algorithm global search characteristics, andthen back to the network repeatedly until you find the best value, theintroduction of neural network algorithm LM optimizationapproach.Elasticformation model, the micro surface wave calculation of the joint inversionusing genetic algorithms and neural networks. The experimental results showthe effectiveness of the method, contrast inversion results, the relative errorsare less than the respective error of the two algorithms, each parameterrelative error does not exceed5%, improve the inversion accuracy, toovercome the respective shortcomings of the two algorithms.更多还原