模糊Petri网自动生成方法研究

为解决模糊Petri网建模效率低、工作量大、易出错等问题,提出了模糊产生式规则自动生成模糊Petri网的方法,并给出了其映射模型。该方法通过模型映射,结合图元生成与定位实现了模糊Petri网的自动建模。避免了模糊Petri网建模的人为失误,提高了建模效率。使知识库与模型库同步更新,保证二者的一致性。有利于充分发挥模糊Petri网的知识表示、模糊信息处理与动态并行推理的优势,对模糊Petri网理论的广泛应用具有推动作用。通过实例表明该方法是可行的。

金川镍矿岩体移动与开采稳定性研究

The two major issues in mining industry are work safety and protection of ground environment when carrying on underground mining activities. Cut-and-fill mining method is increasingly applied in China owing to its advantages of controlling ground pressure and protecting the ground environment effectively. However, some cut-and-fill mines such as Jinchuan nickel mine which has big ore body, broken rock mass and high geostress have unique characteristics on the law of ground pressure and rock mass movement that distinguish from other mining methods. There are still many problems unknown and it is necessary for the further analysis. In this dissertation, vast field survey, geology trenching and relative data analysis are carried out. The distribution of ground fissures and the correlation of the fissures with the location of underground ore body is presented. Using of monitoring data by three-dimension fissure meter and GPS in Jinchuan Deposit Ⅱ, the rule of the surface deformation and the reason of ground fissures generation are analyzed. It is shown that the stress redistribution in surrounding rocks resulting from mining, the existence of the void space underground and the influence of on-going mining activities are three main reasons for the occurrence of ground fissures. Based on actual section planes of No.1 ore body, a large-scale 3D model is established. By this model, the complete process of excavation and filling is simulated and the law of rock mass movement and stability caused by Cut-and-fill Mining is studied. According to simulation results, it is concluded that the deformation of ground surface is still going on developing; the region of subsidence on the ground surface is similar with a circle; the area on the hanging wall side is larger than one on the lower wall side; the contour plots show the centre of subsidence lay on the hanging wall side and the position is near the ore body boundary of 1150m and 1250m where ore body is the thickest. Along strike-line of Jinchuan Deposit Ⅱ, the deformation at the middle of filling body is larger than that in the two sides. Because of the irregular ore body, stress concentrates at the boundary of ore body. With the process of excavation and filling, the high stress release and the stress focus disappear on the hanging wall side. The cut-and-fill mechanism is studied based on monitoring data and numerical simulation. The functions of filling body are discussed. In this dissertation, it is concluded that the stress of filling body is just 2MPa, but the stress of surrounding rock mass is 20MPa. We study the surface movement influenced by the elastic modulus of backfill. The minimal value of the elastic modulus of backfill which can guarantee the safety production of cut-and-fill mine is obtained. Finally, based on the real survey results of the horizontal ore layer and numerical simulation, it is indicated that the horizontal ore layer has destroyed. Key words: cut-and-filling mining, 3D numerical simulation, field monitoring, rock mass movement, cut-and-filling mechanism, the elastic modulus of backfill, the horizontal ore layer

多波多分量地震资料走时近似方法的研究与应用

Compared with the conventional P wave, multi-component seismic data can markedly provide more information, thus improve the quality of reservoir evaluation like formation evaluation etc. With PS wave, better imaging result can be obtained especially in areas involved with gas chimney and high velocity formation. However, the signal-to-noise of multi-component seismic data is normally lower than that of the conventional P wave seismic data, while the frequency range of converted wave is always close to that of the surface wave which adds to the difficulty of removing surface wave. To realize common reflection point data stacking from extracted common conversion point data is a hard nut to crack. The s wave static correction of common receiver point PS wave data is not easy neither. In a word, the processing of multi-component seismic data is more complicated than P wave data. This paper shows some work that has been done, addressing those problems mentioned above. (1) Based on the AVO feature of converted wave, this paper has realized the velocity spectrum of converted waves by using Sarkar’s generalized semblance method taking into account of AVO factor in velocity analysis. (2)We achieve a method of smoothly offset division normal method.Firstly we scan the stacking velocities in different offset divisions for a t0, secondly obtain some hyperbolas using these stacking velocities, then get the travel time for every trace using these hyperbolas; in the end we interpolate the normal move out between two t0 for every trace. (3) Here realize a method of stepwise offset division normal moveout.It is similar to the method of smoothly offset division normal moveout.The main difference is using quadratic curve, sixth order curve or fraction curve to fit these hyperbolas. (4)In this paper, 4 types of travel time versus distance functions in inhomogeneous media whose velocity or slowness varies with depth and vertical travel time have been discussed and used to approximate reflection travel time. The errors of ray path and travel time based on those functions in four layered models were analyzed, and it has shown that effective results of NMO in synthetic or real data can be obtained. (5) The velocity model of converted PS-wave can be considered as that of P -wave based on the ghost source theory, thus the converted wave travel time can be approximated by calculation from 4 equivalent velocity functions: velocity or slowness vary linearly with depth or vertical travel time. Then combining with P wave velocity analysis, the converted wave data can be corrected directly to the P-wave vertical travel time. The improvements were shown in Normal Move out of converted waves with numerical examples and real data. (6) This paper introduces the methods to compute conversion point location in vertical inhomogeneous media based on linear functions of velocity or slowness versus depth or vertical travel time, and introduce three ways to choose appropriate equivalent velocity methods, which are velocity fitting, travel time approximation and semblance coefficient methods.

地震波谱元法数值模拟和偏移研究

With the development of both seismic theory and computer technology, numerical modeling technology of seismic wave has achieved great advancement during the past half century. The current methods under development include finite differentiation method (FDM), finite element method (FEM), pseudospectral method (PSM), integral equation method (IEM) and spectral element method (SEM). They exert their very important roles in every corner of seismology and seismic prospecting. Large quantity of researches towards spectral element method in the end of last century bring this method to a new era, which results in perfect solution of many difficult problems. However, parts of posterior works such as seismic migration and inversion which base on spectral element method have never been studied widely at least up to the present whereas are of importance to seismic imaging and seismic wave propagation. Based on previous work, this paper uses spectral element method to investigate the characteristics and laws of the seismic wave propagation in isotropic and anisotropic media. By thoroughly studying this high-accuracy method, we implement a kind of reverse-time pre- and post-stack migration based on SEM. In order to verify the validity of the SEM method, we have simulated the propagation of seismic wave in several different models. The simulation results show that: (1) spectral element method can be used to model any complex models and the computational results are comparable with the expected results and the analytic results; (2) the optimum accuracy can be achieved when the rank is between 4 and 9. When it is below 4, the dispersion may occur; and when it is above 9, the time step-length will be changed accordingly with the reducing space step-length in order to keep the computation stability. This will exponentially increase the computation time and at the same time the memory even if simulating the same media. This paper also applies explosive reflection surface imaging technology, time constancy principle of wave-filed extrapolation and least travetime raytracing technology of surface source to SEM pre- and post-stack migration of isotropic and anisotropic media. All imaging results derived by the above methods agree well with the real geological models and the position of interface and inflexions can also return to their right location well. This indicates that the method proposed in this paper is a kind of technology with high accuracy and robust stability. It can serve as an alternative method in real seismic data processing. All these work can boost the development of high-accuracy seismic imaging, and therefore have significant inference value.

环渤海花岗岩体地下油库建设的工程地质力学研究

As a kind of strategic resource，petroleum play an very important role in current social stability, economic development and state safety. Since 1993 China has turned from a net oil exporter into a net oil importer, the figure of imported oil increased from then on. In 2004 China's total energy consumption exceeded Japan’s, and ranked in the second place, just inferior to America. Today China is the world’s third-largest importing nation, accounting for 6% of world imports and 8% of world consumption. Comparing with other strategic petroleum reserve schemes, underground oil storage possess many advantages, such as security, economy, less pollution, save land, suited for strategic reserve and so on, so it is the most ideal form for strategic petroleum reserve. In the background of China Strategic Petroleum Reserve Program started just now, this paper choose Circum-Bo sea region as a study area, and do some system study on the underground oil storage caverns constructed in inter-large granite rock masses in Circum-Bo sea region. On the foundation of a great amount of information come from both home and abroad, firstly this paper analysed the principle, economy, cavern shape, profile dimension, and gain some cognizances and logos, as follows: ①Hard rock mass such as granite is the major rock, in which underground oil storage are constructed; ②Unlined underground oil storage caverns had been wide spread used as a sort of oil storage form abroad, there already exist a suit of skilled experience and technologies to prevent oil product from leaking; ③Compared with surface tanks, underground oil storage cavern possess predominance in economy clearly. In general, it will be more economical when the storage capacity exceed 50000m3. The quality of rock mass is the most important factor for underground storage cost, however such as hydrogeology, storage capacity, the number of storage galleries, the length, storage product, mechanical equipments, geographic location also influent the cost. In designed depth of the underground storage, the rock mass of Jinzhou mainly belong to class Ⅱ, but parts with dykes, clayization alteration, and dense joints are Ⅲ, Ⅳ; ④Now, there are few underground oil storages span more than 25m in both abroad and home. The examples of some ancient underground works and modern underground excavation with wide span surely give us many precious elicitations to construct more great unlined storage caverns, when the rock mass quality is good, cavern shape and construction method also are proper, it is quite possible to construct underground oil storage cavern with span more than 30m . The main axis orientation of Jinzhou underground oil storage cavern is NW direction, the cavern's elevation locate between -53msl and -76msl. The storage's total volume is about 3×106m3, composed of 8 parallel galleries with 950m length, the pillars between them are 45m, and every two of galleries form one unit, which can deposit 75×104m3 for each unit. The product will be stored are Saudi light and Saudi medium crude oil, the main cavern's section is 411.5m2, with 23m height and 19m width. According to the principle and technique of engineering geomechanics, this study supply a sort of system scientific thinking and method for sitting location of underground oil storage in granite region: ① On the foundation of the earth crust stability sub-zone appraise of Circum-Bo sea region, farther research concerning granite distribution, genesis, geological period and fault structure are conducted in stable areas, generally, this paper select Liaoxi, east shore of Liaotung peninsula and Jiaotung peninsula as target areas for underground oil storage regions, where Mesozoic granite is magnitude; ②After roundly comparison in facts of geologic structure, engineering geology, hydrogeology, topography, transportation and so on of three granite distributed areas, at last, selecting Jingzhou granite zone in Liaoxi out as an ideal construction area; ③ Detailed investigation is conducted in the southeast of Baimashi in Jingzhou development district, the final field. Ultrasonic Borehole Television, as a major way to collect original information of borehole rock mass were used, which is very effective to appraise the quality of deep rock mass; ④ According to the field data of tectonic stress, rock mass quality, the spatial distribution of fracture water, some optimum designs in cross section, axial direction and cavern span have been designed for the underground oil storage cavern layout in Jinzhou. To understand the characteristics of swelling alteration rock in Jinzhou granite mass, collected abundant swelling alteration rock engineering examples in granite, which study them in detail, concluded the swelling alteration rock distribute nearly everywhere in China, intruded medium-basic dykes alteration, along discontinuities and mineral hydrothermal alteration with genesis of granite are three main forms clayization alteration rock in granite rock mass. In Jinzhou field, from macro to micro studied the swelling rock which induced by mid-basic dyke intrusion, with weak swelling. In conclusion, this paper conclude the distribution rule and features of expansion alteration rock in filed, and advise some technical suggestions for excavation at swelling alteration rock part. The main features of this paper: ①In the process of site selection, investigation and design, a suit of technique and method of engineering geomechanics metasynthesis were formed, which is significative to guide the large scale underground oil storage cavern sitting location, investigation and design in granite rock mass; ②The detailed discussion on the engineering geology problems in granite mass, such as weathering crust, faults, dykes and clayization alteration rock, are useful for other projects in aspects of site selection, engineering geology evaluation and stability estimation; ③The summary and integration of the genesis, type, countermeasure relate to swelling alteration rock, also is likely to be used for other underground oil storage caverns constructed in swelling alteration granite. In conclusion, this study is meaningful for guiding the large scale underground oil storage for site selection, investigation and design in granite rock mass.

MB倒转在中国黄土中的记录及海陆对比

Chinese loess preserved in northwest and north China are famous for its fine grain size, high accumulate rate and high community and can be good archives for paleoclimate and paleomagnetic variation over the later Cenozoic, with which can be correlated well between marine sediments. Major geomagnetic chrons and long term paleoclimate changes in Quaternary are successfully extracted from Chinese loess-paleosols, as well as short-term geomagnetic excursions and climate instability of high resolution. Magneticstratigraphy based on paleogeomagnetic polarity reversal recorded in Chinese loess is a basic project in loess research since decades ago. True geomagnetic records and exact location of geomagnetic reversal boundary in section is the foundation of magneticstratigraphy. Matuyama-Brunhes (MB) reversal as the youngest one still remains divarication about exact location of its boundary (MBB). L8 and S8 of Luochuan and Xifeng located in the interior of Chinese Loess Plateau(CLP) and Mangshan in southeast part of CLP are chosen to make clear some problems which include magnetic mineral, process of MB reversal, location of MBB, downward displacement scale of magnetic reversal boundary, time lag of paleoclimate record by marine and loess, new correlation between Chinese loess and marine sediments. Rock-magnetic investigations carried on L8 and S8 show that the main mineral are ferrimagnetic assemblage consists of magnetite and maghemite in Luochuan and Xifeng, and magnetite in Mangshan, which all contains little hematite belongs to antiferromagnetic phases. The main carrier of nature remanet magnetism (NRM) is detrital magnetite with pseudo-single domain. Detailed paleomagnetic investigations display that there are several rapid reversals in direction during the process of MB reversal which started at the upper part of S8 and finished at the lower part of L8, and lasted about decades of centimeters to more than 100cm correspond to about 104 years. On the assumption that MBB is located in the middle part of the layer which recorded the very reversal, 11cm is considered as the scale of downward displacement for the MBB in Chinese loess after estimation through correlation between Luochuan and Mangshan records. So this study denies the theory of large scale displacement of MBB and large scale Lock-in depth of NRM acquired from Chinese loess. Time lag of paleoclimate records in terrestrial sediments and marine sediments is considered after reassessment of correlation between low field susceptibility of Chinese loess with marine oxygen isotope in benthic foraminifera. On the basis of traditional correlation between Chinese loess and marine oxygen isotope, this study document a new scheme which correlates L8 and S8 to MIS18 and MIS19, respectively.

用宽频带远震体波波形研究中国及邻近地区地壳构造及地质意义初探

The theory and approach of the broadband teleseismic body waveform inversion are expatiated in this paper, and the defining the crust structure's methods are developed. Based on the teleseismic P-wave data, the theoretic image of the P-wave radical component is calculated via the convolution of the teleseismic P-wave vertical component and the transform function, and thereby a P-wavefrom inversion method is built. The applied results show the approach effective, stable and its resolution high. The exact and reliable teleseismic P waveforms recorded by CDSN and IRIS and its geodynamics are utilized to obtain China and its vicinage lithospheric transfer functions, this region ithospheric structure is inverted through the inversion of reliable transfer functions, the new knowledge about the deep structure of China and its vicinage is obtained, and the reliable seismological evidence is provided to reveal the geodynamic evolution processes and set up the continental collisional theory. The major studies are as follows: Two important methods to study crustal and upper mantle structure -- body wave travel-time inversion and waveform modeling are reviewed systematically. Based on ray theory, travel-time inversion is characterized by simplicity, crustal and upper mantle velocity model can be obtained by using 1-D travel-time inversion preliminary, which introduces the reference model for studying focal location, focal mechanism, and fine structure of crustal and upper mantle. The large-scale lateral inhomogeneity of crustal and upper mantle can be obtained by three-dimensional t ravel-time seismic tomography. Based on elastic dynamics, through the fitting between theoretical seismogram and observed seismogram, waveform modeling can interpret the detail waveform and further uncover one-dimensional fine structure and lateral variation of crustal and upper mantle, especially the media characteristics of singular zones of ray. Whatever travel-time inversion and waveform modeling is supposed under certain approximate conditions, with respective advantages and disadvantages, and provide convincing structure information for elucidating physical and chemical features and geodynamic processes of crustal and upper mantle. Because the direct wave, surface wave, and refraction wave have lower resolution in investigating seismic velocity transitional zone, which is inadequate to study seismic discontinuities. On the contrary, both the converse and reflected wave, which sample the discontinuities directly, must be carefully picked up from seismogram to constrain the velocity transitional zones. Not only can the converse wave and reflected wave study the crustal structure, but also investigate the upper mantle discontinuities. There are a number of global and regional seismic discontinuities in the crustal and upper mantle, which plays a significant role in understanding physical and chemical properties and geodynamic processes of crustal and upper mantle. The broadband teleseismic P waveform inversion is studied particularly. The teleseismic P waveforms contain a lot of information related to source time function, near-source structure, propagation effect through the mantle, receiver structure, and instrument response, receiver function is isolated form teleseismic P waveform through the vector rotation of horizontal components into ray direction and the deconvolution of vertical component from the radial and tangential components of ground motion, the resulting time series is dominated by local receiver structure effect, and is hardly irrelevant to source and deep mantle effects. Receiver function is horizontal response, which eliminate multiple P wave reflection and retain direct wave and P-S converted waves, and is sensitive to the vertical variation of S wave velocity. Velocity structure beneath a seismic station has different response to radial and vertical component of an accident teleseismic P wave. To avoid the limits caused by a simplified assumption on the vertical response, the receiver function method is mended. In the frequency domain, the transfer function is showed by the ratio of radical response and vertical response of the media to P wave. In the time domain, the radial synthetic waveform can be obtained by the convolution of the transfer function with the vertical wave. In order to overcome the numerical instability, generalized reflection and transmission coefficient matrix method is applied to calculate the synthetic waveform so that all multi-reflection and phase conversion response can be included. A new inversion method, VFSA-LM method, is used in this study, which successfully combines very fast simulated annealing method (VFSA) with damped least square inversion method (LM). Synthetic waveform inversion test confirms its effectiveness and efficiency. Broadband teleseismic P waveform inversion is applied in lithospheric velocity study of China and its vicinage. According to the data of high quality CDSN and IRIS, we obtained an outline map showing the distribution of Asian continental crustal thickness. Based on these results gained, the features of distribution of the crustal thickness and outline of crustal structure under the Asian continent have been analyzed and studied. Finally, this paper advances the principal characteristics of the Asian continental crust. There exist four vast areas of relatively minor variations in the crustal thickness, namely, northern, eastern southern and central areas of Asian crust. As a byproduct, the earthquake location is discussed, Which is a basic issue in seismology. Because of the strong trade-off between the assumed initial time and focal depth and the nonlinear of the inversion problems, this issue is not settled at all. Aimed at the problem, a new earthquake location method named SAMS method is presented, In which, the objective function is the absolute value of the remnants of travel times together with the arrival times and use the Fast Simulated Annealing method is used to inverse. Applied in the Chi-Chi event relocation of Taiwan occurred on Sep 21, 2000, the results show that the SAMS method not only can reduce the effects of the trade-off between the initial time and focal depth, but can get better stability and resolving power. At the end of the paper, the inverse Q filtering method for compensating attenuation and frequency dispersion used in the seismic section of depth domain is discussed. According to the forward and inverse results of synthesized seismic records, our Q filtrating operator of the depth domain is consistent with the seismic laws in the absorbing media, which not only considers the effect of the media absorbing of the waves, but also fits the deformation laws, namely the frequency dispersion of the body wave. Two post stacked profiles about 60KM, a neritic area of China processed, the result shows that after the forward Q filtering of the depth domain, the wide of the wavelet of the middle and deep layers is compressed, the resolution and signal noise ratio are enhanced, and the primary sharp and energy distribution of the profile are retained.

利用宽频带台阵资料研究华北地区地壳及上地幔间断面结构

Intense tectonic renovation has occurred in the eastern continent of china since Mesozoic, as evidenced by the high heat flow, widespread magma extrusion and volcanic activities, and development of large sedimentary basins. To explain the cause and mechanism for the tectonic process in this period, some researchers have put forward various models, such as mantle plume, subduction of the Pacific slab, Yangtze Block-North China Block collision, etc. Their seismological evidence, however, is still scarce..During the period from 2000 to 2003, large temporary seismic arrays were established in North China by the Institute of the Geology and Geophysics, Chinese Academy of Sciences. Total 129 portable seismic stations were linearly emplaced across the western and eastern boundaries of the Bohai Bay Basin, and accumulated a large amount of high-quality data. Moreover, abundant data were also collected at the capital digital seismic network established in the ninth five-year period of national economic and social development. These provide an unprecedented opportunity for us to study the deep structure and associated geodynamic mechanism of lithospheric processes in North China using seismological techniques.Seismology is a kind of observation-based science. The development of seismic observations greatly promotes the improvement of seismologic theory and methodology. At the beginning of this thesis, I review the history of seismic observation progress, and present some routine processing techniques used in the array seismology. I also introduce two popular seismic imaging methods (receiver function method and seismic tomography).Receiver function method has been widely used to study the crustal and upper mantle structures, and many relevant research results have been published. In this thesis I elaborate the theory of this method, including the basic concept of receiver functions and the methodology for data pre-processing, stacking and migration. I also address some problems often encountered in practical applications of receiver function imaging.By using the teleseismic data collected at the temporary seismic arrays in North China, in particular, the traveltime information of P-to-S conversion and multiple reverberations of the Moho discontinuity, I obtain the distributions of the crustal thickness and the poisson ratio at the northwest boundary area of the Bohai Bay Basin and discuss the geological implications of the results.Through detailed intestigations on the crustal structural feature around the middle part of the Tanlu fault, considerable disparity in poisson ratios is found in the western and eastern sides of the Tanlu fault. Moreover, an obvious Moho offset is coincidently observed at the same surface location. A reasonable density model for the Tanlu fault area is also derived by simulating the observed gravity variations. Both receiver function study and gravity anomaly modeling suggest that the crustal difference between the western and eastern sides of the Tanlu fault is mainly resulted from their different compositions.With common conversion point imaging of receiver functions, I estimate the depths of the upper and lower boundaries of the mantle transition zone, i.e., the 410 and 660 km discontinuities, beneath most part of the North China continent The thickness of the transition zone (TTZ) in the study area is calculated by subtracting the depth of .410 km discontinuity from that of the 660km discontinuity. The resultant TTZ is 10-15 km larger in the east than in the west of the study area. Phase transitions at the 410 km and the 660 km discontinuities are known to have different Clapeyron slopes. Therefore, the TTZ is sensitive to the temperature changes in the transition zone. Previous studies have shown that the TTZ would be smaller in the mantle plume areas and become larger when the remnants of subducted slabs are present The hypothesis of mantle plume cannot give a reasonable interpretation to the observed TTZ beneath North China, Instead, the receiver function imaging results favor a dynamic model that correlates the thermal structure of the mantle transition zone and associated upper mantle dynamics of North China to the Pacific plate subduction process.

An Efficient Correspondence Based Algorithm for 2D and 3D Model Based Recognition

A polynomial time algorithm (pruned correspondence search, PCS) with good average case performance for solving a wide class of geometric maximal matching problems, including the problem of recognizing 3D objects from a single 2D image, is presented. Efficient verification algorithms, based on a linear representation of location constraints, are given for the case of affine transformations among vector spaces and for the case of rigid 2D and 3D transformations with scale. Some preliminary experiments suggest that PCS is a practical algorithm. Its similarity to existing correspondence based algorithms means that a number of existing techniques for speedup can be incorporated into PCS to improve its performance.

Basis Reduction Algorithms and Subset Sum Problems

This thesis investigates a new approach to lattice basis reduction suggested by M. Seysen. Seysen's algorithm attempts to globally reduce a lattice basis, whereas the Lenstra, Lenstra, Lovasz (LLL) family of reduction algorithms concentrates on local reductions. We show that Seysen's algorithm is well suited for reducing certain classes of lattice bases, and often requires much less time in practice than the LLL algorithm. We also demonstrate how Seysen's algorithm for basis reduction may be applied to subset sum problems. Seysen's technique, used in combination with the LLL algorithm, and other heuristics, enables us to solve a much larger class of subset sum problems than was previously possible.

Model Selection for Solving Kinematics Problems

There has been much interest in the area of model-based reasoning within the Artificial Intelligence community, particularly in its application to diagnosis and troubleshooting. The core issue in this thesis, simply put, is, model-based reasoning is fine, but whence the model? Where do the models come from? How do we know we have the right models? What does the right model mean anyway? Our work has three major components. The first component deals with how we determine whether a piece of information is relevant to solving a problem. We have three ways of determining relevance: derivational, situational and an order-of-magnitude reasoning process. The second component deals with the defining and building of models for solving problems. We identify these models, determine what we need to know about them, and importantly, determine when they are appropriate. Currently, the system has a collection of four basic models and two hybrid models. This collection of models has been successfully tested on a set of fifteen simple kinematics problems. The third major component of our work deals with how the models are selected.

Combining Associational and Causal Reasoning to Solve Interpretation and Planning Problems

This report describes a paradigm for combining associational and causal reasoning to achieve efficient and robust problem-solving behavior. The Generate, Test and Debug (GTD) paradigm generates initial hypotheses using associational (heuristic) rules. The tester verifies hypotheses, supplying the debugger with causal explanations for bugs found if the test fails. The debugger uses domain-independent causal reasoning techniques to repair hypotheses, analyzing domain models and the causal explanations produced by the tester to determine how to replace faulty assumptions made by the generator. We analyze the strengths and weaknesses of associational and causal reasoning techniques, and present a theory of debugging plans and interpretations. The GTD paradigm has been implemented and tested in the domains of geologic interpretation, the blocks world, and Tower of Hanoi problems.

Probabilistic Solution of Inverse Problems

In this thesis we study the general problem of reconstructing a function, defined on a finite lattice from a set of incomplete, noisy and/or ambiguous observations. The goal of this work is to demonstrate the generality and practical value of a probabilistic (in particular, Bayesian) approach to this problem, particularly in the context of Computer Vision. In this approach, the prior knowledge about the solution is expressed in the form of a Gibbsian probability distribution on the space of all possible functions, so that the reconstruction task is formulated as an estimation problem. Our main contributions are the following: (1) We introduce the use of specific error criteria for the design of the optimal Bayesian estimators for several classes of problems, and propose a general (Monte Carlo) procedure for approximating them. This new approach leads to a substantial improvement over the existing schemes, both regarding the quality of the results (particularly for low signal to noise ratios) and the computational efficiency. (2) We apply the Bayesian appraoch to the solution of several problems, some of which are formulated and solved in these terms for the first time. Specifically, these applications are: teh reconstruction of piecewise constant surfaces from sparse and noisy observationsl; the reconstruction of depth from stereoscopic pairs of images and the formation of perceptual clusters. (3) For each one of these applications, we develop fast, deterministic algorithms that approximate the optimal estimators, and illustrate their performance on both synthetic and real data. (4) We propose a new method, based on the analysis of the residual process, for estimating the parameters of the probabilistic models directly from the noisy observations. This scheme leads to an algorithm, which has no free parameters, for the restoration of piecewise uniform images. (5) We analyze the implementation of the algorithms that we develop in non-conventional hardware, such as massively parallel digital machines, and analog and hybrid networks.