This invention relates generally to inversion methods for fitting a model to data and, more specifically, to methods of determining values of undetermined parameters that model a geophysical environment using data measured in the geophysical environment.
Conventional inversion methods can be applied to response data measured in an actual system or environment (or a modeled system or environment) to determine values of undetermined model parameters. These inversion methods allow interpretation of the data in terms of the values of the undetermined parameters that characterize the environment or system in which the data was taken.
In conventional inversion, the values of the undetermined parameters are determined by mathematically minimizing the difference between a calculated response of a modeled system or environment (known as a forward model calculation) and the response data. Conventional inversion typically requires many forward modeling calculations and is hence processing-intensive and slow. This is especially true where the data include many points and modeled responses include multiple undetermined parameters. More undetermined parameters moreover increase the possibility of non-unique inversion results.
Another example of an inversion method is turbo-boosting (see U.S. Pat. No. 6,098,019 to Hakvoort et al.). In this method, the end-result is reached by executing a limited number of forward modeling steps or iterations. The model parameter values chosen at each iteration depend in a prescribed way on the mismatch between calculated and actual data responses of the previous iteration, and no explicit mathematical minimization procedures are used. Requiring only a limited number of forward modeling steps, turbo-boosting is fast when compared to conventional inversion.
Both conventional inversion as well as turbo-boosting may not converge, and the resulting model may not be the best representative of the actual system or environment. Conventional inversion, in particular, can be very slow for situations in which the response data depends (to a significant degree) on a large number of parameters of which values are undetermined. In this patent application, this number will be denoted as the dimensionality of the data. Moreover, for many situations, the dimensionality of the data may differ from data-point to data-point, and the existing inversion methods do not make use of this fact.
It would be useful to have a method of quickly and robustly determining values of undetermined parameters, particularly in cases where not all data-points have the same dimensionality. For example, when applying the technique of Transient ElectroMagnetics (TEM) in a geosteering scenario, it would be useful to accurately invert TEM measurements that are obtained during drilling so as to be able to form an image of the subsurface around the drill bit. Therefore, a need exists in the industry to address the aforementioned deficiencies and inadequacies.
The various embodiments of the present invention overcome the shortcomings of the prior art by providing an inversion method denoted as “layer stripping” that quickly and robustly determines values of undetermined parameters that model a system or environment. The layer stripping method is well suited for data that is measured over time and has dimensionality that changes over time, such as that of deep transient electromagnetic (TEM) measurements. The layer stripping method may also be applied to other systems and environments.
Generally, the layer stripping method can be applied to cases where all points of the response data do not have the same dimensionality, that is, when some subsets of the measured data depend only on subsets of the parameters on which other subsets of the data depend. In such cases, the values of a specific subset of undetermined parameters may be determined by applying an inversion method to an appropriate modeled response and an appropriate subset of the response data. These values can then be used to reduce the dimension of the space spanned by the undetermined model parameters of modeled responses (denoted in this application as the “inversion space”) that correspond to other subsets of the data, resulting in a faster, more robust, and more accurate inversion process.
According to an exemplary embodiment of the disclosure, a method of determining values of at least two parameters that model an earth formation includes operating a measurement device to obtain data representing a measured response of the earth formation and operating a computing unit to process the data. The steps of operating the measurement device can include deploying the measurement device in a borehole formed in the earth formation, inducing a magnetic field in the earth formation using a transmitter, removing the transmitter as a source, and measuring the signals using a receiver. The processing steps include determining at least one point that divides the data into at least two subsets, generating at least two modeled responses that correspond to the at least two subsets, determining an order of applying inversion methods to the at least two subsets and corresponding at least two modeled responses, and applying an inversion method to each of the at least two subsets and corresponding at least two modeled responses in the determined order. The order is determined such that a value of at least one of the at least two model parameters determined through a first application of an inversion method to a first of the at least two subsets and a first of the at least two modeled responses can be used to reduce the dimension of the inversion space (i.e., the space spanned by undetermined model parameters on which a modeled response depends) of a second of the at least two modeled responses preceding a second application of an inversion method to a second of the at least two subsets and the second of the at least two modeled responses.
According to one aspect of the invention, a computer readable medium including computer executable instructions is adapted to perform the exemplary method described above.
In certain embodiments, the magnetic field induced by the transmitter is a static (i.e., DC) field, the measured response is an electromagnetic response measured as a function of time, the at least two subsets correspond to at least two time intervals, and the order is a chronological order. Here, the at least one point that divides the data into at least two subsets corresponds to a so-called boundary time. In these embodiments, the method can further include calculating a value of at least one boundary distance using the value of an electromagnetic parameter determined through a preceding application of an inversion method, wherein the value of the at least one boundary distance can be used to reduce the dimension of the inversion space of a modeled response to which, along with a corresponding data subset, an inversion method is subsequently applied.
In certain embodiments, the measured response is a geophysical log recorded as a function of along-hole-depth, and the at least two subsets correspond to length intervals. Here, the at least one point that divides the data into at least two subsets may correspond to at least one boundary between layers.
In certain embodiments, the inversion method is based on mathematical minimization of a misfit function or cost function using an approximation of a derivative. In other embodiments, the inversion method is based on turbo-boosting. In still other embodiments, alternative inversion methods may be used.
According to an exemplary embodiment of the disclosure, a method of determining values of parameters that model a layered earth formation includes operating a measurement device to obtain data representing a measured response of the layered earth formation and operating a computing unit to process the data. The data is measured, for example, over time and the steps of operating the measurement device can include deploying the measurement device in a borehole formed in the earth formation, inducing a magnetic or an electric or an electromagnetic field in the earth formation using a transmitter, removing the transmitter as a source, and measuring the signals received using a receiver.
The steps of operating the computing unit to process the data can include selecting a first subset of the data that corresponds to a first time interval, generating a first modeled response that is dependent on the resistivity of a first layer of the formation, and applying an inversion method to the first subset of the data and to the first modeled response to determine a first value for the first layer resistivity. Once the first value of the first layer resistivity has been determined, the value is input for the first layer resistivity into the first modeled response to provide a first calculated response. The data and the first calculated response are compared to one another to determine a first boundary time where the data deviates significantly from the first calculated response. A second value of a first boundary distance is determined using the first value of the first layer resistivity and the first boundary time. As the data deviates from the first calculated response, a second subset of the data is selected after the first boundary time. A second modeled response is generated that is dependent on the first layer resistivity, the first boundary distance, and the resistivity of a second layer. The first value is input to the first layer resistivity and the second value is input to the first boundary distance to reduce the inversion space of the second modeled response. An inversion method is applied to the second subset of the data and the corresponding second modeled response to determine a third value for the second layer resistivity. The inputting and comparing steps are repeated to determine a second calculated response. Each time the calculated response deviates from the data, a boundary time is determined, a boundary distance is determined, a subset of the data is selected, a modeled response is generated, the subset of the data and corresponding modeled response are inverted to determine a value of a parameter, the value of the parameter is used to provide a calculated response, and the calculated response is compared to the data.
The comparing step can include calculating a ratio curve using the calculated response and the data; selecting the point in time as a boundary time if the value of the ratio curve at the point in time substantially deviates from a value of one.
The foregoing has broadly outlined some of the aspects and features of the present invention, which should be construed to be merely illustrative of various potential applications of the invention. Other beneficial results can be obtained by applying the disclosed information in a different manner or by combining various aspects of the disclosed embodiments. Accordingly, other aspects and a more comprehensive understanding of the invention may be obtained by referring to the detailed description of the exemplary embodiments taken in conjunction with the accompanying drawings, in addition to the scope of the invention defined by the claims.
As required, detailed embodiments of the present invention are disclosed herein. It must be understood that the disclosed embodiments are merely exemplary of the invention that may be embodied in various and alternative forms, and combinations thereof. As used herein, the word “exemplary” is used expansively to refer to embodiments that serve as illustrations, specimens, models, or patterns. The figures are not necessarily to scale and some features may be exaggerated or minimized to show details of particular components. In other instances, well-known components, systems, materials, or methods have not been described in detail in order to avoid obscuring the present invention. Therefore, specific structural and functional details disclosed herein are not to be interpreted as limiting, but merely as a basis for the claims and as a representative basis for teaching one skilled in the art to variously employ the present invention.
The invention is taught in the context of methods that are used to determine values of parameters that fit a model of an environment or system to data that is measured in the actual environment or system. The inversion method that is incorporated into the layer stripping method may be conventional inversion, turbo boosting, combinations thereof, and alternatives thereto. Conventional inversion often utilizes a minimization technique to aid in determining values of model parameters that minimize the difference between a calculated response of the modeled environment and a measured response of the actual environment. Turbo-boosting, in one possible application, iterates estimated values of model parameters a fixed number of times without actively minimizing a mismatch between calculated and measured response.
As used herein, the term “modeled response” will refer to a function of undetermined model parameters that is used to calculate the response of a system or environment. Certain of the undetermined parameters characterize the environment, and the function may be analytical, incorporated in a computer program, etc. The term “calculated response” will refer to a calculation of the response of the modeled system where estimated or determined values of the model parameters are input into the modeled response. The term “measured response” refers to data measured in the actual environment. It should be understood that where the calculated response fits the measured response, the parameter values may be used to characterize the actual system or environment in which the data is measured.
For reference, terms that relate the measured data and the modeled response are now described. As mentioned earlier, the number of parameters of an actual environment that significantly affect a measured response is termed the dimensionality of the data. The parameters that are presumed to significantly affect a measured response are generally those that are used in a well-chosen modeled response. Certain, if not all, of these parameters are undetermined. The number of yet undetermined parameters of a modeled response is termed the inversion space of that modeled response. For well-chosen models, the dimension of the inversion space of the modeled response is less than or equal to the dimensionality of the data. If this were not the case, an inversion method would attempt to find optimum values for parameters that do not actually significantly affect the corresponding data, implying that the model is not a good representation of the actual system.
According to conventional inversion methods, a single modeled response of the entire environment and all the measured data are used to simultaneously determine values for all the undetermined parameters. That is to say, conventional inversion methods use an inversion space of which the dimension is equal to the largest or collective dimensionality of all of the measured data-points. For cases where a first subset of the data has smaller dimensionality than a second subset of the data, such conventional inversion methods are hence not optimal since, for this first subset of the data, unnecessary computing cycles would be spent on trying to find values for undetermined parameters in the inversion space that do in fact not affect this first subset of the data.
The data is measured as function of a parameter that is incremented or changes as the data is measured. For example, the incremented parameter can be time or position. For purposes of teaching, the data will be described as being measured as a function of time.
For certain data measurements, the dimensionality of the data can change over time. The data collected at one point in time may have a different dimensionality than data collected at another point in time. Such data can collectively have a high dimensionality that is equal to the subset of the data with the highest dimensionality although other subsets of the data have a lower dimensionality. Such data is not effectively fit with conventional inversion methods but can be efficiently and robustly fit with a layer stripping method, according to the present disclosure.
Generally described, a first exemplary layer stripping method applies a series of inversion methods to the measured data subset by subset, and uses the parameter values determined through preceding applications of inversion methods to reduce the inversion spaces of modeled responses to which inversion methods are subsequently applied. Inversion methods are typically applied to the subsets of the measured data in order of increasing dimensionality since the lowest dimensionality subsets correspond to modeled responses that have smaller inversion spaces (i.e., inversion spaces of lower dimension) that do not first have to be reduced to be efficiently and robustly fit. Alternatively described, the layer stripping method includes dividing the data into subsets and applying an inversion method to the subsets in a selected order such that the inversion spaces of modeled responses that correspond to certain of the subsets are reduced before the inversion methods are applied. The order can be selected such that the dimension of the inversion spaces of the modeled responses are minimized. The relatively high speed of the application of each inversion method is facilitated by the reduced inversion space.
Referring to
Referring to
To drill borehole 12, a drill bit 16 is positioned at the end of a series of tubular elements, referred to as a drill string 18. Drill bit 16 can be directed by a steering system, such as a rotatable steering system or a sliding steering system. In certain applications, measurements facilitate directing drill bit 16, for example, toward a hydrocarbon fluid reservoir.
Measurement system 10 includes a measurement device 24 that is generally described as an array of transmitters and receivers and a corresponding support structure. Here, measurement device 24 includes a transmitter 26 and a receiver 28. Referring to
In the exemplary embodiment, each of transmitter 26 and receiver 28 includes a coil antenna. Transmitter 26 and receiver 28 are arranged to be substantially coaxial. This arrangement is used for purposes of teaching. However, in alternative embodiments, transmitters and/or receivers can be those other than coil-antennas, and/or multi-axial so as to send and receive signals along multiple axes.
Measurement system 10 further includes a data acquisition unit 40 and a computing unit 50. Data acquisition unit 40 controls the output of transmitter 26 and collects the response at receiver 28. The response and/or data representative thereof are provided to computing unit 50 to be processed Computing unit 50 includes computer components including a data acquisition unit interface 52, an operator interface 54, a processor unit 56, a memory 58 for storing information, and a bus 60 that couples various system components including memory 58 to processor unit 56.
Computing unit 50 can be positioned at the surface or at a remote location such that information collected by measurement device 24 while in borehole 12 is readily available. For example, a telemetry system can connect measurement device 24, data acquisition unit 40, and computing unit 50. In alternative embodiments, data acquisition unit 40 and/or computing unit 50 is combined with or integral to measurement device 24 and processes signals while in borehole 12.
An exemplary method of measuring a transient electromagnetic (TEM) response of formation F with measurement device 24 is now described. A TEM response is useful, for example, in deep reading electromagnetic (DEM) well logging applications to identify the boundaries and properties of layers of formation F at relatively large distances from borehole 12. TEM measurements can be made with measurement device 24 by inducing a static or DC magnetic field with transmitter 26, removing transmitter 26 as a source, and measuring the electromagnetic signals arriving at receiver 28 from regions of formation F. As previously mentioned, here, response data d is measured as a function of time. In alternative embodiments, the magnetic field may be non-static or non-DC.
By measuring response data d over time, data d is physically related to formation F. Different subsets d1, d2, d3 of data d inherently include information about regions of formation F of different extent. For example, early subset d1 represents signals received from regions of formation F that are close to measurement device 24 whereas late subset d3 represents signals that have also traveled through regions of formation F that are farther away from measurement device 24. As described in further detail below, this physical relation facilitates selection of an order in which an inversion method is applied to subsets d1, d2, d3 of data d.
Referring to
At early time y1, a formation response signal S1 received by receiver 28 has only traveled through first layer L1 in which measurement device 24 is located. Formation response signal S1 for early time y1 therefore is influenced by first layer conductivity σ1, but is uninfluenced by properties of layers L2 and L3. Accordingly, early subset d1 has a dimensionality of one and can be modeled as the response of a homogeneous formation.
At intermediate time y2, a formation response signal S2 will have traveled through first and second layers L1, L2 and will be influenced by first and second layer conductivities σ1, σ2 as well as first boundary distance H1. Intermediate subset d2 has a dimensionality of three and can be modeled as the response of a two layer formation. In this example, the modeled response of a two layer formation can successfully be used to fit both early subset d1 and intermediate subset d2.
At late time y3, a formation response signal S3 will have traveled through all three layers L1, L2, L3 and will be influenced by first, second, and third layer conductivities σ1, σ2, σ3 as well as first and second boundary distances H1, H2. Late subset d3 has a dimensionality of five and can be modeled as the response of a three layer formation. In this example, the modeled response of a three layer formation can successfully be used to fit all of data d and represents the response of formation F. Were formation F to have additional layers, the dimensionality of the data would increase for each additional layer and the modeled response would correspond thereto.
Dividing Data into Subsets
To quantitatively parse or divide data d into subsets d1, d2, d3 that correspond to early time y1, intermediate time y2, and late time y3, the following method can be used. Measured data d has characteristics that can be used to identify the number of layers L1, L2, L3 and number of boundaries B1, B2 of formation F. For example, since the response of a homogeneous formation decays at a generally constant slope for later time (when plotted on a double-logarithmic graph), deviations and shifts from a constant slope indicate the presence of boundaries B1, B2 and layers L1, L2, L3. Referring to
One method of determining points P1, P2 is to select boundary times t1, t2 on the basis of changes from a constant value of the slope of the curve representing data d on a double logarithmic scale. Early time y1 can be selected as time interval t<t1, intermediate time y2 can be selected as time interval t1<t<t2, and late time y3 can be selected as time interval t>t2. Here, early subset d1 is data d for early time y1, intermediate subset d2 is data d for intermediate time y2, and late subset d3 is data d for late time y3. A method of determining points P1, P2 using ratio curves x is described in further detail below. The ratio curves method can also be used to adjust or update points P1, P2 that are found using the previous method.
Referring to
Referring momentarily to
Value v1 of first layer conductivity σ1 can then be used to estimate a value v2 for first boundary distance H1 from boundary time t1, using an appropriate inversion method. This inversion method may make use of an equation resembling H12=0.5*8*t1/(σ1*μ0), where μ0 indicates the magnetic permeability of L1, but may also rely on numerical techniques. In subsequent applications of inversion methods, values v1, v2 of parameters σ1, H1 can be input into subsequent modeled responses that include parameters σ1, H1, for example to reduce the dimension of the inversion space of modeled responses m2, m3 that correspond to data subsets d2, d3.
According to the exemplary layer stripping method, a second step is applying an inversion method to intermediate subset d2. A second modeled response m2 is generated that relates to intermediate subset d2. Second modeled response m2 is that of a two layer formation and hence depends on first layer conductivity σ1, first boundary distance H1, and second layer conductivity σ2. Without any additional knowledge, the dimension of the inversion space of d2 would therefore be equal to three.
Initially, the dimension of the inversion space of second modeled response m2 is equal to the dimensionality of intermediate subset d2. However, since values v1, v2 of first layer conductivity σ1 and first boundary distance H1 have been determined through the first step of the layer stripping method, the dimension of the inversion space of second modeled response m2 is reduced to one, being spanned only by the second layer conductivity σ2. Consequently, the dimension of the resulting inversion space of second modeled response m2 is less than the dimensionality of intermediate subset d2. The reduced dimension of the inversion space thus allows an inversion method to be efficiently and robustly applied to second modeled response m2 and intermediate subset d2 to determine a value v3 of second layer conductivity σ2. A second calculated response c2 is second modeled response m2 with value v1 input to first layer conductivity σ1, value v2 input to first boundary distance H1, and value v3 input to second layer conductivity σ2 and is calculated over all times y1, y2, y3. Referring to
A value v4 of second boundary distance H2 can be determined by appropriate conventional or other inversion methods, as value v1 of first layer conductivity σ1, value v2 of first boundary distance H1, value v3 of second layer conductivity σ2, and second boundary time t2 are known. For example, use can be made of an equation resembling H122=0.5*8*t2/(σ12N12*σ21(1−N12)*μ0) where N12≅0.0351, σ12 is the smaller of σ1 and σ2, σ21 is the larger of σ1 and σ2, H12=cos(a)*H1−sin(a)*H2, and a≅0.55* arctan(2*log(σ1/σ2))+270.
A third step of the layer stripping method is the application of an inversion method to late subset d3. A third modeled response m3 is generated that relates to late subset d3. Third modeled response m3 is that of a three layer formation and hence depends on five parameters, namely first layer conductivity σ1, first boundary distance H1, second layer conductivity σ2, second boundary distance H2, and third layer conductivity σ3.
Without any prior knowledge of σ1, σ2, σ3, H1 and H2, the dimension of the inversion space of third modeled response m3 is equal to the dimensionality of late subset d3, i.e., it is equal to five. However, since value v1 of first layer conductivity σ1 and value v2 of first boundary distance H1 have been determined through the first step of the layer stripping method, and value v3 of second layer conductivity σ2 and value v4 of second boundary distance H2 have been determined through the second step of the layer stripping method, the inversion space of third modeled response m3 is reduced to a one dimensional space spanned by third layer conductivity σ3.
Consequently, the dimension of the resulting inversion space of third modeled response m3 is much less than the dimensionality of late subset d3. This allows an inversion method to be efficiently and robustly applied to third modeled response m3 and late subset d3 to determine a value v5 of third layer conductivity σ3. Third calculated response c3 is third modeled response m3 with values v1, v2, v3, v4, v5 input to parameters σ1, H1, σ2, H2, σ3.
Referring to
Values v1, v2, v3, v5 of parameters σ1, H1, σ2, H2, σ3 found at each step of the layer stripping method characterize formation F.
Although the layer stripping method is illustrated with respect to three layer formation F, the layer stripping method is equally applicable to alternative formations and other environments.
In contrast to the first exemplary layer stripping method described above, data d need not be clearly divided into distinct subsets with related modeled responses at the outset. For example, boundary times t1, t2 may be difficult to discern with the above described method. Accordingly, modeled responses m1, m2, m3 corresponding to different subsets d1, d2, d3 of the data d are not known at the outset.
In such instances, referring to
First modeled response m1 is that of a homogeneous formation, as described above. Value v1 of first layer conductivity σ1 can be determined as described above by applying an inversion method to first subset d1 and first modeled response m1. The second exemplary layer stripping method continues as first calculated response c1 is compared to measured data d, where it is understood that calculated response c1 was chosen so as to substantially fit measured data d at least within time interval z1. As above, boundary time t1 is determined where first ratio curve x1 deviates from a value of one by a selected threshold value. The deviation of first ratio curve x1 from a value of one indicates the presence of second layer L2. Boundary time t1 can be used, along with value v1 of first layer conductivity σ1, to determine value v2 of first boundary distance H1 as described above.
Once it is determined that first ratio curve x1 deviates from a value of one, a second modeled response m2 is generated. In general, each time a ratio curve substantially deviates from a value of one, the updated modeled response is that of a formation with an additional layer. Accordingly, second modeled response m2 is selected to be that of a two layer formation. Also, referring to
The second exemplary layer stripping method continues as second calculated response c2, made to substantially fit measured data d up to largest time in interval z2, is compared to measured data d. As described above, boundary time t2 is determined at a time where ratio curve x2 substantially deviates from a value of one by a selected threshold value. The deviation of ratio curve x2 from a value of one indicates the presence of third layer L3. Boundary time t2 can be used, along with value v1 of first layer conductivity σ1, value v2 of first boundary distance H1, and value v3 of second layer conductivity σ2, to determine value v4 of second boundary distance H2, as described above.
Since ratio curve x2 deviates from a value of one, third modeled response m3 is generated to update second modeled response m2. Third modeled response m3 is that of a three layer formation. Referring to
Third calculated response c3, made to substantially fit measured data d up to the largest time in interval z3, is compared to measured data d. In this example, a third ratio curve x3 does not substantially deviate from a value of one in the time-range of interest.
Values of certain parameters that have been found through the application of inversion methods or through calculation can be allowed to adjust with the application of an inversion method. In general, the earlier found values should not change dramatically but should rather be fine tuned.
The layer stripping methods of the present disclosure are useful in a variety of applications. For example, the layer stripping method may be applied to acoustic bond logging applications to evaluate multiple cement jobs during a single logging run or to increasing dimensionality logs that measure the inflow into a well along a specific flow path that intersects regions of different permeability.
The above-described embodiments are merely exemplary illustrations of implementations set forth for a clear understanding of the principles of the invention. Variations, modifications, and combinations may be made to the above-described embodiments without departing from the scope of the claims. All such variations, modifications, and combinations are included herein by the scope of this disclosure and the following claims.
Filing Document | Filing Date | Country | Kind | 371c Date |
---|---|---|---|---|
PCT/US09/45104 | 5/26/2009 | WO | 00 | 1/3/2011 |
Number | Date | Country | |
---|---|---|---|
61056301 | May 2008 | US |