FR2481811A1 - Computer processing of ground formation diagraphic data - uses statistical model to supply corrections for data from sensors - Google Patents

Abstract

Probes traverse the ground formations in bore holes in the search for fossilised fuel or minerals. The characteristics of the formations are studied by combining data with that from a statistical model to replace missing data. A probe assembly (12) contains probes and sensors to measure porosity (18a) and density (20) as well as an acoustic sensor (16) to determine acoustic transit time. The sensor outputs are applied to amplifiers (42,36,24,26) and calculators (28,38) to determine the density and porosity. These values with the transit time data (delta-t) are applied via analogue-digital converters (46,48,50) to a transmission unit (52). The received (56) information is correlated (58) for probe depth and processed to effect zero shifts and to group the data. A CPU (64) provides a statistical model for use in applying corrections to the data.

Description

 The invention relates to methods for studying traversed earth formations by soundings. It relates more particularly to logging data processing for the calibration of logging measurements and the reconstruction in a well of a missing logging parameter.

Boreholes are drilled into the ground to search for formations containing, for example, fossil fuels, coal or hydrocarbons or minerals. To locate and evaluate such training, it is necessary to determine its characteristics. Important characteristics are the lithology or mineral composition of the formations, the structure of the grains, the porosity or volume of the pores, the content of these pores and the permeability.

The study of these training characteristics is described in particular in the book by EJ Lynch entitled "Formation Evaluation" and published by Harper and
Row in 1962.

 To obtain information on the characteristics of the formations crossed by a borehole we use logging devices lowered into the borehole on a cable. These measurements are recorded on a strip of film or paper, which provides a diagram as a function of depth. The measurements can also be recorded in an appropriate memory. There are three types of exploration devices: electric, acoustic and nuclear.

These techniques are diorites for example in the book of Hubert Guyod and Lemay
Shane titled "Geophysical Well Logging", published in 1969 for Hubert Guyot, Houston, Texas. The interpretation of the logs is described for example in the book by SJ Pirson, entitled "Handbook for Well Log Analysis", published by
Prentice-Hall in 1963.

 Although the accuracy of the measurements is gradually improving, modern interpretation techniques are creating very demanding new requirements. Complex interpretation techniques, such as interpretation techniques for clay sands and for multiple lithologies require a precision which is greater than that encountered after calibration in the field of measurement devices.

 The importance of log calibration has been recognized and various techniques have been developed for this purpose. For example, manual calibration against benchmarks is described in an article by Cochrane J.

E., titled "Principles of Log Calibration and their Application to Log
Accuracy ", published in" Journal of Petroleum Technology ", in July 1966,
This technique involves calibration at the surface, at one, two or more points, using reference signals. The exploration device is placed on the surface in an environment of known characteristics, and the signals supplied by the device are recorded at one or more points on the response curve. The differences between the actual and theoretical signals are used to correct the measurements in the survey. A similar technique is described in an article by EA Maciula and JE Cochrane entitled "Quantitative Use of Calibration Data to Correct Miscalibrated Well Legs" published in "Journal of Petroleum Technology" in July 1968, pages 663-670.

This technique involves using reference signals to determine the offset of the measurements from specific measurements, and uses this offset to convert the measurements to true values.

Another technique is described by FS Jeffries and EM Kemp in "Computer Reconciliation of Sonic Log and Core Analysis in the Boundary Lake wn Rhrlrth Annual Jogging Symposium Transactions, May 23-24, 1963, Oklahoma
City, Oklahoma, pages IV-I to IV-18. This technique involves the calibration of acoustic logs by comparison to porosity information obtained from the analysis of cores.

Another approach to calibrating logs uses statistical analysis of the measurements. A field model is established and it is investigated whether a slight offset for a log from a field survey would better match the model. The model corresponds to a subjective estimate of the lithology of the field by an interpretation specialist. It can be obtained from porosity measurements, as described by JA Beurk et al in "The Mtho-Porosity Cross-Plot" SPWLA, Tenth Annual Jogging Symposium, May 25-28, 1969, or on the basis of other information such as logs in field surveys. This technique requires very experienced specialists, and involves a great deal of subjectivity. This technique is also very time consuming and its extension to more than two or three logs is extremely difficult.

 The invention relates to methods for the study of formations traversed by soundings, and in particular the processing of logging data.

More particularly, the invention relates to the calibration of logs and the reconstruction of missing logs.

 According to a characteristic of the invention, a statistical model is created in a field by the following steps: producing several measurements relating respectively to different characteristics of the formations crossed by one or more surveys of the field; forming groups of data respectively comprising a defined combination of said measurements; and combine the data groups to form a statistical model in the field.

According to another characteristic of the invention, we calibrate the logs obtained in a survey of the same field, by producing similar measurements originating from this survey and by forming groups of similar data from these latter measurements. These new groups of data are statistically combined with the statistical model of the field to determine modifications to be made to the last groups of data so that they correspond to the model of the field.

According to yet another characteristic of the invention, a missing log is reconstituted by forming groups of similar data for a field survey, with the difference however that each group does not contain a representation of the sought log. These reduced groups of data are statistically combined with the field model to obtain the measurements representing the missing log.

 More particularly, in order to obtain the statistical model, at each level of one or more surveys, several measures representative of different characteristics of the training courses at this level are produced. By exeinplen at each level, the different measurements may include a neutron log, a density log and an acoustic log. Groups of data are then formed, each group comprising the measurements recorded at a given level. The data groups are combined statistically to obtain a statistical model of the field. We form a three-dimensional memory made up of cubic cells and whose dimensions correspond to the three measures. Each memory cell has for address a particular combination of the three measures, and each group of data can therefore be used as address of a cell We examine all the data groups and the number of occurrences of the group with the corresponding address is stored in each box. When all the groups of data have been examined, the cells of the general memory contain count values whose distribution constitutes the statistical model of the Before other field used as addresses, the measurements of the groups of data may be subjected to processing preliminary to improve their quality. Groups of data that are visibly incorrect or of questionable quality may be excluded.

 When the statistical model has been produced, it is possible to calibrate a log corresponding to one of the measurements of the data groups for a particular survey of the same field, by statistically combining groups of similar data obtained in this survey with the model carried out previously.

Each of these latter groups of data is used as the address of a box in the model. The content of the box having this address is recorded in a central accumulator, and the contents of different adjacent boxes of the model according to the size of the log to be calibrated are recorded in respective lateral accumulators. When all the groups of data from the particular survey have been examined, the contents of the central accumulator and of the lateral accumulators have a distribution representative of the calibration correction sought. In particular, the offset with respect to the central accumulator of the peak of the distribution curve thus obtained represents the calibration correction. If there is no offset, the calibration correction is zero.

 To reconstruct in a survey a log representative of a particular characteristic, at each level of a survey, groups of data are produced which include several measurements but obviously do not contain a measurement representative of the particular characteristic. For example, the data group can comprise only a neutron log measurement and a density measurement but no acoustic log measurement.

The reconstruction of the missing log (for example the acoustic log) consists in statistically combining these reduced groups of data, comprising two elements, with the three-element model previously produced.

For this purpose, each group of two-element data is used as the address of a row of boxes in the three-dimensional memory forming the model. The row is taken along the dimension corresponding to the missing log. Each of the boxes in the row corresponds to a three-element data group, two elements of which are identical to the elements of the reduced data group. The boxes in the row are examined and the box with the highest content is chosen. The measurement for the missing log is obtained by replacing the reduced group of data by that, with three elements, corresponding to the address of the selected box.

The method of the invention is applicable to multidimensional data groups, comprising four or more measurements.

The invention will moreover be better understood with the aid of the following description of several embodiments given by way of nonlimiting examples with reference to the appended drawings in which
- Figure 1 is a diagram of a composite exploration device for studying formations crossed by a borehole and of a device for processing log signals to develop a statistical model of a field, for calibrating logs on the based on this statistical model and to reconstruct logs on the basis of this model;
- Figure 2 shows the main steps of a process for developing a statistical model of a field;
- Figure 3 is a diagram of a three-dimensional memory representing a statistical model of a field;
FIG. 4 represents the main steps of a method for calibrating a log,
- Figure 5 is a diagram of accumulators used in the process of Figure 4;
- Figure 6 is an example of use of the accumulators of Figure 5;
- Figure 7 is a graph obtained by the method of Figure 4;
- Figure 8 shows in more detail the process steps for creating a statistical model of a field;
- Figure 9 shows in detail steps of the method for calibrating a log;
- Figure 10 is a variant of the embodiment of Figure 9;
- Figure 11 shows steps of a method for reconstructing a log.

 The log data processed according to 1 invention to calibrate a log or reconstruct a missing log, are obtained from an apparatus of the type shown schematically in FIG. 1.

With reference to FIG. 1, an exploration device 10 is placed in a borehole 12 at the end of a multi-conductor reinforced table 14 which is wound on a winch (not shown). The exploration apparatus 10 includes an acoustic probe 16 for measuring the acoustic transit time of the formation traversed by the borehole 12. Such a probe is described for example in American patents No. 2,938,592 (CJ Charske) and No .). 231.041 (F, P. I (okesh).

The exploration apparatus 10 also comprises a neutron detection probe 18 comprising a source and a radiation detector mounted in a pad 18a for measuring the hydrogen content of the formationsg and therefore their porosity. Devices of this type are described for example in American patent No. 2,769,918 (C.w. Tittle) and in French patent No.

1.583.809 (H. Sherman and J. Tittman). Instead of the probe 18, it is also possible to use a more conventional probe with neutron emission and detection of gamma capture rays. The scanning apparatus 10 also includes a density measurement sande 20 having a pad 20a which contains a source and two gamma ray detectors. A probe of this type is described in an article by
JS wahl, J. Tittman, CW Johnstone and RP Alger, entitled "Dual Spacing rormation Density Log" and published in "Journal of Petroleum Technology" December 1964, pages 1411 - 1416; in an article by J. Tittman and JS Wahl, entitled "The Phystcal Foundations of Formation Density Logging (Gamma-Gamma)" published in "Geophysics" in April 1965, pages 284 - 294; and in U.S. Patent No. 3,321,625 (JS Wahl). To keep the exploration device 10 centered in the borehole, two extensible members 18b and 20b coming into contact with the wall are respectively provided opposite the pads 18a and 20a.

A diameter may be combined with the arms of the pads 18a and 20a to provide a signal representative of the diameter of the borehole .12. To keep the upper part of the exploration device 10 centered, several elastic spacing members 22 are provided.

 The detectors near and far from the density measurement probe 20 supply signals G1 and G2 representative of the counting rates on these detectors. The signals G1 and G2 are transmitted to the surface by the cable 14, and amplified respectively by two amplifiers 24 and 26. The output signals from the amplifiers 24 and 26 are applied to a computer 28 which calculates the apparent density p3 of the formations. of the diameter may Other applied to the calculator 28 for Other used in the calculation of the apparent density PB- The signal p3 is applied to an appropriate memory 30 coninanded as a function of the depth, by a shaft 32 coupled to a wheel 34 driven by the target 14.

 The signal supplied by the neutron probe is in the form of a series of pulses whose counting rate is proportional to the hydrogen content, and therefore to the porosity of the formations. This signal N is applied, via the cable, to an amplifier 36 whose output is connected to a porosity calculator 38 which converts the counting rate N into a continuous signal proportional to the porosity ssN. A porosity calculator 38 is described in French patent No. 1,583,809 mentioned above. The porosity signal N is applied to a memory 40 driven by the shaft 32. The memory 40 is similar to the memory 30 and operates so as to readjust the porosity signal and the density signal in depth.

The transit time measurements #t supplied by the acoustic probe 16 are applied, via the cable 14, to an amplifier 42, the output of which is connected to a memory 44 similar to memories 30 and 40.

The memory 44 driven by the shaft 32 reads the acoustic signals in depth with respect to the density signals and to the "neutron" porosity signals.

 The three measurement signals are applied respectively to analog-digital converters 46, 48 and 50 to be converted into digital signals then applied to a transmitter 52 to be transmitted, via a link 54, to a place where s 'performs the processing.

 The signals received by a receiver 56 can be applied to a depth correlator 58 to establish a finer depth correlation for each level of the borehole 12 between the various measurements. If desired, the output signal from the depth correlator 58 may further be applied to a preliminary processing device 60 to effect, for example, a preliminary zero offset or a multiplication of scale by a known factor. The output signal from the preliminary processing device 60 is made up of groups of three values called data groups. Each group is a combination of the three measurements (neutron, acoustics and density) recorded at the same level of the survey 12 For example, it can there is a group of data every 15 cm along the borehole 12. The groups of data are stored in a conventional memory 62 with drum, tape or disc. Two or more bores 12 or more located in the same field can be explored as described above, and the groups of data originating therefrom stored in memory 62. The groups of data are then processed under the control of a unit central processing unit 64 to provide a statistical model of the field, and this statistical model is memorized in a memory 66. The memory 66 can title a conventional memory with magnetic cores, with tam bour or with disc When a logging coming from a sounding given 12 being in the same field should other calibrated, groups of data stored in memory 62 are combined with the statistical model of memory 66 in order to determine the offset necessary to bring this logging into statistical conformity with the model. The result of this determination can also be displayed on a display device 68 which can title a conventional computer printer. Under the control of the central unit 64, it is then possible to carry out the calibration offset thus obtained - for the log concerned.

The calibration of a concerned log from a given field therefore comprises two main phases: forming a statistical model of the field by combining a certain number of logs recorded in surveys of the same field; and combine the log which is to be calibrated with the statistical model to determine the correction necessary to bring this log into conformity with the field model. The method of the invention is based on the general principle according to which an analyst who compares logs to an a priori model of a field, and finds that a slight shift made to a log makes it conform to this a priori model, assumes that this shift corresponds to a calibration error (zero or scale error ). The statist model of a field obtained according to the invention corresponds to this general principle, but is only applicable in cases where it would be impossible or implausible for an analyst to process the data available in a field. In addition, the method of the invention eliminates human error and the subjectivity of the analyst.

 The example described includes three porosity logs. The method of the invention is also applicable to other logs and in different numbers, for example two or four, or more.

We assume that there is an a priori probability of finding a particular group of three porosity measurements. That is, for each group of measures there is a probability function n (. PB. At)
P (N PB- At) = ~ N n being the number of occurrences of the particular group of measures and N the total number. It is then assumed that a calibration error in one of the logs shifts the groups in which this measurement is found towards a position of lower probability. For a new survey of the field, there is a corresponding probability function, and the probability at each level can be used as a measure of conformity between the measurement groups of the new survey and the a priori model. Then, we can perform the calibration by maximizing the probability by a systematic search around an initial point.

 The choice of an a priori probability is a critical phase. There is little chance of finding a universal probability function for all logging conditions. Even if it were possible to find one, it would be a fairly flat and not very interesting function because we need a function with significant variations. However, for a certain field, and even probably for a certain type of sediment, such a probability function must exist and allow calibration.

One method for obtaining a probability function consists in selecting a set of logs recorded under good conditions in a field and in counting the number of occurrences of each combination of three measurements. Probability is the ratio of the number of occurrences to the total number of data groups. All mineral combinations normally encountered in the field should preferably be included in the log set, but this is not absolutely necessary. Small calibration errors can slightly flatten the probability function, but do not affect its main properties. Small deviations in depth between logs will also flatten the function a bit. The probability function can be improved gradually by incorporating new information from new logs in the same field. A borehole crossing new minerals will not be a problem since a large part of this borehole corresponds to current lithology.

The main stages of the process for creating a distal sta model of a field are shown in FIG. 2, this process being able to be implemented by a computer program. The computer may include the central processing unit 64 shown in FIG. 1 as well as the memories 62 and 66 and the display device 68. An example of a suitable machine is an IBM 360/65 universal digital computer of conventional configuration and has a core memory of 512 K bytes.

With reference to FIG. 2, the first step to establish a statistical model of a field consists (block 70) of reading a group of data made up of three porosity measurements recorded at a given level of a survey of the field.

The data group is examined (block 72) to determine whether each of the measurements is within defined limits. A measurement that is abnormally out of a reasonable range of values is probably wrong and should be eliminated. Consequently, if the answer to block 72 is NO, we go back to step 70 to read another group of data. The groups of data which are not within the limits defined in step 72 are thus discarded.

If the answer to block 72 is YES, a box address is calculated (block 74) corresponding to the particular combination of measurements constituting the data group. The statistical model can be represented by a three-dimensional memory made up of cubic boxes, each of which corresponds to a particular combination of the three porosity measurements. Such a memory is represented in FIG. 3 with each box identified by three digits chosen between O and 3. In the example shown, the dimension X of the memory can correspond to the neutron log, the dimension Y to the density measurement and the dimension Z to the acoustic measurement. The designation of a box can be considered as its address. The cell designations can be given directly in porosity units, or as defined functions of the measurements in the data groups.

 After having used the three measurements of a data group to calculate the address of a box, the content of the box having this address is increased by one unit (block 76). Initially all the memory boxes contain a null value or any other known arbitrary value.

If there are any groups of data (block 78), return to step 0 to examine the next group of data. If none remains, a distribution list (block 80) is printed by the display device 68 (FIG. 1).

In general, most of the memory boxes are empty.

 In fact, the three-dimensional memory represented in FIG. 3 can be simulated on a two-dimensional memory such as the fast memory (toroid) 66 of FIG. 1. The only important fact is that its capacity is sufficient to contain as many word locations as of cells in the three-dimensional memory. The necessary number of locations is the maximum possible number of different groups of data. For example, if there are 50 possible values for each of the three measurements, the memory 66 must include at least 125,000 locations (that is to say 50 × 50 × 50).

 When all of the data from one or more surveys has been processed, memory 66 contains information that represents a statistical model of the field. This statistical model can be used to calibrate a log from a survey of the same field according to the process represented in FIG. 4.

With reference to FIG. 4, steps 82, 84 and 86 are respectively the same as steps 70, 72 and 74 of FIG. 2, except that the data groups now contain a measurement to be calibrated. For example, each group of data processed in steps 82, 84 and 86 contains a neutron log measurement to be calibrated.

 We read a group of data (block 82) from memory 62, and if the data is between defined limits (block 84) this block is not eliminated and we use it (block 86) to calculate an address as in step 74.

 The content of the box whose address comes from another calculated one is read in a non-destructive way (block 88) and it is recorded in an accumulator called "central" accumulator. The central accumulator can other a register, or simply a memory location in which we keep the sum of the contents of the boxes whose addresses are calculated one after the other.

 In block 90, the content of the different adjacent boxes is read according to the size of the log to be calibrated (for example the neutron log) and this content is respectively recorded in several accumulators called "lateral" accumulators.

 Figure 5 shows a central accumulator 100, two right accumulators 102 and 104 called "+ 1" and "2" and two left accumulators 106 and 108 called "- 1" and "- 2". The central accumulator 100 receives the bare tale of the box whose address is calculated in 3 'step 86. The right accumulators 102 and îo4 receive the content of the boxes immediately adjacent in the positive direction in the direction corresponding to the log to bottom , and the left accumulators 106 and 108 the contents of the immediately adjacent boxes in the negative direction. Other lateral accumulators can be added with a similar function. In fact, a particular embodiment of the invention uses five left and five right accumulators.

 FIG. 6 represents an example in which the box whose address was found in step 86 is the box (2, 3, 1). The content of this box is entered in the central accumulator 100. The box immediately adjacent according to the dimension of the log to be calibrated (that is to say according to the dimension & is the box (3, 3, 1) and its content is entered in the accumulator 102. Similarly, the content of the box (1, 3, 1) which immediately precedes the box (2, 3, 1) is added to the content of the left accumulator 106 and the content of the box (0, 3, 1) is written in the lateral accumulator 108. It has also been assumed that the following data group has the address (2, 1, O).

When all the groups of waves have been examined (block 92), a distribution curve (block 94) is formed using the contents of the accumulators. The distribution curve is a representation of the content of the accumulators as a function of their distance from the central accumulator. An example of a distribution curve is shown in FIG. 7 in which the number of the accumulators is plotted horizontally and their content vertically. In reality, the percentage ratio of the content of each accumulator to the total content of the various accumulators L 'is plotted on the ordinate. central accumulator 100 in FIG. 5 is designated by 0000 in FIG. 7, the left accumulator 106 is designated by - 001, the left accumulator 108 is designated by - 002, etc., 1 t right accumulator 102 is designated by 0001 on In FIG. 7, the right accumulator 104 is designated by 0002, etc. The line designated by "probability" in FIG. 7 gives, above the corresponding accumulator, the ratio of its content to the total of all the accumulators, and the upper line designated by "accumulated contents" gives the actual contents respective accumulators.

 One can draw a curve 110 of the percent contents of the accumulators with respect to their positions. The maximum 11Oa of this curve 110, has, with respect to the central accumulator, an offset position which indicates the correction of calibration to be carried out in order to bring the log under study in accordance with the statistical model of the field. In the example in Figure 7, the top of the curve 110 is shifted to the left by 1.61 boxes. This means that a zero offset corresponding to 1.61 boxes on the memory porosity scale must another applied to each #N porosity measurement. Referring again to FIG. 4, the top of the curve 110 (block 94) is determined then the offset of this top with respect to the central accumulator (block 96) and the log N is calibrated to the using the offset thus obtained (block 98). A printing step can be provided (block 100) for drawing, for example, the graph in FIG. 7.

 A similar calibration can obviously also be performed for density logging or for acoustic logging. When calculating the address of a box, the adjacent boxes whose contents are read and written to the lateral accumulators are arranged according to the size of the log to be calibrated in the memory of FIG. 3.

 A specific 11-step embodiment concerning the creation of a statistical model of the field is shown in detail in FIG. 8. This method can be implemented in the form of a program for a universal digital computer, such as the model. IBM 360/65 with a core memory of 512 K bytes.

 First of all, one or more surveys of the field Have been explored and groups of data have been stored in memory 62. In some cases, a single survey of the field may be sufficient, but it is preferable to have groups of data coming from several surveys, for example three or four. It is also preferable to have reasonably reliable data groups, for example whose calibration has been checked manually.

 With reference to FIG. 8, limits are defined for each of the three porosity logs (block 112). As seen above, this eliminates erroneous measurements. For example, for a particular field, the measurements must be between - 0.10 and + 0.40 porosity units, the density measurements between 1 and 3.5 and the acoustic measurements At between 50 and 150.

 A group of data is read from memory 62 (block 114) and it is determined (block 116) whether a preliminary zero offset has been specified. If this is the case, each measurement of a determined log is shifted (block 118).

It is then examined (block 120) if a preliminary modification of scale has been specified. If so, this change is made (block 122). It is then determined (block 124) whether the data group (after possible zero shift or change of scale) is within defined limits. If the data group is not within the defined limits, it is sought whether there are other data groups (block 126) in which case the next data group is examined.

 If the data group is within the defined limits, each data group (block 128) is converted to obtain a group of three integers (X, Y, Z). For example, the range of variation of each log is associated with a series of integers ranging from O to 49. Thus, if the data groups contain neutron log measurements between - 0.10 and + 0.40 units of porosity, the integer 0 is associated with the measurements from - 0.10 to -0.09, the last tette being excluded. The integer 1 is associated with the measurements from - 0.09 to - 0.08, the latter value being excluded and so on. Likewise, if the density varies between 1.00 and 3.50, the numbers 0, 1, 2 are. will be associated with the measurements 1.00 to 1.05, 1.05 to 1.10, 1.10 to 1.15 etc. The acoustic measurements will be cut in the same way. Each data group is thus represented by three whole numbers (X, Y, Z) between 0 and 49. For example, the data group (- 0.10; 1; 50) corresponds to the set (0, 0 , O), the data group (- 0.05; 3; 100) to the set (5, 40, 25) and the data group (+ 0.40; 3.5; 149) to the set (49, .49, 49).

 In the following operation (block 130), the three integers obtained previously are used to calculate a box address corresponding to the group of data examined. For example, if each whole number is between 0 and 49, it takes 125,000 addresses to have an address corresponding to each set of three integers (i.e. 50 x 50 x 50 box addresses).

If the memory 66 of FIG. 1 (for example the toroidal memory of a universal digital computer) comprises at least 125 K locations of words numbered sequentially, a box address CA can be calculated by the following formula
CA = X + 50Y + 2500Z
Thus, 11 set (O, 0, 0) corresponds to address 0, the set (1, 1, O) at address 51, and the set (1, 1, 1) at address 2551 .

 In step 132, the content of the box whose address has been calculated is increased by one.

After step 132, the next group of data is read when there are still such troops in memory 62. When all the groups of data from the first survey have been processed, it is examined whether there are additional groups of data coming from another poll (block 134) in which case they are read in block 114. If the answer to block 134 is NO, go to step 136 which consists in printing a distribution list of the statistical model. An example of distribution list is given in Table 1 below in which the column designated by "class" indicates the content of the memory cells, the column designated by "frequency of occurrences" indicates the number of cells having the content indicated in the first column, and the next two columns have legends that explain themselves.

Table 1
Frequency Frequency list
Cumulative frequency occurrence class
(C) (F) (C x F)
0 121 430 0 0
1 1953 1953 1953
2,780 1,560 3,513
3,413 1,239 4,752 4,281 1,124 5,876
5,205 1,025 6,901
6,133,793 7,699
7,106,742 8,441
8 82 656 9097
9 79 711 9808
10 46 460 10 268
11 45 495 10763
12 46 552 11315
13 27 351 11 666
14 24 336 12002
15 19 285 12287
16 17 272 12550
17 11 187 12746
18 12 216 12962
19 8 152 13 114
20 10 200 13314
21 3 63 13 377
22 2 44 13 421 23 @ 6 O 138 13 559
24 3 72 13 631
25 2 60 13 681
26 5 130 13311
27 1 27 13838
28 1 28 13 366
29 3 87 13 953
31 1 31 13 984
32 1 32 14016
33 1 33 14049
34 1 34 14063
36 2 72 14 155
41 1 41 14 196
FIG. 9 represents a detailed embodiment of the step of the method relating to the calibration of a logging. The first ten operations represented by blocks 138, 140, 142, 144, 146, 148, 150, 152, 154, 156 correspond respectively to operations 112, 114, 116, 118, 120, 122, 124, 126, 128, 130 of FIG. 8 The only difference resides in the fact that the groups of data examined now contain a measure of a diagram to calibrate Suppose, for example, that you want to calibrate the neutron log
After block 156, the content of the box whose address has just been calculated is read and added to the content of a central accumulator (block 158).

The central accumulator can be a register of the digital computer or simply a memory location.

We also calculate (block 160) ten additional addresses by the expression:
CA = X1 + 50Y + 2500Z
X1 = X + D with #
D = - 5, -4, ... - 1, 1, 2, ... 5
For example, if the address calculated in block 156 is 20, the ten additional cdresses are respectively 15, 16, 17, 18, 19, 21, 22, 23, 24 and 25
The contents of the boxes having the addresses thus calculated are added (block 162) to the contents of the respective lateral accumulators taken in ascending order
We determine (block 152) if there are groups of data for the same survey and if the answer is YES, we return to block 140 to read the next group of data. If there are no more groups of data, we determines the function representing the number of the accumulators in relation to their content (block 164), that is to say a function of the type represented in FIG. 7.

 The top of the curve thus obtained (block 166) is determined, as in FIG. 7, and the offset of the top of this curve relative to the central accumulator (block 168).

We calculate the offset of zero (block 170) in logging units, by performing the operation im @ rse of block 154. In the example of Figure 7, the offset of the top of the curve is - 1.61 box which corresponds to a zero offset of 0.0161 unit porosity for neutron logging.

We then print (block 172) a histogram representing the previously determined function and the values calculated in blocks 168 and 170.

 The particular log (i.e., neutron log) can be calibrated (block 174) by adding the calculated zero offset algebraically to each measurement.

 If another log has to be calibrated (block 176), we start again at block 140 in order to calibrate this other log in the data groups.

If the density log is to be calibrated, the addresses calculated in operation 160 are obtained according to the following formula
CA = X + 50Y '+ 2500Z
Y '= Y + D with #
D = - 5, -4, ... - 1, 1, 2, ... 5
These addresses correspond to adjacent boxes in the direction of the log to be calibrated. A similar technique can be used to calibrate the acoustic logging.

 When it is desired to calibrate more than one log, the method can be modified as shown in FIG. 10. In particular, if it is desired to calibrate each of the three logs, step 158 of FIG. 9 is followed simultaneously or sequentially through steps 160a, 160b and 160c.

 Step 160a corresponds exactly to operation 160 of FIG. 9.

 Step 16Ob is intended for calibrating the density log and involves calculating the addresses for the adjacent cells according to the size of this density log. Step 160c is intended to calibrate the acoustic diagram. The adjacent boxes are taken according to the size of this acoustic log. For each calibration there is a set of ten side accumulators. in addition, in block 154 three different functions are determined which are treated separately in the following steps.

The invention is applicable to groups of data comprising more than three different measurements with, for example, an additional resistivity measurement. The statistical model is therefore four-dimensional. Such a statistical model can be written into a two-dimensional torus memory by calculating the addresses of the boxes by an expression of the form
CA = X + aY + bZ + cR a, b, c being integer constants and X, Y, Z and R being the measures converted to whole numbers. If the integers representing each measurement are between O and 24, the four-dimensional statistical model requires a memory with 590,625 locations. If one byte is assigned to each location, a torus memory of 512 K bytes is sufficient h If one assigns a word of two bytes in each cell, a torized memory of 10,024 K bytes is required.

 There are cases where certain logs in a borehole are of good quality but where a particular log is of poor quality or non-existent. The invention makes it possible to reconstruct a missing or poor quality log using other logs in this same borehole and by means of a statistical model of the field in which the borehole is located. A particular embodiment of this technique is shown in FIG. 11.

Beforehand, it is assumed that a statistical model of the field has been produced as described with reference to FIG. 8, and that there are at least two logs from the particular borehole for which a log is missing.

For example, for a given borehole, neutron and density logs are available, but the acoustic log is missing. To reconstruct the latter, groups of data are formed which have only two elements, namely a neutron measurement and a density measurement. Next, a series of operations similar to those of bloos 138 to 154 in FIG. 9 is carried out, the only difference being that the data groups comprise only two elements instead of three.

In the example cited, each group of data is replaced by two whole numbers (X, Y). These two whole numbers are used (block 178) to calculate addresses of boxes according to the expression:
CA = X + 50Y + 2500Z
with Z = 0, 1, 2, .... 49 one thus obtains 50 addresses of boxes, that is to say a row of boxes according to the dimension of the missing log.

 The contents of these 50 boxes are examined (block 180) to determine the box with the highest content.

 This box has an address whose third element (which is an integer between O and 49) is transformed back into acoustic measurement units. The value thus obtained, expressed in acoustic units, is combined (block 184), with the two other elements of the data group read in step 140 to form a data group with three elements - (N PB At). This three element data group is stored (block 186) in an appropriate memory, and block 140 is started again to read another two element data group.

Each group of data with two elements of a survey is associated with a third element which represents the missing log.

Another embodiment of the method can be used to reconstruct a missing log, in order to determine a characteristic of formations crossed by m borehole when this characteristic cannot be obtained from data originating from this borehole. This is the case, for example, when the number of logs recorded in the survey is too low. In an example of this other embodiment, it is assumed that the characteristic to be reconstructed in the borehole is the porosity of the formations, and that in this borehole an acoustic log is available.
At and of a GR gamma ray log. In one or more other field surveys the number of logs is sufficient to calculate the porosity, level by level, by processing the log data according to a well known method, such as that described. in French patent No. 2,102,380. Note that the characteristic we are looking for may have been obtained in another field survey from sources other than logs such as core analyzes, test results or production data. We can, for example, determine the permeability of formations from core analyzes. Acoustic logs and gamma rays are also available in the field soundings for which the porosity PI has been calculated.

In this embodiment, a statistical model of the field is created as described with reference to FIG. 8, except that the data groups read include the three elements (At, GR) in step 114, and that only two elements ( At, GR) are used to calculate a box address in block 130. In step 132, two numbers are stored in each box. As described above, a first number is increased by one for each group and is representative of the total number n of data groups entered in the box. A second number is increased by the porosity value and is representative of the total # # of the porosity values entered in the box. When all the data has been processed, the total number or second number is divided by the first number in each box to provide the average value i / n so that the statistical model can be represented by a two-dimensional memory of boxes, plow box containing the average thief of pl corresponding to a specific address (At, GRt.

To determine values of the characteristic # at each level of the survey studied, form roups of one from said survey, as described with reference to Figure 9 mi @ apart from the fact that the data groups contain respectively only da @@ elements. see an acoustic measurement #t and a gamma ray measurement GR corresponding to the same level of the survey studied. The following steps are the same as blocks 138 to 154 of FIG. 9, the only difference being that the groups of data contain only two elements as in the method described above for recons titating a missing log, opposite to the three elements of the data groups used in the method illustrated in FIG. 9. After step 154 the groups of data used to determine the characteristic # consist of two whole numbers. In the example given here, each group of data is converted into a pair of whole numbers (X, Y) which is used to calculate a box address according to the expression:
CA = X + 50Y
The content (average of #) of the box whose address was calculated in the previous step is read in a non-destructive way and is taken as the value of the characteristic corresponding to the group of data examined (#t, GR). We return to the first operation to read another group of data with two elements. Once this process has been repeated for all the two-element data groups from the survey in question, a list of porosity values # of the formation is obtained, level by level. These values can be recorded in the form of a porosity log calculated for the survey concerned according to the depth. As seen above, the statistical model of the field can be created using boxes with three or more dimensional addresses, the average of the characteristic value being calculated for each box. The reduced data groups then include a number of elements equal to the number of dimensions to calculate a box address.

Claims (16)

CLAIM 1. Method for automatic processing of logging data for the study  of formations crossed by surveys, characterized in that it comprises  the following steps: produce several parameters respectively represented  sensitive to several characteristics of the forinatons for at least  sounding sections crossing said formations in a field; form  first groups of data, each comprising a defined combination  said parameters; combine at least some of the said groups  data to form a statistical model of the formations crossed; and  combine tenth groups of data from a survey in the  same field with said statistical model to determine which modification  must be brought to said second groups of data to bring them into  statistical compliance with said statistical model. 2. Treatment method according to claim 1, characterized in that  said data groups include values of said parameters taken  at different levels of a ringing. 3. Treatment method according to claim 2, characterized in that the  data groups include each of the values of said parameters taken  at the same level of a survey. 4. Treatment method according to claim 3, characterized in that  the step of combining the data groups to form a modular statistic  that includes the operation of forming a statistical distribution of the number  of occurrences of said defined combinations of parameters. 5. Treatment method according to claim 4, characterized in that  the step of combining the groups of data to form a statistical model that includes the following operations S focus a multidimensional memory-  number of boxes, each dimension of which corresponds to one of the different parameters included in a data group, and each box of which corresponds  to a defined combination of said parameters ;; and memorize in each of  boxes the number of occurrences of the data groups including the combi-  the parameters corresponding to this box in order to form said  statistical distribution and thus the statistical model. 6. Treatment method according to one of claims 4 and 5, characterized  in that it comprises the 17 step of converting each of the parameters constituting  a group of data in an integer.  7. Treatment method according to one of claims 4 to 6, characterized in that it includes the step of pre-processing groups of data  to improve the quality.  8. Treatment method according to claim 7, characterized in that said prior processing comprises the operation of applying an offset of  zero to at least one parameter selected from at least several groups of data.  9. Treatment method according to one of claims 7 and 8, characterized  in that said prior processing includes the operation of modifying  the scale by a given multiplication factor of at least one of the parameters  very in at least several groups of data.  10. Treatment method according to one of claims 1 to 9, characterized in that the step of combining the second data groups has completed the statistical model includes the following operations  forming second groups of data from said survey; cha one of the second groups of data comprising the same parameters as  said first groups of data; and,  combine the second data groups with the statistical model  to determine a calibration correction to be made to at least one  said parameters to bring the second data groups into conformity statistical tee with the statistical model.  11. Treatment method according to claims 5 and 10 taken in combi  naison, characterized in that the step of combining the data groups  with the statistical model includes the following have a central accumulator and several accumulators lateral; for every two th group of data from said survey in the same field, enter the content of the ase in the central accumulator corresponding to the combination of parameters constituting said troop of data, and write the contents of each side accumulator  respective adjacent boxes according to a defined dimension; and  det @ @ mine the top of a curve corresponding to position of said accumulators according to their content, and determining the offset of said  summit by sport to the central accumulator to obtain a correction  calibration -i must be anport4e to the second data groups for bring them into statistical conformity with the statistical model. l9. Treatment method according to one of claims 10 and 11, character  se in that the step of combining the second data groups com takes an integer conversion of each of the parameters constituting  a second group of data. 13. Treatment method according to one of claims 10 to 12, characterized  in that it includes the step of submitting the second datasets has a prior treatment before said combining step. 14. Treatment method according to claim 13, characterized in that said prior processing includes the operation of applying an offset  from zero or a change of scale to at least one selected parameter  in several of the second data groups. 15. Treatment method according to one of claims 1 to 9, characterized  in that the step of combining the second data groups with the  statistical model includes the following operations  form second groups of data from said survey, each  that second group of data being formed by a reduced combination of  parameters of the first data groups from which a parameter is removed missing; and  combine the second reduced data groups with the model statistics to provide, for each of the second data groups, a value of said missing parameter in order to reconstruct a log  missing in said survey. 16. Treatment method according to claims 5 and 15 taken in combi  naison, characterized in that the operation of combining the groups reduced data includes for each second data group the opera following:  define several boxes which correspond to the reduced combination  of parameters of the data group t which are found according to the dimension  the missing parameter;  determine the box with the highest content; and  replace the reduced data group with the corresponding data group  laying in the box with the highest content, in order to provide said missing parameter in this box to reconstruct the missing log. 17. Treatment method according to claim 15, characterized in that the step of combining the first groups of data to form a statistical model includes the following operations: form a multidimensional memory of boxes of which each dimension corresponds to one of the parameters of the reduced combination of the second  data groups and each box of which corresponds to a reduced combination  definition of said parameters included in a first group of data; and  memorize in each of the boxes a statistical optimum of the parameter  missing for data groups including the reduced combination  defined parameters corresponding to this box in order to form said model  the statistic. 18. Pr @ cede de solen of claim 17, characterized in that  the step of combining the second data groups with the model  statistics including the following operations for each second group  of data : define a box in the statistical model that corresponds to the  reduced combination of parameters from the second data group; and read the content of the box thus defined to provide said value @  of the missing parameter in order to reconstruct the missing log. 19. Treatment method according to one of claims 15 to 18, characterized we want to produce several representative parameters of charac  training details including the operation of combining a plurality  of logging measurements along said borehole sections to obtain at least one of said parameters.