CN110852144A - A DTW-based intelligent stratigraphic comparison method and system - Google Patents

Abstract

The invention provides an intelligent stratum comparison method and system based on DTW (delay tolerant W). firstly, original logging data are obtained and normalized preprocessing is carried out to form normalized logging data; then extracting variable point information by adopting a variable point detection algorithm to layer the logging curves, thereby dividing each well into stratum units of different levels according to the stratum sequence; and finally, carrying out stratum comparison and matching: and calculating the form similarity information of the logging curves of the two selected wells based on a dynamic time warping algorithm (DTW), calculating and processing the curve characteristics by at least combining the formation depth, thickness and mean value information through a dynamic waveform matching algorithm to determine a minimum matching path, backtracking the minimum matching path to match the formation units and judging whether the formation is missing or not and outputting. The intelligent comparison of the stratum can be more accurately carried out, and the stratum comparison efficiency is improved.

Description

A DTW-based intelligent stratigraphic comparison method and system

technical field

The invention relates to the technical field of formation data processing, in particular to an intelligent formation comparison method and system based on DTW.

Background technique

With the gradual deepening of oilfield exploration work, the exploration objects are becoming more and more complex, and the research work on the selection of exploration targets and the deployment of exploration wells is also becoming more and more refined. Oil and gas exploration and development is a process of continuously collecting, processing and converting data of underground oil and gas reservoirs into information. There is information in the data and oil and gas resources in the information. How to maximize the use of advanced software and hardware tools in a unified computing environment, so that these data reflecting complex geological objects can be integrated comprehensively and efficiently and displayed in front of researchers in a visual way intuitively and conveniently, so as to understand the geological Data can be more deeply analyzed and applied, and then provide decision support for exploration work, which has become the development needs of the oil industry.

Stratigraphic correlation is one of the basic tasks for the study of stratigraphic structure, structure and sedimentary environment, and it is an indispensable work in oil and gas exploration and development. The quality of manual comparison depends on people's knowledge structure and experience accumulation. Therefore, in the same block, the interpretation of different personnel is also different, the repeatability is poor, and the multi-solution is strong. When logging geologists perform stratigraphic comparison, they consider not only the curve. shape, but a variety of information of the comprehensive curve, and the optimal solution is obtained based on rich knowledge, but this comparison is labor-intensive and low-efficiency; the existing computer programs are compared based on parameters such as depth, thickness, mean and variance. The comparison effect is extremely dependent on information such as depth and thickness. When comparing large cycles, the lithological characteristics cannot be used, and the geological significance reflected by the curve shape information is not considered. Machine learning and other methods are used to cluster and classify the strata. , this method does not consider the Walsau facies law, sequence stratigraphy and other geological laws, but only considers the similarity of curve parameters, which lacks interpretability.

SUMMARY OF THE INVENTION

In order to solve the problems of high intensity, low efficiency, poor repeatability, and lack of interpretability of existing computer program comparison methods in the prior art, the present invention provides an intelligent formation comparison method based on DTW, which is based on DTW (dynamic Time warping) algorithm, which integrates the characteristic parameters such as log curve shape similarity, depth, thickness and mean value to carry out the log curve matching cost, and then uses the dynamic waveform matching algorithm to perform stratigraphic comparison to match the stratigraphic unit, which can be performed more accurately. The intelligent comparison of strata improves the efficiency of stratum comparison. The invention also relates to an intelligent stratigraphic comparison system based on DTW.

Technical scheme of the present invention:

A DTW-based intelligent stratigraphic correlation method, comprising the following steps:

The first step: obtaining the original logging data and performing normalization preprocessing to form normalized logging data;

The second step: using a sliding window-based change point detection algorithm to extract the change point information of the normalized logging data, so as to layer the logging curves of each well to obtain stratigraphic units of different levels;

The third step: extracting the log curve features of the wells to be compared, the log curve features at least include log curve shape similarity, depth, thickness and mean information, and the log curve shape similarity information is obtained through the DTW algorithm and then determine the matching path of the formation by calculating and processing the log curve feature through the dynamic waveform matching algorithm, and backtracking the matching path to match the formation unit to judge whether the formation is missing and output.

The third step includes the following steps:

S31 : extracting the log curve shape similarity and at least depth, thickness and mean value information of the well to be compared, and the log curve shape similarity is calculated and obtained by the DTW algorithm;

S32: Calculate the distance coefficient/local matching cost of each formation unit of the comparison well in turn to establish a local matching cost matrix;

S33: Based on the local matching cost matrix, obtain a global matching cost matrix through a dynamic waveform matching algorithm, and filter out the minimum global matching cost in the global matching cost matrix;

S34: Backtracking the minimum global matching cost to obtain the shortest path, and matching the formation units according to the judgment criterion to judge whether there is a missing formation and output it.

The first step includes the following steps: for the logging data from different sources, firstly extract the data accounting for 95% of the logging data from each source as the maximum value and the data accounting for 5% as the minimum value, and then combine the normalization data. The normalized logging data of each source is obtained by using the normalized calculation formula, so that the numerical range of the logging data from each source is distributed between [0, 1].

The second step includes the following steps: selecting two sliding windows of equal length along the data stream, and comparing each statistical characteristic of the two sliding windows through a change point detection algorithm to obtain a difference metric; according to the difference metric and the data stream Draw a difference curve according to the relationship; detect the peak value of the difference curve, repeat the above process along the data stream until the change point detection in all data streams in each well is completed, and stratify the log curve at each change point, Obtain different levels of stratigraphic units.

The change point detection algorithm calculates the difference metric as the difference between the loss function in the selected two sliding windows and the loss function of each sliding window alone, and the loss function detects the projection of the Gaussian function to the high-dimensional regeneration Hill. The mean change of the signal in Bert space.

A DTW-based intelligent stratigraphic comparison system, comprising a first device, a second device and a third device connected in sequence,

The first device obtains original logging data and performs normalization preprocessing to form normalized logging data;

The second device adopts a sliding-window-based change-point detection algorithm to extract the change-point information of the normalized logging data, so as to stratify the logging curves of each well to obtain formation units of different levels;

The third device extracts the log curve features of the wells to be compared, the log curve features at least include log curve shape similarity, depth, thickness and mean information, and the log curve shape similarity information is obtained through DTW Algorithm acquisition; then calculate and process the log curve features through a dynamic waveform matching algorithm to determine a matching path, backtrack the matching path to match formation units to determine whether there is a missing formation and output.

The first device includes a data acquisition device and a normalization preprocessing device connected in sequence, the data acquisition device receives the original logging data and sends it to the normalization preprocessing device, and the normalization preprocessing The device converts raw data from different sources into normalized logging data distributed in the range of [0,1].

The second device includes a change point detection device and a layering device connected in sequence, the change point detection device includes a data selection device, a difference characteristic calculation device, a curve generation device and a peak detection device connected in sequence, the data selection device It is connected with the normalization preprocessing device and is used for receiving normalized logging data and selecting two sliding windows of equal length along the data stream. The difference characteristic calculating device calculates each statistical characteristic of the two sliding windows. Comparing the difference metric value calculated by the change point detection algorithm, the curve generating device draws the difference metric along the data stream to draw a difference curve, the peak detection device detects the peak value of the difference curve, and the layering device is related to the peak value. The detection device is connected for layering the log curve at each change point to obtain different levels of formation units.

The third device includes a curve shape similarity calculation device, a feature extraction device, a matching device, a missing judgment device and an output device connected in sequence, and the curve shape similarity calculation device is connected with the layering device, and calculates based on the DTW algorithm. The morphological similarity of the logging curves between any two formation units of the two logging curves to be compared, the feature extraction device extracts the morphological similarity of the logging curves and each of the two logging curves to be compared At least the depth, formation thickness and mean value information of the formation unit, the matching device sequentially calculates the distance coefficient/local matching cost of each formation unit of the comparison well to establish a local matching cost matrix, and based on the local matching cost matrix, through dynamic waveform matching The algorithm obtains a global matching cost matrix, filters out the minimum global matching cost in the global matching cost matrix, backtracks the minimum global matching cost to obtain the shortest path, and the missing judging device matches the formation unit according to the judgment criterion and judges whether there is any If the formation is missing, the output device outputs the result.

The matching device includes a local matching cost matrix generating device, a global matching cost matrix generating device and a shortest path backtracking device connected in sequence, and the local matching cost matrix generating device sequentially calculates the distance coefficient/local matching of each formation unit of the comparison well The cost establishes a local matching cost matrix, and the global matching cost matrix generating device obtains a global matching cost matrix through a dynamic waveform matching algorithm based on the local matching cost matrix, and selects the minimum global matching cost in the global matching cost matrix. The shortest path backtracking device backtracks the shortest matching path with the minimum global matching cost, and matches the formation unit to determine whether there is a missing formation.

Beneficial technical effects of the present invention:

A DTW-based intelligent stratigraphic comparison method of the present invention first obtains the original logging data and performs normalization preprocessing to form normalized logging data, so that the numerical range of the logging data from each source is distributed in [0,1 ]; then use the change point detection algorithm to extract the change point information of the normalized logging data of each well to stratify the logging curve, so that each well is divided into different levels of stratigraphic units according to the stratigraphic sequence ; Finally, stratigraphic comparison and matching are carried out: based on the dynamic time warping algorithm (DTW), the log curve shape similarity information of the two selected wells is calculated, and at least combined with the formation depth, thickness and mean information, the dynamic waveform matching algorithm is used to calculate and process all the data. The curve feature is used to determine the minimum matching path, and the minimum matching path is backtracked to match the formation unit and determine whether the formation is missing and output. The method of the invention uses the dynamic time warping algorithm for each stratum unit after stratification, obtains the log curve shape similarity information and excavates the geological law contained in the log curve shape, and simultaneously calculates the depth, thickness, lithology and other parameters. The local matching cost between each stratigraphic unit between the wells to be compared is established, and a local matching cost matrix is established. Based on the local matching cost matrix, the global matching cost matrix is preferably obtained according to the shortest path model of stratigraphic comparison. The minimum global matching cost is screened out from the global matching cost matrix, and each item accumulated on the acquisition path of the shortest global matching cost is backtracked to obtain the shortest matching path. One-to-one correspondence, the two match each other. If one formation unit of one well matches multiple formations of another well, the formation unit with the smallest local matching cost among the multiple formations is the matched formation unit, and the others are missing formation units. To sum up, this method integrates the log curve shape similarity, thickness, depth and other information, and based on the geological understanding that adjacent wells should not have a large difference in distance, the dynamic waveform matching algorithm is used to judge the matching results, and the minimum global matching is used. The shortest path corresponding to the cost is used as the matching path, so that the stratigraphic comparison can be performed more accurately. Compared with the manual method and the existing computer program method, it is fast, efficient and highly interpretable.

The present invention also relates to a DTW-based intelligent formation comparison system, which corresponds to the intelligent formation comparison method of the present invention, and can also be understood as a device for implementing the intelligent formation comparison method of the present invention. A device, a second device and a third device, each device works together, the second device can perform change point detection on the normalized data obtained by the first device, and layer the logging curve to obtain different levels of stratigraphic units. The device first extracts at least the log curve shape similarity, depth, formation thickness and mean value information of the comparison well, wherein the log curve shape similarity is obtained by a dynamic time warping algorithm, and then the dynamic waveform matching algorithm is used to calculate the comparison of the two wells. The local matching cost between different formation units, establishing a local matching cost matrix, obtaining a global matching cost matrix based on the local matching cost matrix, filtering out the minimum global matching cost value in the global matching cost matrix, and obtaining the minimum global matching cost value from the minimum matching cost matrix. The shortest matching path is obtained by backtracking at the global matching cost value. If the formations of the two wells are not missing, each formation unit between the two wells is matched one-to-one, and each point on the shortest matching path corresponds to the matching two. If there are missing formations in two wells, one formation unit of one well matches multiple formation units of another well on the shortest matching path. At this time, the matching principle is: the one with the smallest global matching cost is the one that matches. Stratigraphic units, those that are not matched are missing stratigraphic units, and finally output the matching results. This method avoids the problems of high intensity, low efficiency, and poor repeatability of artificial stratigraphic correlations in the prior art, and can automatically and intelligently complete stratigraphic correlations, improving the Stratigraphic comparison accuracy and work efficiency.

Description of drawings

Fig. 1 is a kind of optimal scheme flow chart of a kind of DTW-based intelligent formation comparison method of the present invention;

Fig. 2 is another preferred solution flow diagram of a DTW-based intelligent formation comparison method of the present invention;

Fig. 3 is each curve generated in the change point detection process;

FIG. 4 is a schematic diagram of an embodiment of a logging curve after layering;

5 is a schematic diagram of an embodiment of a global matching cost matrix with a label;

FIG. 6 is a schematic diagram of an embodiment of a DTW-based intelligent stratigraphic comparison system of the present invention.

Reference number:

1- logging curve; 2-marker formation; 3-other formations.

Detailed ways

In order to introduce the present invention in detail, specific embodiments and accompanying drawings will be described in detail.

As shown in Figure 1, a DTW-based intelligent stratigraphic comparison method includes the following steps: the first step: obtaining original logging data and performing normalization preprocessing to form normalized logging data; The point detection algorithm extracts the change point information of the normalized logging data of each well, and stratifies the logging curves to obtain different levels of stratigraphic units; the third step: extracts the logging curve characteristics of the comparison wells, so The curve features at least include log curve shape similarity, depth, thickness and mean value information, the log curve shape similarity information is obtained by a dynamic time warping algorithm, and the curve features are calculated and processed by a dynamic waveform matching algorithm to determine a path pattern diagram. And find the optimal matching path, backtrack the optimal matching path, match the formation unit, judge whether the formation is missing and output. That is, in order to obtain more accurate stratigraphic comparison results for petroleum exploration and development, the present invention extracts the morphological similarity of the logging curves of each stratigraphic unit of Well A and Well B in the pre-comparison based on the dynamic time warping algorithm, fuses other conventional information, The dynamic waveform matching algorithm calculates the distance coefficient between the formation units of Well A and Well B to establish a local matching cost matrix, obtains the global matching cost matrix, filters out the minimum global matching cost, and traces the optimal matching path to obtain matching and missing results. , the morphological similarity information of the logging curve is considered in the judgment. When comparing, firstly the local and then the global, and then reverse from the global to the local to ensure the optimal matching path. At this time, the difference between the contrasting wells is the smallest and the matching degree is high, and it cannot be matched. If there are missing stratigraphic units, this method is more accurate and efficient, and the stratigraphic comparison results are more interpretable.

Preferably, a preferred embodiment of a DTW-based intelligent stratigraphic comparison method of the present invention is shown in FIG. 2 . Include the following steps:

S11: Obtain original logging data from different sources;

S12: Perform normalization preprocessing on the original logging data, so that the numerical ranges of the logging data from each source are distributed between [0, 1] to form normalized data;

The normalization preprocessing is mainly to solve the problem that the numerical range of the logging data results measured by different instruments will be too large and the error will be too large for the subsequent lithology prediction of the whole area. The preprocessing is to target logging data from different sources, so that all logging data are distributed in the range of [0, 1]. The normalization process does not take the maximum and minimum values of the entire data, but for all data. Make a cumulative probability distribution map, for example, it is preferable to first extract the number that accounts for 95% of all logging data from each source as the maximum value, and then extract the number that accounts for 5% as the minimum value (of course, you can also extract other The maximum and minimum values are used to normalize all the logging curve data to obtain the normalized results of the logging data from each source, so that the logging data from each source can be normalized. The numerical range of well data is distributed between [0, 1], and the calculation formula of the normalized preprocessing is as follows:

Among them, X represents the logging curve data to be normalized, and X' represents the normalized logging curve data, and the calculation process will traverse all the logging curve data.

S2: using a change point detection algorithm to extract the change point information of the normalized logging data, so as to layer the logging curves of each well to obtain stratigraphic units of different levels;

The change point detection algorithm based on sliding window is used to divide different normalized logging curve combinations into multiple layers, which is fast and applicable to multi-variables, and can expand any single-change-point detection algorithm to a multi-variable-point detection algorithm . The algorithm uses two windows that slide along the data stream. The statistical properties of the signal within each window are compared with the variance measure. For a given loss function c( ), a difference measure d( , ) is derived:

d(y _u..v ,y _v..w )=c(y _u..w )-c(y _u..v )-c(y _v..w ) (2)

where {y _t } _t is the input signal, u < v < w is the index, c( ) is the loss function, and d( , ) is a measure of the difference between the two windows.

In this paper, the kernel function is selected as the loss function to project the low-dimensional linear inseparable and highly nonlinear logging signals into the high-dimensional reproducing Hilbert space, English is reproducing Hilbert space, English abbreviation is RKHS, thus achieving linearity divisible purpose.

和其映射关系

H是一个合适的希尔伯特空间，我们的损失函数检测投影之后的信号{Φ(y_t)}_t的均值变化，定义为：Given a positive semi-definite kernel

and its mapping

H is a suitable Hilbert space, and our loss function detects the mean change of the signal {Φ(y _t )} _t after projection, defined as:

为信号{Φ(y_t)}_t∈I的均值，此处所用的核函数为径向基核函数(rbf)：In formula (3), I is the interval of interest,

is the mean value of the signal {Φ(y _t )} _t∈I , and the kernel function used here is the radial basis kernel function (rbf):

k(x,y)=exp(-γ||xy|| ² ) (4)

In the formula, ||·|| is the Euclidean norm, and γ>0 is the so-called bandwidth parameter, which is determined according to the median heuristic, which is equal to the reciprocal of the median of all pairwise distances.

According to the following principle: if the sliding windows u..v and v..w are located in the same layer, their statistical properties are similar, and the difference between the two windows is small; if the sliding windows are located in two different layers, The difference increases significantly, which indicates that the boundary between the two windows is a change point, and the difference curve is defined as:

(t,d(y _tw/2..t ,y _t..t+w/2 )) (5)

Where t is the index, between w/2 and n-w/2, n is the number of sample points; w is the length of the sliding window;

According to the relationship between the difference metric and the data flow, a difference curve is drawn, as shown in FIG. 3, based on the difference curve, the peak detection is continued, and the peak point corresponds to the change point of the original signal, that is, the layer interface of the logging curve;

The above process is repeated along the data stream until the change point detection within the entire data stream in each well is completed, and the logging curves are layered to obtain formation units of different levels. As shown in Fig. 4, for the layered results of logging curve 1 for some signal sources in a well, the position of the black vertical dotted line is the change point. Change point, it can be seen that the layering effect of change point detection is relatively accurate, among which CALI means caliper logging signal source, GR means natural gamma logging signal source, SP means natural potential logging signal source, RT means deep direction sounding resistivity measurement Well signal source, RI stands for shallow direction finding resistivity logging signal source, and of course other signal sources can be included. In theory, the logging curves of all signal sources in a well need to be uniformly layered;

S31 : extracting the log curve shape similarity and at least depth, thickness and mean information of the comparison well, where the log curve shape similarity is calculated and obtained by a dynamic time warping algorithm;

In addition to the above-mentioned curve features, other features such as mean square error can also be selected.

(1) Depth f1

f1=(y1+y2)/2 (6)

(2) Formation thickness f2

f2=y2-y1 (7)

y1: stratum top depth of any stratum unit of each well, y2: stratum bottom depth of any stratum unit of each well;

(3) Mean value of logging response f3

f3=(a1+a2+…ai…+an)/n (8)

ai: logging curve value of the i-th point;

(4) Curve shape similarity f4

Before the similarity of the two curves, this embodiment uses the DTW algorithm to warp and distort one or two sequences under the time axis to achieve better alignment, and then calculate the similarity between the two time series.

Suppose the time series of wells A and B are Q and C, respectively, and their lengths are n and m, respectively: the value of each point in the sequence is the eigenvalue of each frame in the sequence:

Q=q1,q2,…,qi,…,qn;

C=c1,c2,…,cj,…,cm;

In order to align the two sequences, we need to construct an n×m matrix grid, the matrix element (i,j) represents the distance d(qi,cj) between the two points qi and cj, that is, each point of the sequence Q The similarity with each point of C, the smaller the distance, the higher the similarity, and the Euclidean distance d(qi,cj)=(qi-cj) ² is generally used, which can also be understood as the degree of distortion. Each matrix element (i,j) represents the alignment of points qi and cj. The DTW algorithm can be summed up as finding a path through several grid points in this grid to minimize the distance, and the grid point passed by the path is the alignment point calculated by the two sequences.

We define this path as a warped regular path and denote it by W, and the kth element of W is defined as wk=(i,j)k, which defines the mapping of sequences Q and C:

W=w1,w2,...,wk max(m,n)≤=K<m+n+1 (9)

This path is not chosen arbitrarily and needs to satisfy the following constraints:

I boundary conditions: w1=(1,1) and wK=(m,n), the speed of change of logging signals of different logging tools may change, but the order of each part cannot be changed, so the selected The path must start from the lower left corner and end at the upper right corner;

II Continuity: If wk-1=(a',b'), then for the next point of the path wk=(a,b) needs to satisfy (a-a')<=1 and (b-b')< =1, that is, it is impossible to match across a certain point, and it can only be aligned with its own adjacent point, which ensures that each coordinate in Q and C appears in W;

III Monotonicity: If wk-1=(a',b'), then for the next point of the path wk=(a,b) needs to satisfy 0<=(a-a') and 0<=(b-b '), which restricts the points above W to be monotonic over time to ensure that the dashed lines in Figure B do not intersect.

Combining continuity and monotonicity constraints, the path of each lattice point has only three directions. For example, if the path has already passed through the lattice point (i, j), the next lattice point to pass through can only be one of the following three cases: (i+1,j), (i,j+1) or (i+ 1,j+1), the number of paths that satisfy the above constraints can be exponential, what we need is the path that minimizes the following regularization cost:

The K in the denominator is mainly used to compensate for regular paths of different lengths. Here we define a cumulative distance. The two sequences Q and C are matched starting from the (0,0) point, and each time a point is reached, the distances calculated by all the previous points will be accumulated. After reaching the end point (n, m), this cumulative distance is the final total distance we mentioned above, that is, the similarity of the curve shapes of the sequences Q and C.

The cumulative distance γ(i,j) can be expressed in the following way, the cumulative distance γ(i,j) is the current grid point distance d(i,j), that is, the Euclidean distance/similarity of points qi and cj and the reachable distance The sum of the cumulative distances of the smallest neighbors of the point:

f ₄ =γ(i,j)=d(q _i ,c _j )+min{γ(i-1,j-1),γ(i-1,j),γ(i,j-1)} (11)

S32: Calculate the distance coefficient/local matching cost of each formation unit of the comparison well in turn to establish a local matching cost matrix;

The formula for calculating the local matching cost is:

d(A, b)=sqrt{[w1×diff(f1)] ² +[w2×diff(f2)] ² +w3×diff(f3)] ² +[w4×

diff(f4)] ² ) (12)

d(A, B) is the local matching cost, which represents the distance coefficient between the logging curve of the stratigraphic unit in Well A and the logging curve of the stratigraphic unit in Well B; the weight corresponding to the i-th attribute of wi is the set value; diff (i) The cost corresponding to the i-th attribute is the absolute value of the difference of fi between Well A and Well B, i=1-3; diff(f4) is the DTW distance f4 of the curve.

According to the above formula, the obtained local matching cost matrix is shown in Table 1. In Table 1, the vertical represents the formation units I to VII in Well A, the horizontal represents the formation units I-VII in Well B, and each point in the matrix represents the formation units I to VII in Well A and the formation units I-VII in Well B, respectively. The local matching cost of , the smaller the local matching cost, the higher the matching degree of the two stratigraphic units. Of course, Table 1 is only a form of local matching cost matrix. The formation units in each well are consecutively arranged from I to VII. It is also possible that the formation units of A and/or B lack one or some formation units, and I to VII may not be continuous. .

Table 1. Local matching cost matrix

S33: Based on the local matching cost matrix, obtain a global matching cost matrix through a dynamic waveform matching algorithm, and filter out the minimum global matching cost in the global matching cost matrix;

The global matching cost calculation formula is as follows:

D(A _m , B _n )=∑d(A _i , B _j )+∑g(A _k )+∑g(B _k ) (13)

D(Am, Bn) is the global matching cost, ∑d(Ai, Bj) is the sum of the distance coefficient/local matching cost between all matching formation units; k is the number of missing formations in the sequence, ∑g(Ak) Represents the sum of the distance coefficients between the missing stratigraphic units in Well A and Well B; ∑g(Bk) represents the sum of the distance coefficients between the missing stratigraphic units in Well B and Well A.

The stratigraphic correlation using the logging curve can be attributed to the optimal matching of ordered elements between systems. The invention uses the shortest path problem in dynamic programming to simulate the stratigraphic correlation process, and uses the dynamic programming optimization algorithm to complete the specific calculation of the stratigraphic correlation. That is, the minimization of the distance coefficient enables formation matching.

When using the dynamic waveform algorithm to solve the optimization results of the multi-stage decision-making process, the optimization calculation of the distance coefficient should be carried out in stages, that is, the optimization of the current stage is carried out on the basis of the optimization results of the previous stage, and the entire solution process of the well is a process. continuous recursive process.

According to the shortest path model of stratigraphic comparison, if the kth stage is to complete the best matching of A1, A2...Am and B1, B2...Bn, it is based on the three states in the first stage, that is, the first three points. After a match is optimized, the three states of the k-1 stage are:

State 1: at point (Ai-1, Bj-1); state 2: at point (Ai-1, Bj); state 3: at point (Ai, Bj-1); these three points to point (Ai, Bj-1) The distances of Bj) are d(Ai, Bj), g(Ai), g(Bj) respectively, so the recursive formula of the kth stage is:

D(Ai,Bj)=min{D(Ai-1,Bj)+g(Ai),D(Ai-1,Bj-1)+d(Ai,Bj),D(Ai,Bj-1)+ g(Bj)}

According to the above formula, the optimal matching distance D(A, B) of the sequence A and B can be obtained, that is, each point in the global matching cost is calculated by the above formula.

As shown in Figure 5, the global matching cost matrix is obtained according to the shortest path model through the dynamic waveform matching algorithm, and the path matrix is recorded during the distance calculation process. The convention is marked as: 1-from the left; 2-from the upper left;

For example, the three points of the first stage in Figure 5:

288.2/1 means 193.0 points (the stratigraphic unit No. I in well A matches the stratigraphic unit No. II in well B) + 95.2 points on the left (the stratigraphic unit No. I in well A matches the stratigraphic unit No. I in well B);

213.4/2 means 118.2 points (the stratigraphic unit No. II in Well A matches the stratigraphic unit No. I in Well B) + 95.2 points on the upper left (the stratigraphic unit No. I in Well A matches the stratigraphic unit No. I in Well B);

210.5/3 means 115.3 (formation unit No. II in well A matches formation unit No. II in well B) + 95.2 above (formation unit No. I in well A matches formation unit No. I in well B);

On the basis of the above three points, according to the shortest path model and so on, the values corresponding to the seventh row and the sixth column in Figure 6 are obtained as global matching cost values, of which 690.7/2 is the minimum global matching cost.

S34: Backtracking the minimum global matching cost to obtain the shortest path, and matching formation units to determine whether there is a missing formation and output;

As shown in Figure 5, the backtracking path of the global minimum matching cost 690.7/2 represented by the reverse arrow is:

690.7/2 (A's VII matches B's VII) → 618.2 (A's VI matches B's VI) → 550/2 (A's V matches B's V) → 407.7/2 (A's IV matches B's V's) IV matches) → 315/2 (A's III matches B's III)

→ 210.5/3 (A's II matches B's I) → 95.5 (A's I matches B's I);

Among them, the stratigraphic units of the two wells corresponding to each point value in 690.7/2, 618.2, 550/2, 407.7/2, and 315/2 correspond one-to-one, then VII-III of A and VII-III of B are mutually in turn The I of B matches both II and I of A, and the local matching cost of I of B and I of A is 95.2, which means that the I of B should match the I of A, and the II of A is missing.

The present invention also relates to an intelligent stratigraphic comparison system based on DTW. The intelligent stratigraphic comparison system corresponds to the intelligent stratigraphic comparison method of the present invention, and can also be understood as a device for realizing the intelligent stratigraphic comparison method of the present invention, as shown in FIG. 6 . , including a first device, a second device and a third device connected in sequence. The first device includes a data acquisition device and a normalization preprocessing device connected in sequence, the data acquisition device receives the original logging data, and sends it to the normalization preprocessing device, and the normalization preprocessing device The raw data from different sources are transformed into normalized logging data distributed in the range of [0,1]. The second device includes a change point detection device and a layered device connected in sequence, the change point detection device includes a data selection device, a difference characteristic calculation device, a curve generation device and a peak detection device, the data selection device and the The normalization preprocessing device is connected, receives the normalized logging data and selects two sliding windows of equal length along the data stream, and the difference characteristic calculation device compares each statistical characteristic of the two sliding windows through change point detection The algorithm calculates the difference metric value, the curve generation device draws the difference metric along the data stream to draw a difference curve, the peak value detection device detects the peak value of the difference curve, the layering device is connected with the peak value detection device, and the Logging curves are layered at each change point to obtain stratigraphic units of different levels.

The third device includes a curve shape similarity calculation device, a feature extraction device, a matching device, a missing judgment device and an output device connected in sequence, and the matching device includes a local matching cost matrix generating device, a global matching cost matrix generating device and a shortest matching cost matrix. A path backtracking device, the curve shape similarity calculation device calculates the log curve shape similarity between each formation unit of the two logging curves of the two wells compared based on a dynamic time warping algorithm, and the feature extraction device takes the The log curve shape similarity and at least depth, formation thickness and mean value information, the local matching cost matrix generating device sequentially calculates the distance coefficient of each formation unit of the comparison well to establish a local matching cost matrix, and the global matching cost matrix is generated. Based on the local matching cost matrix, the device obtains a global matching cost matrix through a dynamic waveform matching algorithm, filters out the minimum global matching cost in the global matching cost matrix, and the shortest path backtracking device backtracks the shortest of the minimum global matching costs Matching paths, backtracking the shortest matching path to match formation units to determine whether there is a missing formation, the output device outputs the above matching results.

It should be pointed out that the above-mentioned specific embodiments can make those skilled in the art understand the present invention more comprehensively, but do not limit the present invention in any way. Therefore, although this specification has described the invention in detail with reference to the accompanying drawings and embodiments, those skilled in the art should understand that the invention can still be modified or equivalently replaced. The technical solutions and improvements of the spirit and scope shall be covered by the protection scope of the invention patent.

Claims (10)

1. a kind of intelligent stratigraphic contrast method based on DTW, is characterized in that comprising the following steps: The first step: obtaining the original logging data and performing normalization preprocessing to form normalized logging data; The second step: using a sliding window-based change point detection algorithm to extract the change point information of the normalized logging data, so as to layer the logging curves of each well to obtain stratigraphic units of different levels; The third step: extracting the log curve features of the wells to be compared, the log curve features at least include log curve shape similarity, depth, thickness and mean information, and the log curve shape similarity information is obtained through the DTW algorithm and then determine the matching path of the formation by calculating and processing the log curve feature through the dynamic waveform matching algorithm, and backtracking the matching path to match the formation unit to judge whether the formation is missing and output. 2. a kind of intelligent formation contrast method based on DTW according to claim 1 is characterized in that the 3rd step comprises the following steps: S31 : extracting the log curve shape similarity and at least depth, thickness and mean value information of the well to be compared, and the log curve shape similarity is calculated and obtained by the DTW algorithm; S32: Calculate the distance coefficient/local matching cost of each formation unit of the comparison well in turn to establish a local matching cost matrix; S33: Based on the local matching cost matrix, obtain a global matching cost matrix through a dynamic waveform matching algorithm, and filter out the minimum global matching cost in the global matching cost matrix; S34: Backtracking the minimum global matching cost to obtain the shortest path, and matching the formation units according to the judgment criterion to judge whether there is a missing formation and output it. 3. A kind of intelligent stratigraphic comparison method based on DTW according to claim 1, is characterized in that described first step comprises the following steps: for logging data of different sources, first extract the proportion of logging data from each source respectively 95% of the data is taken as the maximum value, and 5% of the data is taken as the minimum value, and then the normalized logging data of each source of logging data are obtained by combining the normalization calculation formula, so that the numerical range of the logging data from each source is equal. Distributed between [0,1]. 4. a kind of intelligent stratigraphic comparison method based on DTW according to claim 1, is characterized in that described second step comprises the following steps: choose two sliding windows of equal length along data flow, by changing point detection algorithm Compare each statistical characteristic of the two sliding windows to obtain a difference measure; draw a difference curve according to the relationship between the difference measure and the data stream; detect the peak of the difference curve, and repeat the above process along the data stream until all data in each well is completed In-stream change point detection, the log is layered at each change point to obtain stratigraphic units of different levels. 5. a kind of intelligent stratigraphic comparison method based on DTW according to claim 4, is characterized in that described change point detection algorithm is with the loss function in the two selected sliding windows and the loss function of each sliding window independently. The difference computes the difference metric, and the loss function detects the mean change of the signal projected by the Gaussian function into the high-dimensional regenerated Hilbert space. 6. an intelligent stratigraphic comparison system based on DTW, is characterized in that comprising the first device, the second device and the third device connected in sequence, The first device obtains original logging data and performs normalization preprocessing to form normalized logging data; The second device adopts a sliding-window-based change-point detection algorithm to extract the change-point information of the normalized logging data, so as to stratify the logging curves of each well to obtain formation units of different levels; The third device extracts the log curve features of the wells to be compared, the log curve features at least include log curve shape similarity, depth, thickness and mean information, and the log curve shape similarity information is obtained through DTW Algorithm acquisition; then calculate and process the log curve features through a dynamic waveform matching algorithm to determine a matching path, backtrack the matching path to match formation units to determine whether there is a missing formation and output. 7. A DTW-based intelligent stratigraphic comparison system according to claim 6, characterized in that the first device comprises a data acquisition device and a normalization preprocessing device connected in sequence, and the data acquisition device receives the original data. The well data is sent to the normalization preprocessing device, and the normalization preprocessing device converts the original data from different sources into normalized logging data distributed in the range of [0,1]. 8. A DTW-based intelligent stratigraphic comparison system according to claim 7, wherein the second device comprises a change point detection device and a layering device connected in sequence, and the change point detection device comprises a change point detection device connected in sequence A data selection device, a difference characteristic calculation device, a curve generation device and a peak detection device, the data selection device is connected with the normalization preprocessing device, and is used for receiving normalized logging data and selecting equal-length logs along the data stream. Two sliding windows, the difference characteristic calculation device compares each statistical characteristic of the two sliding windows to obtain a difference metric value through a change point detection algorithm, and the curve generation device draws the difference metric along the data flow to draw the difference The peak value detection device detects the peak value of the difference curve, and the layering device is connected to the peak value detecting device for layering the logging curve at each change point to obtain formation units of different levels. 9. A DTW-based intelligent stratigraphic comparison system according to claim 8, wherein the third device comprises a curve shape similarity calculation device, a feature extraction device, a matching device, a missing judgment device and an output connected in sequence The device, the curve shape similarity calculation device is connected with the layering device, and the log curve shape similarity between any two formation units of the two logging curves to be compared is calculated based on the DTW algorithm, and the feature extraction The device extracts the morphological similarity of the logging curves and the at least depth, formation thickness and mean value information of each formation unit of the two logging curves to be compared, and the matching device sequentially calculates the distance coefficient of each formation unit of the comparison well / Local matching cost Establish a local matching cost matrix, based on the local matching cost matrix, obtain a global matching cost matrix through a dynamic waveform matching algorithm, filter out the minimum global matching cost in the global matching cost matrix, and backtrack the minimum global matching cost The shortest path is obtained by the cost, the missing judging device matches the formation units to determine whether there is a missing formation, and the outputting device outputs the result. 10. A DTW-based intelligent stratigraphic comparison system according to claim 9, characterized in that the matching device comprises a local matching cost matrix generating device, a global matching cost matrix generating device and a shortest path backtracking device connected in sequence, The local matching cost matrix generating device sequentially calculates the distance coefficient/local matching cost of each formation unit of the contrast well to establish a local matching cost matrix, and the global matching cost matrix generating device is based on the local matching cost matrix, through the dynamic waveform matching algorithm. Obtaining a global matching cost matrix, filtering out the minimum global matching cost in the global matching cost matrix, the shortest path backtracking device backtracking the shortest matching path with the minimum global matching cost, and matching the formation units to determine whether there is a missing formation .

Images