A high water cut reservoir production curve stage diagnosis method based on decreasing constraint and trend feature fusion

Abstract

The application discloses a high water cut reservoir production curve stage diagnosis method based on decreasing constraint and trend feature fusion; the method constructs a trend semantic recognition channel and a decreasing physical constraint channel; point-level trend semantic labels are generated by extracting curve shape features through a sliding window; a local decreasing rate is calculated, combined with long-range decreasing constraints of a near end and a far end and a valley value recovery mechanism to filter a decreasing skeleton conforming to a physical law; then, a continuous segment outside the skeleton is recognized as a suspected measure response segment, a decreasing violation score is calculated, and trend labels are fused to generate engineering diagnosis labels such as natural decreasing, stable production platform, measure response and abnormal fluctuation; the application further introduces a skeleton blocking mechanism based on index interval, time interval, normalized Euclidean distance and cross-measure segment, which effectively avoids unreasonable hard connection of long-distance points; the application can realize automatic and batch diagnosis of high water cut reservoir production dynamics, output structured results and visualized drawings, and significantly improve analysis efficiency.

Description

Technical Field

[0001] This invention belongs to the technical fields of dynamic analysis of oil and gas field development, production curve identification, response diagnosis of measures and intelligent data processing, and specifically relates to a method for stage diagnosis of production curves in high water-cut reservoirs based on the fusion of decreasing constraints and trend characteristics. Background Technology

[0002] In oil and gas field development, single-well production curves are crucial data sources for evaluating reservoir fluid supply capacity, the rationality of production systems, and the effectiveness of development measures. Production wells often experience multiple dynamic stages in actual production, including natural decline, stable production adjustments, production enhancement through measures, abnormal fluctuations, and subsequent decline. Traditional production dynamic analysis typically relies on manual experience or a single decline fitting method. While manual interpretation can be combined with engineering context for judgment, it is inefficient in scenarios involving multiple wells, long periods, and large volumes of data. Furthermore, different personnel may have significant differences in their judgments of anomalies, response segments of measures, and boundaries of decline segments. Simply using methods such as exponential decline, hyperbolic decline, or adjacent point difference is easily affected by short-term fluctuations, sudden increases in production, well shutdowns followed by restarts, operational measures, and missing data, leading to distortions in the decline rate calculation.

[0003] Existing trend identification methods mostly start from the curve shape, classifying curves into categories such as decreasing, increasing, stable, or fluctuating. However, these methods lack the constraint of the decline law in reservoir development, and are prone to misclassifying recovery segments after measures, short-term anomalies, or long-distance fault points as continuous decline segments. Conversely, methods that rely solely on decline rate constraints, while having some physical meaning, are difficult to express the engineering diagnostic meanings of stable production platforms, abnormal fluctuations, and response to measures, and are also difficult to output diagnostic labels that can be directly used for production management.

[0004] Furthermore, existing dynamic analysis techniques, after removing outliers or filtering decreasing points, often forcibly connect discontinuous, distant points into a decreasing framework, resulting in unreasonable curve interpretation. This is especially problematic when there are intervention operations, well shutdowns followed by restarts, sudden production surges, or prolonged data gaps; without segmenting the decreasing framework, the final result will show hard connections across intervention sections.

[0005] Therefore, there is an urgent need for an intelligent diagnostic method for production curves that simultaneously possesses the ability to identify trend features, reduce physical constraints, identify candidate segments of measures, and divide the reducing skeleton into blocks. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for diagnosing the production curve stages of high water-cut reservoirs based on the fusion of decreasing constraints and trend characteristics.

[0007] To achieve the above objectives, the present invention provides the following technical solution: This application provides a method for diagnosing the production curve stages of high water-cut reservoirs based on the fusion of decreasing constraints and trend characteristics, including: The production dynamic time series data of production wells in high water-cut reservoirs are acquired, and the data is cleaned and missing values ​​are processed to reconstruct a standardized production curve sequence. Construct a trend semantic recognition channel: Extract local morphological features from the standardized production curve sequence based on a sliding window, assign window-level trend semantic labels according to the extracted morphological features, and map the window-level trend semantic labels to data point-level trend semantic labels; Constructing a decreasing physical constraint channel: Calculate the local decreasing rate between adjacent data points in the standardized production curve sequence, construct a multi-level decreasing constraint containing near-end long-range constraints and far-end long-range constraints based on the local decreasing rate, retain data points that satisfy the decreasing relationship through iterative screening, and merge them to form a candidate decreasing skeleton; Identify engineering response segments and construct decreasing violation scores: Identify continuous non-skeleton data segments outside the candidate decreasing skeleton as suspected engineering response candidate segments, and calculate the decreasing violation score for each data point based on the local decreasing rate, long-range decreasing constraints at all levels, and membership status in the candidate decreasing skeleton. Fusion diagnosis and discrimination: Based on the trend semantic labels at the data point level, the decreasing violation score, and the identification results of the suspected engineering response candidate segments, a comprehensive discrimination is performed to generate engineering diagnostic labels that characterize the dynamic stage features of the production well; Decreasing skeleton dynamic segmentation: Based on at least one segmentation criterion among data point index interval, time interval, normalized Euclidean distance, and whether it crosses the suspected engineering response candidate segment, the candidate decreasing skeleton is broken into segments, and the effective decreasing skeleton after segmentation and the reason for segmentation break are output.

[0008] Furthermore, the local morphological feature extraction of the standardized production curve sequence based on a sliding window includes: The standardized production curve sequence is normalized to obtain a normalized output sequence, as shown in formula (1). Based on a preset window length or a window length adaptively determined by the number of data points, multiple sliding windows are generated on the normalized output sequence with a fixed step size. Extract the morphological features of each sliding window, including window slope, change at the beginning and end, range, standard deviation, maximum single-step increase, maximum single-step decrease, and number of peaks and valleys; The window slope is obtained by linear fitting using the least squares method, and the change at the beginning and end is the normalized difference in output between the end point and the beginning point within the window, as shown in formula (2). (1) (2) In formula (1), Indicates the first Normalized output value for each data point; Indicates the first The original output value of each data point; This represents the minimum output value for the current data segment; This indicates the maximum output value of the current data segment; In formula (2), Indicates the sliding window number; This indicates the total number of data points contained within the sliding window; Indicates the sequence number of the data point within the window; Indicates the first in the window Normalized output corresponding to each data point; and These represent the slope parameter and intercept parameter of the linear fitting model, respectively. Indicates the first The slope of the fitted production curve within a sliding window; This represents the corresponding intercept parameter; Indicates the first The first and last changes in normalized output within a sliding window; and These represent the normalized output values ​​at the end and start points of the sliding window, respectively.

[0009] Further, mapping the window-level trend semantic labels to data point-level trend semantic labels includes: For the same production data point that is covered by multiple sliding windows, the category and number of votes of the window-level trend semantic labels assigned by each window are counted, and voting mapping is performed according to the preset priority rules to generate point-level initial trend labels. Short interval filtering is performed on the initial trend labels at the point level, and continuous intervals with a length less than the preset minimum segment length are marked as transition labels; The trend labels of the same type separated by the transition labels are merged to form the final data point-level trend semantic label sequence.

[0010] Further, calculating the local decline rate between adjacent data points in the standardized production curve sequence includes: In the standardized production curve sequence, the local decline rate is calculated for adjacent data points. Specifically, as shown in formula (3);

[0011] in, Indicates the first Local decrease rate corresponding to each data point; Indicates the first Production value for each data point; Indicates the first Production value for each data point; Indicates the first The time corresponding to each data point; Indicates the first The time corresponding to each data point; When the time interval is zero, the output value is not greater than zero, or the calculated local decline rate is... If the value is not greater than zero, the data point is marked as an invalid point of local decline rate; otherwise, it is marked as a valid point of local decline rate.

[0012] Furthermore, the multi-level decreasing constraints include proximal long-range decreasing constraints and far-range long-range decreasing constraints; The near-end long-range decreasing constraint includes: maintaining a stack of retained points in chronological order; when a new data point is added, calculating the slope of the line connecting the last two retained points in the stack; if the slope of the line is not less than zero, deleting the second-to-last retained point in the stack, and continuing to iterate forward until the two nearest retained points satisfy the decreasing relationship. The long-range decreasing constraint includes: maintaining a current retention point in chronological order, comparing the slope of the line connecting the current data point and the latest retention point, and if the slope of the line is less than zero, retaining the current data point; if the slope of the line is not less than zero, discarding the current data point and not backtracking to modify the historical retention point. The slope of the line connecting two data points is calculated as shown in formula (4): (4) in, Indicates the first The data point and the first The slope of the line connecting the data points; and They represent the first The and the first Production value for each data point; and They represent the first The and the first The time corresponding to each data point; when When, it indicates the first Point 1 to the 2nd The points satisfy a decreasing trend; when When the two points are in a state of long-range decreasing constraint violation, it is determined that the relationship between them violates the constraint.

[0013] Furthermore, the merging to form a candidate decreasing skeleton includes: The candidate decreasing skeleton is obtained by taking the union of the skeleton points preserved by the proximal long-range decreasing constraint and the skeleton points preserved by the distal long-range decreasing constraint. The candidate decreasing skeleton is segmented by dividing the large increasing intervals, excluding non-decreasing continuous interruptions, and retaining only the skeleton clusters that meet the minimum cluster length requirement to form the final decreasing skeleton.

[0014] Furthermore, it also includes a valley recovery mechanism: During the execution of the proximal long-range decreasing constraint, consecutively deleted data point blocks are extracted as proximal consecutive deletion blocks; For the near-end continuous deletion block, the far-end long-range decreasing constraint is re-run. If the trough data point in the deletion block still meets the far-end long-range decreasing constraint relationship, then the trough data point is restored to the final decreasing skeleton.

[0015] Furthermore, the identification of engineering response segments and the construction of decreasing violation scores include: When the continuous length of the non-skeleton data points outside the final decreasing skeleton reaches a preset threshold, they are identified as suspected engineering response candidate segments. For each production data point, construct multiple binary indicators, including local decline rate validity indicator, near-end long-range decline constraint validity indicator, far-end long-range decline constraint validity indicator, final skeleton membership indicator, and suspected engineering response candidate segment membership indicator. Based on the aforementioned multiple binary indicators, the decreasing violation score ViVi for this data point is calculated, as shown in formula (5): (5) in, For the first Decreasing violation score for each data point; This indicates an invalid local decrease rate indicator. Indicator that the local decline rate is valid but fails the proximal long-range decline constraint; An indicator that the local decline rate is valid but fails to pass the long-range decline constraint at the far end; Indicator that has not entered the final decreasing skeleton; Indicates the indicator items that belong to the suspected engineering response candidate segments; all of the above indicator items are binary variables, when the first... The value is 1 if the data point meets the corresponding condition, otherwise it is 0.

[0016] Further, the process of performing segmentation on the candidate decreasing skeleton based on at least one segmentation criterion—data point index interval, time interval, or normalized Euclidean distance—includes: In the candidate decreasing skeleton, for adjacent skeleton points and Determine whether the index interval exceeds the maximum index interval threshold, whether the time interval exceeds the time interval threshold, or whether the normalized Euclidean distance exceeds the distance threshold. If adjacent skeleton points cross the suspected engineering response candidate segment, the skeleton connection is directly disconnected at that point. The normalized Euclidean distance da,b is calculated as shown in formula (6): (6) in, Indicates the first The first skeleton point and the second Normalized Euclidean distance between skeleton points; and They represent the first The first skeleton point and the first The time corresponding to each skeleton point; and They represent the first The first skeleton point and the second The output value corresponding to each skeleton point; This represents the total time span of the production curve data sequence; This represents the total range of production output for the production curve data sequence.

[0017] Furthermore, the method also includes multi-dimensional batch output of diagnostic results: The output includes a point-level result table containing point-level trend semantic labels, point-level engineering diagnostic labels, and decreasing constraint features; Output a range-level results table containing trend ranges and fusion diagnostic ranges; Output a list of candidate measures containing the suspected engineering response candidate segments; The output includes skeleton block data containing the effective decreasing skeleton blocks and the reasons for the breakage; It also outputs visualization results including trend recognition charts, decreasing constraint block charts, fused diagnostic charts, and simplified diagnostic charts, enabling automated batch diagnosis of production dynamic data of single and multiple wells in high water-cut reservoirs.

[0018] Compared with the prior art, this application has the following beneficial effects: This invention proposes a stage diagnosis method for production curves in high water-cut reservoirs based on the fusion of decline constraints and trend characteristics. It organically combines trend characteristic identification with decline physical constraints. Through parallel analysis of three types of decline constraints—local D, near-end long-range decline, and far-end long-range decline—it can identify curve morphologies such as decline, stable production, increase, and fluctuation, while effectively reducing the sensitivity of single decline rate methods to noise or short-term fluctuations, thus solving the problem of existing trend identification methods lacking physical constraints. Simultaneously, it identifies suspected response segments through continuous non-skeleton points and, combined with decline violation scoring and fusion diagnostic rules, quantifies numerical characteristics into natural decline, stable production, and flat production. Engineering semantic tags such as stage, high-confidence measure response, and abnormal fluctuations compensate for the lack of engineering semantics in single decline rate analysis. In addition, by introducing a skeleton block mechanism based on index interval, time interval, normalized distance, and cross-measure candidate segments, it effectively avoids the forced connection of distant points or cross-measure segment points into a single decline curve, solving the problem of unreasonable curve interpretation caused by hard connection of decline skeleton in existing technologies. Finally, through multi-well batch processing and structured multi-level diagnostic results output, the workload of manual well-by-well interpretation is significantly reduced, and the automation, consistency, and engineering interpretability of production dynamic analysis of high water-cut reservoirs are improved. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the overall processing of the present invention.

[0020] Figure 2 This is a schematic diagram of the four-layer diagnostic mechanism of the present invention.

[0021] Figure 3 This is a schematic diagram of the trend feature recognition results of the present invention.

[0022] Figure 4 This is a schematic diagram of the decreasing constraint diagnostic channel of the present invention.

[0023] Figure 5 This is a schematic diagram of the fusion diagnostic results of the present invention.

[0024] Figure 6 This is a schematic diagram of the effective decreasing skeleton block division of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] Furthermore, in this invention, an element referred to as fixed to or disposed on another element may be directly disposed on the other element, or there may be an intermediate element. When an element is considered to be connected to another element, it may be directly connected to the other element, or there may be an intermediate element present simultaneously. The terms vertical, horizontal, left, right, and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0027] The following detailed description, in conjunction with the accompanying drawings of the embodiments of the present invention, is provided. The described embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0028] Figure 1 The overall processing flowchart provided by the present invention is shown, and its technical process is as follows: Figure 1 As shown, the process includes the following steps: S1 Reading and preprocessing production curve data; S2 Constructing a trend semantic recognition channel to obtain a trend label sequence; S3 Constructing a decreasing constraint diagnosis channel to obtain decreasing constraint features and the final decreasing skeleton; S4 Integrating trend semantic labels and decreasing constraint features to generate engineering diagnosis labels; S5 Performing block connection discrimination on the decreasing skeleton to avoid unreasonable connections between distant points or points across measure segments; S6 Constructing diagnosis intervals and performing statistical summaries; S7 Generating a visualization of the diagnosis results; S8 Outputting a structured diagnosis result file; S9 Using the diagnosis results to serve natural decreasing identification, stable production platform identification, measure response diagnosis, and abnormal fluctuation diagnosis.

[0029] First, in step S1, during the reading and preprocessing of production curve data, the system acquires the dynamic time-series data of production wells in high water-cut reservoirs. For example, the input file is usually in Excel spreadsheet format and can contain multiple columns of production data. To adapt to different template data formats, the system has a built-in column name fuzzy matching mechanism that can automatically identify and extract target production indicator columns such as average monthly oil production, daily oil production, and daily liquid production. If there is no explicit date and time column in the file, the system will default to generating equally spaced time steps according to the order in which the data points appear. Subsequently, the system cleans the data, removes invalid or missing production data, and reconstructs a standardized production curve sequence {(t i ,q i ), i=1,2,...,N}, to prepare high-quality data input for subsequent dual-channel analysis.

[0030] In the process of constructing the trend semantic recognition channel in step S2, the system uses the sliding window technique to extract local curve morphology features; specifically, the standardized sequence is first subjected to Min-Max normalization to obtain the normalized output sequence, as shown in formula (1): (1) in, Indicates the first Normalized output value for each data point; Indicates the first The original output value of each data point; and These represent the minimum and maximum output values ​​of the current data segment, respectively. In actual processing, the size of the sliding window can be adaptively determined according to the total number of data points. For example, the window length can be set to 30 points for long-cycle data and 10 points for short-cycle data. For each sliding window, the system extracts its morphological features such as window slope, first and last changes, range, standard deviation, maximum single-step increase, maximum single-step decrease, and number of peaks and valleys. The window slope is obtained by least squares linear fitting, and the calculation method is shown in formula (2). (2) In the formula, Indicates the sliding window number; Indicates the total number of data points within the window; Indicates the first in the window Normalized output for each data point; and These are the slope and intercept of the linear fit, respectively; This represents the normalized change at the beginning and end of the window; it is determined by setting specific threshold rules (e.g., when the slope...). Less than -0.05 and the change at the beginning and end. When the value is negative and the maximum reverse increase does not exceed the allowed threshold, it is judged as decreasing. The system assigns an initial trend semantic label (decreasing, increasing, stable, fluctuating, transition) to each window; subsequently, the system maps the window labels to point-level labels. For data points covered by multiple windows, a preset priority voting mechanism is used to handle parallel cases, and abnormal intervals that are too short are filtered out. Similar labels separated by short transition segments are merged to generate the final point-level trend semantic label sequence; such as Figure 3 The trend feature recognition result chart shown clearly shows that the original curve has been automatically divided into trend intervals such as the decreasing stage (blue), the stable production platform stage (green), and the fluctuation stage (orange) through lines of different colors.

[0031] Parallel to trend pattern recognition is the construction of decreasing physical constraint channels in steps S3 to S5. First, the local decrease rate between adjacent points is calculated, as shown in formula (3): (3) in For local decrease rate, , and , These represent adjacent data points; when the time interval is 0, the output is not greater than 0, or the calculated value is... When the value is ≤0, the point is determined to be a point that violates the decreasing constraint, thereby eliminating abnormal data. On this basis, in order to ensure the rationality of the natural decreasing skeleton, this invention further introduces long-range decreasing constraints at the near and far ends. The slope formula of the line connecting the two points is shown in (4): (4) in the formula Indicates the first Point and the The slope of the line connecting the points, when When the value is ≥0, it is determined that the constraint is violated; the near-end constraint adopts a backtracking iteration mechanism. If the addition of a new point causes the two most recent points in the stack to no longer decrease, the second to last point is deleted and the backtracking continues to ensure the continuity of the recent skeleton; the far-end constraint only compares the current point with the most recently retained point and does not backtrack to the past, so as to prevent excessive deletion; in the candidate skeleton obtained by combining the two, the points that cross the significantly increasing interval are further divided, and the skeleton clusters that meet the minimum cluster length are retained to form the final decreasing skeleton.

[0032] It is particularly important to note that the valley recovery mechanism designed in this embodiment is key to solving engineering misjudgments: because the near-end long-range constraint uses a backtracking algorithm, it is easy to mistakenly delete valley points when encountering a bimodal structure; this invention packages points continuously deleted by the near-end constraint into a near-end continuous deletion block, and re-executes the far-end long-range decreasing constraint on it. If the valley still conforms to the far-end relationship, it is restored; Figure 4 The diagram showing the diagnostic results of the decreasing constraint channel illustrates the discrete decreasing skeleton that is ultimately formed after near-end and far-end constraint screening and valley value recovery, effectively avoiding pseudo-decreasing points caused by sudden increases in output due to operational measures.

[0033] After completing the decreasing skeleton constraint, the system enters step S6 to identify suspected engineering response segments and calculate the decreasing violation score; since the final decreasing skeleton represents the baseline of natural production decrease, non-skeleton data points outside the skeleton, especially those that form a segment and exceed the preset length threshold, are judged as suspected engineering response candidate segments (for example, frequent production rebounds usually correspond to the effects of fracturing, acidizing and other production increase operations); subsequently, the system establishes a scoring mechanism for each data point containing 5 binary indicators, and the calculation formula is shown in (5): (5) in To reduce the violation score; This is an invalid indicator for the local decrease rate. The proximal constraint indicator failed. The remote constraint indicator failed. This item has not yet entered the final skeleton indicator. These are suspected candidate engineering response section indicators; it is particularly important to emphasize that these indicators... The weight is set to 2. This weighted design is intended to significantly improve the score of the data point in the section when multiple fracturing, pump replacement and other production increase measures occur on site and the output rebounds significantly, so as to keenly capture engineering intervention activities.

[0034] Steps S4 and S7 are the core of the method of this invention, namely, fusion diagnosis and discrimination; the system deeply fuses the acquired point-level trend semantic labels, decreasing violation scores, and identification results of suspected engineering response candidate segments; in specific implementation, this invention follows a set of explicit fusion rules with engineering semantics, typical rules are shown in Table 1: Table 1 Typical fusion diagnosis rules

[0035] By combining monotonic trend patterns (such as increasing) with physical constraint scores (such as high violation) using the above logic, intuitive engineering diagnostic labels such as high-confidence measure response and stable production platform can be deduced; for example... Figure 5 The fusion diagnostic results of the production curve of a certain production well shown clearly indicate which intervals are automatically marked as naturally decreasing (blue), which segments are marked as stable production platforms (green), and the fluctuating parts in the middle are identified as high-confidence response (red) or suspected response (dashed box), intuitively demonstrating the application effect of the method.

[0036] While fusing and identifying the engineering label, step S8 performs the crucial dynamic block processing of the decreasing skeleton. To prevent the forced connection of distant points or points across different measures into a single pseudo-decreasing curve in long-term production curves, this invention introduces a block mechanism. The specific judgment logic is as follows: for adjacent skeleton points in the candidate decreasing skeleton... and If the index interval between the two exceeds the maximum index interval threshold, the time interval exceeds the threshold, the normalized Euclidean distance exceeds the distance threshold, or they cross a suspected engineering response candidate segment, then the skeleton connection is forcibly disconnected at that point. The formula for calculating the normalized Euclidean distance is shown in (6): (6) In the formula, Represents skeleton points and Normalized Euclidean distance between them; , Indicates the corresponding time; , Indicates output; and These represent the total time span and total output range of the production curve, respectively; in this embodiment, if... If the value is greater than 0.3, the connection is immediately broken; this mechanism effectively ensures that a single physically decreasing segment does not exceed a reasonable spatiotemporal span, such as... Figure 6 The diagram shown is the final effective decreasing skeleton extraction result of a certain production well. It demonstrates that the skeleton points after block processing have formed discrete but reasonable decreasing segments, avoiding the engineering interpretation defects of hard connections across large segments.

[0037] Step S9 is the process of the system outputting diagnostic results in batches from multiple dimensions; in order to support engineers to use and manage them effectively, the system outputs corresponding files according to the data processing dimensions.

[0038] Example The present invention will be described below using monthly production data from a single well. This embodiment is only used to explain the technical solution of the present invention and does not limit the scope of protection of the present invention.

[0039] This embodiment uses monthly oil production data from a production well in the Penglai 19-3 oilfield as an example. The data is only used to illustrate the implementation process of the method of this invention and does not limit the scope of application of this invention. In this embodiment, the input data consists of several single-well Excel files stored in a designated data directory. Each file contains at least one column of production indicators, such as average monthly oil production (m³). 3 / d); When a time column exists, it is read as the time axis; when no time column exists, the data point order is used as the time axis; the program automatically creates an output directory by well name and executes the diagnostic process independently for each well.

[0040] First, the q(t) production curve is read for each well, and the sliding window length is automatically selected based on the number of points. For each sliding window, the slope, first and last changes, range, standard deviation, maximum rise, maximum fall, and number of peaks and troughs are calculated to generate window-level trend labels. Then, point-level trend labels are obtained through voting. After short-segment elimination and gap merging, trend intervals such as decreasing, stable, increasing, fluctuating, and transitional are obtained. Figure 3 The figure shown is a trend feature identification result of the oil production curve of a certain production well in this embodiment.

[0041] Secondly, the local decline rate D is calculated on the original production curve. Points with zero time interval, output not greater than zero, or D not greater than zero are marked as invalid points of local decline rate. Then, the local D valid point sequence is subjected to proximal long-range decline constraint and far-end long-range decline constraint respectively, and the proximal continuous deletion block is subjected to secondary far-end recovery to reduce the false deletion of troughs. Figure 4 The diagram shows the diagnostic results of the decreasing constraint channel of a production well in this embodiment.

[0042] Next, the near-end and far-end retained points are merged to form a candidate decreasing skeleton; the significantly increasing intervals in the candidate skeleton are segmented to obtain the final decreasing skeleton; if the continuous points outside the final decreasing skeleton reach the minimum length threshold, they are identified as suspected measure response segments.

[0043] Finally, the trend semantic labels, decreasing skeleton status, decreasing violation score and measure candidate label are integrated to output diagnostic labels such as natural decreasing, decreasing pending verification, stable production platform, stable production, increasing, suspected measure response, high confidence measure response, abnormal fluctuation, fluctuation and transition. Figure 5 This is a fusion diagnostic result diagram of the oil production curve of a certain production well in this embodiment.

[0044] For each well, the output directory includes the following files: clean_q_data.csv (for storing time and production data after cleaning); trend_windows.csv (for storing window-level trend features); fusion_points.csv (for storing point-level trend labels, fusion labels, and decreasing constraint features); decline_constraint_features.csv (for storing features such as local D, proximal retention, distal retention, final skeleton, action candidates, and violation scores); trend_segments.csv (for storing trend intervals); fusion_segments.csv (for storing fusion diagnostic intervals); measure_segments.csv (for storing suspected action segments); curve_blocks.csv (for storing decreasing skeleton blocks); and curve_breaks.csv (for storing the reasons for skeleton breaks).

[0045] Simultaneously, it outputs image files 07_trend_stage.png, 01_decline_constraint_split.png, 08_fusion_diagnosis.png, and 09_fusion_simple.png, respectively displaying trend recognition results, decreasing constraint skeleton segmentation results, fusion diagnosis results, and simple curve diagnosis results. After batch execution, it also generates global summary files such as run_summary.csv, all_fusion_segments.csv, all_measure_segments.csv, and all_fusion_points.csv; among them, Figure 6 The final effective decreasing skeleton extraction result of a production well in this embodiment is shown in the figure; the specific uses of the diagnostic result output file in this embodiment are explained in Table 2.

[0046] Table 2. Output files and uses of diagnostic results

[0047] This embodiment demonstrates that the method can automatically obtain trend intervals, natural decline frameworks, suspected response segments, and abnormal fluctuation intervals from single-well production curves, generating diagnostic results that can be summarized in batches. Engineers can further interpret the fused diagnostic intervals and candidate response segments in conjunction with well history, operational records, and changes in production systems, thereby improving the efficiency of response evaluation and dynamic analysis.

[0048] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0049] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for diagnosing the production curve stages of high water-cut reservoirs based on the fusion of decreasing constraints and trend characteristics, characterized in that, include: The production dynamic time series data of production wells in high water-cut reservoirs are acquired, and the data is cleaned and missing values ​​are processed to reconstruct a standardized production curve sequence. Construct a trend semantic recognition channel: Extract local morphological features from the standardized production curve sequence based on a sliding window, assign window-level trend semantic labels according to the extracted morphological features, and map the window-level trend semantic labels to data point-level trend semantic labels; Constructing a decreasing physical constraint channel: Calculate the local decreasing rate between adjacent data points in the standardized production curve sequence, construct a multi-level decreasing constraint containing near-end long-range constraints and far-end long-range constraints based on the local decreasing rate, retain data points that satisfy the decreasing relationship through iterative screening, and merge them to form a candidate decreasing skeleton; Identify engineering response segments and construct decreasing violation scores: Identify continuous non-skeleton data segments outside the candidate decreasing skeleton as suspected engineering response candidate segments, and calculate the decreasing violation score for each data point based on the local decreasing rate, long-range decreasing constraints at all levels, and membership status in the candidate decreasing skeleton. Fusion diagnosis and discrimination: Based on the trend semantic labels at the data point level, the decreasing violation score, and the identification results of the suspected engineering response candidate segments, a comprehensive discrimination is performed to generate engineering diagnostic labels that characterize the dynamic stage features of the production well; Decreasing skeleton dynamic segmentation: Based on at least one segmentation criterion among data point index interval, time interval, normalized Euclidean distance, and whether it crosses the suspected engineering response candidate segment, the candidate decreasing skeleton is broken into segments, and the effective decreasing skeleton after segmentation and the reason for segmentation break are output.

2. The method according to claim 1, characterized in that, The step of extracting local morphological features from the standardized production curve sequence based on a sliding window includes: The standardized production curve sequence is normalized to obtain a normalized output sequence, as shown in formula (1). Based on a preset window length or a window length adaptively determined by the number of data points, multiple sliding windows are generated on the normalized output sequence with a fixed step size. Extract the morphological features of each sliding window, including window slope, change at the beginning and end, range, standard deviation, maximum single-step increase, maximum single-step decrease, and number of peaks and valleys; The window slope is obtained by linear fitting using the least squares method, and the change at the beginning and end is the normalized difference in output between the end point and the beginning point within the window, as shown in formula (2). (1) (2) In formula (1), Indicates the first Normalized output value for each data point; Indicates the first The original output value of each data point; This represents the minimum output value for the current data segment; This indicates the maximum output value of the current data segment; In formula (2), Indicates the sliding window number; This indicates the total number of data points contained within the sliding window; Indicates the sequence number of the data point within the window; Indicates the first in the window Normalized output corresponding to each data point; and These represent the slope parameter and intercept parameter of the linear fitting model, respectively. Indicates the first The slope of the fitted production curve within a sliding window; This represents the corresponding intercept parameter; Indicates the first The first and last changes in normalized output within a sliding window; and These represent the normalized output values ​​at the end and start points of the sliding window, respectively.

3. The method according to claim 2, characterized in that, The step of mapping the window-level trend semantic labels to data point-level trend semantic labels includes: For the same production data point that is covered by multiple sliding windows, the category and number of votes of the window-level trend semantic labels assigned by each window are counted, and voting mapping is performed according to the preset priority rules to generate point-level initial trend labels. Short interval filtering is performed on the initial trend labels at the point level, and continuous intervals with a length less than the preset minimum segment length are marked as transition labels; The trend labels of the same type separated by the transition labels are merged to form the final data point-level trend semantic label sequence.

4. The method according to claim 1, characterized in that, The calculation of the local decline rate between adjacent data points in the standardized production curve sequence includes: In the standardized production curve sequence, the local decline rate is calculated for adjacent data points. Specifically, as shown in formula (3); （3） in, Indicates the first Local decrease rate corresponding to each data point; Indicates the first Production value for each data point; Indicates the first Production value for each data point; Indicates the first The time corresponding to each data point; Indicates the first The time corresponding to each data point; When the time interval is zero, the output value is not greater than zero, or the calculated local decline rate is... If the value is not greater than zero, the data point is marked as an invalid point of local decline rate; otherwise, it is marked as a valid point of local decline rate.

5. The method according to claim 4, characterized in that, The multi-level decreasing constraints include proximal long-range decreasing constraints and distal long-range decreasing constraints; The near-end long-range decreasing constraint includes: maintaining a stack of retained points in chronological order; when a new data point is added, calculating the slope of the line connecting the last two retained points in the stack; if the slope of the line is not less than zero, deleting the second-to-last retained point in the stack, and continuing to iterate forward until the two nearest retained points satisfy the decreasing relationship. The long-range decreasing constraint includes: maintaining a current retention point in chronological order, comparing the slope of the line connecting the current data point and the latest retention point, and if the slope of the line is less than zero, retaining the current data point; if the slope of the line is not less than zero, discarding the current data point and not backtracking to modify the historical retention point. The slope of the line connecting two data points is calculated as shown in formula (4): (4) in, Indicates the first The data point and the first The slope of the line connecting the data points; and They represent the first The and the first Production value for each data point; and They represent the first The and the first The time corresponding to each data point; when When, it indicates the first Point 1 to the 2nd The points satisfy a decreasing trend; when When the two points are in a state of long-range decreasing constraint violation, it is determined that the relationship between them violates the constraint.

6. The method according to claim 5, characterized in that, The merging process forms a candidate decreasing skeleton, including: The candidate decreasing skeleton is obtained by taking the union of the skeleton points preserved by the proximal long-range decreasing constraint and the skeleton points preserved by the distal long-range decreasing constraint. The candidate decreasing skeleton is segmented by dividing the large increasing intervals, excluding non-decreasing continuous interruptions, and retaining only the skeleton clusters that meet the minimum cluster length requirement to form the final decreasing skeleton.

7. The method according to claim 5, characterized in that, It also includes a valley recovery mechanism: During the execution of the proximal long-range decreasing constraint, consecutively deleted data point blocks are extracted as proximal consecutive deletion blocks; For the near-end continuous deletion block, the far-end long-range decreasing constraint is re-run. If the trough data point in the deletion block still meets the far-end long-range decreasing constraint relationship, then the trough data point is restored to the final decreasing skeleton.

8. The method according to claim 6, characterized in that, The identification of engineering response segments and the construction of decreasing violation scores include: When the continuous length of the non-skeleton data points outside the final decreasing skeleton reaches a preset threshold, they are identified as suspected engineering response candidate segments. For each production data point, construct multiple binary indicators, including local decline rate validity indicator, near-end long-range decline constraint validity indicator, far-end long-range decline constraint validity indicator, final skeleton membership indicator, and suspected engineering response candidate segment membership indicator. Based on the aforementioned multiple binary indicators, the decreasing violation score ViVi for this data point is calculated, as shown in formula (5): (5) in, For the first Decreasing violation score for each data point; This indicates an invalid local decrease rate indicator. Indicator that the local decline rate is valid but fails the proximal long-range decline constraint; An indicator that the local decline rate is valid but fails to pass the long-range decline constraint at the far end; Indicator that has not entered the final decreasing skeleton; Indicates the indicator items that belong to the suspected engineering response candidate segments; all of the above indicator items are binary variables, when the first... The value is 1 if the data point meets the corresponding condition, otherwise it is 0.

9. The method according to claim 1, characterized in that, The process of breaking down the candidate decreasing skeleton into blocks based on at least one block criterion, including data point index interval, time interval, and normalized Euclidean distance, includes: In the candidate decreasing skeleton, for adjacent skeleton points and Determine whether the index interval exceeds the maximum index interval threshold, whether the time interval exceeds the time interval threshold, or whether the normalized Euclidean distance exceeds the distance threshold. If adjacent skeleton points cross the suspected engineering response candidate segment, the skeleton connection is directly disconnected at that point. The normalized Euclidean distance The calculation method is shown in formula (6): (6) in, Indicates the first The first skeleton point and the second Normalized Euclidean distance between skeleton points; and They represent the first The first skeleton point and the first The time corresponding to each skeleton point; and They represent the first The first skeleton point and the second The output value corresponding to each skeleton point; This represents the total time span of the production curve data sequence; This represents the total range of production output for the production curve data sequence.

10. The method according to claim 1, characterized in that, The method also includes multi-dimensional batch output of diagnostic results: The output includes a point-level result table containing point-level trend semantic labels, point-level engineering diagnostic labels, and decreasing constraint features; Output a range-level results table containing trend ranges and fusion diagnostic ranges; Output a list of candidate measures containing the suspected engineering response candidate segments; The output includes skeleton block data containing the effective decreasing skeleton blocks and the reasons for the breakage; It also outputs visualization results including trend recognition charts, decreasing constraint block charts, fused diagnostic charts, and simplified diagnostic charts, enabling automated batch diagnosis of production dynamic data of single and multiple wells in high water-cut reservoirs.

Images