Method and system for multi-dimensional quality checking of oil and gas field production data

Abstract

The application relates to the technical field of data processing, and discloses an oil and gas field production data multi-dimensional quality checking method and system. The method comprises the following steps: collecting production parameter values of each link node and link source identification, and encapsulating the production parameter values and the link source identification into link marked data records; performing four-dimensional checking processing on the link marked data records to obtain an abnormal marked data set; extracting time dimension, space dimension and form dimension features from the abnormal marked data set to form an abnormal feature vector; matching the abnormal feature vector with a cause knowledge base to obtain a cause confidence distribution vector of five types of link causes, generating a fixed-point repair instruction, and writing a repair confirmation result back to the cause knowledge base. The application improves the automation degree of abnormal root positioning of oil and gas field production data and the accuracy of initial cause attribution of new facilities.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for multidimensional quality verification of oil and gas field production data. Background Technology

[0002] During the production and operation of floating production facilities in oil and gas fields, production parameters such as output, wellhead pressure, production flow temperature, water cut, and gas-oil ratio are continuously collected. These parameters are transmitted to the production management system through five links: direct sensor acquisition, forwarding by edge transmission nodes, output by protocol conversion gateways, manual entry into the system, and push via cross-system synchronization interfaces, forming a massive production data set. Existing solutions typically introduce the following four dimensions of verification rules for the quality control of this production data: first, missing data detection, checking for null values ​​or interrupted records; second, threshold boundary judgment, judging whether the value of a single parameter exceeds a preset engineering physical range; third, cross-parameter constraint verification, checking the consistency of the logical relationships between parameters such as output, water cut, and gas-oil ratio; and fourth, time-series trend verification, checking whether the changing trend of parameter values ​​over time shows any abnormal deviations. These four dimensions of verification rules are executed independently through preset thresholds, marking data exceeding the threshold as abnormal and outputting anomaly labels and quality scores, providing data quality references for production dynamic analysis.

[0003] However, the existing verification method, after an anomaly is detected, uniformly labels all types of abnormal data generated through five links—sensor direct sampling, edge transmission node forwarding, protocol conversion gateway output, system manual input, and cross-system synchronization interface push—as an anomaly tag, without distinguishing which link the abnormal data originated from. This causes the quality verification output to terminate at the anomaly detection and labeling stage. In actual production and maintenance, the five types of link causes—sensor drift, transmission packet loss, protocol truncation, manual input deviation, and cross-system synchronization errors—exhibit fundamentally different abnormal characteristics in their data. The existing solution does not establish a mapping relationship between abnormal data characteristics and link causes. After receiving an anomaly label, maintenance engineers still need to manually check all five link links one by one, unable to directly locate the root cause based on the verification results, resulting in a large amount of repetitive troubleshooting work.

[0004] The fundamental reason why the causes of anomalies cannot be automatically located lies in the lack of a structured extraction mechanism for the multi-dimensional features of anomaly data in the existing verification system. Anomalies caused by different links have different characteristics in terms of temporal distribution, parameter co-occurrence patterns, and numerical forms. If these characteristics can be structured into computable feature vectors and mapped to historically confirmed causes, then the cause location problem is technically solvable. However, even if the above mapping mechanism is established, newly commissioned facilities lack historically confirmed cause records in the initial stage, and the mapping model cannot be directly used. This cold start problem further restricts the applicability of the cause location mechanism to new facilities, and existing solutions do not have a corresponding solution for this. Summary of the Invention

[0005] This application provides a multi-dimensional quality verification method and system for oil and gas field production data, which solves the problems in existing oil and gas field production data quality verification methods, such as the inability to automatically locate the causes of anomalies, the need for maintenance engineers to manually check each link in the entire process, and the cold start problem where the cause knowledge base lacks historical records in the early stages of newly commissioned facilities, resulting in the inability to activate the cause location mechanism. This improves the automation level of locating the root causes of anomalies in oil and gas field production data and the accuracy of attributing causes in the early stages of new facilities.

[0006] Firstly, this application provides a method for multidimensional quality verification of oil and gas field production data, the method comprising: Step S1: Collect the production parameter values ​​and link source identifiers of each link node of the floating production facility, and encapsulate the production parameter values ​​and the link source identifiers into a link tag data record; Step S2: The link tag data records are processed by missing detection, threshold boundary judgment, cross-parameter constraint verification and time series trend test to obtain an anomaly tag dataset carrying anomaly type and anomaly value; Step S3: Extract the anomaly duration, the link source identifier and parameter co-occurrence anomaly set, the anomaly value sequence morphology code and boundary regularity index from each anomaly record in the anomaly label dataset, and merge them to form an anomaly feature vector; Step S4: Perform similarity matching between the abnormal feature vector and historical feature records in the causal knowledge base to obtain the causal confidence distribution vectors corresponding to the five types of link causes. Generate a fixed-point repair instruction based on the link cause corresponding to the largest component in the causal confidence distribution vector, and write the repair confirmation result and the abnormal feature vector back to the causal knowledge base. When the number of local records in the causal knowledge base is lower than a preset threshold, the historical causal records of similar floating production facilities are weighted and merged according to parameter similarity to form the initial content of the causal knowledge base.

[0007] Secondly, this application provides a multi-dimensional quality verification system for oil and gas field production data, the multi-dimensional quality verification system for oil and gas field production data comprising: The encapsulation module is used to collect the production parameter values ​​and link source identifiers of each link node of the floating production facility, and encapsulate the production parameter values ​​and the link source identifiers into a link tag data record. The verification module is used to process the link tag data records through missing detection, threshold boundary judgment, cross-parameter constraint verification and time series trend verification to obtain an anomaly tag dataset carrying anomaly type and anomaly value; The merging module is used to extract the duration of the anomaly, the link source identifier and the set of co-occurrence anomalies of parameters, the anomaly value sequence morphology encoding and boundary regularity index from each anomaly record in the anomaly label dataset, and merge them to form an anomaly feature vector; The generation module is used to perform similarity matching between the abnormal feature vector and historical feature records in the causal knowledge base to obtain the causal confidence distribution vectors corresponding to the five types of link causes. The module generates a fixed-point repair instruction based on the link cause corresponding to the largest component in the causal confidence distribution vector, and writes the repair confirmation result and the abnormal feature vector back to the causal knowledge base. When the number of local records in the causal knowledge base is lower than a preset threshold, the historical causal records of similar floating production facilities are weighted and merged according to parameter similarity to form the initial content of the causal knowledge base.

[0008] Thirdly, a multi-dimensional quality verification device for oil and gas field production data is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the multi-dimensional quality verification device for oil and gas field production data to execute the aforementioned multi-dimensional quality verification method for oil and gas field production data.

[0009] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, cause the computer to perform the above-described method for multi-dimensional quality verification of oil and gas field production data.

[0010] The technical solution provided in this application forcibly encapsulates the link source identifier into the production parameter values ​​of each link node during the data acquisition stage. This ensures that each link-marked data record carries traceable link coordinates at the time of its generation, unlike existing solutions that only record parameter values ​​without retaining the link information that generated the data. This provides an indispensable spatial dimension data foundation for subsequent anomaly attribution. Based on this, the link-marked data records undergo four-dimensional verification processing—missing data detection, threshold boundary judgment, cross-parameter constraint verification, and time-series trend testing—to obtain an anomaly-marked dataset. These four verifications are independent yet synergistic: missing data detection identifies missing data, threshold boundary judgment identifies single parameter out-of-bounds errors, cross-parameter constraint verification identifies logical contradictions between parameters, and time-series trend testing identifies gradual shifts. The merged results of these four verifications enable the anomaly-marked dataset to carry both anomaly type and anomaly magnitude information, providing structured, fine-grained input for subsequent feature extraction. This differs from existing solutions that only output binary anomaly labels.

[0011] The abnormal duration and abnormal time density are extracted from the abnormal marker dataset to form a time dimension feature, the link source identifier and parameter co-occurrence anomaly set to form a spatial dimension feature, and the outlier sequence morphology code and boundary regularity index to form a morphological dimension feature. The three types of features are merged into an abnormal feature vector. This vector structures the differential traces left on the data by different link causes into a computable numerical representation, so that the gradual and continuous characteristics of sensor drift, the instantaneous missing characteristics of transmission packet loss and the fixed boundary truncation characteristics of protocol truncation can be distinguished in the same vector space. This is an abnormal feature structuring mechanism that has not been established in existing solutions. After matching the abnormal feature vectors with historical feature records in the causal knowledge base, the causal confidence distribution vectors corresponding to the five types of link causes are obtained. The link cause corresponding to the largest component is used to generate a fixed-point repair instruction, so that the output of quality verification extends directly from anomaly labeling to root cause location and repair guidance, which is different from the limitation of existing solutions where the verification result ends at anomaly discovery. At the same time, the repair confirmation result is written back to the causal knowledge base, and when the number of local records is lower than a preset threshold, the historical causal records of similar floating production facilities are fused according to parameter similarity weights as the initial content, which solves the problem of cold start of the causal knowledge base of newly commissioned facilities, and makes the causal location mechanism available in the early stage of facility commissioning. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1This is a schematic diagram of an embodiment of the multidimensional quality verification method for oil and gas field production data in this application. Figure 2 This is a schematic diagram of one embodiment of the multi-dimensional quality verification system for oil and gas field production data in this application. Figure 3 This is a schematic block diagram of the structure of the multi-dimensional quality verification device for oil and gas field production data in this embodiment of the invention. Detailed Implementation

[0014] This application provides a method and system for multi-dimensional quality verification of oil and gas field production data. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0015] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the multi-dimensional quality verification method for oil and gas field production data in this application includes: Step S1: Collect the production parameter values ​​and link source identifiers of each link node of the floating production facility, and encapsulate the production parameter values ​​and link source identifiers into a link tag data record; Specifically, the production data from the floating production facility undergoes five physical link nodes during its transmission from sensors to the application system: direct sensor acquisition, edge transmission node forwarding, protocol conversion gateway output, manual system entry, and cross-system synchronization interface push. Each node has an independent data writing agent responsible for data encapsulation. When data is generated, the agent encapsulates the current node type in the link source identifier field, which is bound to the production parameter value as a link marker data record. The link source identifier is encapsulated in real-time when the data is generated, rather than being added retroactively, thus ensuring a one-to-one correspondence between the identifier and the data, providing traceable link coordinates for the extraction of spatial features in step S3.

[0016] Step S2: The link tag data records are processed by missing detection, threshold boundary judgment, cross-parameter constraint verification and time series trend test to obtain an anomaly tag dataset carrying anomaly type and anomaly value; Specifically, the constraint equations used in cross-parameter constraint verification are derived from the physical conservation relationships in oil and gas field production. Liquid production equals oil production divided by the water cut complement, and gas production equals the product of oil production and the gas-oil ratio. These two equations constitute hard constraints between parameters. The measured parameter values ​​at any given time point are substituted into these equations to calculate the constraint residuals. The residual threshold is set at 15%, determined based on industry experience regarding measurement errors in oil and gas field engineering. Exceeding this threshold indicates a logical conflict between parameters that cannot be explained by measurement errors. In the time-series trend verification, the trend slope threshold is set independently based on the upper limit of the engineering physical change rate of each parameter. For example, the wellhead pressure trend slope threshold is set at 0.5 MPa / h; pressure changes exceeding this rate are not physically feasible under normal production conditions.

[0017] Step S3: Extract the anomaly duration, link source identifier and parameter co-occurrence anomaly set, anomaly value sequence morphology encoding and boundary regularity index from each anomaly record in the anomaly label dataset, and merge them to form an anomaly feature vector; Specifically, morphological encoding classifies and assigns values ​​to the first-order difference signs of the difference between the outlier sequence and the baseline sequence, mapping five morphological types to integer codes 1 to 5. The truncated code 5 corresponds to the case where the absolute value of the difference is concentrated at a fixed value. This characteristic is a direct data representation of the fixed-bit truncation generated by the protocol conversion gateway when the field lengths do not match. The boundary regularity index calculates the percentage of records in the outlier sequence that satisfy the fixed-byte boundary divisibility condition. A high percentage indicates that the outlier originates from protocol field length limitations, showing a statistically distinguishable difference from the morphological characteristics of sensor drift and transmission packet loss.

[0018] Step S4: Perform similarity matching between the abnormal feature vector and the historical feature records in the cause knowledge base to obtain the cause confidence distribution vectors corresponding to the five types of link causes. Generate a fixed-point repair instruction based on the link cause corresponding to the largest component in the cause confidence distribution vector, and write the repair confirmation result and the abnormal feature vector back to the cause knowledge base. When the number of local records in the cause knowledge base is lower than a preset threshold, the historical cause records of similar floating production facilities are fused together according to parameter similarity weights to form the initial content of the cause knowledge base.

[0019] Specifically, the trigger condition for cross-facility migration is that the number of local records is less than 50. This threshold is determined based on the minimum sample size required for nearest neighbor matching; below this number, the representativeness of the nearest neighbor search results is insufficient. Reference facilities are selected based on the Euclidean distance ranking of four normalized parameters: design capacity, water depth, average wellhead pressure, and average water cut. The top three most similar facilities are selected, and their historical causal records are weighted and fused using similarity as the initial content of the knowledge base. As the number of local records increases, the proportion of local knowledge in the fusion weight dynamically increases by the ratio of the number of local records to the number of local records plus 50. When the number of local records reaches 200, it completely switches to local knowledge, and the migrated knowledge exits the fusion calculation, ensuring that the knowledge base content continuously aligns with the actual operating conditions of the target facility.

[0020] In one specific embodiment, step S1 includes: At the wellhead, separator, metering manifold, and external pump nodes of the floating production facility, production rate, wellhead pressure, production flow temperature, water cut, and gas-oil ratio are collected to obtain production parameter values. Based on the data writing agent program of each link node, the production parameter values ​​are encapsulated with five types of link source identifiers: direct sensor acquisition, edge transmission node forwarding, protocol conversion gateway output, system manual input, and cross-system synchronous interface push, to obtain parameter acquisition records carrying link source identifiers; The parameter acquisition records are written into the time series database using a dual index of oil and gas field identifier and UTC timestamp to obtain the link tag data records.

[0021] Specifically, the data acquisition nodes for five types of production parameters—production rate, wellhead pressure, production flow temperature, water cut, and gas-oil ratio—correspond to key process locations in the floating production facility that perform metering functions: the wellhead node collects wellhead pressure and production rate, the separator node collects water cut and gas-oil ratio, the metering manifold node aggregates and measures the production rate of each well, and the export pump node collects export production rate and production flow temperature. Each node sensor continuously outputs data at a time resolution of 1 minute per record. This resolution is determined based on the minimum response cycle for dynamic monitoring of oil and gas field production; a resolution lower than this will distort the calculation benchmark for the expected number of records in step S2 (missing record detection). The data writing agent is deployed at each link node. Within the same database transaction where data is generated, the current node type is written to the link source identifier field. If the transaction is successfully committed, both the parameter value and the link source identifier are written to the database simultaneously; if the transaction is rolled back, neither is written, thus ensuring a strict binding relationship between the link source identifier and the parameter value. The link source identifier records the direct node type where the data was generated. When the data is transmitted through multiple link nodes, the node type that is finally written into the time series database shall prevail. The value shall be one of five categories: direct sensor acquisition, edge transmission node forwarding, protocol conversion gateway output, system manual entry, and cross-system synchronization interface push.

[0022] The parameter type coding assigns a unique integer code to five types of parameters: production rate, wellhead pressure, production flow temperature, water cut, and gas-oil ratio. These codes are uniformly managed by the parameter dictionary of the oil and gas field information system and are used to distinguish different parameter records collected at the same node and time in the time-series database. Parameter acquisition records are written to the time-series database using a dual index of oil and gas field identifier and UTC timestamp. The oil and gas field identifier distinguishes data attribution in scenarios of concurrent acquisition from multiple facilities, while the UTC timestamp, with second-level precision, ensures time-series alignment of data across nodes. The complete field structure of the link tag data record consists of five fields: oil and gas field identifier, UTC timestamp, parameter type code, parameter value, and link source identifier. These five fields constitute the basic data unit for anomaly verification and cause attribution in subsequent steps.

[0023] In one specific embodiment, step S2 includes: Within a preset time window, the ratio of the actual number of records for each production parameter in the link marker data record to the expected number of records is calculated. Records with a ratio lower than the missing rate threshold are marked with missing type and missing rate to obtain the integrity verification result. Based on the engineering physical boundaries of each production parameter, the parameter values ​​in the link marking data records are compared with the engineering physical boundaries. Records that exceed the boundaries are marked with the out-of-bounds type and deviation amount to obtain the rationality verification results. Based on the cross-parameter constraint equation between production, water content and gas-oil ratio, the constraint residual is calculated for the parameter values ​​at each time point in the link marker data record. Records whose constraint residual exceeds the residual threshold are marked with logical conflict type and residual value to obtain the consistency verification result. Based on a preset trend window, linear regression is performed on the time-series data of each production parameter in the link marker data record to calculate the trend slope and residual standard deviation. Records whose trend slope or residual standard deviation exceeds the corresponding threshold are marked with trend anomaly type and slope value. The integrity verification results, rationality verification results, consistency verification results and trend anomaly type are merged to obtain an anomaly marker dataset carrying anomaly type and anomaly value.

[0024] Specifically, the preset time window is set to 60 minutes, and the expected number of records is determined to be 60 records based on the acquisition resolution of 1 record per minute in step S1. When the ratio of the actual number of records to the expected number of records is lower than the missing rate threshold of 0.85, it is determined that there are missing records. The missing rate threshold of 0.85 is determined based on the engineering reliability standard of the oil and gas field production data acquisition system. A ratio lower than this indicates that the degree of data missing has exceeded the normal communication jitter range.

[0025] The engineering physical boundary is set independently based on the production engineering constraints of each parameter. Taking the wellhead pressure as an example, its reasonable range is set to 2.0MPa to 35.0MPa. If it is lower than 2.0MPa, it indicates that the wellhead may be in a shut-in or abnormal pressure relief state. If it is higher than 35.0MPa, it exceeds the design pressure limit of the floating production facility pipeline. The deviation is defined as the absolute value of the difference between the measured value and the nearest boundary value, and is written into the rationality verification result.

[0026] The cross-parameter constraint equations are constructed based on the material conservation relationships in oil and gas fields, assuming an oil production rate of... The moisture content is The gas-oil ratio is GOR, and the measured liquid production is... Actual production volume The liquid production volume record value is directly collected by the metering manifold sensor, transmitted through the link, and stored in the time-series database, along with the liquid production volume constraint calculation value. for: Constrained residuals The calculation formula is: The residual threshold is set to 0.15, which corresponds to the maximum allowable comprehensive measurement error of the oil and gas field metering system. Exceeding 0.15 indicates that there is a logical contradiction between the parameters that cannot be explained by the measurement error.

[0027] The preset trend window is set to 24 hours. Least squares linear regression is performed on the time-series data of each production parameter within this window, using timestamps as independent variables and parameter values ​​as dependent variables. Assume there are a total of [number missing] parameters within the trend window. Record number 1 The timestamp of the record is The corresponding parameter value is trend slope Solve using the least squares method: The trend slope threshold is set independently based on the upper limit of the engineering physical change rate of each parameter. Taking wellhead pressure as an example, the trend slope threshold is set to 0.5 MPa / h. This value corresponds to the maximum reasonable rate of pressure change under normal production conditions. Exceeding this rate indicates an abnormal trend deviation in pressure. The residual standard deviation threshold is set based on the fluctuation statistical characteristics of each parameter under historical normal operating conditions. Exceeding this threshold indicates an abnormal trend in the parameter. The integrity check results, rationality check results, consistency check results, and trend anomaly types are merged at the record level. Each anomaly record carries an anomaly type field and a corresponding anomaly value field. The anomaly value field stores the missing rate, deviation, residual value, or slope value according to the anomaly type. After merging, an anomaly-marked dataset is obtained. Each record in this dataset maintains a correspondence with the link-marked data record output in step S1 through a joint index of three fields: oil and gas field identifier, UTC timestamp, and parameter type code.

[0028] In one specific embodiment, step S3 includes: The duration of an anomaly is obtained by subtracting the start and end timestamps of consecutive anomalies for each anomaly record in the anomaly-marked dataset; the time density of anomalies is obtained by counting the number of times an anomaly record appears within a preset time window; and the time density of anomalies is combined to obtain the time dimension feature. Based on the link source identifier, other production parameter numbers that occur simultaneously with the current abnormal record are collected and merged within a preset time window to obtain a parameter co-occurrence anomaly set; the link source identifier and the parameter co-occurrence anomaly set are merged to obtain spatial dimension features; The outlier sequence of each outlier record in the outlier-marked dataset is subtracted from the baseline sequence. The first-order difference sign of the difference sequence is encoded to obtain the outlier sequence morphological code. The proportion of records in the outlier sequence that meet the fixed-byte boundary divisibility condition is calculated to obtain the boundary regularity index. The outlier sequence morphological code and the boundary regularity index are combined to obtain the morphological dimension feature. The time dimension features, spatial dimension features, and morphological dimension features are concatenated in the order of the fields to obtain the abnormal feature vector.

[0029] Specifically, the duration of an anomaly is obtained by subtracting the start and end timestamps of consecutively anomaly-marked record sequences with the same parameter in the anomaly-marked dataset, with the unit being minutes. The criteria for determining consecutive anomaly sequences is that the time interval between two adjacent anomaly records does not exceed the acquisition resolution of 1 minute set in step S1. If the interval exceeds this, they are considered two independent anomaly sequences, and their durations are calculated separately. Anomaly time density is defined as the number of occurrences of the same type of anomaly record within the 60-minute preset time window used in step S2. It is obtained by grouping and statistically analyzing the anomaly-marked dataset according to three fields: parameter type encoding, anomaly type, and time window, with the unit being times per hour. This indicator reflects the concentration of anomalies on the time axis and, together with the anomaly duration, constitutes a time-dimensional feature.

[0030] The parameter co-occurrence anomaly set is obtained by searching other production parameter records with overlapping timestamps and non-empty anomaly types within a 60-minute preset time window, extracting their parameter type codes, and then performing set merging. The merging operation removes duplicate codes, retaining the unique code set of all co-occurring anomaly parameters within the window. The link source identifier directly reads the encapsulated field value, which takes one of five categories: sensor direct acquisition, edge transmission node forwarding, protocol conversion gateway output, system manual entry, and cross-system synchronization interface push. It is merged with the parameter co-occurrence anomaly set to form a spatial dimension feature. The spatial dimension feature is stored in the form of a structure, containing two subfields: the link source identifier field and the parameter code set field.

[0031] Suppose that there are M non-abnormal records for the parameter to which the current abnormal record belongs within the last 72 hours, and the parameter values ​​of each record are denoted in chronological order as follows: benchmark value Defined as the mean of the values ​​of the above non-abnormal records: Based on the benchmark value For reference, suppose the outlier sequence has L records, and the values ​​of each record are denoted in chronological order as follows: The difference between each record and the benchmark value is calculated one by one to form a deviation sequence. Performing a first-order difference on the deviation sequence yields a sequence of changes between adjacent differences. : For the series of changes The sign of each element in the sequence is determined: all positive values ​​are encoded as 1, all negative values ​​as 2, alternating positive and negative values ​​as 3, monotony in the first part and convergence in the second part as 4, and the absolute values ​​of all differences concentrated at a fixed value as 5, thus obtaining the outlier sequence morphology code. The boundary regularity index is obtained by performing a fixed-byte boundary divisibility test on the values ​​of each record in the outlier sequence, using... The remainder is calculated for the divisor. Records with a remainder of zero are counted as satisfying the divisibility condition. The ratio of this count to the total number of records in the outlier sequence is the boundary regularity index, ranging from 0 to 1. A higher value indicates that the outlier sequence is closer to the protocol field length truncation feature. After merging the outlier sequence morphological encoding and the boundary regularity index to form a morphological feature, the time-dimensional feature, spatial-dimensional feature, and morphological feature are concatenated in a fixed order: time-dimensional feature fields first, spatial-dimensional feature fields in the middle, and morphological feature fields last. This results in an outlier feature vector with fixed dimensions. The categorical fields in this vector include the link source identifier and parameter co-occurrence anomaly set, while the numerical fields include anomaly duration, anomaly time density, outlier sequence morphological encoding, and boundary regularity index.

[0032] In one specific embodiment, step S4 involves performing similarity matching between the abnormal feature vector and historical feature records in the causal knowledge base to obtain the causal confidence distribution vectors corresponding to each of the five types of link causes, including: The distance between the abnormal feature vector and each historical feature record in the causal knowledge base is calculated using the Euclidean distance formula. The ten historical feature records with the smallest distance values ​​are selected to obtain the nearest neighbor feature record set. Based on the distance values ​​between each record in the nearest neighbor feature record set and the abnormal feature vector, the similarity weight of each record is calculated according to the distance inverse normalization formula to obtain the nearest neighbor similarity weight set. After multiplying the repair confirmation results of each record in the nearest neighbor similarity weight set with the similarity weight, the results are accumulated according to the five types of link causes: sensor failure, transmission packet loss, protocol truncation, manual input deviation and cross-system synchronization error. The accumulated results are then normalized according to the total weight to obtain the cause confidence distribution vector.

[0033] Specifically, the numerical fields in the anomaly feature vector include anomaly duration, anomaly time density, anomaly value sequence morphology encoding, and boundary regularity indicators. The categorical fields include link source identifier and parameter co-occurrence anomaly set. Before calculating the Euclidean distance, the numerical fields are subjected to max-min normalization based on the difference between the maximum and minimum values ​​of each field in the full historical record of the causal knowledge base, mapping the values ​​of each field to the interval between 0 and 1. Among the categorical fields, the link source identifier is encoded as a five-dimensional 0-1 vector using one-hot encoding based on five categories of values. The parameter co-occurrence anomaly set is encoded as a five-dimensional 0-1 vector based on five categories of parameters: production rate, wellhead pressure, production flow temperature, water cut, and gas-oil ratio, with 1 indicating presence and 0 indicating absence.

[0034] After concatenating all the above fields into a unified numerical vector, let the unified numerical vector of the current abnormal feature vector after the above processing be . The unified numerical vector corresponding to the s-th historical feature record in the causal knowledge base is: ,in Let be the total dimension of the vectors, and be the Euclidean distance between them. The calculation formula is: Pick The ten smallest records constitute the nearest neighbor feature record set. The number of nearest neighbors is determined based on the scale of valid confirmed records that the causal knowledge base can accumulate within a normal operation and maintenance cycle of a single facility. This number ensures coverage of representative samples from the five causal categories while controlling computational overhead. For the tenth record in the nearest neighbor feature record set... records ( Euclidean distance value Take the reciprocal, sum the reciprocals of all ten records, and then divide the reciprocal of each record by the sum to obtain the similarity weight. .

[0035] Each historical feature record in the causal knowledge base includes a knowledge record confidence field. The assignment rule is as follows: when the engineer enters the repair validity flag as valid, it indicates that the correspondence between the anomaly feature vector and the link cause confirmation value has been verified as correct through actual repair, and the value is assigned. When the repair validity flag is invalid, the record still carries exclusion information, which helps to reduce the confidence contribution of similar causes in subsequent similarity matching. Instead of 0, to distinguish it from unverified records, the confidence field ranges from 0 to 1.

[0036] For the h-th record in the nearest neighbor feature record set, assign its similarity weight. With confidence in knowledge records Multiply to obtain the weighted contribution value Let the cumulative score of the t-th type of link cause (t=1,2,3,4,5 corresponding to sensor failure, packet loss, protocol truncation, manual input error, and cross-system synchronization error, respectively) be... for: in For the first The repair confirmation result of each nearest neighbor record belongs to the cause type number. Normalize the cumulative scores of each category using the sum of the cumulative scores of the five categories as the denominator to obtain the t-th component of the cause confidence distribution vector. : The sum of the five components is strictly equal to 1, and the link cause corresponding to the largest component in the vector is the principal cause of the current anomaly record.

[0037] For example, suppose the Euclidean distance values ​​of 10 records in the nearest neighbor feature record set are 0.20, 0.25, 0.30, 0.35, 0.40, 0.45, 0.50, 0.55, 0.60, and 0.65, respectively. After normalization by the inverse of the distance, the precise similarity weight values ​​of each record are 0.185625, 0.148500, 0.123750, 0.106072, 0.092813, 0.082500, 0.074250, 0.067500, 0.061875, and 0.057115 (all rounded to six decimal places). The approximate values ​​are then listed below with three decimal places for each weight to facilitate calculation and display. Each approximate value is obtained by directly truncating the fourth decimal place and subsequent values ​​from the exact weight. The approximate display values ​​are 0.185, 0.148, 0.123, 0.106, 0.092, 0.082, 0.074, 0.067, 0.061, and 0.057, respectively. Since the last digit of each value is discarded during truncation, the sum of the ten approximate display values ​​is approximately 0.995, not strictly equal to 1. However, the subsequent normalization calculation uses the sum of the cumulative scores of each cause as the denominator, which is irrelevant to whether the sum of the weights is 1. Therefore, it does not affect the correctness of the final cause confidence distribution vector.

[0038] The repair confirmation results and knowledge record confidence levels for the 10 records are as follows: Records 1, 2, and 3 are confirmed as sensor faults, with a confidence level of 1.0 each; Records 4 and 5 are confirmed as packet loss, with a confidence level of 1.0 each; Record 6 is confirmed as a sensor fault, with a confidence level of 0.3; Records 7 and 8 are confirmed as protocol truncation, with a confidence level of 1.0 each; Record 9 is confirmed as manual input error, with a confidence level of 1.0; and Record 10 is confirmed as a cross-system synchronization error, with a confidence level of 1.0. The cumulative score for each cause is calculated using the above formula: Sensor failure Packet loss during transmission Protocol truncation Manual input error Cross-system synchronization error The sum of the cumulative scores of the five categories is approximately 0.938. After normalization, the cause confidence distribution vector is (0.513, 0.211, 0.150, 0.065, 0.061). The maximum component 0.513 corresponds to sensor failure. A fixed-point repair instruction is generated with sensor failure as the main cause. The transmission packet loss component 0.211 exceeds the threshold of 0.15 and constitutes a candidate cause.

[0039] In one specific embodiment, step S4, which generates a targeted repair instruction based on the link cause corresponding to the largest component in the cause confidence distribution vector, includes: Compare the values ​​of each component in the cause confidence distribution vector, take the link cause corresponding to the largest component as the main cause, and take the remaining link causes with component values ​​exceeding 0.15 to form the candidate cause set. Based on the primary and alternative cause sets, the following steps are taken: for sensor faults, check the sensor zero-point drift and range calibration records; for packet loss, check the network link status of the edge transmission node; for protocol truncation, check the protocol conversion gateway field mapping configuration; for manual input deviation, retrieve the input terminal operation log; and for cross-system synchronization errors, check the mapping rules of the synchronization interface version number and timestamp alignment status. The candidate cause components in the cause confidence distribution vector are sorted in descending order to obtain the fixed-point repair instructions.

[0040] Specifically, the cause confidence distribution vector includes five components: sensor failure, packet loss, protocol truncation, manual input error, and cross-system synchronization error. The values ​​of each of the five components are compared one by one, and the link cause corresponding to the component with the highest value is selected as the primary cause. The screening threshold for candidate causes is set to 0.15. This value is based on the fact that the sum of the five components of the cause confidence distribution vector is 1, and when the five types of causes are equally distributed, the mean of each component is 0.2. Setting the threshold to 0.15 indicates that only causes with confidence significantly higher than the random baseline are retained as candidates. Components with confidence values ​​lower than 0.15 have insufficient supporting samples in the nearest neighbor feature record set and lack statistical basis for repair reference. Link causes with component values ​​exceeding 0.15 in the cause confidence distribution vector (excluding the primary cause) are extracted one by one and arranged in descending order of component value to form a candidate cause set.

[0041] The targeted repair command is generated based on the primary cause and alternative cause sets according to the following mapping rules: When the primary cause is sensor failure, the zero-point drift record and range calibration record of the device pointed to by DeviceID are checked. The zero-point drift check is confirmed by comparing the current sensor output value with the output value of redundant sensors in the same node. If the deviation exceeds 2% of the full range, the sensor replacement process is triggered. When the primary cause is packet loss, the packet loss rate of the network link between the edge transmission node and the upper-level data aggregation node is checked. If the packet loss rate exceeds 5%, the transmission proxy service is restarted. When the primary cause is protocol truncation, the field mapping configuration file of the protocol conversion gateway is checked. The current data truncation position is compared with the protocol definition field length. If they are inconsistent, the protocol parsing template is updated. When the primary cause is manual input deviation, the operation log of the system input terminal within 10 minutes before and after the abnormal timestamp is retrieved. After locating the input personnel number and the input value, manual review is performed. When the primary cause is cross-system synchronization error, the data version number of the cross-system synchronization interface and the UTC timestamp alignment status are checked. If there is a version conflict, incremental synchronization compensation is performed. The fixed-point repair instructions are output in the following order: the instructions corresponding to the main cause are arranged first, and the instructions corresponding to the alternative causes are arranged in descending order of component values. Each instruction is accompanied by the component value of the corresponding cause, which constitutes a complete fixed-point repair instruction sequence.

[0042] In one specific embodiment, step S4, writing the repair confirmation result and the abnormal feature vector back to the cause knowledge base, includes: Input the link cause confirmation value and repair validity flag entered by the engineer into the cause knowledge base. When the repair validity flag is valid, assign a confidence value of 1.0 to the knowledge record. When the repair validity flag is invalid, assign a confidence value of 0.3 to the knowledge record to obtain a new knowledge record. After binding new knowledge records with abnormal feature vectors, they are written into the causal knowledge base. The total number of knowledge records in the causal knowledge base is compared with the capacity limit. When the total number of knowledge records exceeds the capacity limit, the 10% of records with the lowest confidence are deleted in ascending order of knowledge record confidence, resulting in an updated causal knowledge base.

[0043] Specifically, after completing the inspection operations corresponding to the targeted repair instruction, the engineer enters two pieces of information into the maintenance terminal: the link cause confirmation value and the repair validity flag. The link cause confirmation value is selected from one of five categories: sensor failure, packet loss, protocol truncation, manual input error, and cross-system synchronization error, based on the actual inspection conclusion. The repair validity flag is either valid or invalid. Valid indicates that the targeted repair operation eliminated the anomaly, while invalid indicates that the anomaly was not eliminated after executing the targeted repair instruction. The knowledge record confidence level is assigned based on the repair validity flag: a value of 1.0 is assigned when the repair validity flag is valid, indicating that the correspondence between the anomaly feature vector and the link cause confirmation value has been verified as correct through actual repair; a value of 0.3 is assigned instead of 0 when the repair validity flag is invalid, because invalid repair records still carry exclusion information, and retaining this record helps reduce the confidence contribution of similar causes in subsequent similarity matching; assigning a value of 0 would make it impossible to distinguish between unverified records and verified invalid records.

[0044] The complete field structure of the new knowledge record consists of three fields: anomaly feature vector, link cause confirmation value, and knowledge record confidence level. The anomaly feature vector field directly stores the vector content output from step S3, and maintains association with the original records in the anomaly tag dataset through a joint index of three fields: oil and gas field identifier, UTC timestamp, and parameter type encoding. The capacity limit is set to 5000 records. This value is determined based on the scale of effective confirmed records that a single floating production facility can accumulate within a three-year standard operation and maintenance cycle. After exceeding this scale, the marginal contribution of newly added records to the nearest neighbor matching results approaches zero. When the total number of knowledge records exceeds 5000, all records are sorted in ascending order by the knowledge record confidence level field value, and the lowest 10% of confidence levels, i.e., 500 records, are deleted. The 10% deletion ratio is based on ensuring that the continuous removal rate of low-quality records in the knowledge base matches the addition rate while controlling the overhead of a single deletion operation. After the deletion is completed, the total number of knowledge records is recounted and the index is updated to obtain the updated cause knowledge base.

[0045] The above describes the multi-dimensional quality verification method for oil and gas field production data in the embodiments of this application. The following describes the multi-dimensional quality verification system for oil and gas field production data in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the multi-dimensional quality verification system for oil and gas field production data in this application includes: The encapsulation module is used to collect the production parameter values ​​and link source identifiers of each link node of the floating production facility, and encapsulate the production parameter values ​​and the link source identifiers into a link tag data record. The verification module is used to process the link tag data records through missing detection, threshold boundary judgment, cross-parameter constraint verification and time series trend verification to obtain an anomaly tag dataset carrying anomaly type and anomaly value; The merging module is used to extract the duration of the anomaly, the link source identifier and the set of co-occurrence anomalies of parameters, the anomaly value sequence morphology encoding and boundary regularity index from each anomaly record in the anomaly label dataset, and merge them to form an anomaly feature vector; The generation module is used to perform similarity matching between the abnormal feature vector and historical feature records in the causal knowledge base to obtain the causal confidence distribution vectors corresponding to the five types of link causes. The module generates a fixed-point repair instruction based on the link cause corresponding to the largest component in the causal confidence distribution vector, and writes the repair confirmation result and the abnormal feature vector back to the causal knowledge base. When the number of local records in the causal knowledge base is lower than a preset threshold, the historical causal records of similar floating production facilities are weighted and merged according to parameter similarity to form the initial content of the causal knowledge base.

[0046] above Figure 2 The multi-dimensional quality verification system for oil and gas field production data in this embodiment of the invention is described in detail from the perspective of modular functional entities. The multi-dimensional quality verification device for oil and gas field production data in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0047] Reference Figure 3 This invention also provides a multi-dimensional quality verification device for oil and gas field production data. This device can be a server, and its internal structure can be as follows: Figure 3 As shown, the multi-dimensional quality verification device for oil and gas field production data includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor in this computer design provides computational and control capabilities. The memory of the multi-dimensional quality verification device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the multi-dimensional quality verification device for oil and gas field production data stores the data corresponding to this embodiment. The network interface of the multi-dimensional quality verification device for oil and gas field production data is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0048] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the multi-dimensional quality verification equipment for oil and gas field production data applied thereto.

[0049] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the multi-dimensional quality verification method for oil and gas field production data.

[0050] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0051] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the existing solution, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an oil and gas field production data multidimensional quality verification device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0052] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for multidimensional quality verification of oil and gas field production data, characterized in that, The method includes: Step S1: Collect the production parameter values ​​and link source identifiers of each link node of the floating production facility, and encapsulate the production parameter values ​​and the link source identifiers into a link tag data record; Step S2: The link tag data records are processed by missing detection, threshold boundary judgment, cross-parameter constraint verification and time series trend test to obtain an anomaly tag dataset carrying anomaly type and anomaly value; Step S3: Extract the anomaly duration, the link source identifier and parameter co-occurrence anomaly set, the anomaly value sequence morphology code and boundary regularity index from each anomaly record in the anomaly label dataset, and merge them to form an anomaly feature vector; Step S4: Perform similarity matching between the abnormal feature vector and historical feature records in the causal knowledge base to obtain the causal confidence distribution vectors corresponding to the five types of link causes. Generate a fixed-point repair instruction based on the link cause corresponding to the largest component in the causal confidence distribution vector, and write the repair confirmation result and the abnormal feature vector back to the causal knowledge base. When the number of local records in the causal knowledge base is lower than a preset threshold, the historical causal records of similar floating production facilities are weighted and merged according to parameter similarity to form the initial content of the causal knowledge base.

2. The multi-dimensional quality verification method for oil and gas field production data according to claim 1, characterized in that, Step S1 includes: At the wellhead, separator, metering manifold, and external pump nodes of the floating production facility, production rate, wellhead pressure, production flow temperature, water cut, and gas-oil ratio are collected to obtain production parameter values. Based on the data writing agent program of each link node, the production parameter values ​​are encapsulated with five types of link source identifiers: direct sensor acquisition, edge transmission node forwarding, protocol conversion gateway output, system manual input, and cross-system synchronization interface push, to obtain parameter acquisition records carrying the link source identifiers; The parameter acquisition records are written into the time series database using a dual index of oil and gas field identifier and UTC timestamp to obtain the link tag data records.

3. The multi-dimensional quality verification method for oil and gas field production data according to claim 1, characterized in that, Step S2 includes: Within a preset time window, the ratio of the actual number of records of each production parameter in the link marker data record to the expected number of records is calculated. Records with a ratio lower than the missing rate threshold are marked with missing type and missing rate to obtain the integrity verification result. Based on the engineering physical boundaries of each production parameter, the parameter values ​​in the link marking data record are compared with the engineering physical boundaries. Records that exceed the boundaries are marked with the out-of-bounds type and deviation amount to obtain the rationality verification result. Based on the cross-parameter constraint equation between production, water content and gas-oil ratio, the constraint residual is calculated for the parameter values ​​at each time point in the link marker data record. Records whose constraint residual exceeds the residual threshold are marked with logical conflict type and residual value to obtain the consistency verification result. Based on a preset trend window, linear regression is performed on the time-series data of each production parameter in the link marker data record to calculate the trend slope and residual standard deviation. Records whose trend slope or residual standard deviation exceeds the corresponding threshold are marked with trend anomaly type and slope value. The integrity verification result, the rationality verification result, the consistency verification result and the trend anomaly type are merged to obtain an anomaly marker dataset carrying anomaly type and anomaly value.

4. The method for multi-dimensional quality verification of oil and gas field production data according to claim 1, characterized in that, Step S3 includes: The duration of an anomaly is obtained by subtracting the start and end timestamps of consecutive anomalies for each anomaly record in the anomaly-marked dataset; the frequency of occurrence of the anomaly record within a preset time window is counted to obtain the anomaly time density; the duration of anomalies and the anomaly time density are combined to obtain a time dimension feature. Based on the link source identifier, other production parameter numbers that occur simultaneously with the current abnormal record are collected and merged within a preset time window to obtain a parameter co-occurrence anomaly set; the link source identifier and the parameter co-occurrence anomaly set are merged to obtain a spatial dimension feature. The outlier sequence of each outlier record in the anomaly-marked dataset is subtracted from the baseline sequence. The first-order difference sign of the difference sequence is encoded to obtain the outlier sequence morphological code. The proportion of records in the outlier sequence that satisfy the fixed-byte boundary divisibility condition is calculated to obtain the boundary regularity index. The outlier sequence morphological code and the boundary regularity index are combined to obtain the morphological dimension feature. The time dimension feature, the spatial dimension feature, and the morphological dimension feature are concatenated in field order to obtain the abnormal feature vector.

5. The multi-dimensional quality verification method for oil and gas field production data according to claim 1, characterized in that, In step S4, the abnormal feature vector is matched with historical feature records in the causal knowledge base to obtain the causal confidence distribution vectors corresponding to the five types of link causes, including: The distance between the abnormal feature vector and each historical feature record in the cause knowledge base is calculated using the Euclidean distance formula. The ten historical feature records with the smallest distance values ​​are selected to obtain the nearest neighbor feature record set. Based on the distance values ​​between each record in the nearest neighbor feature record set and the abnormal feature vector, the similarity weight of each record is calculated according to the inverse distance normalization formula to obtain the nearest neighbor similarity weight set. After multiplying the repair confirmation results of each record in the nearest neighbor similarity weight set with the similarity weight, the results are accumulated according to the five types of link causes: sensor failure, transmission packet loss, protocol truncation, manual input deviation, and cross-system synchronization error. The accumulated results are then normalized according to the total weight to obtain the cause confidence distribution vector.

6. The multi-dimensional quality verification method for oil and gas field production data according to claim 5, characterized in that, Step S4, which generates a targeted repair instruction based on the link cause corresponding to the largest component in the cause confidence distribution vector, includes: The component values ​​in the cause confidence distribution vector are compared, and the link cause corresponding to the largest component is taken as the main cause. The remaining link causes with component values ​​exceeding 0.15 are taken as the candidate cause set. Based on the primary cause and the set of alternative causes, the following steps are taken: checking sensor zero-point drift and range calibration records for sensor faults; checking the network link status of edge transmission nodes for packet loss; checking the protocol conversion gateway field mapping configuration for protocol truncation; retrieving the input terminal operation log for manual input deviations; and checking the mapping rules for synchronization interface version number and timestamp alignment status for cross-system synchronization errors. The results are then sorted in descending order of the candidate cause component values ​​in the cause confidence distribution vector to obtain the fixed-point repair instruction.

7. The method for multidimensional quality verification of oil and gas field production data according to claim 6, characterized in that, Step S4 involves writing the repair confirmation result and the abnormal feature vector back to the cause knowledge base, including: The link cause confirmation value and repair validity flag entered by the engineer are input into the cause knowledge base. When the repair validity flag is valid, the confidence level of the knowledge record is assigned to 1.

0. When the repair validity flag is invalid, the confidence level of the knowledge record is assigned to 0.3, thus obtaining a new knowledge record. After binding the new knowledge record with the abnormal feature vector, it is written into the cause knowledge base. The total number of knowledge records in the cause knowledge base is compared with the capacity limit. When the total number of knowledge records exceeds the capacity limit, the 10% of records with the lowest confidence are deleted in ascending order of knowledge record confidence, and the updated cause knowledge base is obtained.

8. A multi-dimensional quality verification system for oil and gas field production data, characterized in that, For implementing the multidimensional quality verification method for oil and gas field production data as described in any one of claims 1-7, the multidimensional quality verification system for oil and gas field production data comprises: The encapsulation module is used to collect the production parameter values ​​and link source identifiers of each link node of the floating production facility, and encapsulate the production parameter values ​​and the link source identifiers into a link tag data record. The verification module is used to process the link tag data records through missing detection, threshold boundary judgment, cross-parameter constraint verification and time series trend verification to obtain an anomaly tag dataset carrying anomaly type and anomaly value; The merging module is used to extract the duration of the anomaly, the link source identifier and the set of co-occurrence anomalies of parameters, the anomaly value sequence morphology encoding and boundary regularity index from each anomaly record in the anomaly label dataset, and merge them to form an anomaly feature vector; The generation module is used to perform similarity matching between the abnormal feature vector and historical feature records in the causal knowledge base to obtain the causal confidence distribution vectors corresponding to the five types of link causes. The module generates a fixed-point repair instruction based on the link cause corresponding to the largest component in the causal confidence distribution vector, and writes the repair confirmation result and the abnormal feature vector back to the causal knowledge base. When the number of local records in the causal knowledge base is lower than a preset threshold, the historical causal records of similar floating production facilities are weighted and merged according to parameter similarity to form the initial content of the causal knowledge base.

9. A multi-dimensional quality verification device for oil and gas field production data, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the multi-dimensional quality verification method for oil and gas field production data as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the multi-dimensional quality verification method for oil and gas field production data as described in any one of claims 1 to 7.

Images