Oil well data intelligent management system and method

Abstract

The invention provides an oil well data intelligent management system and method, and relates to the technical field of oil well production. Comprising an edge layer and a platform layer. The edge layer is used for performing anomaly analysis on the oil well data; the platform layer is used for formulating an intervention control strategy according to abnormal oil well data; the edge computing node is used for performing feature extraction on the original oil well data; the model building module is used for building an anomaly detection model in combination with a historical oil well database; the anomaly recognition module is used for inputting the oil well data features into an anomaly detection model for anomaly recognition. According to the method, the edge computing nodes are combined with the lightweight LSTM model, so that the problem of high delay of cloud centralized processing in a traditional scheme is solved, and sudden equipment faults can be accurately captured in a low response period; a dynamically optimized intervention strategy generation mechanism is adopted, core parameters are adaptively adjusted according to different well conditions, a closed-loop control link is formed in cooperation with a cloud continuous iteration diagnosis knowledge base, and the manual inspection frequency and the operation and maintenance cost are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of oil well production, and particularly to an intelligent management system and method for oil well data. Background Art

[0002] With the continuous growth of global energy demand and the increasing difficulty of oil and gas resource development, oil well production management is facing unprecedented technical challenges. In the traditional oil well development mode, the average daily production cost per well exceeds 20,000 yuan, and the unplanned shutdowns caused by equipment failures result in annual losses of over 10 billion yuan for the industry. In complex operating environments such as shale gas fields and high-sulfur oil fields, existing production management systems are difficult to meet the core requirements of safe production, cost reduction, and efficiency improvement. The 2023 report of the International Energy Agency (IEA) pointed out that the digital penetration rate of the global oil and gas industry is less than 35%, and the data utilization rate is lower than 20%. There is an urgent need to build a new generation of intelligent production management systems. The continuous growth of global energy demand and the continuous upgrading of the difficulty of oil and gas resource development are driving technological innovation in the field of oil well production management. According to the "Global Energy Outlook" released by the International Energy Agency (IEA) in 2023, the global crude oil demand will increase at an average annual rate of 0.8% in the next decade. However, due to the increasing reservoir complexity and lagging development technology, the oil and gas resource recovery rate is expected to drop to 45%-50%. This contradiction is particularly prominent in the development of unconventional resources such as deepwater oil and gas, shale oil, and tight gas. The average burial depth of newly discovered oil fields has increased from 2,000 meters in 2000 to 3,500 meters in 2023, and the drilling cost per well has soared by 40%. The proportion of high-temperature and high-pressure wells (temperature > 150°C, pressure > 70 MPa) exceeds 25%, and traditional monitoring technologies are difficult to accurately perceive downhole conditions. Taking a certain ultra-deepwater oil field in the Middle East as an example, its daily production per well reaches 80,000 barrels, but the unplanned shutdowns caused by equipment failures result in direct economic losses of over 200 million US dollars annually. The International Energy Agency predicts that if intelligent transformation is not promoted, the global oil and gas industry will face an annual production efficiency loss of 150 billion US dollars by 2030.

[0003] Generally, the oil well data acquisition can only obtain surface parameters such as wellhead pressure and flow rate, lacking real-time monitoring of key data such as wellbore dynamics and formation seepage. And a linear analysis model is mostly used, which is difficult to handle high-dimensional non-linear problems. Equipment maintenance mostly adopts the "repair after failure" mode, and the coverage rate of predictive maintenance is low. Summary of the Invention

[0004] The present invention provides an intelligent management system and method for oil well data to solve the defects existing in the prior art.

[0005] On the one hand, the present invention provides an intelligent management system for oil well data, including: Edge layer and platform layer; the edge layer is used for anomaly analysis of well data and outputs abnormal well data; the platform layer is used for formulating intervention control strategies based on the abnormal well data; the edge layer includes: Downhole sensor array for real-time collection of original well data; Edge computing node for feature extraction of the original well data and outputting well data features; Model construction module for establishing an anomaly detection model in combination with the historical well database; Anomaly identification module for inputting the well data features into the anomaly detection model for anomaly identification and outputting abnormal well data.

[0006] According to an intelligent management system for well data provided by the present invention, the edge computing node includes a data preprocessing unit and a feature extraction unit. The preprocessing unit is used for data preprocessing of the original well data to generate preprocessed well data; the feature extraction unit is used for feature extraction of the preprocessed well data and outputting well data features.

[0007] According to an intelligent management system for well data provided by the present invention, the steps for establishing an anomaly detection model include: Normalizing and serializing the data in the historical well database to construct a training set, a test set, and a validation set; Constructing an LSTM model network; Training the LSTM model according to the training set using the cross-entropy loss function and the Adam optimization algorithm and outputting the training result; Validating the training result using the test set until the loss of the validation set does not improve in consecutive validations and then terminating the training.

[0008] According to an intelligent management system for well data provided by the present invention, the specific steps for normalizing and serializing the data include: Using the continuity of the time series to fill in the missing data in the historical well database; Correcting the outlier data that exceeds the reasonable value range; Calculating the mean and standard deviation of the data.

[0009] According to an intelligent management system for well data provided by the present invention, the anomaly identification module includes a window division unit and an anomaly determination unit; the window division unit is used for dividing the well data features into multiple detection windows with a fixed length, and the anomaly determination unit is used for inputting the well data features in the multiple detection windows into the anomaly detection model for detection and outputting abnormal well data.

[0010] According to an intelligent management system for well data provided by the present invention, the platform layer includes: A cloud data processing module for analyzing the causes of anomalies in abnormal oil well data and generating an anomaly diagnosis report; An intelligent decision-making module for generating an intervention control strategy based on the anomaly diagnosis report.

[0011] For an intelligent oil well data management system provided by the present invention, the specific steps for analyzing the cause of the anomaly include: Spatiotemporally aligning the abnormal oil well data with the same-type working condition data in the historical oil well database and outputting an abnormal parameter combination; Based on the association rule mining algorithm and in combination with the abnormal parameter combination, generating a structured fault mapping table; According to the abnormal parameter combination and in combination with the equipment failure database, locating the abnormal propagation path and potential interference factors; Generating an anomaly diagnosis report based on the abnormal propagation path and potential interference factors.

[0012] For an intelligent oil well data management system provided by the present invention, the specific steps for generating an abnormal parameter combination include: Setting the parameters of the association rule mining algorithm according to the abnormal parameter combination and generating a high-frequency parameter combination; Generating candidate rules based on the high-frequency parameter combination and calculating the screening rules for the candidate rules to obtain a list of highly correlated rules; Constructing a triple mapping based on the list of highly correlated rules to generate a structured fault mapping table.

[0013] For an intelligent oil well data management system provided by the present invention, the specific steps for outputting the abnormal parameter combination include: Extracting the working condition characteristics of the abnormal oil well data and the historical oil well data and outputting a working condition characteristic matrix; Performing abnormal parameter clustering on the working condition characteristic matrix based on the random forest algorithm and outputting an abnormal parameter combination.

[0014] On the other hand, the present invention also provides an intelligent oil well data management method, including: Real-time collecting raw oil well data; Performing feature extraction on the raw oil well data and outputting oil well data features; Performing abnormal feature detection on the oil well data features based on the LSTM algorithm model and outputting abnormal oil well data; Analyzing the causes of anomalies in the abnormal oil well data and generating an anomaly diagnosis report; Generating an intervention control strategy based on the anomaly diagnosis report.

[0015] An intelligent management system and method for oil well data provided by the present invention can autonomously complete data cleaning, feature extraction, and initial anomaly screening at the well site through embedded edge computing nodes in the edge layer combined with a lightweight LSTM model, effectively solving the high latency problem of cloud centralized processing in traditional solutions, enabling sudden equipment failures to be accurately captured within a millisecond response cycle. Through spatio-temporal alignment technology and association rule mining algorithms, not only can the non-linear association patterns between key parameters be automatically decoupled from massive historical data, but also the anomaly conduction path can be visually presented through a three-dimensional fault mapping table, significantly improving the root cause diagnosis accuracy under complex working conditions. By adopting an intervention strategy generation mechanism based on dynamic optimization, core parameters such as injection pressure and oil production rate can be adaptively adjusted according to different well conditions, and combined with the continuously iterative diagnostic knowledge base in the cloud, a closed-loop control link of "perception - analysis - decision - execution" is formed, greatly reducing the frequency of manual inspections and operation and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] Figure 1 FIG. 1 is a schematic structural diagram of an intelligent management system for oil well data provided in Embodiment 1 of the present invention; Figure 2 FIG. 2 is a step diagram for establishing an anomaly detection model in Embodiment 1 of the present invention; Figure 3 FIG. 3 is a step diagram of an intelligent management method for oil well data provided in Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0019] The following will describe Figures 1-3 an intelligent management system and method for oil well data of the present invention.

[0020] Embodiment 1: Figure 1 FIG. 1 is a schematic structural diagram of an intelligent management system for oil well data provided in an embodiment of the present invention.

[0021] Figure 2 It is a step diagram for establishing an anomaly detection model in the first embodiment of the present invention.

[0022] As Figures 1-2 shown, an intelligent management system and method for oil well data provided by an embodiment of the present invention may have an execution entity as a big data analysis system of cloud computing, including: An edge layer and a platform layer. The edge layer is used for anomaly analysis of oil well data and outputs abnormal oil well data. The edge layer includes: An underground sensor array for real-time collection of raw oil well data. It includes new Internet of Things devices such as triaxial accelerometers (monitoring vibration), fiber Bragg grating sensors (measuring strain), multi-parameter probes (obtaining temperature / pressure / flow), and acoustic sensors (capturing leakage signals). These sensors can monitor key parameters such as vibration, strain, temperature, pressure, flow, and leakage signals of oil wells.

[0023] An edge computing node for feature extraction of raw oil well data and outputs oil well data features. The edge computing node includes a data preprocessing unit and a feature extraction unit. The preprocessing unit is used for data preprocessing of raw oil well data to generate preprocessed oil well data. Data preprocessing is an important step in data analysis. It can improve the availability and reliability of data and provide a good data foundation for subsequent feature extraction and model establishment. Usually, a sliding window mechanism is used to implement data cleaning, cubic spline interpolation is used to fill time series gaps, and the K-means clustering method is used to identify outliers and the median is used for substitution and other methods for data preprocessing. The feature extraction unit is used for feature extraction of preprocessed oil well data and outputs oil well data features. By converting complex raw data into analyzable numerical features. By fusing time-frequency analysis means, an improved wavelet packet transform is used to extract energy entropy features in the frequency band of 0.1 - 100 kHz, and the principal component analysis method is combined to reduce the dimension to a 20-dimensional feature vector and other methods for feature extraction.

[0024] A model construction module for randomly sampling a fixed number from the working condition data covering 100,000 oil wells within a 5-year cycle in combination with a historical oil well database. The steps for establishing an anomaly detection model include: Normalizing and serializing the data in the historical oil well database to construct a training set, a test set, and a validation set. The specific steps for normalizing and serializing the data in the historical oil well database include: Utilizing the continuity of the time series to fill in the missing data in the historical oil well database. A bidirectional LSTM network is used for cross-cycle data completion, and multi-source data fusion compensation is initiated for sequences with a missing rate exceeding 15%. This method can effectively utilize the information in the time series and improve the accuracy of data completion.

[0025] Outlier correction is performed on data outside the reasonable value range. An improved Grubbs test method combined with Markov chain Monte Carlo simulation is used to determine the confidence interval. Outliers can be accurately identified and corrected, improving the reliability of the data.

[0026] Calculate the mean and standard deviation of the data. When calculating the mean and standard deviation of the data, robust statistics are introduced, and the Huber loss function is used to replace the traditional variance calculation, reducing the influence of outliers on the statistical results and improving the robustness of the statistical results. A hybrid neural network architecture consisting of two layers of bidirectional LSTM layers (128 units) and an attention mechanism layer is constructed to effectively extract features in the time series and improve the prediction performance of the model.

[0027] Construct an LSTM model network. Use deep learning frameworks such as Keras or TensorFlow to construct a hybrid neural network architecture consisting of two layers of bidirectional LSTM layers (128 units) and an attention mechanism layer. The LSTM layer can effectively extract features in the time series, and the attention mechanism layer can further highlight important features, improving the prediction performance of the model.

[0028] According to the training set, train the LSTM model with an improved Focal Loss function (adjustment factor γ = 2, α = 0.85) and an AdamW optimizer (weight decay 0.01), and output the training results. The improved Focal Loss function can effectively solve the class imbalance problem and improve the generalization ability of the model. The AdamW optimizer can effectively improve the convergence speed and performance of the model.

[0029] Use the test set to verify the training results until the loss of the validation set does not improve during consecutive validations. That is, use the dynamic time warping algorithm to perform variable-length alignment on the test set, and terminate the training when the MAE of the validation set decreases by less than 0.1% for 3 consecutive epochs, ensuring the generalization ability of the model and improving the prediction performance of the model.

[0030] An outlier recognition module is used to input the oil well data features into an outlier detection model for outlier recognition and output the outlier oil well data. The outlier recognition module includes a window division unit and an outlier determination unit. The window division unit adopts a variable-length window strategy (128 sampling points for the basic window, expansion coefficient 1.5), combined with a dynamic threshold adjustment mechanism; the outlier determination unit introduces a multi-modal decision fusion mechanism, integrating the LSTM prediction probability (threshold 0.8), reconstruction error (threshold 3σ), and SHAP value interpretability results, and outputs the outlier oil well data.

[0031] The platform layer is used to formulate intervention control strategies based on the outlier oil well data. The platform layer includes: Cloud data processing module, which is used to analyze the abnormal causes of abnormal oil well data and generate an abnormal diagnosis report. The specific steps for analyzing the abnormal causes include: Perform spatio-temporal alignment of the abnormal oil well data with the same-type working condition data in the historical oil well database, use the dynamic time warping algorithm to eliminate clock deviation, and output an abnormal parameter combination. Specifically, the module will perform spatio-temporal alignment of the abnormal oil well data with the same-type working condition data in the historical oil well database. This step is to ensure the comparability between data, and spatio-temporal alignment can help us more accurately identify and analyze abnormal situations. During the spatio-temporal alignment process, the module will use the dynamic time warping algorithm to eliminate clock deviation. The dynamic time warping algorithm is a non-linear time series analysis method that can effectively handle non-linear changes in time series, thereby improving the accuracy of alignment.

[0032] The specific steps for outputting the abnormal parameter combination include: Extract the working condition characteristics of the abnormal oil well data and the historical oil well data, and output a working condition characteristic matrix. In order to transform complex oil well data into analyzable numerical features. The extracted working condition characteristics may include key parameters such as the pressure, flow rate, and temperature of the oil well. Through these characteristics, we can more accurately analyze and identify abnormal situations. Obtain the intrinsic mode functions by VMD decomposition, and construct a 32×32-dimensional working condition characteristic matrix through the correlation coefficient matrix. VMD is a non-recursive and adaptive signal processing method that can effectively decompose complex signals into multiple intrinsic mode functions (IMFs). Through VMD decomposition, we can obtain IMF components that reflect the working condition characteristics.

[0033] Perform abnormal parameter clustering on the working condition characteristic matrix based on the random forest algorithm, optimize the number of clusters using the silhouette coefficient index, and output an abnormal parameter combination including key parameters such as excessive pressure fluctuations and abnormal pump valve strokes. The silhouette coefficient is a comprehensive index for measuring the clustering effect, which can reflect the degree of tightness within the cluster and the degree of separation between clusters. By optimizing the number of clusters, we can obtain a more reasonable abnormal parameter clustering result.

[0034] Based on the association rule mining algorithm, combine the abnormal parameter combination to generate a structured fault mapping table. The abnormal parameter combination can help us more accurately identify and analyze abnormal situations, providing a basis for subsequent fault diagnosis and intervention control. The specific steps for generating the structured fault mapping table include: Extract the core elements in the high-correlation rule list and output a preliminary rule set. Use the FP-Growth algorithm to mine strong association rules between parameters (minimum support 0.05, confidence 0.9), and extract the high-correlation rule list through frequent item set analysis. The specific methods for obtaining high-correlation rules include: Discretize continuous parameters (quantile grading) and encode (One-Hot / binary identification) the data in the abnormal parameter combination.

[0035] Use the FP-Growth algorithm to extract parameter combinations that meet the minimum support.

[0036] Generate candidate rules A→B for each frequent itemset. Specific method: For the frequent itemset X={a,b,c}, generate all non-empty proper subsets A of X, and the remaining part is B = X / A. For each subset A, form the rule A→B. If the support of A or B is lower than the preset threshold (e.g., min support =0.05), then discard this rule.

[0037] For example: The frequent itemset X={a,b,c} has a support of 0.10, and the generated candidate rules include: A→BC (support ≥ 0.05), AB→C (support ≥ 0.05), AC→B, B→AC, BC→A, C→AB.

[0038] Calculate confidence, lift, and leverage, and filter out low-strength rules (e.g., Lift < 1.2). Confidence represents the probability that rule B also appears when rule A appears; lift represents the ratio of the probability that rules A and B appear simultaneously to the probability that A and B appear independently; leverage reflects the difference between the simultaneous appearance of rules A and B and their independent appearance. The formula for calculating confidence is expressed as:

[0039] Represents the probability that event B will surely occur when time A occurs.

[0040] The formula for calculating lift is expressed as:

[0041] Used to measure the correlation between event A and event B. If Lift > 1, then positive correlation (A and B tend to occur simultaneously). If Lift = 1, then independent (unrelated). If Lift < 1, then negative correlation (A and B tend to be mutually exclusive).

[0042] The formula for calculating leverage is expressed as:

[0043] Used to measure the deviation between the actual co-occurrence frequency of event A and event B and the independent situation: If Leverage > 0, then positive correlation (the co-occurrence frequency is higher than the independent expectation). If Leverage = 0, then independent. If Leverage < 0, then negative correlation.

[0044] Filter out the low-intensity rules with lift less than 1.2 and retain the rules with Lift≥1.2 (to avoid weak correlations). Retain the rules with Confidence≥0.85 (to avoid low confidence). For rules with lift close to the threshold, further screen them in combination with leverage (such as Leverage>0.05).

[0045] Then sort the rule strengths in descending order of lift (prioritize strongly correlated rules). If the lift is the same, sort them in descending order of confidence.

[0046] Retain the rules that meet the preset weights (such as reducing the confidence threshold for high-risk faults). Prune the inclusion relationship rules and merge semantic duplicates (such as A→B and A∧C→B).

[0047] Statistical tests (chi-square / Fisher test) confirm significance. Prioritize in combination with domain knowledge (such as rules for decreasing oil production taking precedence over rules for equipment efficiency), and output a list of highly correlated rules.

[0048] Perform semantic parsing on the rules, and use natural language processing technology (BERT model) to identify core fault elements (such as the combination of "abnormal pump power↑+fluctuating oil pressure↓"), and output a preliminary rule set containing weight coefficients. Association rule mining is a data mining method that can discover potential relationships and rules in data.

[0049] The specific steps for outputting the structured fault mapping table include: Map the preliminary rule set to the ontology predefined in the knowledge graph, map the parameters in the preliminary rule set to ontology attributes, standardize the relationship types to ontology relationships such as isCausedBy and affects, and annotate the rule confidence through RDF-star, that is, based on the verification success rate of historical cases, and output the standardized rule data.

[0050] Convert each rule in the standardized rule data into a triple, and output a list of standardized triples. Design a multi-hop relationship encoding strategy to convert the standardized rules into the form of <subject, predicate, object> triples.

[0051] Generate a structured fault mapping table based on the standardized triple list and related table fields. Develop a dynamic field mapping engine to automatically match database fields according to the device type. Use a columnar storage database (ClickHouse) to build the structured fault mapping table, supporting millisecond-level query responses.

[0052] Based on the abnormal parameter combination, combined with the equipment failure database, locate the abnormal propagation path and potential interference factors. Based on the equipment failure database (including a knowledge graph of more than 3,000 failure cases), use a graph convolutional network (GCN) to rank the importance of nodes. Design a random walk algorithm (restart probability 0.15) to identify the abnormal parameter association path in the knowledge graph (such as "downhole vibration anomaly → drive shaft eccentricity → ground torque fluctuation"). Introduce an LSTM-Autoencoder model to analyze the time-series interference signal and detect potential interference factors such as sudden changes in formation pressure and equipment aging trends.

[0053] Generate an abnormal diagnosis report based on the abnormal propagation path and potential interference factors. Construct a Bayesian network inference engine to calculate the root cause probability ranking (such as the bearing wear probability reaching 87.3%). Predict the production loss (confidence interval 95%) and environmental impact level through Monte Carlo simulation. Finally, output an interactive diagnosis report containing a fault tree analysis diagram and a three-dimensional heat map.

[0054] An intelligent decision-making module for generating intervention control strategies based on the abnormal diagnosis report. It includes an adaptive regulation strategy library (presetting various typical working condition response plans), a digital twin simulation verification module (using a physical information neural network to accelerate the simulation), and can dynamically optimize control parameters through a reinforcement learning framework, and output the final control strategy including throttle valve adjustment instructions, water injection plan adjustment, and equipment start-stop plan.

[0055] In summary, the present embodiment provides a big data analysis system for cloud computing. Through the embedded edge computing nodes in the edge layer combined with a lightweight LSTM model, it can autonomously complete data cleaning, feature extraction, and abnormal preliminary screening at the well site, effectively solving the high-latency problem of centralized processing in the cloud in the traditional solution, enabling sudden equipment failures to be accurately captured within a millisecond-level response cycle. Through the spatio-temporal alignment technology and the association rule mining algorithm, it can not only automatically decouple the non-linear association patterns between key parameters from a large amount of historical data, but also intuitively present the abnormal conduction path through a three-dimensional fault mapping table, significantly improving the root cause diagnosis accuracy under complex working conditions. By adopting an intervention strategy generation mechanism based on dynamic optimization, it can adaptively adjust core parameters such as water injection pressure and oil production rate according to different well conditions, and cooperate with the continuously iterated diagnostic knowledge base in the cloud to form a closed-loop control link of "perception - analysis - decision - execution", greatly reducing the frequency of manual inspections and operation and maintenance costs.

[0056] Example 1: Data acquisition shows that the downhole triaxial accelerometer detects an abnormal vibration frequency value of 25 kHz continuous pulse. The fiber Bragg grating sensor shows that the pump barrel strain value exceeds the threshold by 15%.

[0057] After edge computing processing, the data is cleaned through a sliding window (128 points / window), and two time gaps are repaired by cubic spline interpolation. The improved wavelet packet is used to extract the energy entropy features from 0.5 - 45 kHz, and the dimension is reduced to 20 by PCA.

[0058] The probability predicted by the trained LSTM prediction model is 0.83. The reconstruction error reaches 3.2σ, and the SHAP analysis shows that the contribution degree of the vibration spectrum features reaches 62%.

[0059] The disposal plan given by the system is as follows: The digital twin simulation shows that an emergency speed reduction of 30% is required. The reinforcement learning outputs a compensation plan with a 15% increase in the water injection volume.

[0060] Embodiment 2: Under the same general inventive concept, the present invention also protects an intelligent management method for oil well data. The following describes an intelligent management method for oil well data provided by the present invention. The intelligent management method for oil well data described below can be correspondingly referred to the big data analysis system of cloud computing described above.

[0061] Figure 3 It is a step diagram of an intelligent management method for oil well data provided by an embodiment of the present invention.

[0062] As Figure 3 shown, an intelligent management method for oil well data provided by an embodiment of the present invention includes: Real-time collection of original oil well data. Through a multi-parameter monitoring device (including pressure transmitters, temperature sensors, flow meters, vibration probes, etc.) deployed in the downhole sensor network, dynamic data of the entire oil well production process is obtained at a millisecond-level sampling frequency. The industrial Internet of Things architecture is used to realize data transmission, and the hybrid networking technology of ZigBee, LoRaWAN and 5G is integrated to ensure data real-time performance. A distributed data cache pool is established to preliminarily verify the original data, and the sliding window algorithm is used to eliminate instantaneous noise interference, and the timestamp alignment and unit standardization processing are completed synchronously. A satellite communication module is configured for oil wells in remote areas as a redundant link to ensure data integrity and continuity.

[0063] Extract features from the original well data and output the well data features. Construct a multi-dimensional feature space based on digital signal processing technology, covering time-domain features (mean, variance, peak-to-peak value), frequency-domain features (FFT spectral peak, harmonic distortion degree), and non-linear dynamics features (Lyapunov exponent, fractal dimension). The peak-to-peak value is the difference between the maximum and minimum values of a signal within one period in signal processing. It is usually used to describe the fluctuation range or amplitude of a signal. Use an improved wavelet packet transform algorithm to extract the energy entropy index of the fault-sensitive frequency band, and combine the principal component analysis method (PCA) for feature dimensionality reduction. Develop an adaptive feature selection model and use a reinforcement learning mechanism to dynamically optimize the combination of feature subsets. Construct a three-dimensional feature mapping matrix including geological structure parameters, equipment working condition parameters, and production operation parameters to realize the correlation mapping between reservoir seepage characteristics and surface production data.

[0064] Perform anomaly feature detection on the well data features based on the LSTM algorithm model and output the abnormal well data. Build a deep long short-term memory neural network architecture, set up a three-layer bidirectional LSTM unit to form a time series prediction network, and the number of input layer nodes corresponds to the optimized feature dimension. Introduce an attention mechanism module in the hidden layer to strengthen the feature response of key time steps. Use the Adam optimizer combined with the Dropout regularization technique to prevent overfitting. The training set uses the normal working condition samples marked in the historical data, and the validation set introduces the synthetic minority over-sampling technique (SMOTE) to balance the class distribution. Design a dynamic threshold determination mechanism, and combine the Mahalanobis distance and the Grubbs test statistic to determine the multi-variable joint anomaly criterion. Develop an incremental learning framework to support online model updates, and automatically retrain the model parameters every 24 hours to adapt to the changes in production conditions.

[0065] Analyze the abnormal causes of the abnormal well data and generate an abnormal diagnosis report. Construct an expert system based on a knowledge graph, integrate equipment manuals, fault case libraries, and workers' experience knowledge, and use natural language processing technology to parse unstructured documents to construct a causal inference chain. Apply a Bayesian network for multi-source heterogeneous data fusion reasoning to quantify the conditional probability relationship between different factors. Develop a hierarchical diagnosis module: the primary diagnosis identifies typical fault modes (such as pump valve failure, tubing leakage) through pattern matching; the advanced diagnosis calls a physical simulation engine to construct a digital twin model and simulate the evolution process of production parameters under different working conditions.

[0066] Generate an intervention control strategy based on the anomaly diagnosis report. Establish a rule-based strategy library that includes three types of knowledge units: equipment maintenance plans, production parameter optimization plans, and emergency response measures. Develop a multi-objective optimization engine that comprehensively considers economic benefits, safety risks, and environmental protection constraints, and uses the NSGA-II genetic algorithm to solve for the optimal control sequence. For deterministic fault types, trigger a preset standard operating procedure (SOP) and send control instructions to the RTU unit through the industrial Internet of Things; for complex working condition problems, call the digital twin platform for virtual verification and generate a decision proposal that includes the parameter adjustment range, implementation time, and expected effects. Introduce a human-machine collaboration mechanism where key decisions need to be reviewed and confirmed by experts before execution, and use blockchain technology to record operation logs to ensure traceability.

[0067] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected according to actual needs to achieve the objectives of the embodiment solution. A person of ordinary skill in the art can understand and implement it without creative effort.

[0068] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware. With such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An intelligent management system for oil well data, characterized in that, Including: Edge layer and platform layer; The edge layer is used to perform anomaly analysis on well data and output abnormal well data; The platform layer is used to formulate intervention control strategies based on the abnormal well data; The edge layer includes: Downhole sensor array, used to collect raw well data in real time; Edge computing node, used to extract features from the raw well data and output well data features; Model construction module, used to establish an anomaly detection model in combination with the historical well database; Anomaly identification module, used to input the well data features into the anomaly detection model for anomaly identification and output the abnormal well data.

2. The intelligent management system for oil well data according to claim 1, wherein, The edge computing node includes a data preprocessing unit and a feature extraction unit. The preprocessing unit is used to perform data preprocessing on the raw well data to generate preprocessed well data; the feature extraction unit is used to extract features from the preprocessed well data and output well data features.

3. An intelligent management system for oil well data according to claim 1, characterized in that, The steps of establishing the anomaly detection model include: Normalize and serialize the data in the historical well database to construct a training set, a test set, and a validation set; Construct an LSTM model network; According to the training set, use the cross-entropy loss function and the Adam optimization algorithm to train the LSTM model and output the training result; Use the test set to verify the training result until the loss of the validation set does not improve in consecutive verifications and terminate the training.

4. An intelligent management system for oil well data according to claim 3, characterized in that, The specific steps of performing the normalization and the serialization processing include: Utilize the continuity of the time series to fill in the missing data in the historical well database; Correct the outlier values of the data that exceed the reasonable value range; Calculate the mean and standard deviation of the data.

5. An intelligent management system for oil well data according to claim 1, characterized in that, The anomaly identification module includes a window division unit and an anomaly determination unit; the window division unit is used to divide the well data features into multiple detection windows of a fixed length, and the anomaly determination unit is used to input the well data features within multiple detection windows into the anomaly detection model for detection and output the abnormal well data.

6. An intelligent management system for oil well data according to claim 1, characterized in that, The platform layer includes: Cloud data processing module, used to analyze the abnormal causes of the abnormal well data and generate an abnormal diagnosis report; Intelligent decision-making module, used to generate an intervention control strategy according to the abnormal diagnosis report.

7. An intelligent management system for oil well data according to claim 6, characterized in that, The specific steps of analyzing the abnormal causes include: Align the abnormal well data with the same-type working condition data in the historical well database in space and time to output an abnormal parameter combination; Based on the association rule mining algorithm, combine the abnormal parameter combination to generate a structured fault mapping table; According to the abnormal parameter combination, combine the equipment fault database to locate the abnormal propagation path and potential interference factors; Generate the abnormal diagnosis report according to the abnormal propagation path and potential interference factors.

8. An intelligent management system for oil well data according to claim 7, wherein, The specific steps of generating the structured fault mapping table include: According to the abnormal parameter combination, set the parameters of the association rule mining algorithm and generate a high-frequency parameter combination; Generate candidate rules according to the high-frequency parameter combination and calculate the screening rules of the candidate rules to obtain a list of highly correlated rules; Construct a triple mapping according to the list of high - correlation rules to generate the structured fault mapping table.

9. An intelligent management system for oil well data according to claim 8, characterized in that, The specific steps for outputting the abnormal parameter combination include: Extract the working condition characteristics of the abnormal well data and the historical well data, and output the working condition characteristic matrix. Based on the random forest algorithm, perform abnormal parameter clustering on the working condition characteristic matrix, and output the abnormal parameter combination.

10. A method for intelligent management of oil well data, for an intelligent management system of oil well data according to any one of claims 1 to 9, characterized in that, Include: Collect the original well data in real - time. Extract the characteristics of the original well data and output the well data characteristics. Based on the LSTM algorithm model, perform abnormal feature detection on the well data characteristics and output the abnormal well data. Analyze the abnormal causes of the abnormal well data and generate an abnormal diagnosis report. Generate an intervention control strategy according to the abnormal diagnosis report.

Images