Multi-source geothermal data intelligent cleaning and normalization method, device and equipment and storage medium
By constructing a hybrid cleaning engine and a unified spatiotemporal reference system, the heterogeneity and physical law violations in multi-source geothermal data processing were solved, achieving data consistency and reliability, and improving the accuracy of geothermal resource exploration and development.
Patent Information
- Application Number
- CN202610391480.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-27
- Publication Date
- 2026-06-23
Smart Images
Figure CN122262485A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of geothermal exploration data processing technology, and in particular to a method, apparatus, equipment and storage medium for intelligent cleaning and normalization of multi-source geothermal data. Background Technology
[0002] With advancements in geothermal exploration and development technologies, data from geological surveys, well logging, geophysical monitoring, distributed fiber optic thermometry, remote sensing thermal anomaly monitoring, and injection-production operations have accumulated, resulting in multi-source geothermal data with diverse origins, varying acquisition cycles, and significant differences in physical meaning. These data exhibit substantial differences across three dimensions: source, acquisition cycle, and physical meaning, highlighting the limitations of existing geothermal data processing methods.
[0003] Fixed threshold methods can only identify anomalies outside the numerical range, making it difficult to detect anomalies that appear normal on the surface but violate geothermal physical relationships. Conventional coordinate transformation and depth benchmark unification processes often rely on manual verification, which is insufficient to handle the batch processing needs when historical data and real-time monitoring data are accessed in parallel. While single anomaly detection models can uncover statistical biases, they cannot directly constrain the restoration results to meet the inherent consistency of formation continuity and geothermal reservoir physical laws. Meanwhile, data from different sources commonly suffer from problems such as missing metadata, inconsistent coordinate systems and elevation benchmarks, mixed units, unclear data lineage, and untraceable versions, leading to the introduction of systematic biases in subsequent resource evaluation, geothermal reservoir modeling, and development dynamic analysis. Therefore, an intelligent cleaning and normalization method for multi-source geothermal data is needed. This method should integrate physical rule verification, anomaly detection, spatiotemporal unification, missing data completion, and version control into a single processing chain while preserving the traceability of the original data, thereby improving data consistency, reusability, and reliability in engineering applications. Summary of the Invention
[0004] This application provides a method, apparatus, equipment, and storage medium for intelligent cleaning and normalization of multi-source geothermal data. It aims to automatically address the problems of geothermal data, such as the lack of physical validity verification, difficulty in achieving unified standardization across time and space dimensions, and the system's lack of self-optimization capabilities as data is updated. This application can achieve efficient cleaning and standardization of multi-source heterogeneous data, improving the accuracy and reliability of data analysis input, and laying a solid data foundation for the precise exploration and efficient development of geothermal resources.
[0005] Firstly, this application provides a method for intelligent cleaning and normalization of multi-source geothermal data, including: Multi-source geothermal data is accessed, and the data undergoes metadata parsing and format standardization to generate an original dataset; A hybrid cleaning engine is constructed, comprising a rule verification subsystem and a deep learning anomaly detection subsystem. The original dataset is input into the rule verification subsystem, which performs primary filtering based on geothermal physical thresholds and multi-physics field coupling logic constraints to generate preliminary cleaned data. The preliminary cleaned data is then input into the deep learning anomaly detection subsystem, which mines high-dimensional nonlinear features in the data through a neural network model, identifies and labels anomalous data, and generates an intermediate dataset labeled with anomalies. A unified four-dimensional spatiotemporal reference system and physical dimension standard library are established. The intermediate dataset is mapped to this unified reference system, and coordinate transformation, stratigraphic depth alignment and physical unit standardization are performed. Missing data is filled in to generate a standardized target dataset. A quality evaluation report is generated based on the target dataset, and an expert interactive feedback interface is provided. The neural network model in the deep learning anomaly detection subsystem is updated and iteratively optimized online based on user feedback obtained through the interface. The target dataset is versioned, classified for credibility, and published via interfaces to provide unified data services to external systems.
[0006] In one possible design, the rule verification subsystem adopts a hierarchical constraint architecture, including single-parameter constraints, gradient continuity constraints, and multi-physics coupling constraints. The single-parameter constraints include theoretical lower and upper limits for parameters based on regional geological background. The gradient continuity constraints include detecting the rate of change of parameters in spatial depth or time series and eliminating noise that does not conform to physical laws. The multi-physics coupling constraints are used to verify the consistency of multi-source data at the same spatiotemporal point, including: When the input data includes reservoir resistivity logging, porosity logging, and water saturation parameters, the relationship between resistivity, porosity, and saturation is verified using the Archie formula. The expression for the Archie formula is: in, Porosity coefficient The bonding index is the number of cementing components. The saturation index. For formation water resistivity, Resistivity Saturation; When the input data includes temperature, pressure, and density parameters of the liquid geothermal fluid, and the temperature and pressure ranges fall within a preset applicable range, the relationship between temperature, pressure, and density is verified using a state equation. The expression for the state equation is: in, For fluid density, For reference density, , For reference temperature and pressure, The coefficient of volumetric expansion. The compression factor is 1. The rule verification subsystem employs an adaptive threshold update strategy based on historical error distribution. of quantiles as the initial threshold After each batch is completed, the distribution of the new samples is considered. Update dynamic threshold The updated formula is: in, For smoothing coefficients, for Quantile operator, For quantiles, and These are the dynamic thresholds for the current batch and the previous batch, respectively.
[0007] In one possible design, the deep learning anomaly detection subsystem includes a denoising autoencoder, a long short-term memory network, or a generative adversarial network. The denoising autoencoder calculates the input vector. With reconstructed vector Reconstruction error between To detect multi-parameter decoupling anomalies; The Long Short-Term Memory network calculates measured values. Compared with the predicted value Prediction error To identify anomalies in time-series data; The generative adversarial network utilizes a generator to realistically reconstruct missing or removed anomalous data segments based on contextual geological features, and a discriminator to ensure that the repaired data conforms to stratigraphic statistical patterns. The generator uses the context of the anomalous point and stratigraphic features as conditions, while the discriminator simultaneously receives real data and repaired data. Through adversarial training, it is ensured that the repair results conform to both data distribution and geological continuity.
[0008] In one possible design, the four-dimensional spatiotemporal normalization includes coordinate and elevation normalization, formation depth alignment, and dimension normalization. The coordinate and elevation normalization uses the Bursa model for coordinate transformation, and the formula is as follows: in, It is a translation vector. It is a rotation matrix consisting of three Euler angles. The scale variation factor. The spatial coordinates after transformation to the target coordinate system. These are the original spatial coordinates in the source coordinate system. The formation depth alignment is based on a cross-correlation function. Calculate depth translation The calculation formula is: in, For reference logging curves at depth The value at that location, For the target logging curve in translation The value at the corresponding depth is then taken. For depth variables, This is the depth translation amount. This is the operator that takes the maximum value of the function corresponding to the independent variable.
[0009] In one possible design, the missing value imputation involves estimating consecutive missing values in a time series using Kalman filtering, with the core recursive formula including: , , , , in, Here is the state transition matrix. For the observation matrix, and For process noise and observation noise, For prior covariance, To observe the noise covariance, T This is the matrix transpose operator. For a moment Posterior state estimation, For a moment Posterior state estimation, For a moment The observation vector, For a moment Kalman gain, For a moment Predicting prior states.
[0010] In one possible design, the missing value completion includes: For the missing key parameters, constrained interpolation is performed using the thermal storage mass balance equation, and its expression is as follows: ( ) in, For water intrusion volume, This represents the cumulative output. To reduce formation pressure, The volume factor is the water volume coefficient. For pore volume, This represents the overall compression ratio.
[0011] In one possible design, the quality assessment report at least includes data completeness. ,accuracy ,consistency With timeliness The indicators are calculated using the following expressions: in, The number of data entries that passed the validation. This represents the total number of entries. The root mean square error, For the range of parameter values, For deviations that violate logical constraints, As the benchmark for logical threshold deviation, Due to data latency, For the maximum allowable delay, To prevent extremely small positive numbers with a denominator of zero.
[0012] Secondly, this application provides a multi-source geothermal data intelligent cleaning and normalization device, the device comprising: The raw data generation module is configured to access multi-source geothermal data, perform metadata parsing and format standardization processing on the data, and generate a raw dataset. The hybrid cleaning module is configured to build a hybrid cleaning engine, which includes a rule verification subsystem and a deep learning anomaly detection subsystem. The original dataset is input into the rule verification subsystem, which performs primary filtering based on geothermal physical thresholds and multiphysics coupling logic constraints to generate preliminary cleaned data. The preliminary cleaned data is then input into the deep learning anomaly detection subsystem, which mines high-dimensional nonlinear features in the data through a neural network model, identifies and marks abnormal data, and generates an intermediate dataset marked with anomalies. The standardization processing module is configured to establish a unified four-dimensional spatiotemporal reference system and a physical dimension standard library, map the intermediate dataset to the unified reference system, perform coordinate transformation, stratum depth alignment and physical unit standardization processing, and complete missing data to generate a standardized target dataset. The update and optimization module is configured to generate a quality evaluation report based on the target dataset and provide an expert interactive feedback interface, and to perform online updates and iterative optimizations on the neural network model in the deep learning anomaly detection subsystem based on user feedback obtained through the interface. The versioning governance module is configured to perform versioning management, credibility classification, and interface-based publishing of the target dataset, providing unified data services to external systems.
[0013] Thirdly, embodiments of this application provide an electronic device, including: at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, causing the at least one processor to perform the intelligent cleaning and normalization method for multi-source geothermal data as described in the first aspect and various possible designs of the first aspect.
[0014] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions. When a processor executes the computer-executable instructions, it implements the intelligent cleaning and normalization method for multi-source geothermal data as described in the first aspect and various possible designs of the first aspect.
[0015] The intelligent cleaning and normalization method, apparatus, equipment, and storage medium for multi-source geothermal data provided in this application have at least the following beneficial effects: 1. This application combines physical rule verification with learning model detection, which can identify anomalies that are difficult to detect by relying solely on numerical thresholds, and reduce the risk of the repair results deviating from the laws of geothermal physics.
[0016] 2. This application unifies the processing of coordinate, elevation, and depth benchmarks and physical units to reduce systematic deviations caused by inconsistencies in reference systems and dimensions during multi-source data fusion.
[0017] 3. This application introduces a quality evaluation, expert review, and version rollback mechanism to make the model update process traceable and reduce the risk that abnormal update results will directly enter downstream business. Attached Figure Description
[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0019] Figure 1 A flowchart illustrating an intelligent cleaning and normalization method for multi-source geothermal data provided in this application embodiment; Figure 2 This is a structural diagram of the intelligent cleaning and normalization device for multi-source geothermal data provided in an embodiment of this application.
[0020] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation
[0021] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.
[0022] The collection, storage, use, processing, transmission, provision, and disclosure of financial data or user data involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.
[0023] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.
[0024] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.
[0025] To address the issues of severe heterogeneity, inconsistent quality, lack of unified spatiotemporal references, and lack of physical constraints in existing geothermal data, this application provides an intelligent cleaning and normalization method for multi-source geothermal data. This method constructs a dual-channel cleaning architecture that integrates rule verification and a learning model: an adaptive rule verification subsystem performs hard constraint verification based on a geothermal multiphysics coupling model, continuity constraints, and regional empirical boundaries; an anomaly detection subsystem identifies anomalies and repairs missing data using at least one model among denoising autoencoders, long short-term memory networks, and generative adversarial networks. For different data types, the system only calls the physical relationship model that matches that type of data; when data exceeds the applicable boundaries of the selected model, the data is marked for review or switched to a preset alternative rule. Based on this, a unified four-dimensional spatiotemporal reference system is established to achieve coordinate transformation, formation depth alignment, and dimensional normalization, and missing values are completed using Kalman filtering and geothermal reservoir material balance constraints. Subsequently, continuous correction and traceability management of the processing results are achieved through quality evaluation, expert review, version rollback, and data fingerprint governance. The method described in this application can be applied to scenarios such as geothermal resource exploration and evaluation, geothermal field development scheme design, and geothermal power plant / heating station operation and maintenance. It is especially suitable for large-scale geothermal digital transformation projects that require the integration of historical data and high-frequency monitoring data.
[0026] like Figure 1 As shown, the intelligent cleaning and normalization method for multi-source geothermal data specifically includes the following steps S10-S50.
[0027] S10: Access multi-source geothermal data, perform metadata parsing and format standardization processing on the data, and generate the original dataset.
[0028] In this embodiment, the multi-source geothermal data accessed in step S10 includes, but is not limited to, geological survey data, geophysical exploration data (gravity, magnetism, seismicity), geochemical data, drilling and logging data, and real-time monitoring data; metadata parsing and format standardization are performed on the multi-source geothermal data; and data lineage records and automated quality labels are introduced in the access process to ensure that the source, acquisition equipment, and acquisition parameters of each original record in the generated raw dataset are traceable.
[0029] In one exemplary embodiment, a unified ubiquitous connectivity adapter is designed to support data acquisition from traditional relational databases or data warehouses, real-time data streams, and file and document systems. Input data formats are limited to structured tables (CSV / Excel), Logging Standard Documents (LAS), Geophysical Standard Format (SEGY), and time-series data streams (JSON / CSV). Required metadata fields include acquisition time, coordinate system, depth reference, unit system, and equipment identifier. Integrated natural language processing technology is used to automatically extract metadata from documents, establishing a data fingerprint database to achieve source deduplication and end-to-end traceability.
[0030] S20: Construct a hybrid cleaning engine, which includes a rule verification subsystem and a deep learning anomaly detection subsystem. Input the original dataset into the rule verification subsystem, and perform primary filtering based on geothermal physical thresholds and multi-physics field coupling logic constraints to generate preliminary cleaned data. Then, input the preliminary cleaned data into the deep learning anomaly detection subsystem, and use a neural network model to mine high-dimensional nonlinear features in the data, identify and label abnormal data, and generate an intermediate dataset labeled with anomalies.
[0031] In this embodiment, step S20 uses the rule verification subsystem to perform primary filtering and correction based on preset geothermal physical thresholds and multi-physics field coupling logic constraints; then, the deep learning anomaly detection subsystem is used to mine high-dimensional nonlinear features and spatiotemporal distribution patterns in the data to identify and mark potential non-physical anomalies, equipment drift or measurement decoupling anomalies.
[0032] In one exemplary embodiment, the rule verification subsystem employs an adaptive threshold update strategy. Specifically, this rule verification subsystem utilizes a geothermal domain knowledge base to construct an expert rule chain and extracts parameter correlations from high-quality historical data to dynamically optimize the verification threshold. For a specific physical relationship model, the corresponding rule is activated only when the input data meets the applicable conditions of the model; for data exceeding the applicable boundaries, a pending review mark or alternative rules are used for further processing. The adaptive threshold update employs a combination of a quantile strategy and Bayesian update: based on the historical error distribution... of quantiles as the initial threshold After each batch is completed, the distribution of the new samples is considered. Updated to: in, For smoothing coefficients, for Quantile operator, For quantiles, and These are the dynamic thresholds for the current batch and the previous batch, respectively.
[0033] In one exemplary embodiment, the rule verification subsystem adopts a hierarchical constraint architecture, including single-parameter constraints, gradient continuity constraints, and multi-physics coupling constraints.
[0034] Among these, multiphysics coupling constraints focus on verifying the consistency of multi-source data at the same spatiotemporal point. For data including reservoir resistivity logging, porosity logging, and water saturation parameters, the Archie formula is used to verify resistivity (…). ), porosity ( ) and saturation ( The nonlinear relationship of ) in, Porosity coefficient The bonding index is the number of cementing components. The saturation index. , , It can be calibrated through core experiments or regional empirical values.
[0035] For data containing temperature, pressure, and density parameters of liquid geothermal fluid, where the temperature and pressure ranges fall within a preset applicable range, the temperature-pressure-density relationship is verified using the equation of state: in, For fluid density, For reference density, , For reference temperature and pressure, The coefficient of volumetric expansion. The compression factor is 1.
[0036] In one exemplary embodiment, the deep learning anomaly detection subsystem includes a denoising autoencoder, a long short-term memory network, or a generative adversarial network. Specifically, this deep learning anomaly detection subsystem employs a cutting-edge generative model to enhance its cleaning capabilities, and it can achieve at least the following four functions: 1) GAN Data Restoration: Generative Adversarial Networks (GANs) are introduced for data augmentation and restoration. A generator realistically reconstructs missing or removed anomalous data segments based on contextual geological features, while a discriminator ensures the restored data conforms to stratigraphic statistical patterns. The generator uses the context of the anomalous points and stratigraphic features as conditions, while the discriminator receives both real and restored data. Adversarial training ensures the restored results conform to both data distribution and geological continuity. The generated restored results are then input into a rule-validation subsystem for secondary verification; data that fails verification is moved to a low-confidence queue.
[0037] 2) Multidimensional Correlation Anomaly Detection: A denoising autoencoder (DAE) is used to learn the joint distribution characteristics of multidimensional logging curves (GR, RT, RHOB, NPHI, DT, etc.). The input vector is calculated. With reconstructed vector Reconstruction error between: , in For the input feature vector, This is the result of model reconstruction.
[0038] when When the dynamic threshold is exceeded, it is determined to be an anomaly in multi-parameter decoupling. For example, the dynamic threshold can be set to... ,in This is the historical average error. The standard deviation is denoted as .
[0039] 3) Temporal anomaly detection: Utilizing LSTM or Transformer models to learn dynamic patterns in time-series data (such as production well monitoring data) and calculate prediction errors: , in These are measured values. For predicted values, This is the length of the continuous time step window. If continuous... Each time step If the value exceeds the confidence interval, it is identified as equipment drift or sudden failure.
[0040] The input and output of a deep learning model are defined as follows: DAE input is multi-dimensional logging curves (GR, RT, RHOB, NPHI, DT), and output is the reconstructed curve and reconstruction error; LSTM input is time series data (…). The output is the predicted value. The input to a GAN is the missing segment context and a noise vector, and the output is the repaired data segment.
[0041] The collaborative mechanism between the rule engine and the model is as follows: First, the data is screened for hard constraints by physical rules and quality labels are generated. Then, the model is trained and fine-tuned with the labeled data. The outliers and repair values output by the model must be verified again by the rule engine. Those that fail are marked as low confidence and enter the expert review queue. The review results will be used to update the threshold and model weights.
[0042] 4) Confidence assessment: Calculate the confidence level of anomaly detection by combining the sampling density of the input data, noise estimation, and instrument condition.
[0043] S30: Establish a unified four-dimensional spatiotemporal reference system and physical dimension standard library, map the intermediate dataset to this unified reference system, perform coordinate transformation, stratum depth alignment and physical unit standardization, and complete the missing data to generate a standardized target dataset.
[0044] In an exemplary embodiment, step S30 may generate a standardized target dataset in the following manner.
[0045] Construct a four-dimensional spacetime reference system; establish a unified three-dimensional spatial coordinate + time dimension reference framework.
[0046] The implementation method for coordinate and elevation normalization is as follows: a built-in Geographic Information System (GIS) conversion engine supports seven-parameter conversion between mainstream coordinate systems. For wellbore data, the depth benchmark (core elevation KB or surface elevation GL) is automatically identified and uniformly corrected to sea level depth (TVDSS). The seven-parameter Bursa-Wolf model conversion formula is as follows: in, It is a translation vector. It is a rotation matrix consisting of three Euler angles. The scale variation factor. The spatial coordinates after transformation to the target coordinate system. These are the original spatial coordinates in the source coordinate system.
[0047] The implementation method for formation depth alignment is as follows: For depth errors from different measurements or using different equipment, correlation matching is performed using characteristic curves such as spontaneous potential (SP) or natural gamma (GR). Based on the cross-correlation function: Calculate the depth shift corresponding to the maximum correlation coefficient to achieve fine alignment of formation depths: in, For reference logging curves at depth The value at that location, For the target logging curve in translation The value at the corresponding depth is then taken. For depth variables, The operator that takes the maximum value of a function is the operator corresponding to the independent variable. Depth translation amount The range of values is .
[0048] Dimensional normalization: A semantic web of geothermal physical parameters is constructed based on ontology. Semantic reasoning algorithms are used to automatically parse the physical meaning of the data column headers (e.g., recognizing "T", "Temp", and "DegC" as temperature), and implicit units are inferred based on context (e.g., determining the pressure unit as MPa or psi based on the numerical range). The data is then automatically normalized to the SI unit system using a standard conversion factor. Unit inference verification employs interval consistency rules. When the inferred unit causes a sample to fall into a physically inaccessible interval, a retry of alternative units is triggered, and a candidate set is recorded. Finally, the unit with the highest rule consistency score is used as the final unit determination.
[0049] Missing value handling falls into two categories:
[0050] Case 1: For discontinuous missing data, use Kriging interpolation (spatial correlation) or spline interpolation (temporal correlation) for completion; for continuous missing data in the time series, use Kalman filtering for estimation. , , , , in, Here is the state transition matrix. For the observation matrix, and For process noise and observation noise, For prior covariance, To observe the noise covariance, T This is the matrix transpose operator. For a moment Posterior state estimation, For a moment Posterior state estimation, For a moment The observation vector, For a moment Kalman gain, For a moment The prior state prediction is performed. For sequences with consecutive missing lengths exceeding a preset upper limit or whose noise covariance cannot be stably estimated, Kalman filtering is not used directly for completion. Instead, missing markers are output and the sequences are then reviewed by experts.
[0051] Scenario 2: For large-scale missing parameters, they are marked and not included in subsequent calculations; in the case of missing key parameters, the cumulative production is calculated based on the thermal reservoir material balance equation. ), formation pressure drop ( ) and water intrusion ( Perform constrained interpolation: ( ) in For water intrusion volume, This represents the cumulative output. To reduce formation pressure, The volume factor is the water volume coefficient. For pore volume, This represents the overall compressibility coefficient. The above constraint interpolation applies to thermal storage units that satisfy the mass balance assumption; when the thermal storage type or production regime does not satisfy this assumption, only missing markers are output, and complete values are not forcibly generated.
[0052] S40: Generate a quality evaluation report based on the target dataset and provide an expert interactive feedback interface. Based on user feedback obtained through the interface, perform online updates and iterative optimizations of the neural network model in the deep learning anomaly detection subsystem.
[0053] In one exemplary embodiment, the quality assessment report includes at least data completeness. ,accuracy ,consistency With timeliness The metric, truncated and limited to between 0 and 1, is calculated using the following expressions: in The number of data entries that passed the validation. This represents the total number of entries. The root mean square error, For the range of parameter values, For deviations that violate logical constraints, As the benchmark for logical threshold deviation, Due to data latency, This represents the maximum allowable delay.
[0054] Data completeness Used to measure the percentage of valid data; accuracy Evaluation based on the root mean square error of the cleaning value and the baseline value; consistency. Used to measure the degree of violation of physical and logical constraints (such as the total amount of inverse gradient deviation as temperature decreases with depth); timeliness Used to assess data processing latency.
[0055] In one exemplary embodiment, a quality report is generated based on the aforementioned indicators, and an interactive visual interface is provided for expert sampling. "False positives" or "false negatives" confirmed by experts are fed back to the deep learning model. The model weights are fine-tuned through online learning or active learning mechanisms to continuously improve cleaning accuracy. A / B evaluation and automatic alarm strategies are introduced into key processes (e.g., processing the same batch of data with both old and new cleaning models, comparing the output quality indicators, and automatically evaluating the model update effect). Batches showing abnormal fluctuations in quality indicators are automatically blocked from downstream processing. The model update and validation constraints are: after each online update, the anomaly detection accuracy is evaluated on a fixed validation set. With recall ,like or If the data falls below a preset threshold, it will be rolled back to the previous version, and the batch of data will be marked as pending review.
[0056] S50: Implement version management, credibility grading, and interface-based publishing of target datasets to provide unified data services to external systems.
[0057] This embodiment, through step S50, performs version management, credibility classification, and interface-based publishing of the cleaned and normalized data, providing unified data services to geological modeling, numerical simulation, and real-time monitoring systems. Quality tracking and auditing are achieved through interface logs and statistical indicators. Specifically, the cleaned and normalized data is version-managed (e.g., v1.0-Raw, v2.0-Cleaned, v2.1-ExpertRevised) to ensure data status traceability. Encrypted storage is achieved using data fingerprinting technology and national cryptographic algorithms; data subscriptions are provided to geological modeling, geothermal reservoir simulation, and real-time monitoring systems through RESTful APIs, OGC standard services, or message queues; a de-identified data sandbox is provided for researchers, and access behavior is audited and tracked.
[0058] The feasibility and progressiveness of the method in this application will be explained in detail below with a specific example.
[0059] For a geothermal field in northern China, hundreds of logging curves (temperature, pressure, flow rate, conventional logging, etc.) from different periods and different logging companies have been accumulated. The method described in this application is used for processing, focusing on a governance strategy that ensures the original data remains immutable and the cleaned data is versioned. The specific implementation steps for this example are as follows.
[0060] Step 1, Data Access and Parsing: The system uses a template matching + manual assistance mode to scan the server directory, identifying over 700 files in LAS, TXT, and Excel formats. The automatic parsing success rate for standard LAS files reached 95%; for non-standard Excel spreadsheets, column mapping relationships were established through an interactive interface, and batch parsing was performed after manual confirmation to ensure accurate metadata extraction. Statistical results from the data access phase are shown in Table 1.
[0061] Table 1 Statistical Results of Data Access and Parsing Phase
[0062] Step 2, rule validation and tagging.
[0063] Basic cleaning: Automatically identifies zero and negative temperature data segments caused by instrument malfunctions, does not delete them directly, but labels them as invalid values and filters them by default in the analysis view.
[0064] Physical correlation verification: Using density-neutron porosity cross plots, density logging distortion sections severely affected by wellbore enlargement are identified; theoretical deviations are calculated by combining resistivity-porosity-saturation relationship (Archie formula), and suggested correction curves are generated as derivative versions for experts to adopt with one click.
[0065] AI-assisted detection and repair, including time-series drift detection, hidden anomaly discovery, and GAN-assisted repair.
[0066] Temporal drift detection: During LSTM model training, a non-physical linear drift was detected in a batch of pressure monitoring data over a specific time period. Upon verification, this was determined to be zero-point drift of the pressure gauge. The system generated a corrected data sequence after degaussing, and simultaneously recorded the correction algorithm and parameters in the metadata, while retaining the original observations for auditing purposes.
[0067] Hidden Anomaly Discovery: Using the DAE model, statistical anomalies were detected in the resistivity curves of several wells in non-permeable sections. After verification by experts in conjunction with the casing log, it was confirmed that the signal interference was caused by casing corrosion. The system marked this data segment as "non-formation response" to prevent it from misleading subsequent inversion.
[0068] GAN-based collaborative restoration: For missing segments of well logging curves, GAN is used to generate complete data. To ensure physical plausibility, the system inputs the generated restoration data into a physical rule engine for secondary verification (such as the matching of porosity and sonic transit time), retaining only the verified samples to ensure that the restoration results conform to the formation depositional patterns.
[0069] Table 2 shows the comparison results of the well temperature logging curves before and after cleaning after step 2.
[0070] Table 2 Comparison of Well Temperature Logging Curves Before and After Cleaning and Anomaly Detection Results
[0071] Step 3, normalization and alignment.
[0072] Coordinate unification: The previously used "Beijing 54" coordinate system was automatically converted to "CGCS2000", and the seven conversion parameters were recorded.
[0073] Depth benchmark correction: The depth benchmark is automatically identified by scanning the well head file. Some old well data with the rotary table surface (RT) as the zero point are converted into a unified elevation depth (TVDSS) using core height (KB) data.
[0074] Stratigraphic alignment: Using the stable mudstone caprock in the whole area as a marker layer, cross-correlation matching was performed on the GR curves of different measurements to correct the average depth systematic error of 1.5 meters.
[0075] The key operational results of the normalization and completion stages are shown in Table 3.
[0076] Table 3 Key operational results of the normalization and completion phases
[0077] Through the above steps, a three-tiered data system of raw, cleaned, and standardized data was finally established. Compared with purely manual processing, the data preparation cycle was shortened by 80%; compared with traditional cleaning that only uses fixed thresholds, the anomaly detection accuracy was improved to 0.92 and the recall to 0.88; the average relative error of missing value completion decreased from 12.6% to 5.4%; and the historical fitting error (RMSE) of the thermal reservoir model built based on the standardized dataset decreased from 5.2℃ to 1.8℃, and all data modifications are traceable, meeting the compliance requirements of reserve review.
[0078] This application also provides an intelligent cleaning and normalization device for multi-source geothermal data, such as... Figure 2 As shown, the intelligent cleaning and normalization device for multi-source geothermal data includes: The raw data generation module 201 is configured to access multi-source geothermal data, perform metadata parsing and format standardization processing on the data, and generate a raw dataset. The hybrid cleaning module 202 is configured to build a hybrid cleaning engine, which includes a rule verification subsystem and a deep learning anomaly detection subsystem. The original dataset is input into the rule verification subsystem, which performs primary filtering based on geothermal physical thresholds and multi-physics field coupling logic constraints to generate preliminary cleaned data. The preliminary cleaned data is then input into the deep learning anomaly detection subsystem, which mines high-dimensional nonlinear features in the data through a neural network model, identifies and marks abnormal data, and generates an intermediate dataset marked with anomalies. The standardization processing module 203 is configured to establish a unified four-dimensional spatiotemporal reference system and a physical dimension standard library, map the intermediate dataset to the unified reference system, perform coordinate transformation, stratum depth alignment and physical unit standardization processing, and complete missing data to generate a standardized target dataset. The update and optimization module 204 is configured to generate a quality evaluation report based on the target dataset and provide an expert interactive feedback interface, and to perform online updates and iterative optimizations on the neural network model in the deep learning anomaly detection subsystem based on user feedback obtained through the interface. The versioning governance module 205 is configured to perform versioning management, credibility classification, and interface-based publishing of the target dataset, providing unified data services to external systems.
[0079] This application provides an electronic device. The electronic device may include a processor and a memory, wherein the processor and the memory can communicate; exemplarily, the processor and the memory communicate via a communication bus.
[0080] The processor executes computer execution instructions stored in memory, causing the processor to perform the scheme in the above embodiments. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.
[0081] The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. Transceivers are used to enable communication between database access devices and other computers (e.g., clients, read-write libraries, and read-only libraries). Memory may include random access memory (RAM) and may also include non-volatile memory.
[0082] The electronic device provided in this application embodiment can be the terminal device described in the above embodiments.
[0083] This application also provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed on a computer, the computer performs the technical solution of the multi-source geothermal data intelligent cleaning and normalization method described in the above embodiments.
[0084] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, it can implement the technical solution of the intelligent cleaning and normalization method for multi-source geothermal data in the above embodiments.
[0085] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.
[0086] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.
[0087] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.
[0088] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.
[0089] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.
[0090] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.
[0091] Buses can be Industry Standard Architecture (ISA) buses, Peripheral Component Interconnect (PCI) buses, or Extended Industry Standard Architecture (EISA) buses, etc. Buses can be categorized into address buses, data buses, control buses, etc.
[0092] The aforementioned storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.
[0093] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. The processor and storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components in an electronic control unit or main control device.
[0094] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.
[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.
Claims
1. A multi-source geothermal data intelligent cleaning and normalization method, characterized in that, The method includes: Multi-source geothermal data is accessed, and the data undergoes metadata parsing and format standardization to generate an original dataset; A hybrid cleaning engine is constructed, comprising a rule verification subsystem and a deep learning anomaly detection subsystem. The original dataset is input into the rule verification subsystem, which performs primary filtering based on geothermal physical thresholds and multi-physics field coupling logic constraints to generate preliminary cleaned data. The preliminary cleaned data is then input into the deep learning anomaly detection subsystem, which mines high-dimensional nonlinear features in the data through a neural network model, identifies and labels anomalous data, and generates an intermediate dataset labeled with anomalies. A unified four-dimensional spatiotemporal reference system and physical dimension standard library are established. The intermediate dataset is mapped to this unified reference system, and coordinate transformation, stratigraphic depth alignment and physical unit standardization are performed. Missing data is filled in to generate a standardized target dataset. A quality evaluation report is generated based on the target dataset, and an expert interactive feedback interface is provided. The neural network model in the deep learning anomaly detection subsystem is updated and iteratively optimized online based on user feedback obtained through the interface. The target dataset is versioned, classified for credibility, and published via interfaces to provide unified data services to external systems.
2. The multi-source geothermal data intelligent cleaning and normalization method according to claim 1, characterized in that, The rule verification subsystem adopts a hierarchical constraint architecture, including single-parameter constraints, gradient continuity constraints, and multi-physics coupling constraints. The single-parameter constraints include theoretical lower and lower limits for parameters based on the regional geological background. The gradient continuity constraints include detecting the rate of change of parameters in spatial depth or time series and eliminating noise that does not conform to physical laws. The multiphysics coupling constraints are used to verify the consistency of multi-source data at the same spatiotemporal point, and call the corresponding physical relationship model according to the data type, including: When the input data includes reservoir resistivity logging, porosity logging, and water saturation parameters, the relationship between resistivity, porosity, and saturation is verified using the Archie formula. The expression for the Archie formula is: wherein, is a porosity factor, is a cementation exponent, is a saturation exponent, is a formation water resistivity, is a resistivity, is a saturation; When the input data includes temperature, pressure, and density parameters of liquid geothermal fluid, the relationship between temperature, pressure, and density is verified using a state equation. The expression for the state equation is: wherein, is the fluid density, is the reference density, , is the reference temperature and pressure, is the coefficient of volume expansion, is the compressibility factor; The rule verification subsystem employs an adaptive threshold update strategy based on historical error distribution. of quantiles as the initial threshold After each batch is completed, the distribution of the new samples is considered. Update dynamic threshold The updated formula is: wherein, is a smoothing coefficient, is a quantile operator, is a quantile point, and are the dynamic thresholds for the current batch and the previous batch, respectively.
3. The multi-source geothermal data intelligent cleaning and normalization method of claim 1, wherein, The deep learning anomaly detection subsystem includes at least one of a denoising autoencoder, a long short-term memory network, or a generative adversarial network, and calls the corresponding model according to the data type. The denoising autoencoder calculates the input vector. With reconstructed vector Reconstruction error between To detect multi-parameter decoupling anomalies; The long short-term memory network identifies the time series data anomaly by calculating a prediction error of a measured value and a predicted value The generative adversarial network utilizes a generator to realistically reconstruct missing or removed anomalous data segments based on contextual geological features. A discriminator ensures that the repaired data conforms to stratigraphic statistical patterns, and the repaired results are re-input into the rule verification subsystem for secondary verification. The generator uses the context of the anomalous points and stratigraphic features as conditions, while the discriminator simultaneously receives real data and repaired data. Through adversarial training, it is ensured that the repaired results conform to both data distribution and geological continuity.
4. The multi-source geothermal data intelligent cleaning and normalization method of claim 1, wherein, The four-dimensional spatiotemporal normalization includes coordinate and elevation normalization, stratigraphic depth alignment, and dimensional normalization; The coordinate and elevation normalization uses the Bursa model for coordinate transformation, and the formula is as follows: wherein, is a translation vector, is a rotation matrix composed of three Euler angles, is a scale change factor, is a converted spatial coordinate under a target coordinate system, is an original spatial coordinate under a source coordinate system; The formation depth alignment is based on a cross-correlation function Computing the depth translation The formula is: wherein, is the value of the reference log curve at depth , is the value of the target log curve at the corresponding depth after translation , is the depth variable, is the depth translation, is the operator that takes the argument corresponding to the maximum value of the function.
5. The intelligent cleaning and normalization method for multi-source geothermal data according to claim 1, characterized in that, The missing value imputation includes estimation using Kalman filtering for consecutive missing values in the time series, and its core recursive formula includes: ， ， ， ， in, Here is the state transition matrix. For the observation matrix, and For process noise and observation noise, For prior covariance, To observe the noise covariance, T This is the matrix transpose operator. For a moment Posterior state estimation, For a moment Posterior state estimation, For a moment The observation vector, For a moment Kalman gain, For a moment Predicting prior states.
6. The intelligent cleaning and normalization method for multi-source geothermal data according to claim 1, characterized in that, The missing value completion includes: When key parameters are missing, and the target object is a hydrothermal storage facility or a production unit that can meet the material balance constraints, constraint interpolation is performed using the storage facility's material balance equation. The expression is as follows: in, For water intrusion volume, This represents the cumulative output. To reduce formation pressure, The volume factor is the water volume coefficient. For pore volume, This represents the overall compression ratio.
7. The intelligent cleaning and normalization method for multi-source geothermal data according to any one of claims 1 to 6, characterized in that, The quality assessment report shall at least include data completeness. ,accuracy ,consistency With timeliness The indicators are calculated using the following expressions: in, The number of data entries that passed the validation. This represents the total number of entries. The root mean square error, For the range of parameter values, For deviations that violate logical constraints, As the benchmark for logical threshold deviation, Due to data latency, For the maximum allowable delay, To prevent extremely small positive numbers with a denominator of zero.
8. A multi-source geothermal data intelligent cleaning and normalization device, characterized in that, The device includes: The raw data generation module is configured to access multi-source geothermal data, perform metadata parsing and format standardization processing on the data, and generate a raw dataset. The hybrid cleaning module is configured to build a hybrid cleaning engine, which includes a rule verification subsystem and a deep learning anomaly detection subsystem. The original dataset is input into the rule verification subsystem, which performs primary filtering based on geothermal physical thresholds and multiphysics coupling logic constraints to generate preliminary cleaned data. The preliminary cleaned data is then input into the deep learning anomaly detection subsystem, which mines high-dimensional nonlinear features in the data through a neural network model, identifies and marks abnormal data, and generates an intermediate dataset marked with anomalies. The standardization processing module is configured to establish a unified four-dimensional spatiotemporal reference system and a physical dimension standard library, map the intermediate dataset to the unified reference system, perform coordinate transformation, stratum depth alignment and physical unit standardization processing, and complete missing data to generate a standardized target dataset. The update and optimization module is configured to generate a quality evaluation report based on the target dataset and provide an expert interactive feedback interface, and to perform online updates and iterative optimizations on the neural network model in the deep learning anomaly detection subsystem based on user feedback obtained through the interface. The versioning governance module is configured to perform versioning management, credibility classification, and interface-based publishing of the target dataset, providing unified data services to external systems.
9. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes the computer execution instructions stored in the memory to implement the intelligent cleaning and normalization method for multi-source geothermal data as described in any one of claims 1-7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the intelligent cleaning and normalization method for multi-source geothermal data as described in any one of claims 1-7.