A reservoir fluid identification method, device, equipment and medium

Abstract

The application provides a reservoir fluid identification method, device, equipment and medium. It relates to the technical field of tight gas reservoir fluid identification. The method comprises the following steps: obtaining well logging data of a target work area; constructing a training sample set based on the well logging data, screening sensitive well logging parameters, and establishing a parameter combination space covering different missing situations; training a missing parameter self-adaptive compensation model for each missing situation; constructing a parameter self-adaptive fluid identification model based on complete samples and compensated samples; obtaining measured parameters of a well to be identified, calling a corresponding compensation model to complete the missing parameters, and then inputting the parameters into the identification model to obtain a fluid type identification result. The application can adapt to changes and missing of well logging parameters, avoid repeated modeling, and improve identification accuracy.

Description

Technical Field

[0001] This invention relates to the field of tight gas reservoir fluid identification technology, and in particular to a reservoir fluid identification method, apparatus, equipment and medium. Background Technology

[0002] Reservoir fluid identification is a key technical step in oil and gas exploration and development, and its accuracy directly affects subsequent reserve assessment and development plan design. Traditional logging fluid identification methods typically establish a discrimination model based on a fixed combination of parameters, requiring complete and fixed types of input parameters.

[0003] Currently, there is a fluid identification method based on pore-throat structure and resistivity. This method proposes to establish a fluid identification chart by coupling pore-throat structure and resistivity to identify the fluid properties of complex reservoirs. The core working principle is as follows: first, obtain the pore-throat structure parameters and resistivity parameters of the reservoir; then, perform coupled analysis on these parameters to construct a two-dimensional or three-dimensional fluid identification chart; finally, determine the fluid type based on the projection position of the sample point to be identified on the chart. This method relies on the joint analysis of the specific combination of pore-throat structure and resistivity, and distinguishes different fluid types such as gas layers and water layers through the division of the chart area. However, this method requires both conventional logging data and two-dimensional NMR logging data to perform fluid identification; it cannot be completed independently using only conventional logging data. If either the pore-throat structure parameter or the resistivity parameter is missing from the logging data of a new well, the pre-established fluid identification chart becomes unusable, requiring the collection of complete sample data and remodeling, resulting in low processing efficiency. Summary of the Invention

[0004] Based on this, it is necessary to provide a reservoir fluid identification method, device, equipment, and medium to address the above-mentioned technical problems, so as to solve the defects of insufficient model generalization ability, low efficiency of repeated modeling, and poor adaptability of incomplete data in the existing technology, and to achieve adaptive adaptation to scenarios of changes and missing logging parameters, so as to complete high-precision fluid identification without repeated modeling.

[0005] The following technical solution is adopted in this specification: This specification provides a method for reservoir fluid identification, including: Obtain well logging data for the target work area; A training sample set is constructed based on well logging data. Sensitivity analysis is performed on the well logging parameters in the training sample set to screen out well logging parameters that are sensitive to fluid type and establish a parameter combination space. The parameter combination space covers different missing scenarios of sensitive well logging parameters. For each missing parameter scenario in the parameter combination space, the existing logging parameters in the complete sample are used as input and the missing logging parameters are used as output to train an adaptive compensation model for missing parameters corresponding to each missing parameter scenario. Based on the complete parameter samples in the training sample set and the compensated samples obtained after compensation by the missing parameter adaptive compensation model, a parameter adaptive fluid recognition model is constructed. Obtain the measured logging parameters of the well to be identified, and determine whether there are any missing parameters: if there are no missing parameters, directly input the measured logging parameters into the parameter adaptive fluid identification model to obtain the fluid type identification result; if there are missing parameters, call the missing parameter adaptive compensation model corresponding to the current missing parameter situation to compensate for the missing logging parameters, form a complete input parameter set, and then input it into the parameter adaptive fluid identification model to obtain the fluid type identification result.

[0006] Furthermore, the construction of the training sample set based on well logging data specifically includes: Well logging data is depth aligned, cleaned, and standardized to remove duplicate sampling points, correct outliers, and unify the dimensions of the data. Based on multi-source data, well logging response characteristics of different fluid types are determined, fluid labels are established, and training sample sets are constructed.

[0007] Furthermore, the sensitivity analysis of the logging parameters in the training sample set specifically includes: The feature importance evaluation method is used to obtain the contribution ranking of each logging parameter to the fluid identification result, and the logging parameters with the ranking higher than the preset threshold are selected as sensitive logging parameters. Alternatively, cross-plot analysis can be used to determine the clustering characteristics of different fluid types under different combinations of logging parameters by creating cross-plot diagrams, thereby identifying sensitive logging parameters.

[0008] Furthermore, the missing parameter adaptive compensation model is constructed using the LightGBM algorithm, which introduces unilateral gradient sampling and mutually exclusive feature binding techniques and is trained using a leaf growth strategy with depth constraints.

[0009] Furthermore, the parameter adaptive fluid identification model is constructed using the random forest algorithm; the random forest algorithm forms an ensemble classifier by constructing multiple independent decision trees and uses a voting mechanism to output the final fluid type prediction result.

[0010] Furthermore, during the construction of the parameter adaptive fluid identification model, the feature importance evaluation mechanism of the random forest algorithm is used to rank the input logging parameters by importance, so as to verify the screening results of the sensitive logging parameters and optimize the model hyperparameters.

[0011] Furthermore, after obtaining the fluid type identification result, the process also includes: Import the fluid type identification results into the well logging software for visualization. For each sampling point, a probability distribution map of various fluid types is generated. When the probability value of a sampling point in a specific fluid type dimension exceeds a preset threshold, the corresponding sampling point is determined to be the corresponding reservoir fluid type.

[0012] This specification provides a reservoir fluid identification device, comprising: The data acquisition module is used to acquire well logging data for the target work area; The parameter filtering module is used to construct a training sample set based on well logging data, perform sensitivity analysis on the well logging parameters in the training sample set, filter out well logging parameters that are sensitive to fluid type, and establish a parameter combination space; the parameter combination space covers different missing scenarios of sensitive well logging parameters; The compensation model construction module is used to train an adaptive compensation model for each missing parameter case in the parameter combination space, taking the existing logging parameters in the complete sample as input and the missing logging parameters as output. The identification model building module is used to build a parameter adaptive fluid identification model based on the complete parameter samples in the training sample set and the compensated samples obtained after compensation by the missing parameter adaptive compensation model. The identification module is used to obtain the measured logging parameters of the well to be identified and determine whether there are any missing parameters. If there are no missing parameters, the measured logging parameters are directly input into the parameter adaptive fluid identification model to obtain the fluid type identification result. If there are missing parameters, the missing parameter adaptive compensation model corresponding to the current missing parameter situation is called to compensate for the missing logging parameters. After forming a complete set of input parameters, the model is input to obtain the fluid type identification result.

[0013] This specification provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described reservoir fluid identification method.

[0014] This specification provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described reservoir fluid identification method.

[0015] The above-mentioned technical solutions adopted in this specification can achieve the following beneficial effects: This invention establishes a holistic technical concept of collaborative work between a classification model and a missing parameter compensation model, forming a dynamic processing path of "parameter adaptive completion followed by classification and recognition." Specifically, this invention constructs a parameter combination space in the early stages and pre-trains corresponding adaptive compensation models for each missing parameter scenario. This allows the system to automatically detect the completeness of input parameters before processing the well to be identified, and selectively invoke the compensation model based on the detection results. This "judgment first, compensation later, recognition later" mechanism establishes a clear sequential connection between the compensation model and the classification model. The output of the compensation model is directly received by the classification model, without manual intervention or retraining of the main model. By decoupling the missing parameter compensation process from the fluid identification process and establishing a standardized interface, this invention transforms the identification task, which was originally unable to run due to incomplete parameters, into a new process that can automatically complete completion and perform identification, thus solving the technical problem of model failure caused by a single missing parameter in traditional methods. The synergistic effect of both allows the model to adaptively adapt to changes and missing scenarios in logging parameters, significantly reducing the cost of repeated modeling due to incomplete parameters and achieving automated processing from incomplete input to accurate output. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0017] Figure 1 A flowchart illustrating a reservoir fluid identification method provided in this specification; Figure 2 This is a schematic diagram illustrating a single-parameter cross-plot method for identifying reservoir fluid type (AC-CNL) provided in this specification; Figure 3 This specification provides a schematic diagram of a single-parameter cross-plot method for identifying reservoir fluid types. Figure 4 This specification provides a schematic diagram of a single-parameter cross-plot method for identifying reservoir fluid types (DEN-CNL). Figure 5 This is a schematic diagram of a model structure provided in this specification; Figure 6 This is a schematic diagram illustrating the distribution of the importance of a parameter feature, as provided in this specification. Figure 7 This is a schematic diagram illustrating the evaluation of stratigraphic classification effectiveness provided in this specification; Figure 8 A schematic diagram of a variable parameter reservoir fluid identification result based on machine learning, provided in this specification; Figure 9This is a schematic diagram of the structure of a reservoir fluid identification device provided in this specification; Figure 10 This is a schematic diagram of a computer device provided for this specification. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this specification clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative effort are within the scope of protection of this application.

[0019] The technical solution provided by this invention can be applied to the exploration and development of tight gas reservoir fluid identification. In the exploration and development of tight sandstone gas reservoirs, the identification of gas-water layers has always been a key focus and challenge in oil and gas exploration and development. Traditional tight reservoir fluid identification technologies suffer from problems such as insufficient model generalization ability, inefficient repeated modeling, and poor adaptability due to incomplete data. Traditional logging fluid identification methods are usually based on fixed parameter combinations. When logging parameters are missing, sample data needs to be collected again and modeled, resulting in low efficiency and poor interpretability. The technical problem this invention aims to solve is: the problems of insufficient model generalization ability, inefficient repeated modeling, and poor adaptability due to incomplete data in traditional tight reservoir fluid identification technologies in oil and gas exploration. Specifically, traditional logging fluid identification methods are usually based on fixed parameter combinations. When logging parameters are missing, sample data needs to be collected again and modeled, resulting in low efficiency and poor interpretability. Existing algorithms, such as multivariate discriminant methods, cannot function properly when sample parameters are limited, and cannot effectively handle situations where parameters are missing. To address the aforementioned issues, this application provides a reservoir fluid identification method.

[0020] The reservoir fluid identification method of the present invention is described below with reference to the accompanying drawings.

[0021] Figure 1 This is a flowchart illustrating a reservoir fluid identification method provided in this specification, such as... Figure 1 As shown, the method includes: S101. Obtain logging data for the target work area.

[0022] For example, when acquiring logging data for the target work area, logging data can be received from on-site logging equipment, or historical logging data can be retrieved from local storage. The specific settings can be configured according to actual needs. Specifically, in this embodiment, logging, well logging, and test conclusion data of the Xujiahe Formation in the Zhongjiang area were collected. The logging parameters included natural gamma (GR), acoustic transit time (AC), compensated neutron (CNL), density (DEN), and deep lateral resistivity (RD).

[0023] S102. Construct a training sample set based on well logging data, perform sensitivity analysis on the well logging parameters in the training sample set, screen out well logging parameters that are sensitive to fluid type, and establish a parameter combination space; wherein, the parameter combination space covers different missing scenarios of sensitive well logging parameters.

[0024] For example, when constructing the training sample set, the original well logging data needs to undergo preprocessing operations such as depth alignment, cleaning, and standardization to remove duplicate sampling points, correct outliers, and unify the data dimensions. Then, based on multi-source data, the well logging response characteristics of different fluid types are determined, and fluid labels are established, thereby constructing a training sample set that can be used for supervised learning. The purpose of this is to provide a standardized, labeled, high-quality data foundation for subsequent sensitivity analysis and model training. Then, combining multi-source data such as gas testing and well logging, the fluid types are divided into five categories: gas layer, poor gas layer, gas-water co-layer, water layer, and dry layer, and fluid labels are established for each depth point.

[0025] Next, sensitivity analysis was performed on each logging parameter in the training sample set. This analysis could employ feature importance evaluation methods or cross-plot analysis to screen logging parameters that contribute significantly to fluid type identification or exhibit clear clustering characteristics as sensitive logging parameters. Based on the screened sensitive logging parameters, considering the different parameter missing scenarios that may occur in new wells during actual production, various possible combinations of missing parameters were enumerated or clustered to establish a parameter combination space covering multiple missing scenarios. The purpose of this was to extend the fixed input parameter pattern to a variable parameter pattern, providing a basis for subsequently constructing specific compensation models for different missing scenarios. The sensitivity analysis used the cross-plot method, and AC-CNL, AC-RD, and DEN-CNL cross-plots were created. Figure 2 This is a schematic diagram illustrating a single-parameter cross-plot method for identifying reservoir fluid type (AC-CNL) provided in this specification; Figure 3 This specification provides a schematic diagram of a single-parameter cross-plot method for identifying reservoir fluid types. Figure 4 This specification provides a schematic diagram of a single-parameter cross-plot method for identifying reservoir fluid types (DEN-CNL); as shown. Figures 2-4As shown, the five types of fluids in the Xujiahe Formation of the Zhongjiang area exhibit distinct clustering characteristics in the cross-plot: gas layers and poorly gas layers have relatively high acoustic transit time (AC), high resistivity (RD), low neutron density (CNL), and low density (DEN); gas-water co-layers show transitional characteristics; water layers have relatively low acoustic transit time and resistivity, high neutron density, and high density; dry layers have high density and resistivity, and low acoustic and neutron density. Based on this, sensitive logging parameters were selected: GR, AC, CNL, DEN, and RD. Based on these five parameters, possible missing scenarios (such as missing CNL, missing DEN, and simultaneous missing CNL and DEN) were enumerated to establish a parameter combination space.

[0026] S103. For each missing parameter scenario in the parameter combination space, use the existing logging parameters in the complete sample as input and the missing logging parameters as output to train an adaptive compensation model for the missing parameters corresponding to each missing parameter scenario.

[0027] For example, for each parameter missing scenario defined in the parameter combination space (e.g., missing density curve DEN, or missing neutron curve CNL), the corresponding "existing parameters" are extracted from the complete samples as input features, and the "missing parameters" are used as prediction targets to independently train a compensation model. The purpose of this is to pre-equip a dedicated parameter incompleter for each possible data incompleteness situation. In this embodiment, the missing parameter adaptive compensation model is constructed using the LightGBM algorithm. Taking well CH100 as an example, the measured parameters of this well are only GR, AC, and RD, lacking CNL and DEN. For the case of "missing CNL and DEN," a LightGBM model is trained using complete samples (including GR, AC, RD, CNL, and DEN) within the work area, with GR, AC, and RD as inputs and CNL and DEN as outputs. LightGBM introduces one-sided gradient sampling (GOSS) and mutually exclusive feature binding (EFB), employing a leaf growth strategy with depth constraints. After training, the model can accurately predict CNL and DEN based on the input GR, AC, and RD, thus completing the missing curves.

[0028] S104. Based on the complete parameter samples in the training sample set and the compensated samples obtained after compensation by the missing parameter adaptive compensation model, construct a parameter adaptive fluid recognition model.

[0029] For example, the original complete parameter samples and the compensated samples generated after processing by the compensation model are merged into a unified training dataset. This dataset is then used to train a classification model that can directly output the probability of fluid type. This model can employ ensemble learning algorithms such as random forest. The purpose of this is to enable the fluid identification model to work not only with complete data but also with compensated "repaired data," thus providing unified model support for the automatic identification of subsequent measured data (whether complete or incomplete) of wells to be identified. In this embodiment, the parameter-adaptive fluid identification model uses the random forest algorithm. Random forests form strong splitters by constructing multiple independent decision trees, and the final output is determined by a voting mechanism.

[0030] S105. Obtain the measured logging parameters of the well to be identified, and determine whether there are any missing parameters: If there are no missing parameters, directly input the measured logging parameters into the parameter adaptive fluid identification model to obtain the fluid type identification result; if there are missing parameters, call the missing parameter adaptive compensation model corresponding to the current missing parameter situation to compensate for the missing logging parameters, form a complete input parameter set, and then input it into the parameter adaptive fluid identification model to obtain the fluid type identification result.

[0031] For example, when the measured logging curve of a new well is obtained, the system first automatically detects the types of parameters it contains and compares them with the sensitive parameter set to determine if any are missing. If all parameters are complete, they are directly fed into the classification model trained in S104 for prediction; if missing parameters exist, the system calls the corresponding compensation model from the model library trained in S103 according to the specific missing combination to complete the curve prediction, and then sends the complete parameter set into the classification model. The purpose of this is to achieve fully automatic processing from "incomplete input" to "accurate output," avoiding model failure or repeated modeling due to missing parameters. Specifically, taking well CH100 as an implementation case, the measured parameters of this well are only GR, AC, and RD. The system detects the missing CNL and DEN, automatically calls the LightGBM compensation model trained in S103 to complete CNL and DEN, forming a complete 5-parameter set (GR, AC, CNL, DEN, RD), and then inputs it into the random forest classification model. The final output is the probability distribution of fluid type at each depth point. When the probability of a certain type exceeds a preset threshold (such as 0.6), it is determined to be that type.

[0032] Based on the above Figure 1 In the embodiment shown, when constructing a training sample set based on well logging data and performing sensitivity analysis on well logging parameters in S102 above, this application provides a possible implementation method.

[0033] S201. Perform depth alignment, cleaning, and standardization on well logging data. Specifically, firstly, match and align curves from different well logging series (such as natural gamma ray GR, sonic transit time AC, compensated neutron CNL, etc.) according to depth points, and remove duplicate sampling points and outliers caused by well diameter changes or measurement errors; then, use the Z-score standardization method to unify the dimensions.

[0034] S202. Determine fluid labels based on multi-source data. Specifically, combining gas testing results, well logging data, and production dynamics data, fluid types are classified into categories such as gas layers, poor gas layers, gas-water co-containment layers, water layers, and dry layers. A corresponding category label is assigned to each depth sample point, thereby constructing the training sample set required for supervised learning. By executing S201 and S202, the raw well logging data can be converted into a standardized training sample set with clearly defined fluid labels, thus providing a high-quality data foundation for subsequent sensitivity analysis and model training.

[0035] S203. Perform sensitivity analysis on the logging parameters in the training sample set. Specifically, feature importance evaluation methods (such as Gini importance or permutation importance built into random forests) can be used to calculate the contribution of each logging parameter to the fluid identification result, and the parameters are selected as sensitive logging parameters by ranking them from high to low according to their contribution and selecting those with a ranking higher than a preset threshold (such as the top 5 or a cumulative contribution exceeding 90%). Alternatively, cross plot analysis can be used to create cross plots of different combinations of logging parameters (such as AC-CNL cross plots, AC-RD cross plots, and DEN-CNL cross plots), and manually or automatically determine the clustering characteristics of different fluid types on the cross plots. Parameters with obvious clustering and strong separability are identified as sensitive logging parameters. By executing S203, the most discriminative sensitive parameters for fluid types can be screened from numerous logging curves, thus providing a screening basis for subsequently establishing a compact and efficient parameter combination space.

[0036] Based on the above Figure 1 In the embodiment shown, in S103 above, for each missing parameter case in the parameter combination space, using the existing logging parameters in the complete sample as input and the missing logging parameters as output, when training the missing parameter adaptive compensation model corresponding to each missing parameter case, this application embodiment provides a possible implementation method: The missing parameter adaptive compensation model is constructed using the LightGBM algorithm. LightGBM is a high-efficiency ensemble learning algorithm based on a gradient boosting decision tree framework. It introduces one-sided gradient sampling and mutually exclusive feature binding techniques, significantly improving training efficiency while maintaining accuracy. Unlike traditional hierarchical growth strategies, LightGBM employs a depth-constrained leaf growth strategy, selecting the leaf with the largest splitting gain from all current leaf nodes for splitting at each step. The formula for calculating the splitting gain is:

[0037] ; in, and These are the first-order and second-order gradient statistics, respectively. and This is the sample set of the left and right leaf nodes after the split. and This is the regularization parameter. This strategy allows the model to achieve lower learning error with the same number of iterations, accurately capturing the nonlinear relationships between logging parameters, such as the mineral response relationship between density and neutron porosity.

[0038] For the well logging parameter completion problem, the model objective can be expressed as: using a complete dataset for training, learning the intrinsic correlation between various well logging parameters, and completing the model within a given subset of well logging parameters. Predict missing parameters under conditions such as GR, AC, and RD. (For example, DEN). LightGBM iteratively learns the negative gradient directions of multiple weak learners (decision trees) to gradually approximate the true mapping relationship. ,in Let T be the input feature vector, and T be the number of iterations. For the first A decision tree. By using LightGBM to construct a compensation model, missing parameters can be predicted efficiently and accurately from existing logging parameters, thus providing a complete set of input parameters for subsequent fluid identification models.

[0039] Based on the above Figure 1 In the embodiment shown, in S104 above, when constructing the parameter adaptive fluid recognition model based on the complete parameter samples in the training sample set and the compensated samples obtained after compensation by the missing parameter adaptive compensation model, this application embodiment provides a possible implementation method: The parameter-adaptive fluid recognition model is constructed using the random forest algorithm. Random forest is a classifier based on a combination of Bagging ensemble learning strategy and the idea of ​​random subspaces. Its core mathematical principle lies in constructing multiple independent decision trees. To form strong splitters, in which They are independent and identically distributed random vectors. The input is a vector of logging parameters. The final fluid type prediction result is determined by a voting mechanism.

[0040] ; in, For the final classification result, Label the fluid type (gas layer, differential gas layer, gas-water co-layer, water layer, dry layer). This is an indicator function. Figure 5 This is a schematic diagram of a model structure provided in this specification. The model structure is as follows: Figure 5 As shown. Using the built-in feature importance evaluation mechanism of random forest, the importance ranking of each parameter is obtained as follows: Figure 6 As shown, Figure 6 This diagram illustrates the distribution of parameter feature importance as provided in this specification. During model construction, the feature importance evaluation mechanism built into random forests (such as the average value based on Gini impurity reduction) is used to rank the importance of the logging parameters input to the model. This verifies the effectiveness of the sensitive parameters selected through the aforementioned sensitivity analysis and optimizes the model's hyperparameters (such as the number of decision trees and maximum depth). By employing random forests to construct an adaptive classifier, the advantages of ensemble learning can be leveraged to improve the accuracy and robustness of fluid identification. Furthermore, feature importance analysis further validates the rationality of the sensitive parameter selection, thus providing a high-precision classification model for subsequent fluid type prediction.

[0041] Based on the above Figure 1 In the illustrated embodiment, after obtaining the fluid type identification result in S105 above, this application embodiment also provides a possible implementation: Fluid type identification results are imported into specialized logging software (such as CIGLog, TechLog, GeoLog, etc.) for visualization, assisting interpreters in intuitively verifying the accuracy of the prediction results. Simultaneously, probability distribution maps for various fluid types are generated for each sampling point. Specifically, for each sample point output by the random forest model, the proportion of each decision tree voting it as a different fluid type (e.g., gas layer, poor gas layer, gas-water co-existing layer, water layer, dry layer) is recorded, and this proportion is used as the probability distribution. When the probability value of a predicted sample point in a specific fluid type dimension exceeds a preset threshold (e.g., 0.6), the sample point is determined to belong to the corresponding reservoir fluid type. Through visualization and probability distribution output, the model's black-box prediction results can be transformed into interpretable and verifiable geological maps, thus providing intuitive and reliable data support for on-site decision-making. Figure 7 This is a schematic diagram illustrating the evaluation of stratigraphic classification effectiveness provided in this specification; Figure 8This is a schematic diagram of a variable parameter reservoir fluid identification result based on machine learning, provided in this specification.

[0042] The reservoir fluid identification device provided by the present invention is described below. The reservoir fluid identification device described below and the reservoir fluid identification method described above can be referred to in correspondence.

[0043] Figure 9 For an example, please refer to the structural schematic diagram of a reservoir fluid identification device provided by the present invention. Figure 9 As shown, the reservoir fluid identification device may include: The data acquisition module is used to acquire well logging data for the target work area.

[0044] The parameter filtering module is used to construct a training sample set based on well logging data, perform sensitivity analysis on the well logging parameters in the training sample set, filter out well logging parameters that are sensitive to fluid type, and establish a parameter combination space; the parameter combination space covers different missing scenarios of sensitive well logging parameters.

[0045] The compensation model construction module is used to train an adaptive compensation model for each missing parameter scenario in the parameter combination space, using existing logging parameters from the complete sample as input and missing logging parameters as output.

[0046] The identification model building module is used to construct a parameter adaptive fluid identification model based on the complete parameter samples in the training sample set and the compensated samples obtained after compensation by the missing parameter adaptive compensation model.

[0047] The identification module is used to obtain the measured logging parameters of the well to be identified and determine whether there are any missing parameters. If there are no missing parameters, the measured logging parameters are directly input into the parameter adaptive fluid identification model to obtain the fluid type identification result. If there are missing parameters, the missing parameter adaptive compensation model corresponding to the current missing parameter situation is called to compensate for the missing logging parameters. After forming a complete set of input parameters, the model is input to obtain the fluid type identification result.

[0048] Specific limitations regarding the reservoir fluid identification device can be found in the limitations on reservoir fluid identification described above, and will not be repeated here. Each module in the aforementioned reservoir fluid identification device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0049] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 The provided method for reservoir fluid identification.

[0050] This instruction manual also provides Figure 10 The schematic diagram of the computer device shown is as follows: Figure 10 At the hardware level, the computer device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 1 The provided method for reservoir fluid identification.

[0051] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0052] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A method for identifying reservoir fluids, characterized in that, include: Obtain well logging data for the target work area; A training sample set is constructed based on well logging data. Sensitivity analysis is performed on the well logging parameters in the training sample set to screen out well logging parameters that are sensitive to fluid type and establish a parameter combination space. The parameter combination space covers different missing scenarios of sensitive well logging parameters. For each missing parameter case in the parameter combination space, the existing logging parameters in the complete sample are used as input and the missing logging parameters are used as output to train an adaptive compensation model for missing parameters corresponding to each missing parameter case. Based on the complete parameter samples in the training sample set and the compensated samples obtained after compensation by the missing parameter adaptive compensation model, a parameter adaptive fluid recognition model is constructed. Obtain the measured logging parameters of the well to be identified, and determine whether there are any missing parameters: if there are no missing parameters, directly input the measured logging parameters into the parameter adaptive fluid identification model to obtain the fluid type identification result; if there are missing parameters, call the missing parameter adaptive compensation model corresponding to the current missing parameter situation to compensate for the missing logging parameters, form a complete input parameter set, and then input it into the parameter adaptive fluid identification model to obtain the fluid type identification result.

2. The reservoir fluid identification method as described in claim 1, characterized in that, The construction of the training sample set based on well logging data specifically includes: Well logging data is depth aligned, cleaned, and standardized to remove duplicate sampling points, correct outliers, and unify the dimensions of the data. Based on multi-source data, well logging response characteristics of different fluid types are determined, fluid labels are established, and training sample sets are constructed.

3. The reservoir fluid identification method as described in claim 1, characterized in that, The sensitivity analysis of logging parameters in the training sample set specifically includes: The feature importance evaluation method is used to obtain the contribution ranking of each logging parameter to the fluid identification result, and the logging parameters with the ranking higher than the preset threshold are selected as sensitive logging parameters. Alternatively, cross-plot analysis can be used to determine the clustering characteristics of different fluid types under different combinations of logging parameters by creating cross-plot diagrams, thereby identifying sensitive logging parameters.

4. The reservoir fluid identification method as described in claim 1, characterized in that, The missing parameter adaptive compensation model is constructed using the LightGBM algorithm, which introduces unilateral gradient sampling and mutually exclusive feature binding techniques and is trained using a leaf growth strategy with depth constraints.

5. The reservoir fluid identification method as described in claim 1, characterized in that, The parameter-adaptive fluid identification model is constructed using the random forest algorithm. The random forest algorithm forms an ensemble classifier by constructing multiple independent decision trees and uses a voting mechanism to output the final fluid type prediction result.

6. The reservoir fluid identification method as described in claim 5, characterized in that, During the construction of the parameter adaptive fluid identification model, the feature importance evaluation mechanism of the random forest algorithm is used to rank the input logging parameters by importance, so as to verify the screening results of the sensitive logging parameters and optimize the model hyperparameters.

7. The reservoir fluid identification method as described in claim 1, characterized in that, After obtaining the fluid type identification result, the process also includes: Import the fluid type identification results into the well logging software for visualization. For each sampling point, a probability distribution map of various fluid types is generated. When the probability value of a sampling point in a specific fluid type dimension exceeds a preset threshold, the corresponding sampling point is determined to be the corresponding reservoir fluid type.

8. A reservoir fluid identification device, characterized in that, include: The data acquisition module is used to acquire well logging data for the target work area; The parameter filtering module is used to construct a training sample set based on well logging data, perform sensitivity analysis on the well logging parameters in the training sample set, filter out well logging parameters that are sensitive to fluid type, and establish a parameter combination space; the parameter combination space covers different missing scenarios of sensitive well logging parameters; The compensation model construction module is used to train an adaptive compensation model for each missing parameter case in the parameter combination space, taking the existing logging parameters in the complete sample as input and the missing logging parameters as output. The identification model building module is used to build a parameter adaptive fluid identification model based on the complete parameter samples in the training sample set and the compensated samples obtained after compensation by the missing parameter adaptive compensation model. The identification module is used to obtain the measured logging parameters of the well to be identified and determine whether there are any missing parameters. If there are no missing parameters, the measured logging parameters are directly input into the parameter adaptive fluid identification model to obtain the fluid type identification result. If there are missing parameters, the missing parameter adaptive compensation model corresponding to the current missing parameter situation is called to compensate for the missing logging parameters. After forming a complete set of input parameters, the model is input to obtain the fluid type identification result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the reservoir fluid identification method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the reservoir fluid identification method as described in any one of claims 1 to 7.

Images