Formation resistivity prediction model training method and formation resistivity prediction method
By using the timing neural network model combined with the knowledge embedding method in oil and gas exploration, the problem of poor correlation between the electromagnetic logging curve and the formation resistivity after the sleeve is solved, and the accuracy and interpretability of the formation resistivity prediction are improved.
Patent Information
- Application Number
- CN202411509782.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-28
- Publication Date
- 2025-02-25
AI Technical Summary
In the prior art, the correlation between the post-shut electromagnetic well logging curve and the formation resistivity is poor, resulting in high fitting difficulty and low accuracy of the formation resistivity prediction model, which affects the accuracy of oil and gas exploration.
By obtaining the conventional logging curves and post-shut electromagnetic logging curves of multiple wells, extracting multi-dimensional feature data, and performing normalization, feature dimensionality reduction and data cleaning, building a training sample set, training based on the timing neural network model, introducing knowledge embedding methods, and using physical constraint formulas as part of the loss function, improving the generalization ability and interpretability of the model.
The fitting effect and accuracy of the formation resistivity prediction model is improved, and the selectivity of the output results is improved, which is more suitable for the needs of oil and gas exploration.
Smart Images
Figure CN119513567A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of oil and gas exploration, and in particular to a formation resistivity prediction model training method, a formation resistivity prediction method, a formation resistivity prediction model training device, an electronic device and a computer-readable storage medium. Background Art
[0002] In the field of oil and gas exploration, accurate prediction of formation resistivity is crucial for evaluating the potential of oil and gas reservoirs, guiding drilling decisions, and optimizing development plans. As a commonly used geophysical logging technology, post-casing transient electromagnetic logging estimates formation resistivity by analyzing the response of the formation to electromagnetic waves. In the prior art, due to the poor correlation between the post-casing electromagnetic logging curve and the formation resistivity and the high difficulty of fitting, the accuracy of the formation resistivity prediction model finally fitted is low, which makes the selected degree of the predicted formation resistivity result low, which is not conducive to oil and gas exploration. Summary of the invention
[0003] The purpose of the present invention is to provide a formation resistivity prediction model training method, a formation resistivity prediction method, a formation resistivity prediction model training device, an electronic device and a computer-readable storage medium, which are applied to the field of oil and gas exploration. The method improves the generalization ability and interpretability of the model, so that the fitting effect of the model is better and the output formation resistivity prediction result is more accurate.
[0004] In order to solve the above technical problems, the present invention provides a formation resistivity prediction model training method, comprising:
[0005] Obtaining conventional logging curves and post-casing electromagnetic logging curves of multiple oil wells, and extracting multi-dimensional feature data of the conventional logging curves and the post-casing electromagnetic characteristic curves at various measuring depths;
[0006] Normalizing, reducing the dimension of the multi-dimensional feature data, and performing data cleaning to obtain target features;
[0007] Taking the true value of the formation resistivity of the oil well at each measurement depth as the first label value of the target feature;
[0008] Determine a first physically constrained formation resistivity of the oil well at each of the measurement depths based on a physical constraint formula of a conventional well logging curve, and use the first physically constrained formation resistivity as a second label value of the target feature;
[0009] Determine the second physically constrained formation resistivity of the oil well at each of the measurement depths based on the physical constraint formula of the post-casing electromagnetic logging curve, and use the second physically constrained formation resistivity as the third label value of the target feature;
[0010] A training sample set is constructed based on the target feature, the first label value, the second label value and the third label value, and a time series neural network model is trained based on the training sample set to obtain a trained formation resistivity prediction model.
[0011] Optionally, the conventional well logging curve physical constraint formula is:
[0012] ;
[0013] In the formula, is the first physically constrained formation resistivity at the measurement depth, is the well logging curve sonic logging curve value at the measuring depth, is the natural gamma curve value at the measurement depth, is the spontaneous potential curve value at the measurement depth, is the induction logging curve value at the measuring depth.
[0014] Optionally, the physical constraint formula of the post-casing electromagnetic logging curve is:
[0015] ;
[0016] In the formula, is the second physically constrained formation resistivity at the measurement depth, is a constraint formula, which is related to the casing medium model. - is the electromotive force response of n transient electromagnetic curves at the measuring depth.
[0017] Optionally, the loss function of the formation resistivity prediction model is:
[0018] ;
[0019] In the formula, is the loss value, is the formation resistivity prediction value output by the formation resistivity prediction model, y is the first label value, is the second label value, is the third label value, is the first weight coefficient, is the second weight coefficient, is the third weight coefficient.
[0020] Optionally, constructing a training sample set based on the target feature, the first label value, the second label value, and the third label value includes:
[0021] Constructing a training sub-sample set for each oil well; the training sub-sample set includes the target feature, the first label value, the second label value and the third label value of the corresponding oil well at each of the measurement depths;
[0022] Sort the samples in each of the training sub-sample sets in the order of the measurement depth from shallow to deep or from deep to shallow;
[0023] Constructing the training sample set based on each of the training sub-sample sets;
[0024] Correspondingly, the training of the time series neural network model based on the training sample set to obtain a trained formation resistivity prediction model includes:
[0025] Constructing an input window; the size of the input window is a preset number of the target features;
[0026] Traversing the target features of each training subsample set based on the input window, obtaining a formation resistivity test value output by the time series neural network model;
[0027] Determine a loss value based on the formation resistivity test value and the first label value, the second label value, and the third label value in the training sample set;
[0028] The model parameters of the time series neural network model are updated based on the loss value to obtain the trained formation resistivity prediction model.
[0029] Optionally, the temporal neural network model is an LSTM model; the LSTM model includes a noise reduction module and an attention module.
[0030] In order to solve the above technical problems, the present invention provides a formation resistivity prediction method, comprising:
[0031] Obtain the conventional logging curve and the post-casing electromagnetic logging curve of the oil well to be tested;
[0032] Extracting multi-dimensional characteristic data of the conventional logging curve and the post-casing electromagnetic characteristic curve at each measuring depth;
[0033] Normalizing, reducing the dimension of the multi-dimensional feature data and performing data cleaning to obtain the features of the target to be measured;
[0034] Inputting the target characteristics to be measured into a formation resistivity prediction model to obtain a formation resistivity prediction value output by the model;
[0035] Wherein, the formation resistivity prediction model is a model trained according to any formation resistivity prediction model training method.
[0036] In order to solve the above technical problems, the present invention provides a formation resistivity prediction model training device, comprising:
[0037] The first module is used to obtain conventional logging curves and post-casing electromagnetic logging curves of multiple oil wells, and extract multi-dimensional feature data of the conventional logging curves and the post-casing electromagnetic characteristic curves at various measurement depths;
[0038] The second module is used to normalize, reduce the dimension of the multi-dimensional feature data and clean the data to obtain the target feature;
[0039] The third module is used to use the true value of the formation resistivity of the oil well at each measurement depth as the first label value of the target feature;
[0040] A fourth module is used to determine a first physically constrained formation resistivity of the oil well at each of the measurement depths based on a physical constraint formula of a conventional logging curve, and use the first physically constrained formation resistivity as a second label value of the target feature;
[0041] A fifth module is used to determine the second physically constrained formation resistivity of the oil well at each of the measurement depths based on the physical constraint formula of the post-casing electromagnetic logging curve, and use the second physically constrained formation resistivity as the third label value of the target feature;
[0042] The sixth module is used to construct a training sample set based on the target feature, the first label value, the second label value and the third label value, and train the time series neural network model based on the training sample set to obtain a trained formation resistivity prediction model.
[0043] In order to solve the above technical problems, the present invention provides an electronic device, comprising:
[0044] Memory, for storing computer programs;
[0045] A processor is used to implement the above-mentioned formation resistivity prediction model training method and the above-mentioned formation resistivity prediction method when executing the computer program.
[0046] In order to solve the above technical problems, the present invention provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, the above-mentioned formation resistivity prediction model training method and the above-mentioned formation resistivity prediction method are implemented.
[0047] It can be seen that the method of the present invention obtains conventional logging curves and post-casing electromagnetic logging curves of multiple oil wells, extracts multidimensional feature data of conventional logging curves and post-casing electromagnetic characteristic curves at each measuring depth; normalizes the multidimensional feature data, performs feature dimension reduction and data cleaning to obtain target features; uses the true value of the formation resistivity of the oil well at each measuring depth as the first label value of the target feature; determines the first physically constrained formation resistivity of the oil well at each measuring depth based on the physical constraint formula of the conventional logging curve, and uses the first physically constrained formation resistivity as the second label value of the target feature; determines the second physically constrained formation resistivity of the oil well at each measuring depth based on the physical constraint formula of the post-casing electromagnetic logging curve, and uses the second physically constrained formation resistivity as the third label value of the target feature; constructs a training sample set based on the target feature, the first label value, the second label value and the third label value, and trains the time series neural network model based on the training sample set to obtain a trained formation resistivity prediction model.
[0048] The present invention introduces a knowledge embedding method and uses the traditional empirical formula as a part of the loss function of the neural network model in single-well resistivity logging prediction, thereby realizing the model's anti-noise ability and prior knowledge utilization, and ultimately improving the model's multi-parameter coupling ability and interpretability. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0050] Figure 1 A flow chart of a formation resistivity prediction model training method provided by an embodiment of the present invention;
[0051] Figure 2 A comparative example of prediction results of a formation resistivity prediction model provided by an embodiment of the present invention;
[0052] Figure 3 A structural block diagram of a formation resistivity prediction model training device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0053] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0054] In oil exploration and development, accurate prediction of formation resistivity is crucial for resource evaluation and well site design. Post-casing transient electromagnetic logging technology is a commonly used geophysical logging method that indirectly obtains formation resistivity information by measuring the response of the formation to electromagnetic pulses. Among them, the logging interpretation of resistivity can provide a basis for the development of oil and gas fields, such as the proportion of underground fluids and the distribution of rock formations, which helps to optimize well site design, development strategies and reservoir management, and improve the efficiency of resource development. However, due to the complexity and uncertainty of underground geological conditions, as well as the noise and limitations of logging data itself, traditional logging curve prediction methods often find it difficult to achieve sufficient accuracy and reliability. Therefore, in order to improve the accuracy and reliability of post-casing transient electromagnetic logging interpretation, it is of great practical significance to study a method for predicting formation resistivity from post-casing transient electromagnetic logging curves based on knowledge embedding.
[0055] However, existing logging interpretation technologies have some limitations when dealing with post-casing electromagnetic logging curves, especially in the resistivity interpretation of complex geology. Among them, traditional methods include: physical model method, statistical analysis method, artificial intelligence-based method, empirical formula method, frequency domain inversion method, which have the following limitations:
[0056] Model complexity: Conventional methods such as physical model method and frequency domain inversion method are computationally complex and sensitive to the selection of initial model and inversion algorithm. In addition, such methods usually require assumptions under ideal conditions, such as assuming that the formation is uniform or can be approximated as a multi-layer medium, so as to simplify the calculation. Therefore, such methods have high model complexity and high computational cost, and are not suitable for interpreting formation resistivity from post-casing electromagnetic logging curves with large data volumes and complex formations.
[0057] Noise resistance: Conventional artificial intelligence methods such as neural network methods and machine learning methods are usually pure data-driven models. Pure data-driven models often cannot handle nonlinear and uncertainty problems in logging data well, are sensitive to noise, lack effective integration of geological knowledge, and have limited generalization capabilities when faced with new geological conditions.
[0058] Model interpretability: Conventional physical simulation methods and artificial intelligence methods are insufficient in interpretability. Because the internal mechanism algorithm has a black box part or a certain degree of randomness, the traditional artificial intelligence method is insufficient in interpretability.
[0059] Accuracy problem: Conventional methods formed by traditional empirical formulas and statistical analysis methods are difficult to apply to new reservoirs or other wells to interpret formation resistivity problems using post-casing electromagnetic logging curves. Since such methods are purely model-driven, they are insufficient in accuracy.
[0060] Multi-parameter coupling: Well logging curves are affected by many factors, such as formation rock type, porosity, saturation, mud content, etc. There are complex coupling relationships between these parameters, which are difficult to separate and quantify.
[0061] Knowledge embedding is a scientific machine learning technique that transfers domain knowledge as prior experience to the learning of neural networks. Among them, domain knowledge plays an important role in the powerless constraints of neural networks. It integrates the physical laws and scientific theories of the scientific field into the design and training process of the model to improve the accuracy, reliability and interpretability of the model. In data-driven models, control equations can be incorporated into the training process of neural network models in the scientific field. These control equations describe the basic laws of physical phenomena such as fluid mechanics and heat conduction. By introducing these equations, the model can follow these physical laws during the training process, thereby improving the prediction accuracy of the model on complex problems. For example, in underground multiphase seepage problems, physical constraints such as mass conservation and energy conservation of the fluid can be introduced into the training of deep learning models, so that the model satisfies physical laws during the learning process. Physical constraints can also be integrated with the knowledge and experience of domain experts. In the process of model design, domain experts can provide important physical constraints and prior knowledge to guide the learning and prediction process of the model. This can make the model more in line with practical problems and enhance the interpretability and applicability of the model in specific fields. Equation-based physical constraints can be implemented in two ways: one is to formalize the physical equations as constraints of the neural network, and the other is to use the physical equations as regularization terms to constrain the training of the neural network. Physical constraints play a bridging role in the fusion of data and knowledge, integrating the physical laws and knowledge in the scientific field into the model, so that the model can better follow the physical laws and scientific theories on the basis of data-driven, thereby improving the performance and interpretability of the model. This effective fusion of data and knowledge makes the model more accurate, reliable and practical when dealing with complex problems and specific fields. The present invention uses the traditional empirical formulas in the oil field as physical equation constraints, and embeds knowledge in the prediction or interpretation of formation resistivity of the post-casing electromagnetic logging curve, thereby realizing the model's anti-noise ability and prior knowledge utilization, and ultimately improving the model's multi-parameter coupling ability and interpretability.
[0062] In view of these shortcomings of traditional methods, this study introduced a knowledge embedding method to overcome the limitations of traditional post-casing electromagnetic logging curve fitting of formation resistivity. By introducing the knowledge embedding method, the neural network achieved the characteristics of retaining interpretable formula fitting and completing data learning in the process of establishing post-casing electromagnetic logging curve mapping formation resistivity, with stronger generalization ability and interpretability. This method provides geological engineers with a more accurate and comprehensive post-casing electromagnetic logging curve fitting tool.
[0063] The following combination Figure 1 , Figure 1 A flow chart of a formation resistivity prediction model training method provided by an embodiment of the present invention, the method may include:
[0064] S101: Obtain conventional logging curves and casing electromagnetic logging curves of multiple oil wells, and extract multi-dimensional feature data of the conventional logging curves and casing electromagnetic characteristic curves at various measurement depths.
[0065] This embodiment can obtain conventional logging curves and post-casing logging curves of multiple oil wells. The multiple oil wells in this embodiment can come from the same oil field or from different oil fields, which is not limited here.
[0066] Conventional logging curves are important tools for identifying and evaluating oil and gas reservoirs in petroleum engineering. In practical applications, these conventional logging curves can solve basic problems such as reservoir division, porosity calculation, and oil and gas layer identification.
[0067] The post-casing logging curve refers to the formation information obtained through specific logging technology after the casing of an oil or gas well is installed. This logging method can evaluate the integrity of the cased well, detect casing damage, cement bonding quality, formation pressure and other important parameters.
[0068] After obtaining the conventional logging curves and post-casing logging curves of each oil well, this embodiment can extract the multi-dimensional feature data of the conventional logging curves and post-casing electromagnetic feature curves at each measurement depth. Specifically, the conventional logging curves and post-casing electromagnetic logging curves can be converted into multi-dimensional feature data by time series conversion.
[0069] In this embodiment, the measurement depths of the oil wells are not necessarily the same. For example, the measurement depth of oil well 1 may be 50m-100m, and the measurement depth of oil well 2 may be 80m-130m. This embodiment does not limit the measurement depths of the oil wells.
[0070] S102: Normalize, reduce the dimension of the multi-dimensional feature data, and clean the data to obtain the target feature.
[0071] Each conventional logging curve and post-casing logging curve can have multi-dimensional data at a certain measurement depth. For example, a conventional logging curve can have 33-dimensional data at a certain measurement depth, and a post-casing logging curve can have 400-dimensional data at a certain measurement depth, then there can be 433-dimensional data at the measurement depth.
[0072] This embodiment can perform data preprocessing on the extracted multidimensional feature data, such as normalization, feature dimensionality reduction and data cleaning to obtain target features, so that data regularization eliminates dimensional differences between features and reduces scale differences between features, and feature dimensionality reduction reduces the dimension of the features, making the model easier to fit.
[0073] S103: Taking the true value of the formation resistivity of the oil well at each measurement depth as the first label value of the target feature.
[0074] In this embodiment, the true value of the formation resistivity of the oil well at each measurement depth may be used as the first label value of the target feature corresponding to the measurement depth.
[0075] S104: Determine a first physically constrained formation resistivity of the oil well at each measurement depth based on a physical constraint formula of a conventional logging curve, and use the first physically constrained formation resistivity as a second label value of the target feature.
[0076] This embodiment can introduce the physical constraint formula of conventional logging curves into the loss function based on the knowledge embedding method, that is, determine the first physical constraint formation resistivity of the oil well at each measurement depth based on the physical constraint formula of conventional logging curves, and use the first physical constraint formation resistivity as the second label value of the target feature.
[0077] The physical constraint formula of the conventional well logging curve in this embodiment can be:
[0078] ;
[0079] In the formula, is the first physical constraint on the formation resistivity at the measured depth, is the acoustic logging curve value of the logging curve at the measured depth, is the natural gamma curve value at the measured depth, is the value of the natural potential curve at the measurement depth, is the induction logging curve value at the measuring depth.
[0080] That is, by inputting the acoustic logging curve value, natural gamma curve value, natural potential curve value and induction logging curve value of the oil well at a certain measuring depth into the above-mentioned conventional logging curve physical constraint formula, the first physical constraint formation resistivity at the corresponding measuring depth can be obtained.
[0081] In this embodiment, the calculated first physically constrained formation resistivity may be used as the second label value of the target feature corresponding to the measurement depth.
[0082] S105: Determine a second physically constrained formation resistivity of the oil well at each measurement depth based on a physical constraint formula of the post-casing electromagnetic logging curve, and use the second physically constrained formation resistivity as a third label value of the target feature.
[0083] This embodiment can introduce the physical constraint formula of the post-casing electromagnetic logging curve into the loss function based on the knowledge embedding method, that is, determine the second physical constraint formation resistivity of the oil well at each measurement depth based on the physical constraint formula of the post-casing electromagnetic logging curve, and use the second physical constraint formation resistivity as the third label value of the target feature.
[0084] The physical constraint formula of the post-casing electromagnetic logging curve in this embodiment can be:
[0085] ;
[0086] In the formula, The second physical constraint is the formation resistivity at the measured depth, is a constraint formula, which is related to the casing medium model. - It is the electromotive force response of n transient electromagnetic curves at the measured depth.
[0087] That is, by inputting the electromotive force response values of n transient electromagnetic curves of the oil well at a certain measuring depth into the above-mentioned physical constraint formula of the post-casing electromagnetic logging curve, the second physical constraint formation resistivity at the corresponding measuring depth can be obtained.
[0088] In this embodiment, the calculated second physical constraint formation resistivity may be used as the third label value of the target feature corresponding to the measurement depth.
[0089] S106: constructing a training sample set based on the target feature, the first label value, the second label value and the third label value, and training a time series neural network model based on the training sample set to obtain a trained formation resistivity prediction model.
[0090] After obtaining the target characteristics and the first label value, the second label value and the third label value of the oil well at each measurement depth, the present embodiment can construct a training sample set based on the target characteristics, the first label value, the second label value and the third label value, and train the time series neural network model based on the training sample set to obtain a trained formation resistivity prediction model.
[0091] Since there are three label values in this embodiment, the loss function of the formation resistivity prediction model is:
[0092] ;
[0093] In the formula, is the loss value, is the formation resistivity prediction value output by the formation resistivity prediction model, y is the first label value, is the second label value, is the third label value, is the first weight coefficient, is the second weight coefficient, is the third weight coefficient. is the first loss value obtained by the first label value, is the second loss value obtained by the second label value, is the third loss value obtained by the third label value.
[0094] Furthermore, this embodiment does not limit specific values of the first weight coefficient, the second weight coefficient, and the third weight coefficient, which may generally be 0.6, 0.2, and 0.2, respectively.
[0095] This embodiment can update the model parameters of the formation resistivity prediction model based on the calculated loss value, and iterate the training until the formation resistivity prediction model converges. This embodiment does not limit the specific method of updating the model parameters, which can generally be back propagation update.
[0096] The temporal neural network model in this embodiment may be an LSTM (Long short-term memory) model, and the LSTM model may include a noise reduction module and an attention module.
[0097] This embodiment does not limit the specific method of model training. Generally, a training sub-sample set is constructed for each oil well; the training sub-sample set contains the target features, the first label value, the second label value and the third label value of the corresponding oil well at each measurement depth; the samples in each training sub-sample set are sorted in order from shallow to deep or from deep to shallow in measurement depth; a training sample set is constructed based on each training sub-sample set, and the sorting method between each training sub-sample set is the same.
[0098] That is, this embodiment can organize the features and labels of each oil well into a training sub-sample set, and each sample in the training sub-sample set can be sorted in the order of measured depth from shallow to deep or from deep to shallow.
[0099] Correspondingly, this embodiment can construct an input window; the size of the input window is a preset number of target features; based on the input window, the samples in each training sub-sample set are traversed to obtain the formation resistivity test value output by the time series neural network model; the loss value is determined based on the formation resistivity test value and the first label value, the second label value and the third label value in the training sample set; the model parameters of the time series neural network model are updated based on the loss value to obtain a trained formation resistivity prediction model.
[0100] That is, this embodiment can input multiple samples each time through the input window to obtain multiple formation resistivity test values, and can determine the loss value based on the obtained formation resistivity test values and the first label value, second label value and third label value in the training sample set to update the model parameters.
[0101] This embodiment does not limit the size of the input window. The input window in this embodiment inputs samples in the form of traversal. For example, if the size of the input window is 30 and the number of samples in the training sub-sample set is 100, the starting position of the input window traverses from the sample ranked 1 in the training sub-sample set to the sample ranked 70, and each training sub-sample set is input 70 times in total, and each input can obtain a corresponding set of formation resistivity prediction values. When a training sub-sample set is traversed, the next training sub-sample set is traversed until all training sub-sample sets are traversed. When the model parameters are not fitted, the step of traversing the training sub-sample set can be re-executed.
[0102] The method of the present invention introduces a knowledge embedding method and uses the traditional empirical formula as a part of the loss function of the neural network model in the single well resistivity logging prediction, thereby realizing the model's anti-noise ability and prior knowledge utilization, and ultimately improving the model's multi-parameter coupling ability and interpretability, making the model-predicted data more accurate.
[0103] The following is a formation resistivity prediction method provided by an embodiment of the present invention, which may include:
[0104] Obtain the conventional logging curve and the post-casing electromagnetic logging curve of the oil well to be tested;
[0105] Extract multi-dimensional characteristic data of conventional logging curves and post-casing electromagnetic characteristic curves at various measurement depths;
[0106] Normalize the multi-dimensional feature data, reduce the feature dimension and clean the data to obtain the target features to be measured;
[0107] The target feature to be measured is input into the formation resistivity prediction model to obtain the formation resistivity prediction value output by the model; wherein the formation resistivity prediction model is a model trained by any formation resistivity prediction model training method.
[0108] This embodiment can obtain conventional logging curves and post-casing electromagnetic logging curves of the oil well to be tested, and extract multi-dimensional feature data of the conventional logging curves and post-casing electromagnetic characteristic curves at each measuring depth, and perform normalization, feature dimension reduction and data cleaning on the multi-dimensional feature data to obtain the target features to be tested. The method of normalizing, feature dimension reduction and data cleaning the multi-dimensional feature data of the oil well to be tested in this embodiment is consistent with the same steps in model training.
[0109] In this embodiment, the target feature to be measured can be input into the formation resistivity prediction model to obtain the formation resistivity prediction value output by the model. Correspondingly, when the input window is used to input samples during the model training process, the samples should also be input in a corresponding manner in this embodiment.
[0110] The formation resistivity prediction value obtained by the method of the present invention is more accurate than the prediction result of the traditional LSTM model. For details, please refer to Figure 2 , where depth is the measurement depth and RT is the formation resistivity.
[0111] The corresponding error index results can be found in Table 1:
[0112]
[0113] Among them, MAE is the mean absolute error, MSE is the mean square error, RMSE is the root mean square error, R 2 is the determination coefficient. It can be seen that the formation resistivity prediction model after the introduction of knowledge embedding method is better than the traditional formation resistivity prediction model.
[0114] The method of the present invention introduces a knowledge embedding method and uses the traditional empirical formula as a part of the loss function of the neural network model in the single well resistivity logging prediction, thereby realizing the model's anti-noise ability and prior knowledge utilization, and ultimately improving the model's multi-parameter coupling ability and interpretability, making the model-predicted data more accurate.
[0115] The following combination Figure 3 , Figure 3 A structural block diagram of a formation resistivity prediction model training device provided by an embodiment of the present invention, the device may include:
[0116] The first module 100 is used to obtain conventional logging curves and post-casing electromagnetic logging curves of multiple oil wells, and extract multi-dimensional feature data of the conventional logging curves and the post-casing electromagnetic characteristic curves at various measurement depths;
[0117] The second module 200 is used to normalize, reduce the dimension of the multi-dimensional feature data, and clean the data to obtain the target feature;
[0118] The third module 300 is used to use the true value of the formation resistivity of the oil well at each measurement depth as the first label value of the target feature;
[0119] The fourth module 400 is used to determine the first physically constrained formation resistivity of the oil well at each of the measurement depths based on the physical constraint formula of the conventional logging curve, and use the first physically constrained formation resistivity as the second label value of the target feature;
[0120] The fifth module 500 is used to determine the second physically constrained formation resistivity of the oil well at each of the measurement depths based on the physical constraint formula of the post-casing electromagnetic logging curve, and use the second physically constrained formation resistivity as the third label value of the target feature;
[0121] The sixth module 600 is used to construct a training sample set based on the target feature, the first label value, the second label value and the third label value, and train a time series neural network model based on the training sample set to obtain a trained formation resistivity prediction model.
[0122] Based on the above embodiments, the method of the present invention introduces a knowledge embedding method and uses the traditional empirical formula as a part of the loss function of the neural network model in single-well resistivity logging prediction, thereby realizing the model's anti-noise ability and prior knowledge utilization, and ultimately improving the model's multi-parameter coupling ability and interpretability, making the model's predicted data more accurate.
[0123] Based on the above embodiment, the conventional well logging curve physical constraint formula is:
[0124] ;
[0125] In the formula, is the first physically constrained formation resistivity at the measurement depth, is the well logging curve sonic logging curve value at the measuring depth, is the natural gamma curve value at the measurement depth, is the spontaneous potential curve value at the measurement depth, is the induction logging curve value at the measuring depth.
[0126] Based on the above embodiments, the physical constraint formula of the post-casing electromagnetic logging curve is:
[0127] ;
[0128] In the formula, is the second physically constrained formation resistivity at the measurement depth, is a constraint formula, which is related to the casing medium model. - is the electromotive force response of n transient electromagnetic curves at the measuring depth.
[0129] Based on the above embodiments, the loss function of the formation resistivity prediction model is:
[0130] ;
[0131] In the formula, is the loss value, is the formation resistivity prediction value output by the formation resistivity prediction model, y is the first label value, is the second label value, is the third label value, is the first weight coefficient, is the second weight coefficient, is the third weight coefficient.
[0132] Based on the above embodiments, the sixth module 600 may include:
[0133] The first unit is used to construct a training sub-sample set for each oil well; the training sub-sample set includes the target feature, the first label value, the second label value and the third label value of the corresponding oil well at each measurement depth;
[0134] A second unit is used to sort the samples in each of the training sub-sample sets in the order of the measurement depth from shallow to deep or from deep to shallow;
[0135] A third unit is used to construct the training sample set based on each of the training sub-sample sets;
[0136] Accordingly, the sixth module 600 may include:
[0137] The fourth unit is used to construct an input window; the size of the input window is a preset number of the target features;
[0138] A fifth unit is used to traverse the target features of each training subsample set based on the input window to obtain a formation resistivity test value output by the time series neural network model;
[0139] A sixth unit, configured to determine a loss value based on the formation resistivity test value and the first label value, the second label value, and the third label value in the training sample set;
[0140] The seventh unit is used to update the model parameters of the time series neural network model based on the loss value to obtain the trained formation resistivity prediction model.
[0141] Based on the above embodiments, the temporal neural network model is an LSTM model; the LSTM model includes a noise reduction module and an attention module.
[0142] The following is a formation resistivity prediction device provided by an embodiment of the present invention, which may include:
[0143] The seventh module is used to obtain the conventional logging curve and the post-casing electromagnetic logging curve of the oil well to be tested;
[0144] An eighth module is used to extract multi-dimensional characteristic data of the conventional logging curve and the post-casing electromagnetic characteristic curve at each measuring depth;
[0145] The ninth module is used to normalize, reduce the dimension of the multi-dimensional feature data, and clean the data to obtain the features of the target to be measured;
[0146] A tenth module is used to input the target characteristics to be measured into the formation resistivity prediction model to obtain the formation resistivity prediction value output by the model;
[0147] Wherein, the formation resistivity prediction model is a model trained by any formation resistivity prediction model training device.
[0148] Based on the above embodiments, the present invention further provides an electronic device, which may include a memory and a processor, wherein the memory stores a computer program, and when the processor calls the computer program in the memory, the steps provided in the above embodiments may be implemented. Of course, the device may also include various necessary network interfaces, power supplies, and other components.
[0149] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by an execution terminal or a processor, the method provided in the embodiment of the present invention can be implemented; the storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.
[0150] In this article, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.
Claims
1. A formation resistivity prediction model training method, characterized in that: include: Obtaining conventional logging curves and post-casing electromagnetic logging curves of multiple oil wells, and extracting multi-dimensional feature data of the conventional logging curves and the post-casing electromagnetic characteristic curves at various measuring depths; Normalizing, reducing the dimension of the multi-dimensional feature data, and performing data cleaning to obtain target features; Taking the true value of the formation resistivity of the oil well at each measurement depth as the first label value of the target feature; Determine a first physically constrained formation resistivity of the oil well at each of the measurement depths based on a physical constraint formula of a conventional well logging curve, and use the first physically constrained formation resistivity as a second label value of the target feature; Determine the second physically constrained formation resistivity of the oil well at each of the measurement depths based on the physical constraint formula of the post-casing electromagnetic logging curve, and use the second physically constrained formation resistivity as the third label value of the target feature; A training sample set is constructed based on the target feature, the first label value, the second label value and the third label value, and a time series neural network model is trained based on the training sample set to obtain a trained formation resistivity prediction model.
2. The formation resistivity prediction model training method according to claim 1, characterized in that: The physical constraint formula of the conventional logging curve is: ； In the formula, is the first physically constrained formation resistivity at the measurement depth, is the well logging curve sonic logging curve value at the measuring depth, is the natural gamma curve value at the measurement depth, is the spontaneous potential curve value at the measurement depth, is the induction logging curve value at the measuring depth.
3. The formation resistivity prediction model training method according to claim 1, characterized in that: The physical constraint formula of the post-casing electromagnetic logging curve is: ； In the formula, is the second physically constrained formation resistivity at the measurement depth, is a constraint formula, which is related to the casing medium model. - is the electromotive force response of n transient electromagnetic curves at the measuring depth.
4. The formation resistivity prediction model training method according to claim 1, characterized in that: The loss function of the formation resistivity prediction model is: ； In the formula, is the loss value, is the formation resistivity prediction value output by the formation resistivity prediction model, y is the first label value, is the second label value, is the third label value, is the first weight coefficient, is the second weight coefficient, is the third weight coefficient.
5. The formation resistivity prediction model training method according to claim 1, characterized in that: The constructing a training sample set based on the target feature, the first label value, the second label value, and the third label value includes: Constructing a training sub-sample set for each oil well; the training sub-sample set includes the target feature, the first label value, the second label value and the third label value of the corresponding oil well at each of the measurement depths; Sort the samples in each of the training sub-sample sets in the order of the measurement depth from shallow to deep or from deep to shallow; Constructing the training sample set based on each of the training sub-sample sets; Correspondingly, the training of the time series neural network model based on the training sample set to obtain a trained formation resistivity prediction model includes: Constructing an input window; the size of the input window is a preset number of the target features; Traversing the target features of each training subsample set based on the input window, obtaining a formation resistivity test value output by the time series neural network model; Determine a loss value based on the formation resistivity test value and the first label value, the second label value, and the third label value in the training sample set; The model parameters of the time series neural network model are updated based on the loss value to obtain the trained formation resistivity prediction model.
6. The formation resistivity prediction model training method according to claim 1, characterized in that: The temporal neural network model is an LSTM model; the LSTM model includes a noise reduction module and an attention module.
7. A formation resistivity prediction method, characterized in that: include: Obtain the conventional logging curve and the post-casing electromagnetic logging curve of the oil well to be tested; Extracting multi-dimensional characteristic data of the conventional logging curve and the post-casing electromagnetic characteristic curve at each measuring depth; Normalizing, reducing the dimension of the multi-dimensional feature data and performing data cleaning to obtain the features of the target to be measured; Inputting the target characteristics to be measured into a formation resistivity prediction model to obtain a formation resistivity prediction value output by the model; Wherein, the formation resistivity prediction model is a model trained according to the formation resistivity prediction model training method according to any one of claims 1 to 6.
8. A formation resistivity prediction model training device, characterized in that: include: The first module is used to obtain conventional logging curves and post-casing electromagnetic logging curves of multiple oil wells, and extract multi-dimensional feature data of the conventional logging curves and the post-casing electromagnetic characteristic curves at various measurement depths; The second module is used to normalize, reduce the dimension of the multi-dimensional feature data and clean the data to obtain the target feature; The third module is used to use the true value of the formation resistivity of the oil well at each measurement depth as the first label value of the target feature; A fourth module is used to determine a first physically constrained formation resistivity of the oil well at each of the measurement depths based on a physical constraint formula of a conventional logging curve, and use the first physically constrained formation resistivity as a second label value of the target feature; A fifth module is used to determine the second physically constrained formation resistivity of the oil well at each of the measurement depths based on the physical constraint formula of the post-casing electromagnetic logging curve, and use the second physically constrained formation resistivity as the third label value of the target feature; The sixth module is used to construct a training sample set based on the target feature, the first label value, the second label value and the third label value, and train the time series neural network model based on the training sample set to obtain a trained formation resistivity prediction model.
9. An electronic device, characterized in that: include: Memory, for storing computer programs; A processor, used to implement the formation resistivity prediction model training method as described in any one of claims 1 to 6 and the formation resistivity prediction method as described in claim 7 when executing the computer program.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, the formation resistivity prediction model training method according to any one of claims 1 to 6 and the formation resistivity prediction method according to claim 7 are implemented.