Prediction method for old well retest interval in low-permeability reservoir
Through the convolutional neural network model, the prediction of the re-examination section of the old well in low-permeability reservoirs is achieved, and the multi-type logging curve data and development dynamic data are used to solve the problems of low interpretation accuracy and weak generalization of the model in the existing technology, and the rapid and accurate prediction of the re-examination section is achieved, which improves the re-examination efficiency and saves costs.
Patent Information
- Application Number
- CN202311382930.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-10-24
- Publication Date
- 2025-04-25
AI Technical Summary
The existing technology has fuzzy geological significance in the prediction of the re-examination layer section of the old well in low-permeability reservoirs, low interpretation accuracy, and weak generalization of the prediction model, resulting in high engineering costs and large differences in re-examination production capacity.
By using the formation cell frame of the target layer of the low-permeability reservoir as a basis, multi-type logging curve data is intercepted based on sedimentary cyclotron constraints, processed into a unified two-dimensional matrix form, and used as input to the convolutional neural network model. The fluid and production layer data identified by dynamic data are developed as dual training labels, and the convolutional neural network model is trained. The trained model is used for the old well retest layer prediction.
The rapid and accurate prediction of the re-examination layer section of the old well in the low-permeability reservoir was achieved, and the problems of low interpretation accuracy and weak generalization of the model in traditional methods were overcome, which significantly improved the re-examination efficiency and saved costs.
Smart Images
Figure CN119884727A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of prediction methods for retesting layers of old wells, and is a prediction method for retesting layers of old wells in low-permeability oil reservoirs. Background Art
[0002] The review of old wells in low-permeability reservoirs is an important scientific research and production work in the middle and late stages of development. At present, the review of old wells mainly adopts the multi-parameter reservoir classification method, through the well logging interpretation model, to determine the lower limit values of parameters such as lithology, physical properties, and oil content, and then analyze the size of the target layer evaluation parameters, and select the layers with parameters above the lower limit as favorable retest layers.
[0003] However, there are two problems in the prediction of retest layers based on the multi-parameter classification method. The first is that the evaluation parameters are obtained from the measured results of the logging sample points, and the spatial correlation between the sample point data is not considered, which makes the geological significance of the interpretation results vague and the interpretation accuracy low. The second is that the prediction of the retest layer section of the old well requires a systematic analysis of the specific target layer. The established prediction model for the retest layer section of the old well has weak generalization, and the labor and time cost of the retest analysis is huge. The old well retest method proposed in the present invention can effectively overcome the limitations of traditional methods and realize rapid and accurate prediction of favorable retest layers in low permeability reservoirs. Summary of the invention
[0004] The present invention provides a method for predicting retesting layers of old wells in low permeability oil reservoirs, which overcomes the shortcomings of the above-mentioned prior art and can quickly and accurately predict favorable retesting layers of low permeability oil reservoirs.
[0005] The reserves and crude oil production of low-permeability reservoirs account for a high proportion of the scale and production of oil and gas reserves in my country. Therefore, improving the development results of low-permeability reservoirs is of great significance to the stability and development of my country's petroleum industry. This type of reservoir has the characteristics of high oil and gas content, multiple oil and gas reservoir types, wide distribution area, and "upper gas and lower oil, marine gas-dominated, and continental oil and gas". However, due to the complex geological conditions of low-permeability reservoirs, the distribution law of oil and water is difficult to clarify. After several years or more than ten years of adjustment, the old wells in the reservoir can still rebuild crude oil production capacity by repeating oil testing in the oil and gas-bearing layers. However, the engineering cost of retesting old wells is high and the retesting capacity varies greatly. Therefore, making full use of the existing logging data of each well and the dynamic data of oil reservoir production, using a fast and accurate prediction method for retesting layers of old wells, and efficiently identifying favorable retesting layers on the well can significantly improve the efficiency of retesting old wells and save costs, providing a scientific basis for the preparation of efficient oil and gas reservoir development adjustment plans, which is conducive to extending the economic development life of low-permeability oil and gas reservoirs in the medium and high water cut period.
[0006] The technical solution of the present invention is achieved through the following measures: a prediction method for retest intervals of old wells in low-permeability oil reservoirs, including taking the stratigraphic unit cell frame of the target layer of the low-permeability oil reservoir as the basis, intercepting multiple types of logging curve data based on the sedimentary cycle as the constraint and processing them into a unified two-dimensional matrix form, and using the data as the input of a convolutional neural network model, and using the fluid and production layer data identified based on the development dynamic data as the dual training labels of the convolutional neural network model to train the convolutional neural network model, and using the trained convolutional neural network model as a convolutional neural network model for predicting retest intervals of old wells, and inputting the logging data corresponding to the well of the retest interval of the old well to be predicted into the convolutional neural network model for predicting retest intervals of old wells, predicting the fluid and production layer of the retest interval of the old well to be predicted, and obtaining a favorable prediction result for the retest interval of the old well.
[0007] The following are further optimizations and / or improvements to the above technical solutions: The above-mentioned multi-type logging curve data include four logging data: natural gamma, sonic time difference, deep resistivity and shallow resistivity curves.
[0008] Each type of logging curve needs to be normalized before being input into the convolutional neural network model.
[0009] The above-mentioned processing of multi-type logging curve data based on sedimentary cycles as constraints into a unified two-dimensional matrix form is as follows: Taking into account the differences in thickness of each cycle in the vertical direction, the number of logging data sample points corresponding to the average thickness of all five-level sedimentary cycles in the study area (oil storage area) is taken as the standard length, denoted as N. The four logging data corresponding to each single layer are interpolated respectively, so that the number of sample points of each logging data after interpolation is N, and each sedimentary cycle corresponds to an N×4 two-dimensional input data.
[0010] The above-mentioned dual training labels also need to be interpolated into N-length data in each five-level sedimentation cycle, and each five-level sedimentation cycle corresponds to N×2 training label data.
[0011] The fluid and production layer data identified based on the development of dynamic data are used as dual training labels for the convolutional neural network model, specifically: The target layer is identified based on production dynamic data, and the layer with a daily output greater than 2t and a water content less than 60% is defined as the producing layer. Based on the identification of the effective producing layer, its discrimination curve is generated, and the producing layer position is assigned a value of 1, and other positions of the target layer are assigned a value of 0. In addition, based on the oil test data, the oil layer section of the target layer is extracted, and the oil layer section is assigned a value of 1, and other positions of the target layer are assigned a value of 0. The producing layer and oil layer discrimination curves are used as dual training labels for the convolutional neural network model.
[0012] Before the well logging data corresponding to the wells in the retest interval of the old wells to be predicted are input into the convolutional neural network model for predicting the retest interval of the old wells, it is necessary to intercept the well logging data corresponding to the wells in the retest interval of the old wells to be predicted (the well logging curves corresponding to the wells for retest prediction) with the five-level sedimentary cycle as the constraint, and process the data corresponding to each five-level cycle into the form of N×4, and then input into the convolutional neural network model for predicting the retest interval of the old wells.
[0013] In the above output prediction results of the old well retest layer section, when the two prediction results of the output well fluid and the production layer are both >0.8, the layer section is a favorable old well retest layer section.
[0014] In the process of predicting the retest interval of old wells, the present invention intercepts input data based on sedimentary cycle constraints, so that the input data has geological genesis significance, and the spatial correlation between each input data point is conducive to the accurate prediction of the retest interval of old wells. The convolutional neural network model training for predicting the retest interval of old wells adopts double labels, so that the convolutional neural network model outputs fluid prediction and production layer prediction results, and only when both results are favorable (>0.8) can it be used as a favorable retest interval of old wells. The present invention overcomes the problem of distinguishing the retest interval based on local hard indicators, and the training model (convolutional neural network model) has scalability. With the increase of training data and the expansion of types, the generalization ability can be continuously improved. Therefore, the present invention realizes the rapid and accurate prediction of the retest interval of old wells in low-permeability oil and gas reservoirs with comprehensive data and knowledge, and outputs the fluid and production layer discrimination results by inputting multi-type logging curve data constrained by sedimentary cycles into the trained retest interval prediction convolutional neural network model of old wells, thereby obtaining favorable prediction results of the retest interval of old wells. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Attached Figure 1 It is the input data and label data of the convolutional neural network model.
[0016] Attached Figure 2 Schematic diagram of training data for the convolutional neural network model.
[0017] Attached Figure 3 This is the prediction result diagram of favorable layer sections in the retest of old wells. DETAILED DESCRIPTION
[0018] The present invention is not limited by the following embodiments, and specific implementation methods can be determined based on the technical solution of the present invention and actual conditions.
[0019] The present invention will be further described below in conjunction with embodiments: Embodiment: An old well production area in a low permeability oil reservoir is used for illustration, and four types of logging curves are selected: natural gamma, acoustic wave time difference, deep resistivity and shallow resistivity curves.
[0020] In the target layer of the low-permeability oil reservoir, the target layer section is divided into several five-level sedimentary cycles based on the rhythmic characteristics of sediments reflected by logging curves such as natural gamma and sonic time difference. The sedimentary cycle will serve as an effective constraint on the input part of the prediction model for the retest layer section of the favorable old well.
[0021] Based on the production dynamic data, the target layer is identified as a production layer, and the layer with a daily production of more than 2t and a water content of less than 60% is considered to be a production layer. Based on the identification of the effective production layer, its discrimination curve is generated, and the production layer position is assigned a value of 1, and the other positions of the target layer are assigned a value of 0. In addition, based on the oil test data, the oil layer section of the target layer is extracted, and the oil layer section is assigned a value of 1, and the other positions of the target layer are assigned a value of 0. The production layer and oil layer discrimination curve will be used as dual training labels for the convolutional neural network model.
[0022] The input part of the convolutional neural network model is the well logging curves available for each well ( Figure 1 ), including four types of logging curves: natural gamma, acoustic time difference, deep resistivity and shallow resistivity. Each type of logging curve needs to be normalized before input.
[0023] The logging data as the input of the convolutional neural network model will be intercepted by the sedimentary cycle constraints, that is, each input logging data and label data pair input into the convolutional neural network model is the response data of a sedimentary cycle.
[0024] Taking into account the differences in thickness of each cycle in the vertical direction, the number of logging data sample points corresponding to the average thickness of all five-level sedimentary cycles in the study area is taken as the standard length, denoted as N. The four logging data corresponding to each single layer are interpolated separately, so that the number of sample points of each logging data after interpolation is N, and each sedimentary cycle corresponds to an N×4 two-dimensional input data.
[0025] The dual training labels also require interpolation of N-length data in each five-level sedimentation cycle, and each five-level sedimentation cycle corresponds to N×2 training label data.
[0026] The data intercepted from wells with favorable production layers are used as training and test data. Among these data, favorable layer segments and unfavorable layer segments are distinguished. 80% of the data are randomly selected from these two categories to form the training set, and the corresponding production layer and fluid discrimination curve data are used as training labels; the remaining data constitute the test set.
[0027] Input the training data into the convolutional neural network model ( Figure 2), where the activation function uses the modified linear unit function, and the loss function uses the mean square error function. The number and size of convolution kernels, pooling method, number of nonlinear layer groups and other parameters are adjusted until the accuracy of the trained convolutional neural network model on the test set is greater than 95%. The convolutional neural network model after training is used as the convolutional neural network model for predicting the retest interval of the old well.
[0028] The well logging curves corresponding to the wells that need to be retested are intercepted with the five-level sedimentary cycle as the constraint, and the data corresponding to each five-level cycle are processed into N×4 form and input into the old well retest layer prediction convolutional neural network model, and the prediction results of the well fluid and production layer are output. Only the layers with favorable results (>0.8) can be used as favorable old well retest layers ( Figure 3 ), thereby realizing the prediction of retest intervals of old wells in low permeability reservoirs.
[0029] In the process of predicting the retest interval of old wells, the present invention intercepts input data based on sedimentary cycle constraints, so that the input data has geological genesis significance, and the spatial correlation between each input data point is conducive to the accurate prediction of the retest interval of old wells. The convolutional neural network model training adopts double labels to output fluid prediction and production layer prediction results. Only when both results are favorable (>0.8) can they be used as favorable old well retest intervals. The present invention overcomes the problem of distinguishing retest intervals based on local hard indicators, and the training model has scalability. With the increase of training data and the expansion of types, the generalization ability can be continuously improved. Therefore, the present invention realizes the rapid and accurate prediction of the retest interval of old wells in low-permeability oil and gas reservoirs with comprehensive data and knowledge. By inputting multi-type logging curve data constrained by sedimentary cycles into the trained convolutional neural network model, the fluid and production layer discrimination results are output, and then a favorable prediction result of the retest interval of old wells is obtained.
[0030] The above technical features constitute the embodiments of the present invention, which have strong adaptability and implementation effect. Non-essential technical features can be added or reduced according to actual needs to meet the requirements of different situations.
Claims
1. A method for predicting the retest interval of old wells in low permeability reservoirs, characterized in that Based on the stratigraphic cell frame of the target layer of the low-permeability oil reservoir, multi-type logging curve data intercepted based on sedimentary cycles as constraints are processed into a unified two-dimensional matrix form and used as the input of the convolutional neural network model. The fluid and production layer data identified based on the development dynamic data are used as the dual training labels of the convolutional neural network model to train the convolutional neural network model. The trained convolutional neural network model is used as the convolutional neural network model for predicting the retest layer section of the old well. The logging data corresponding to the wells of the retest layer section to be predicted are input into the convolutional neural network model for predicting the retest layer section of the old well. The fluid and production layer of the retest layer section to be predicted are predicted, and a favorable prediction result for the retest layer section of the old well is obtained.
2. The method for predicting the retest interval of old wells in low permeability reservoirs according to claim 1, characterized in that The multi-type logging curve data include four logging data: natural gamma, sonic time difference, deep resistivity and shallow resistivity curves.
3. The method for predicting the retest interval of an old well in a low permeability reservoir according to claim 1 or 2, characterized in that Based on the constraints of sedimentary cycles, multiple types of logging curve data are intercepted and processed into a unified two-dimensional matrix form, specifically: Taking into account the differences in thickness of each cycle in the vertical direction, the number of logging data sample points corresponding to the average thickness of all five-level sedimentary cycles in the oil reservoir area is taken as the standard length, denoted as N. The four logging data corresponding to each single layer are interpolated respectively, so that the number of sample points of each logging data after interpolation is N, and each sedimentary cycle corresponds to 1 N×4 two-dimensional input data.
4. The method for predicting the retest interval of an old well in a low permeability reservoir according to claim 1, 2 or 3, characterized in that The dual training labels also require interpolation of N-length data in each five-level sedimentation cycle, and each five-level sedimentation cycle corresponds to N×2 training label data.
5. The method for predicting the retest interval of old wells in low permeability reservoirs according to claim 1 or 2, characterized in that Based on the development of dynamic data identification, fluid and production layer data are used as dual training labels for the convolutional neural network model, specifically: Based on the production dynamic data, the target layer is identified as a producing layer, and the layer section with a daily production of more than 2t and a water content of less than 60% is regarded as a producing layer. Based on the identification of the effective producing layer, a discriminant curve is generated, and the producing layer position is assigned a value of 1, and other positions of the target layer are assigned a value of 0; in addition, based on the oil test data, the oil layer section of the target layer is extracted, and the oil layer section is assigned a value of 1, and other positions of the target layer are assigned a value of 0; The pay zone and oil layer discrimination curves are used as dual training labels for the convolutional neural network model.
6. The method for predicting the retest interval of an old well in a low permeability reservoir according to claim 3 or 4, characterized in that Based on the development of dynamic data identification, fluid and production layer data are used as dual training labels for the convolutional neural network model, specifically: Based on the production dynamic data, the target layer is identified as a producing layer, and the layer section with a daily production of more than 2t and a water content of less than 60% is regarded as a producing layer. Based on the identification of the effective producing layer, a discriminant curve is generated, and the producing layer position is assigned a value of 1, and other positions of the target layer are assigned a value of 0; in addition, based on the oil test data, the oil layer section of the target layer is extracted, and the oil layer section is assigned a value of 1, and other positions of the target layer are assigned a value of 0; The pay zone and oil layer discrimination curves are used as dual training labels for the convolutional neural network model.
7. The method for predicting the retest interval of old wells in low permeability reservoirs according to claim 1 or 2, characterized in that Before the well logging data corresponding to the wells in the old well retest interval to be predicted are input into the old well retest interval prediction convolutional neural network model, it is necessary to intercept the well logging data corresponding to the wells in the old well retest interval to be predicted with the five-level sedimentary cycle as the constraint, and process the data corresponding to each five-level cycle into the form of N×4, and then input into the old well retest interval prediction convolutional neural network model.
8. The method for predicting the retest interval of an old well in a low permeability reservoir according to claim 3, 4 or 5, characterized in that Before the well logging data corresponding to the wells in the old well retest interval to be predicted are input into the old well retest interval prediction convolutional neural network model, it is necessary to intercept the well logging data corresponding to the wells in the old well retest interval to be predicted with the five-level sedimentary cycle as the constraint, and process the data corresponding to each five-level cycle into the form of N×4, and then input into the old well retest interval prediction convolutional neural network model.
9. The method for predicting the retest interval of an old well in a low permeability reservoir according to claim 6, characterized in that Before the well logging data corresponding to the wells in the old well retest interval to be predicted are input into the old well retest interval prediction convolutional neural network model, it is necessary to intercept the well logging data corresponding to the wells in the old well retest interval to be predicted with the five-level sedimentary cycle as the constraint, and process the data corresponding to each five-level cycle into the form of N×4, and then input into the old well retest interval prediction convolutional neural network model.
10. The method for predicting the retest interval of old wells in low permeability reservoirs according to any one of claims 1 to 9, characterized in that In the output prediction results of the old well retest layer section, when the two prediction results of the output well fluid and the production layer are both >0.8, the layer section is a favorable old well retest layer section.