Intelligent optimization method, system and medium for oilfield infill well location

Abstract

The present invention discloses an intelligent optimization method, system, and medium for infill well locations in an oil field, relating to the field of oil and gas field development. By randomly determining the spatial positions of infill wells within an effective grid range and performing numerical simulation in combination with well condition data, the method obtains simulated data of the total cumulative oil production or economic net present value of a target reservoir at different spatial positions of the infill wells based on the results of the numerical simulation. The simulated data of the total cumulative oil production or economic net present value are used as labels to train a machine learning model, thereby using the trained prediction model to predict the predicted data of the total cumulative oil production of the target reservoir at each effective grid position where no wells have been drilled at the target time, and then determining the optimal infill well location based on the predicted data. The present invention can efficiently and accurately determine the optimal infill well location based on the above-mentioned optimization process of the optimal infill well location.

Description

Technical Field

[0001] The present invention relates to the field of oil and gas field development, and in particular to an intelligent optimization method, system and medium for infill well locations in an oil field. Background Art

[0002] Over 70% of my country's oil reserves and production come from water-driven oilfields. After more than 50 years of development, these fields have generally entered a high-water-cut phase. Well pattern intensification improves the sweep of injected water and is a key method for flow field control in high-water-cut oilfields. Well pattern intensification involves adjusting the well pattern within an oilfield block based on a known number of infill wells and their respective injection and production well types, effectively increasing the recovery rate of water-driven oilfields.

[0003] Commonly used methods for determining infill well placement include empirical methods, reservoir engineering methods, vector well pattern methods, and optimization methods based on numerical simulation. Empirical methods rely on manual experience to screen the optimal solution, which may miss the optimal solution and make it difficult to comprehensively consider the influence of multiple factors. Reservoir engineering methods offer fast solutions, but their assumptions are overly ideal, limiting their application scope and primarily applicable to homogeneous reservoirs and simple regular well patterns. The vector well pattern method aims to maximize balanced displacement, but it lacks comprehensive considerations. Optimization methods based on numerical simulation require multiple calls to the reservoir simulator during the optimization process, resulting in high computational complexity, a tendency to fall into local optimal solutions, and inconvenient application. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent optimization method, system and medium for oil field infill well locations, which can quickly and accurately determine the location of the optimal infill wells.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] The present invention provides an intelligent optimization method for infill well locations in an oil field, comprising:

[0007] Randomly select a preset number of grids within the effective grid range of the target oil reservoir where no wells have been drilled, and obtain multiple spatial location layout plans for infill wells;

[0008] For each of the spatial location layout schemes of the infill wells, numerical simulation is performed under preset infill well conditions to obtain a variety of different infill well numerical simulation results; based on the infill well numerical simulation results, simulated data on the cumulative oil production of the entire area of the target oil reservoir at a preset time or simulated data on the economic net present value;

[0009] Determine the factors affecting infill well locations in the target reservoir and select key factors affecting infill well locations;

[0010] For each of the spatial location layout plans of the infill wells, the key infill well location influencing factors are used as input, and the corresponding simulated data of the cumulative oil production of the entire area or the simulated data of the economic net present value is used as output to train a machine learning model to obtain a prediction model;

[0011] Obtaining all valid grid locations of the oil reservoir to be optimized at the target time where no wells have been drilled, inputting the key infill well location influencing factors at each valid grid location into the prediction model, and obtaining the area-wide cumulative oil production prediction data or economic net present value prediction data corresponding to each valid grid location;

[0012] The optimal infill well location is determined based on the cumulative oil production forecast data or the economic net present value forecast data for the entire area.

[0013] The present invention also provides an intelligent optimization system for infill well locations in an oil field, comprising:

[0014] A layout scheme determination module is used to randomly select a preset number of grids within the effective grid range of the target reservoir where no wells have been drilled, and obtain spatial location layout schemes for multiple infill wells;

[0015] A numerical simulation module is used to perform numerical simulation on the spatial location layout of each infill well under preset infill well conditions to obtain a variety of different infill well numerical simulation results; and to determine, based on the infill well numerical simulation results, simulated data on the cumulative oil production of the entire area of the target oil reservoir at a preset time or simulated data on the economic net present value;

[0016] A key influencing factor determination module is used to determine the influencing factors of the infill well locations in the target oil reservoir and screen out the key influencing factors of the infill well locations;

[0017] A model training module is used to train a machine learning model for the spatial location layout plan of each infill well, using the key infill well location influencing factors as input and the corresponding simulated data of cumulative oil production in the entire area or the simulated data of economic net present value as output to obtain a prediction model;

[0018] A prediction module is used to obtain all valid grid locations of the oil reservoir to be optimized at the target time, input the key infill well location influencing factors at each valid grid location into the prediction model, and obtain the predicted data of the cumulative oil production of the entire area or the predicted data of the economic net present value corresponding to each valid grid location;

[0019] The optimal location determination module is used to determine the optimal infill well location based on the cumulative oil production forecast data or the economic net present value forecast data of the entire area.

[0020] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the intelligent optimization method for infilling well locations in high-water-cut oil fields.

[0021] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0022] The present invention provides an intelligent optimization method, system, and medium for infill well locations in an oil field. The method randomly determines the spatial locations of infill wells within an effective grid range and performs numerical simulations in combination with well condition data. Based on the results of the numerical simulations, simulated data for the cumulative oil production of the entire area (or simulated data for the economic net present value (NPV)) of a target reservoir at different spatial locations of the infill wells are obtained. The simulated data for the cumulative oil production of the entire area (or simulated data for the economic net present value (NPV)) and corresponding key infill well location influencing factors are then used to train a machine learning model to obtain a prediction model. The prediction model is then used to predict the cumulative oil production (or economic net present value (NPV)) of the entire area of the target reservoir at each effective grid location where no wells have been drilled at a target time. The optimal infill well location is then determined based on the predicted data. The present invention, based on the above-mentioned optimization process for the optimal infill well location, can efficiently and accurately determine the optimal infill well location. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0024] Figure 1 This is a flow chart of an intelligent optimization method for infill well locations in an oil field provided in Example 1 of the present invention.

[0025] Figure 2 This is a schematic diagram of the design of a numerical simulation model for well pattern densification provided in Example 1 of the present invention.

[0026] Figure 3 This is a flow chart of the process for screening factors affecting key encrypted well locations provided in Example 1 of the present invention.

[0027] Figure 4 This is the R-type clustering process provided in Example 1 of the present invention.

[0028] Figure 5 This is a NPV comparison chart of the predicted results and actual results of reservoir model C provided in Example 1 of the present invention.

[0029] Figure 6This is the NPV diagram of the transfer learning prediction and actual results of reservoir model B provided in Example 1 of the present invention.

[0030] Figure 7 An intelligent optimization system for infill well locations in an oil field is provided in the second embodiment of the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0032] The technical idea of the present invention is:

[0033] 1. Construct a sample set for optimizing infill well locations in high-water-cut oilfields;

[0034] (1) Generate an infill well numerical simulation plan and conduct infill well numerical simulation in the target reservoir:

[0035] A predetermined number of grid cells are randomly selected within the target reservoir's valid, undrilled grid range. These random grid coordinates form a spatial location scheme for infill wells. These spatial location schemes, combined with pre-determined infill well conditions (including, but not limited to, infill well operating procedures), form different numerical simulation schemes for infill wells. Numerical simulations are then performed based on these schemes. The simulation results then determine simulated cumulative oil production data (or NPV data) for the infill wells at a predetermined time in the area for each simulation scheme.

[0036] (2) A sample set is established based on the numerical simulation results of the infill wells in the target reservoir. The input variables of the data set are: overall water content (the overall water content here refers to the overall water content of the reservoir used when constructing the sample set), the location of the infill wells, the type of the infill wells, the perforation layer of the infill wells, the location of the old wells, the well spacing, the porosity, permeability, saturation, pressure, thickness, water content of each old well, cumulative oil production of each old well, cumulative water production of each old well and cumulative water injection volume of each old well, etc. The response variable of the data set is: the simulated data of the cumulative oil production of the entire area at the preset time corresponding to the infill wells (or the economic net present value). The variables of different numerical simulation schemes correspond to the response variables one by one and together constitute the sample set. The sample set is preprocessed and divided into a training set and a validation set according to a preset ratio.

[0037] 2. Analysis of the main controlling factors for infill well locations;

[0038] Based on the optimized sample set of infill well locations, the key control factors of infill well locations are determined using feature selection method.

[0039] 3. Using the training data set to train a machine learning model;

[0040] Based on the constructed infill well sample library, six variables (key control factors identified in the previous step) were used as model inputs: overall water content, infill well coordinates, water content of each old well, cumulative oil production of each old well, cumulative water production of each old well, and cumulative injection volume of each old well. The cumulative oil production (or economic net present value) of the entire area at different times was used as the output. Ultimately, a production dynamics prediction model based on machine learning was trained. The machine learning model can be an XGBoost model, a random forest, a support vector machine, or a multivariate linear regression model. The choice is not limited here and should be based on actual needs.

[0041] 4. Use the prediction model to predict the cumulative oil production (or economic net present value NPV) of the entire area corresponding to the infill wells of the target oil reservoir, and then obtain the optimal infill well location of the target oil reservoir at any time.

[0042] The purpose of the present invention is to provide an intelligent optimization method, system and medium for infill well locations in an oil field, which randomly determines the spatial position of infill wells within an effective grid range and performs numerical simulation in combination with well condition data, thereby obtaining the simulated data of the total cumulative oil production of the target reservoir at different spatial positions of the infill wells (or the simulated data of the economic net present value (NPV)) based on the results of the numerical simulation, and then uses the simulated data of the total cumulative oil production (or the simulated data of the economic net present value (NPV)) and the corresponding key infill well location influencing factors to train an XGBoot prediction model, thereby using the trained XGBoot prediction model to predict the predicted data of the total cumulative oil production (or the predicted data of the economic net present value (NPV)) of the target reservoir at each effective grid position where no wells are drilled at the target time, and then determines the optimal infill well location based on the predicted data. The present invention can efficiently and accurately determine the optimal infill well location based on the optimization process of the above-mentioned optimal infill well location.

[0043] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] Example 1

[0045] like Figure 1 As shown, this embodiment provides an intelligent optimization method for infilling well locations in an oil field, which can be specifically applied to high-water-cut oil fields, such as oil fields with an overall water cut of more than 80%, and can also be applied to oil fields with other water cut values. The intelligent optimization method for infilling well locations in an oil field includes:

[0046] S1: Randomly selecting a preset number of grids within the effective grid range of the target reservoir without wells, obtaining multiple spatial layout plans for infill wells, and determining the spatial layout plan for each infill well. The preset number of randomly selected grids is no more than 20% of the effective grid range of the target reservoir without wells. In this embodiment, the preset number is 10% of the effective grid range of the target reservoir without wells.

[0047] S2: Numerical simulation is performed under preset infill well conditions to obtain multiple different infill well numerical simulation results. Based on the infill well numerical simulation results, simulated cumulative oil production data (or simulated net present value (NPV) data) for the entire target reservoir at a preset time is determined. Cumulative oil production is directly simulated, while NPV data is calculated based on the simulated cumulative oil and water production data.

[0048] Through numerical simulation, a sample set for optimizing the infill well locations in high-water-cut oil fields is constructed. For a typical oil reservoir, a numerical simulation model of a typical well group in the reservoir is designed and established for different injection-production well patterns and locations of new infill wells, taking into account the influence of geological factors, development factors and other factors. An interface between the numerical simulator and the machine learning platform is compiled to carry out numerical simulation research and quickly construct a sample library for infill well locations in old oil fields.

[0049] In order to make the sample library more representative, a large number of samples need to be constructed for different actual reservoirs. At the same time, the infill well location is also affected by different infill timings. Therefore, for different reservoir models, different water cuts are set as infill timings. The infill well locations are randomly determined from the model and reservoir numerical simulations are performed. Based on the simulation results, the NPV is calculated to construct a sample library for machine learning. The model scheme is as follows: Figure 2 shown.

[0050] In this example, we used two reservoir models and set four different water cuts as infill timings. We performed multiple numerical simulations with 20 different initial conditions to construct an infill well location sample set, as shown in Table 1.

[0051] Table 1 Composition of the encrypted well sample library

[0052]

[0053]

[0054] S3: Determine the factors affecting the infill well locations in the target reservoir and select the key factors affecting the infill well locations.

[0055] Reservoir physical properties, seepage parameters, and reservoir production control parameters all influence the location of infill wells during development, and therefore require comprehensive consideration. Based on the data available during on-site infill well planning, the influencing factors initially identified are overall water cut, infill well location, infill well type, perforation layer location, old well location, well spacing, porosity, permeability, saturation, pressure, thickness at the infill well location, and production performance data for old wells (including cumulative oil production, cumulative water production, cumulative water injection volume, and water cut for each old well), totaling 15 parameters.

[0056] First, data collection and preprocessing are performed on the 15 identified influencing factors. After obtaining the data of each influencing factor, the influencing factors of the infill well location are determined by the feature selection method. The specific process of the feature selection method is as follows: Figure 3 As shown in the figure, a multivariate selection method is first used to comprehensively rank the scores of each factor, selecting the top-ranked influencing factors. Next, the R-type clustering method is used to analyze the independence of each influencing factor, eliminating factors with strong correlations, and ultimately determining the factors affecting the infill well location. Feature selection methods can generally be divided into three types based on their feature selection methods: filtering, embedding, and packaging.

[0057] The filtering method statistically scores and ranks each feature, setting a threshold to select and delete some features. Among the methods suitable for analyzing dominant factors, linear correlation (C), Pearson correlation test (PCC), and maximum information coefficient (MIC) are used to rank features by measuring the strength of nonlinearity between the independent and dependent variables.

[0058] Embedding methods use machine learning algorithms and models for training to determine weight coefficients for each feature, then select features based on these coefficients in descending order. Embedding methods include linear regression (LR), L1 regularization (Lasso), L2 regularization (Ridge), and random forest regression (RFR). These methods can be used to evaluate the relationship between independent and dependent variables. The first three primarily exploit the linear or nonlinear regression relationship between the target and features, using mathematical models to solve the problem. The last, based on the concept of decision trees, utilizes the random forest method in machine learning and, based on the bagging concept, uses information gain to determine the relationship between each feature and the target. The greater the information gain, the more selective the feature.

[0059] The wrapper method uses machine learning predictions to score or demonstrate the objective function, selecting or excluding certain feature variables. Recursive feature selection (RFE_lr) is used here to analyze the relationship between each feature and the target. The core idea is to use a base model for multiple rounds of training. After each round of training, features with certain weight coefficients are removed, and the next round of training is performed based on the new feature set.

[0060] For the above three types of feature selection methods, the filtering method mainly considers the influence of a single feature variable on the target, with fast operation speed and simple model, but it cannot consider the correlation between multiple variables; while the packaging method and the embedding method both process multivariate data through machine learning, and can perform multivariate and big data calculations, but the calculation process is relatively complex, and the calculation results depend on the accuracy of machine learning. In this embodiment, all of the above methods, namely C, PCC, MICLR, LR, Lasso, Ridge, RFR and RFE_lr, are used to obtain comprehensive feature selection results, and the main control factor variables are effectively screened to balance the advantages and disadvantages of each method, so that the obtained main control factors are more reasonable.

[0061] Taking the above-mentioned reservoir model C as an example, data were collected for 15 influencing factors. Since the model has multiple wells, a total of 67 influencing variables were obtained. A certain influencing factor may correspond to multiple influencing variables. For example, for the influencing factor "water content of old wells", the 9 old wells in model C correspond to one water content respectively. Therefore, this influencing factor corresponds to the water content of the 9 old wells, that is, 9 influencing variables. After calculating the scores of each variable under different methods (referring to the scoring results of evaluating the "importance" of a certain factor), each score is normalized, and then all the scores of each variable under the same influencing factor are added together to obtain the influence coefficient S of the influencing factor x on the NPV of the infill well evaluation index. ex , as shown in the following formula. Furthermore, based on the influence coefficient of each variable, we can preliminarily screen out the main control factors affecting infill well locations. The screening criterion used here is that the influence coefficient of a variable is greater than 0.2.

[0062]

[0063] Among them, S ix S is the result of normalizing the evaluation score of the influencing factor x by the i-th feature selection model; ex is the influence coefficient of influencing factor x on the NPV of infill well location evaluation index; n is the number of feature selection models.

[0064] Initially screened factors with influence coefficients greater than 0.2 included: overall water cut, infill well location, infill well type, infill well perforation layer, well spacing, cumulative oil production of old wells, cumulative water production of old wells, cumulative water injection of old wells, and water cut of old wells. Further R-type cluster analysis was performed on these variables. Because the production performance data of old wells had been normalized based on well spacing (i.e., production performance was divided by well spacing), well spacing and production performance of old wells showed a strong correlation during the clustering process. Therefore, well spacing was removed from the list of influencing factors. Ultimately, eight factors influencing infill well location were identified: overall water cut, infill well location, infill well type, infill well perforation layer, water cut of old wells, cumulative oil production of old wells, cumulative water production of old wells, and cumulative water injection of old wells. Table 2 shows some of the scoring results for factors influencing infill well location.

[0065] Table 2. Some scoring results of factors affecting infill well locations

[0066]

[0067]

[0068] Therefore, step S3 specifically includes:

[0069] (1) Preliminarily determine the factors affecting the infill well location in the target reservoir.

[0070] (2) Using a comprehensive feature selection method to select features of various influencing factors of the infill well location in the target oil reservoir, and obtain preliminary influencing factors; the comprehensive feature selection method includes linear correlation degree, Pearson correlation test, maximum information coefficient method, linear regression method, L1 regularization, L2 regularization, random forest and recursive feature selection method.

[0071] (3) Using the R-type clustering algorithm, the correlation between the influencing factors of the encrypted well locations after the initial selection is analyzed, and the influencing factors with correlations higher than the preset correlation threshold are eliminated to obtain the key influencing factors of the encrypted well locations.

[0072] Among them, a comprehensive feature selection method is used to select the influencing factors of the infill well location in the target reservoir, and the preliminary influencing factors are obtained, which specifically include:

[0073] Analyzing each influencing factor of the infill well location in the target reservoir using each feature selection method, and obtaining a score for each influencing factor corresponding to each feature selection method;

[0074] Using each feature selection method, summing the scores of the same influencing factor of the infill well locations in the target oil reservoir to obtain a comprehensive score of each influencing factor of the infill well locations in the target oil reservoir;

[0075] The preliminary influencing factors are obtained by comparing the comprehensive score of each influencing factor of the infill well location in the target oil reservoir with a preset threshold.

[0076] Regarding the R-type clustering method: First, determine the similarity measure of each influencing factor. The similarity measure is expressed as a correlation coefficient, and the single sample x in a single influencing factor is recorded as ij The value of (x 1j ,x 2j ,…,x nj ) T ∈R n (j=1,2,…,m). Then the two influencing factors x ij with x ik The correlation coefficient is calculated as follows:

[0077]

[0078] Where r jk is the correlation coefficient between the jth influencing factor and the kth influencing factor, x ij is the i-th sample in the j-th influencing factor, x ik is the i-th sample in the k-th influencing factor, is the arithmetic mean of the total samples in the j-th influencing factor, is the arithmetic mean of the total samples in the kth influencing factor, n is the total number of samples in a single influencing factor, and each of the m influencing factors has n samples.

[0079] Secondly, the correlation coefficient is used to calculate the shortest distance between multiple influencing factors. The shortest distance is calculated as follows:

[0080]

[0081] Next, construct m classes, each class contains a single influencing factor (G1, G2, G3, ..., G m ), merge the classes with the shortest distance in sequence until m classes are merged into one class, and the clustering is completed.

[0082] Finally, the clustering results were analyzed and the correlation coefficients of the influencing factors in the same category were calculated. ij The influencing factors with a value of ≥0.8 were eliminated.

[0083] Specifically, the R-type clustering method was used to analyze the independence of various influencing factors, eliminate factors with strong correlation, and finally determine the influencing factors of infill well locations, including:

[0084] (1) Determine the correlation coefficient between each two primary influencing factors.

[0085] (2) Treat each primary election influencing factor as a class.

[0086] (3) Determine the distance between each two classes based on the correlation coefficients between the primary influencing factors contained in each class; and consider the two classes corresponding to the shortest distance among all distances as the same class.

[0087] Return to step (3) "determine the distance between each two categories based on the correlation coefficients between the primary influencing factors contained in the current categories" until the final clustering is completed into one category.

[0088] (4) Eliminate the primary influencing factors whose correlation coefficients are greater than a preset correlation threshold in the results of each clustering to obtain the key encrypted well location influencing factors.

[0089] The following is an example process of a specific clustering method:

[0090] First, each influencing factor is regarded as a class; for example, the six influencing factors shown in Table 3 constitute six classes.

[0091] Table 3 Initially, the six influencing factors constitute six categories

[0092] kind Influencing factors G1 X1 G2 X2 G3 X3 G4 X4 G5 X5 G6 X6

[0093] The correlation coefficient is symmetrical, so you can only look at half of it. In addition, the table here directly assumes all the data, and the actual data needs to be calculated based on r jk Table 4 shows the correlation coefficients among the six influencing factors.

[0094] Table 4 Correlation coefficients among the 6 influencing factors

[0095] X1 X2 X3 X4 X5 X6 X1 1 0.7 0.8 0.4 0.2 0.9 X2 0.7 1 0.5 0.8 0.4 0.3 X3 0.8 0.5 1 0.5 0.5 0.4 X4 0.4 0.8 0.5 1 0.9 0.5 X5 0.2 0.4 0.5 0.9 1 0.7 X6 0.9 0.3 0.4 0.5 0.7 1

[0096] First judgment: Directly determine the minimum value in all data and find that the distance between X1 and X5 is the shortest. After combining X1 and X5, recalculate the distance between clusters and combinations, and between combinations (hereinafter also called correlation coefficient). The calculation method is shown in the following formula:

[0097]

[0098] Table 5. Clustering process based on correlation coefficient

[0099] (X1, X5) X2 X3 X4 X6 (X1, X5) 1 0.55 0.65 0.65 0.8 X2 0.55 1 0.5 0.8 0.3 X3 0.65 0.5 1 0.5 0.4 X4 0.65 0.8 0.5 1 0.5 X6 0.8 0.3 0.4 0.5 1

[0100] Second evaluation: First, we need to update the data in the table because we have reduced the number of clusters from the initial six to five. We then continue to determine the minimum value among all the data. We find that X2 and X6 have the shortest distance, so we merge them. This reduces the number of clusters from five to four. Table 6 shows the secondary clustering process based on the correlation coefficient.

[0101] Table 6 Secondary clustering process based on correlation coefficient

[0102] (X1, X5) (X2, X6) X3 X4 (X1, X5) 1 0.675 0.65 0.65 (X2, X6) 0.675 1 0.45 0.65 X3 0.65 0.45 1 0.5 X4 0.65 0.65 0.5 1

[0103] In the third judgment, similarly, it is found that X3 and X4 have the shortest distance, so we continue to merge, reducing the number of categories from 4 to 3. Table 7 shows the three-level clustering process based on the correlation coefficient.

[0104] Table 7 Three-dimensional clustering process based on correlation coefficient

[0105] (X1, X5) (X2, X6) (X3, X4) (X1, X5) 1 0.675 0.65 (X2, X6) 0.675 1 0.55 (X3, X4) 0.65 0.55 1

[0106] In the fourth judgment, similarly, it is found that (X2, X6) and (X3, X4) have the shortest distances, so continue to merge and reduce the number of categories from 3 to 2.

[0107] In the fifth judgment, (X2, X3, X4, X6) and (X1, X5) are combined into 1 category, and the clustering is completed. Figure 4 shown.

[0108] Furthermore, “Finally, the clustering results were analyzed and the correlation coefficients of the influencing factors in the same category were calculated. ij Factors with a correlation coefficient of ≥0.8 are eliminated. ” For example, in the fourth judgment, (X2, X3, X4, X6) are merged, but in the original matrix, the correlation coefficient between X2 and X4 is 0.8, so X2 or X4 is eliminated.

[0109] S4: For each infill well spatial location plan, the XGBoot model is trained using the key infill well location influencing factors as input and the corresponding simulated cumulative oil production data (or simulated NPV data) for the entire area as output to obtain a prediction model. Here, the XGBoost model is used as an example of a machine learning model.

[0110] The model used eight variables as inputs: overall water content, infill well location, infill well type, infill well perforation layer, water content of each old well, cumulative oil production of each old well, cumulative water production of each old well, and cumulative injection volume of each old well. The output was the cumulative oil production (or net present value (NPV)) of the entire region at different times. Ultimately, an XGBoost-based prediction model was constructed. After data processing, the input data (overall water content, infill well location, infill well type, infill well perforation layer, water content of each old well, cumulative oil production of each old well, cumulative water production of each old well, and cumulative injection volume) totaled eight variables, and the data labels were the cumulative oil production at 10 moments, ultimately generating 4,816 sets of sample data.

[0111] The XGBoost-based prediction model was trained with the above model input and output. The model hyperparameter values are shown in Table 8. After 20 training cycles, a prediction model for predicting the total cumulative oil production (or economic net present value NPV) of the entire area at each valid grid location was obtained.

[0112] Table 8 XGBoost prediction model parameters

[0113] Parameter name Value Number of trees 374 Learning rate 0.06 The maximum depth of each tree 5 The proportion of data used by each tree 0.7 The proportion of features used by each tree 0.9 Minimum splitting loss for each tree 0.2

[0114] In this embodiment, after executing step S4, the following steps are further performed:

[0115] (1) Inputting the verification samples of the factors affecting the key infill well locations into the prediction model to obtain the corresponding verification data of the cumulative oil production of the entire area or the verification data of the economic net present value;

[0116] (2) Comparing the verification data of the total oil production of the entire area with the simulation data of the total oil production of the entire area at the same time, or comparing the verification data of the economic net present value with the simulation data of the economic net present value at the same time, to determine the prediction effect of the prediction model.

[0117] (3) If the prediction effect of the prediction model does not reach the preset effect, the number of spatial location layout schemes for the infill wells is increased, and the process returns to step S1 of "randomly selecting a preset number of grids within the effective grid range of the target oil reservoir where no wells have been drilled" to obtain more spatial location layout schemes for the infill wells.

[0118] (4) If the prediction effect of the prediction model reaches the preset effect, the prediction model is used to predict the cumulative oil production data or economic net present value of the entire area of the oil reservoir to be optimized.

[0119] S5: Obtain all valid grid positions of the oil reservoir to be optimized at the target time where no wells have been drilled, input the key infill well location influencing factors at each valid grid position into the prediction model, and obtain the predicted data of the total cumulative oil production or the economic net present value corresponding to each valid grid position. Figure 5 The NPV diagram of the predicted and actual results of reservoir model C is shown on the left. The economic net present value corresponding to each effective grid position obtained by numerical simulation is shown on the right. The economic net present value corresponding to each effective grid position obtained by the trained prediction model is shown on the right. For reservoir model C, the R 2 The two data are basically consistent, indicating that the method proposed by the present invention has high accuracy. In addition, the computational complexity of the method proposed by the present invention is only 10% of that of the numerical simulation method, and the computational efficiency is high.

[0120] Before executing step S5 of "obtaining all valid grid positions of the oil reservoir to be optimized at the target time without drilling wells", the following steps are included:

[0121] (1) Determine whether the oil reservoir to be optimized and the target oil reservoir are in the same block.

[0122] (2) If yes, obtain the valid grid positions of all undrilled wells in the reservoir to be optimized at the target time.

[0123] (3) If not, randomly select a preset number of grids within the effective grid range of the oil reservoir to be optimized without drilling wells, obtain the corresponding spatial position layout plans of multiple infill wells, and perform numerical simulation on the spatial position layout plan of each infill well corresponding to the oil reservoir to be optimized, and obtain the simulated data of the cumulative oil production or the economic net present value simulation data of the whole area corresponding to the oil reservoir to be optimized; use the key infill well location influencing factors as input and the simulated data of the cumulative oil production or the economic net present value simulation data of the whole area of the corresponding oil reservoir to be optimized as output to train the prediction model, obtain the prediction model after transfer learning, and use the prediction model after transfer learning as the prediction model; the prediction model after transfer learning is used to predict the predicted data of the cumulative oil production or the economic net present value prediction data of the whole area of the oil reservoir to be optimized. The preset number of grids here can be grids within 10% of the number of all effective grids without drilling wells.

[0124] For the reservoir blocks corresponding to the sample set used during prediction model training, subsequent predictions can be made directly using the prediction model. However, for reservoir blocks outside the sample set, a small number of numerical simulation plans for infill wells in the reservoir blocks to be optimized are regenerated and simulated. Based on the numerical simulation results, a supplementary sample data set is established. This supplementary sample data set is then used to retrain and fine-tune the parameters of the trained prediction model, enabling the model to determine the optimal infill well locations for the reservoir blocks to be optimized outside the sample set.

[0125] For Reservoir Model B, based on the prediction model trained with Reservoir Model C, 1% data from Reservoir Model B at different water cuts was added as a supplementary sample set, totaling 892 sets of sample data. The already trained prediction model for Reservoir Model C was then retrained and parameter fine-tuned with this 892-set sample data. This resulted in a prediction model after transfer learning. The hyperparameters of the prediction model after transfer learning are shown in Table 9.

[0126] Table 9 Prediction model parameters after transfer learning

[0127] Parameter name Value Number of trees 300 Learning rate 0.1 The maximum depth of each tree 5 The proportion of data used by each tree 0.8 The proportion of features used by each tree 0.7 Minimum splitting loss for each tree 0.4

[0128] Figure 6The figure shows the comparison between the transfer learning prediction results and the actual NPV of reservoir model B. For reservoir model B, although only 1% of the data of reservoir model B at different water cuts in the sample set in Table 1 was added as a supplementary sample set for transfer learning, a high accuracy was still achieved. The economic net present value of its numerical simulation and the economic net present value predicted by transfer learning R 2 Therefore, the method proposed in the present invention can still ensure accurate and fast prediction when adding very little data, and has high portability.

[0129] S6: Determine the optimal infill well location based on the area-wide cumulative oil production forecast data or the economic net present value forecast data.

[0130] The infill well location corresponding to the maximum value among all the area-wide cumulative oil production forecast data or economic net present value forecast data is determined as the optimal infill well location.

[0131] Obtain all valid grid locations of the target reservoir that have not been drilled at the target time, and use the XGBoost prediction model to predict the economic net present value of the infill wells for all locations one by one. The input parameters of the XGBoost prediction model are the x and y coordinates of each location, the well type of the infill well, the perforation layer of the infill well, the overall water content, the water content of each old well, the cumulative oil production of each old well, the cumulative water production of each old well, and the cumulative injection volume of each old well. The output parameter is the cumulative oil production (or economic net present value) of the infill wells at each location at the preset time. After completing the prediction of all locations of the target reservoir, the location corresponding to the maximum value of all predicted values is determined as the optimal infill well location of the target reservoir at the target time.

[0132] The advantages of the present invention are:

[0133] (1) By using the prediction model based on the machine learning model trained by the present invention, the optimal infill well location of other oil reservoirs can be efficiently determined through transfer learning. Simply put, for example, for a block, if you want to obtain an XGBoost prediction model for infill well locations, you may need to perform numerical simulations of 5,000 sets of scenarios to construct a sample set. However, assuming that there is already an XGBoost model trained for other blocks, then based on this model, you only need to add 50 sets of samples, and then train the XGBoost model again through transfer learning to get better results. Therefore, the method provided by the present invention has high portability;

[0134] (2) The key infill well location influencing factors input into the XGBoost prediction model of the present invention, such as the overall water content, the location of the infill wells, the water content of each old well, the cumulative oil production of each old well, the cumulative water production of each old well, and the cumulative water injection volume of each old well, are all production dynamic parameters that are easy to obtain on site. Therefore, when the trained XGBoost prediction model is used, it is only necessary to input the production dynamic parameters directly obtained on site, and no numerical simulation is required. In this way, even if there is no numerical simulation model for other oil reservoir blocks, the model can still be used to determine the optimal infill well location, so this method has stronger operability.

[0135] Example 2

[0136] like Figure 7 As shown, this embodiment provides an intelligent optimization system for infill well locations in high-water-cut oil fields, comprising:

[0137] The layout scheme determination module 100 is used to randomly select a preset number of grids within the effective grid range of the target oil reservoir where no wells have been drilled, and obtain spatial position layout schemes for multiple infill wells.

[0138] Numerical simulation module 200 is configured to perform numerical simulations on each of the spatial layout schemes for the infill wells under pre-set infill well conditions, thereby obtaining a variety of different infill well numerical simulation results. Based on the infill well numerical simulation results, simulated cumulative oil production data or economic net present value data for the target reservoir at a pre-set time are determined.

[0139] The key influencing factor determination module 300 is used to determine the influencing factors of the infill well location in the target reservoir and screen out the key infill well location influencing factors.

[0140] The model training module 400 is used to train a machine learning model for the spatial location layout plan of each of the infill wells, using the key infill well location influencing factors as input and the corresponding cumulative oil production simulation data of the entire area or the economic net present value simulation data as output to obtain a prediction model.

[0141] The prediction module 500 is used to obtain all valid grid locations of the oil reservoir to be optimized at the target time, input the key infill well location influencing factors at each valid grid location into the prediction model, and obtain the cumulative oil production prediction data or economic net present value prediction data for the entire area corresponding to each valid grid location.

[0142] The optimal location determination module 600 is used to determine the optimal infill well location based on the cumulative oil production prediction data or the economic net present value prediction data of the entire area.

[0143] Example 3

[0144] An embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, the method for intelligently optimizing infill well locations in an oil field according to the first embodiment is implemented.

[0145] In addition, this embodiment also provides an electronic device, including a memory and a processor, the memory is used to store computer programs, and the processor runs the computer programs to enable the electronic device to execute the oil field infill well location intelligent optimization method of embodiment one.

[0146] Optionally, the above-mentioned electronic device may be a server.

[0147] Embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0148] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0149] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0150] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1A step that specifies a function in one or more boxes.

[0151] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0152] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. An intelligent optimization method for infill well locations in an oil field, characterized in that: include: Randomly select a preset number of grids within the effective grid range of the target oil reservoir where no wells have been drilled, and obtain multiple spatial location layout plans for infill wells; For each of the spatial layout schemes of the infill wells, numerical simulations are performed under preset infill well conditions to obtain a variety of different infill well numerical simulation results; Determine the simulated data of the cumulative oil production of the entire area of the target oil reservoir or the simulated data of the economic net present value at a preset time according to the numerical simulation results of the infill wells; Determine the factors affecting infill well locations in the target reservoir and select key factors affecting infill well locations; For each of the spatial location layout plans of the infill wells, the key infill well location influencing factors are used as input, and the corresponding simulated data of the cumulative oil production of the entire area or the simulated data of the economic net present value is used as output to train a machine learning model to obtain a prediction model; Obtaining all valid grid locations of the oil reservoir to be optimized at the target time where no wells have been drilled, inputting the key infill well location influencing factors at each valid grid location into the prediction model, and obtaining the area-wide cumulative oil production prediction data or economic net present value prediction data corresponding to each valid grid location; Determine the optimal infill well location based on the cumulative oil production forecast data or the economic net present value forecast data for the entire area; Among them, determining the factors affecting the infill well location in the target reservoir and screening out the key infill well location influencing factors specifically includes: Preliminarily determine the factors affecting the infill well location in the target reservoir; Using a comprehensive feature selection method to select features of various influencing factors of infill well locations in the target reservoir, and obtain preliminary influencing factors; the comprehensive feature selection method includes linear correlation degree, Pearson correlation test, maximum information coefficient method, linear regression method, L1 regularization, L2 regularization, random forest and recursive feature selection method; Analyze the correlation between the primary influencing factors using the R-type clustering algorithm, eliminate the influencing factors with correlations higher than a preset correlation threshold, and obtain the key infill well location influencing factors; Among them, a comprehensive feature selection method is used to select the influencing factors of the infill well location in the target reservoir, and the preliminary influencing factors are obtained, which specifically include: Analyzing each influencing factor of the infill well location in the target reservoir using each feature selection method, and obtaining a score for each influencing factor corresponding to each feature selection method; Using each feature selection method, summing the scores of the same influencing factor of the infill well locations in the target oil reservoir to obtain a comprehensive score of each influencing factor of the infill well locations in the target oil reservoir; The preliminary influencing factors are obtained by comparing the comprehensive score of each influencing factor of the infill well location in the target oil reservoir with a preset threshold.

2. The intelligent optimization method for infill well locations in an oil field according to claim 1, characterized in that: The correlation between the primary influencing factors is analyzed using the R-type clustering algorithm, and the influencing factors with correlations higher than the preset correlation threshold are eliminated to obtain the key infill well location influencing factors, which specifically include: Determine the correlation coefficient between each two primary election influencing factors; Each primary election influencing factor is considered as a class; According to the correlation coefficients between the primary influencing factors contained in each category, the distance between each two categories is determined, and the two categories corresponding to the shortest distance among all distances are regarded as the same category; Return to step "Determine the distance between each two classes based on the correlation coefficients between the primary influencing factors contained in each class" until the cluster is finally clustered into one class. The primary influencing factors with correlation coefficients greater than a preset correlation threshold in the results of each clustering are eliminated to obtain the key encrypted well location influencing factors.

3. The intelligent optimization method for infill well locations in an oil field according to claim 2, characterized in that: The key factors affecting the location of infill wells include: overall water cut, location of infill wells, type of infill wells, perforation layer of infill wells, water cut of old wells, cumulative oil production of old wells, cumulative water production of old wells and cumulative water injection of old wells.

4. The intelligent optimization method for infill well locations in an oil field according to claim 1, characterized in that: Determine the optimal infill well location based on the cumulative oil production forecast data or economic net present value forecast data for the entire area, specifically including: The infill well location corresponding to the maximum value among all the area-wide cumulative oil production forecast data or economic net present value forecast data is determined as the optimal infill well location.

5. The intelligent optimization method for infill well locations in an oil field according to claim 1, characterized in that: Before executing the step "Obtaining all valid grid locations of undrilled wells in the reservoir to be optimized at the target time", include: Determining whether the oil reservoir to be optimized and the target oil reservoir are in the same block; If so, obtain the valid grid positions of all undrilled wells in the reservoir to be optimized at the target time; If not, a preset number of grids are randomly selected within the effective grid range of the oil reservoir to be optimized where no wells have been drilled, and corresponding spatial position layout schemes of multiple infill wells are obtained. A numerical simulation is then performed on the spatial position layout scheme of each infill well corresponding to the oil reservoir to be optimized, and simulated data of cumulative oil production or economic net present value for the entire area corresponding to the oil reservoir to be optimized is obtained. The prediction model is trained with the key infill well location influencing factors as input and the corresponding simulated data of the cumulative oil production in the entire area or the economic net present value simulation data of the oil reservoir to be optimized as output to obtain a prediction model after transfer learning, and the prediction model after transfer learning is used as the prediction model; the prediction model after transfer learning is used to predict the predicted data of the cumulative oil production in the entire area or the economic net present value prediction data of the oil reservoir to be optimized.

6. The intelligent optimization method for infill well locations in an oil field according to claim 1, characterized in that: After executing the step of "training a machine learning model for each of the spatial location layout plans of the infill wells, using the key infill well location influencing factors as input and the corresponding simulated data of the cumulative oil production of the entire area or the simulated data of the economic net present value as output to obtain a prediction model", the method further includes: Input the verification samples of the factors affecting the key infill well locations into the prediction model to obtain the corresponding verification data of the cumulative oil production of the entire area or the verification data of the economic net present value; Comparing the verified data of the total oil production of the entire area with the simulated data of the total oil production of the entire area at the same time, or comparing the verified data of the economic net present value with the simulated data of the economic net present value at the same time, to determine the prediction effect of the prediction model; If the prediction effect of the prediction model does not reach the preset effect, the number of spatial location layout schemes for infill wells is increased, and the process returns to step "randomly selecting a preset number of grids within the effective grid range of the target reservoir where no wells have been drilled"; If the prediction effect of the prediction model reaches the preset effect, the prediction model is used to predict the cumulative oil production data or the economic net present value of the entire area of the oil reservoir to be optimized.

7. An intelligent optimization system for infill well locations in oil fields, characterized by: The system comprises: A layout scheme determination module is used to randomly select a preset number of grids within the effective grid range of the target reservoir where no wells have been drilled, and obtain spatial location layout schemes for multiple infill wells; A numerical simulation module is used to perform numerical simulation on the spatial location layout of each infill well under preset infill well conditions to obtain a variety of different infill well numerical simulation results; and to determine, based on the infill well numerical simulation results, simulated data on the cumulative oil production of the entire area of the target oil reservoir at a preset time or simulated data on the economic net present value; A key influencing factor determination module is used to determine the influencing factors of the infill well locations in the target oil reservoir and screen out the key influencing factors of the infill well locations; Among them, determining the factors affecting the infill well location in the target reservoir and screening out the key infill well location influencing factors specifically includes: Preliminarily determine the factors affecting the infill well location in the target reservoir; Using a comprehensive feature selection method to select features of various influencing factors of infill well locations in the target reservoir, and obtain preliminary influencing factors; the comprehensive feature selection method includes linear correlation degree, Pearson correlation test, maximum information coefficient method, linear regression method, L1 regularization, L2 regularization, random forest and recursive feature selection method; Analyze the correlation between the primary influencing factors using the R-type clustering algorithm, eliminate the influencing factors with correlations higher than a preset correlation threshold, and obtain the key infill well location influencing factors; Among them, a comprehensive feature selection method is used to select the influencing factors of the infill well location in the target reservoir, and the preliminary influencing factors are obtained, which specifically include: Analyzing each influencing factor of the infill well location in the target reservoir using each feature selection method, and obtaining a score for each influencing factor corresponding to each feature selection method; Using each feature selection method, summing the scores of the same influencing factor of the infill well locations in the target oil reservoir to obtain a comprehensive score of each influencing factor of the infill well locations in the target oil reservoir; Determining the preliminary influencing factors based on the comparison of the comprehensive score of each influencing factor of the infill well location in the target oil reservoir with a preset threshold; A model training module is used to train a machine learning model for the spatial location layout plan of each infill well, using the key infill well location influencing factors as input and the corresponding simulated data of cumulative oil production in the entire area or the simulated data of economic net present value as output to obtain a prediction model; A prediction module is used to obtain all valid grid locations of the oil reservoir to be optimized at the target time, input the key infill well location influencing factors at each valid grid location into the prediction model, and obtain the cumulative oil production prediction data or economic net present value prediction data corresponding to each valid grid location; The optimal location determination module is used to determine the optimal infill well location based on the cumulative oil production forecast data or the economic net present value forecast data of the entire area.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the intelligent optimization method for infill well locations in an oil field as claimed in any one of claims 1 to 6 is implemented.

Images