Formation pressure prediction method, device and system and storage medium
By preprocessing the raw logging data and building a training model, and using machine learning algorithms to predict formation pressure, the problem of low accuracy in carbonate formation pressure prediction was solved, and high-precision drilling safety and efficiency were improved.
Patent Information
- Application Number
- CN202410622489.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-20
- Publication Date
- 2025-11-21
AI Technical Summary
Existing technologies have low accuracy and poor universality in predicting pressure in carbonate formations, are greatly affected by human factors, and are difficult to meet the requirements of engineering applications.
By preprocessing the raw logging data, a training model is constructed, a target logging pressure dataset is selected, and XGBoost, LightGBM, BP neural network and LSTM algorithm are used for model analysis to establish a target prediction model. The logging data to be predicted is then imported to predict the formation pressure.
It improves the accuracy of formation pressure prediction, enhances drilling efficiency and safety, and prevents complex situations such as blowouts and well leakage. The prediction accuracy is high, unaffected by geological environment, and does not require manual parameter adjustment.
Smart Images
Figure CN120993510A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application mainly relates to the technical field of oil and gas exploration, and particularly relates to a formation pressure prediction method, device, system and storage medium. BACKGROUND
[0002] The carbonate rock formation has great oil and gas accumulation and development potential. Research on the pore pressure of the carbonate rock formation can not only provide a basis for well site selection in a new exploration area, but also has an important role in the selection of drilling fluid density and the design of wellbore structure to achieve safe and rapid drilling.
[0003] Due to the complex formation mechanism of the carbonate rock formation pressure, the prediction accuracy of the current equivalent depth method, effective stress method, seismic data and logging data and other methods is difficult to meet the engineering requirements and has poor universality. Considering the sedimentary characteristics and cementing properties of the carbonate rock reservoir, a normal compaction curve cannot be established, from different angles, a method for predicting the formation pressure according to the relationship curve between the acoustic velocity and the difference between the confining pressure and the pore pressure, the sensitivity of the longitudinal wave velocity to the change of the formation pressure and the acoustic velocity of the rock skeleton and the formation pore fluid is proposed, but the above method can predict the carbonate rock formation pressure, but in actual engineering application, there are problems such as low precision, great influence of human factors and weak pertinence. SUMMARY
[0004] The technical problem to be solved by the present application is to provide a formation pressure prediction method, device, system and storage medium to solve the problems of the prior art.
[0005] The technical scheme for solving the above technical problem is as follows: a formation pressure prediction method, comprising the following steps:
[0006] Import a plurality of original logging data and a plurality of historical formation pressure values, respectively preprocess each of the original logging data to obtain preprocessed logging data corresponding to each of the original logging data;
[0007] Construct an original logging pressure data set from all the preprocessed logging data and all the historical formation pressure values;
[0008] Select a plurality of target logging pressure data from the original logging pressure data set, and obtain a target logging pressure data set by collecting all the target logging pressure data;
[0009] Construct a training model, perform model analysis on the training model through the target logging pressure data set to obtain a target prediction model;
[0010] Import the to-be-predicted logging data, predict the to-be-predicted logging data through the target prediction model, and obtain a formation pressure prediction result.
[0011] Another technical solution to solve the above technical problems is as follows: a formation pressure prediction device, comprising:
[0012] An import module is configured to import a plurality of original logging data and a plurality of historical formation pressure values;
[0013] A preprocessing module is configured to preprocess each of the original logging data to obtain preprocessed logging data corresponding to each of the original logging data;
[0014] A construction module is configured to construct an original logging pressure data set by using all the preprocessed logging data and all the historical formation pressure values;
[0015] A screening module is configured to screen a plurality of target logging pressure data from the original logging pressure data set and obtain a target logging pressure data set by combining all the target logging pressure data;
[0016] A model analysis module is configured to construct a training model, perform model analysis on the training model by using the target logging pressure data set, and obtain a target prediction model;
[0017] The import module is further configured to import to-be-predicted logging data;
[0018] A prediction result obtaining module is configured to predict the to-be-predicted logging data by using the target prediction model and obtain a formation pressure prediction result.
[0019] Based on the above-mentioned formation pressure prediction method, the present application further provides a formation pressure prediction system.
[0020] Another technical solution to solve the above technical problems is as follows: a formation pressure prediction system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, when the processor executes the computer program, the formation pressure prediction method described above is realized.
[0021] Based on the above-mentioned formation pressure prediction method, the present application further provides a computer readable storage medium.
[0022] Another technical solution to solve the above technical problems is as follows: a computer readable storage medium, the computer readable storage medium stores a computer program, when the computer program is executed by a processor, the formation pressure prediction method described above is realized.
[0023] The beneficial effects of the present application are: the original logging data is preprocessed to obtain preprocessed logging data, the preprocessed logging data and historical formation pressure values are used to construct an original logging pressure data set, the target logging pressure data set is screened from the original logging pressure data set, the target prediction model is obtained by analyzing the model of the training model through the target logging pressure data set, and the formation pressure prediction result is obtained by predicting the to-be-predicted logging data through the target prediction model, thereby improving the prediction accuracy of the formation pressure, and improving the efficiency and safety of drilling, meeting the engineering technology application requirements, preventing the occurrence of blowout, lost circulation and other complex situations, and at the same time, being not affected by the geological environment, not needing artificial parameter adjustment, having high prediction accuracy, and providing a new path for the prediction of carbonate formation pressure. BRIEF DESCRIPTION OF DRAWINGS
[0024] Figure 1 A flowchart of a formation pressure prediction method provided by an embodiment of the present application is shown.
[0025] Figure 2 A correlation diagram of input parameters and formation pressure of a formation pressure prediction method provided by an embodiment of the present application is shown.
[0026] Figure 3 A Pearson correlation coefficient thermogram of input parameters of a formation pressure prediction method provided by an embodiment of the present application is shown.
[0027] Figure 4 A genetic algorithm optimization flowchart of a formation pressure prediction method provided by an embodiment of the present application is shown.
[0028] Figure 5 A model performance evaluation block diagram of a formation pressure prediction method provided by an embodiment of the present application is shown.
[0029] Figure 6 A module block diagram of a formation pressure prediction device provided by an embodiment of the present application is shown. DETAILED DESCRIPTION
[0030] The principles and characteristics of the present application are described below in combination with the drawings, and the examples are only used to explain the present application and are not used to limit the scope of the present application.
[0031] Figure 1 A flowchart of a formation pressure prediction method provided by an embodiment of the present application is shown.
[0032] As shown in Figure 1 , a formation pressure prediction method comprises the following steps:
[0033] Import a plurality of original logging data and a plurality of historical formation pressure values, respectively preprocess each of the original logging data to obtain preprocessed logging data corresponding to each of the original logging data;
[0034] Construct an original logging pressure data set through all the preprocessed logging data and all the historical formation pressure values;
[0035] Screen a plurality of target logging pressure data from the original logging pressure data set, and obtain a target logging pressure data set by collecting all the target logging pressure data;
[0036] Construct a training model, perform model analysis on the training model through the target logging pressure data set to obtain a target prediction model;
[0037] Import the to-be-predicted logging data, and predict the to-be-predicted logging data through the target prediction model to obtain a formation pressure prediction result.
[0038] It should be understood that the original logging data of a plurality of wells in the Lunnan low salient of the Tarim Basin and the formation pressure (i.e. historical formation pressure value) adjusted by the measured pressure points are collected and arranged, and the original logging data (i.e. original logging data) is preprocessed to establish a data set (i.e. original logging pressure data set) for intelligent prediction of formation pressure.
[0039] In the above embodiment, the preprocessed logging data is obtained by preprocessing the original logging data, the original logging pressure data set is constructed by the preprocessed logging data and the historical formation pressure value, the target logging pressure data set is screened from the original logging pressure data set, the target prediction model is obtained by model analysis of the training model through the target logging pressure data set, and the formation pressure prediction result is obtained by predicting the to-be-predicted logging data through the target prediction model, which improves the prediction accuracy of the formation pressure, and further improves the efficiency and safety of drilling, meets the engineering and technical application requirements, prevents the occurrence of complex conditions such as blowout and lost circulation, is not affected by the geological environment, does not need artificial parameter adjustment, has high prediction accuracy, and provides a new path for prediction of carbonate formation pressure.
[0040] Optionally, as an embodiment of the present application, the process of respectively preprocessing each of the original logging data to obtain preprocessed logging data corresponding to each of the original logging data comprises:
[0041] Respectively clean the abnormal values of each of the original logging data to obtain preprocessed logging data corresponding to each of the original logging data;
[0042] The missing values of each of the processed logging data are filled respectively to obtain the pretreated logging data corresponding to each of the original logging data.
[0043] It should be understood that the data preprocessing operation mainly includes abnormal value cleaning and missing value supplementing.
[0044] Specifically, abnormal value cleaning is an important step in data analysis, which involves identifying and processing outliers or outliers in a data set. Outliers refer to data points that are significantly different from most values in the data set, which may be caused by measurement errors, data input errors, or real variations. The existence of outliers can have a significant impact on the results of data analysis.
[0045] It should be understood that the filling of missing values, i.e. missing value processing, is an important part of data preprocessing, which aims to solve the missing or unknown data points in the data set. The method of processing missing values needs to be selected according to the characteristics of the data, the type of missing data and the analysis target.
[0046] In the above embodiment, the original logging data is pretreated to obtain pretreated logging data, which improves the accuracy of predicting the formation pressure, and further improves the efficiency and safety of drilling, meets the engineering and technical application requirements, and prevents the occurrence of blowout, leakage and other complex situations.
[0047] Optionally, as one embodiment of the present application, as shown in Figures 1 to 3 The process of screening a plurality of target logging pressure data from the original logging pressure data set includes:
[0048] Each of the pretreated logging data in the original logging pressure data set and any of the remaining original logging pressure data in the original logging pressure data set is taken as a group, thereby obtaining a plurality of original logging pressure data groups;
[0049] The correlation coefficient of each of the original logging pressure data groups is calculated by the first formula, thereby obtaining the Spearman correlation coefficient corresponding to each of the original logging pressure data groups, and the first formula is:
[0050]
[0051] Wherein,
[0052] Wherein, is the Spearman correlation coefficient corresponding to the a-th original logging pressure data group, cov(x a ,y a) is the covariance of the original logging pressure data x and the original logging pressure data y in the a-th original logging pressure data set, and n is the total number of the original logging pressure data sets, is the standard deviation of the original logging pressure data x in the a-th original logging pressure data set, is the standard deviation of the original logging pressure data y in the a-th original logging pressure data set, is the Pearson correlation coefficient corresponding to the a-th original logging pressure data set, x a is the original logging pressure data x in the a-th original logging pressure data set, y a is the original logging pressure data y in the a-th original logging pressure data set;
[0053] When the Spearman correlation coefficient is greater than or equal to a preset correlation coefficient threshold, all of the original logging pressure data in the original logging pressure data set corresponding to the Spearman correlation coefficient are taken as target logging pressure data, thereby obtaining a plurality of target logging pressure data.
[0054] It should be understood that, by the degree of association, the parameters such as “DEPTH”, “DEN”, “DT”, “SP”, “DEVI”, “AZIM”, “CAL”, “GR”, “RT”, and “RM” in the logging data of multiple wells in a block are selected, the Spearman function and the Pearson function are used to calculate the correlation coefficients, the optimal parameters having a stronger correlation with the formation pressure and a weaker linear relationship with each other are obtained, and it is ensured that overfitting or unstable parameter values do not occur.
[0055] Specifically, the calculation formulas of the Spearman function and the Pearson function are as follows:
[0056]
[0057]
[0058] In the formula, ρ pearson , ρ Spearson is the Pearson correlation coefficient and the Spearman correlation coefficient; x and y are variables, cov(x, y) is the covariance of x and y, σ x , σ y is the standard deviation of x and y.
[0059] In the above embodiments, the plurality of target logging pressure data are selected from the original logging pressure data set, it is ensured that overfitting or unstable parameter values do not occur, the accuracy of predicting the formation pressure is improved, and the efficiency and safety of drilling are improved.
[0060] Optionally, as an embodiment of the present application, the process of constructing a training model, performing model analysis on the training model by the target logging pressure data set, and obtaining a target prediction model comprises:
[0061] dividing the target logging pressure data set into a logging pressure training set and a logging pressure test set according to a preset ratio;
[0062] constructing a first original prediction model based on an XGBoost algorithm, a second original prediction model based on a LightGBM algorithm, a third original prediction model based on a BP neural network algorithm, and a fourth original prediction model based on an LSTM algorithm;
[0063] training the first original prediction model according to the logging pressure training set to obtain a first trained prediction model;
[0064] training the second original prediction model according to the logging pressure training set to obtain a second trained prediction model;
[0065] training the third original prediction model according to the logging pressure training set to obtain a third trained prediction model;
[0066] training the fourth original prediction model according to the logging pressure training set to obtain a fourth trained prediction model;
[0067] testing the first trained prediction model, the second trained prediction model, the third trained prediction model, and the fourth trained prediction model according to the logging pressure test set, respectively, to obtain a plurality of first prediction values corresponding to the first trained prediction model, a plurality of second prediction values corresponding to the second trained prediction model, a plurality of third prediction values corresponding to the third trained prediction model, and a plurality of fourth prediction values corresponding to the fourth trained prediction model;
[0068] importing a plurality of true values, scoring the plurality of first prediction values corresponding to the first trained prediction model, the plurality of second prediction values corresponding to the second trained prediction model, the plurality of third prediction values corresponding to the third trained prediction model, and the plurality of fourth prediction values corresponding to the fourth trained prediction model according to all the true values, respectively, to obtain a first score corresponding to the first trained prediction model, a second score corresponding to the second trained prediction model, a third score corresponding to the third trained prediction model, and a fourth score corresponding to the fourth trained prediction model;
[0069] Screen the maximum value among the first score, the second score, the third score and the fourth score, and take the first trained prediction model or the second trained prediction model or the third trained prediction model or the fourth trained prediction model corresponding to the screened result as the target prediction model.
[0070] Preferably, the preset ratio can be 7:3 or 8:2.
[0071] It should be understood that the XGBoost algorithm, the LightGBM algorithm, the BP neural network algorithm and the LSTM algorithm learner (i.e. the LSTM algorithm) are trained.
[0072] It should be understood that the data processing process of training the first original prediction model according to the logging pressure training set is the same as the data processing process of training the second original prediction model according to the logging pressure training set, training the third original prediction model according to the logging pressure training set and training the fourth original prediction model according to the logging pressure training set, only the processed prediction model is different.
[0073] It should be understood that the XGBoost algorithm (eXtreme Gradient Boosting) is based on the CART regression tree model, and the final prediction result is the weighted sum of the outputs of each CART model. Taking logging data as input, the CART tree model in the XGBoost model is established, and the CART tree calculates the weight corresponding to each leaf node through training.
[0074] Specifically, the LightGBM algorithm (Light Gradient Boosting Machine) is a machine learning algorithm based on gradient boosting trees, which has advantages in training speed and memory usage. It is particularly suitable for training tasks with large-scale data sets and high-dimensional features, and can obtain good model performance in a relatively short time.
[0075] It should be understood that the BP neural network (Backpropagation Neural Network) (i.e. the BP neural network algorithm) is a common artificial neural network model used to solve classification, regression and function approximation problems. It uses the backpropagation algorithm for training, and adjusts the weights and bias terms in the network to reduce the error between the predicted value and the true value.
[0076] Specifically, the LSTM algorithm (Recurrent Neural Network, RNN) is an improved recurrent neural network that is very sensitive to short-term inputs, but easily forgets longer information when faced with long-term inputs. To perform better on long sequence problems, LSTM adds memory units to each neuron in the hidden layer, thus making the memory information on the time series controllable.
[0077] In the above embodiments, the target prediction model is obtained by analyzing the training model using the target logging pressure dataset, which improves the accuracy of formation pressure prediction, thereby improving drilling efficiency and safety, meeting the requirements of engineering and technical applications, preventing complex situations such as blowouts and well leakage. At the same time, it is not affected by the geological environment, does not require manual parameter adjustment, has high prediction accuracy, and can provide a new path for the prediction of carbonate rock formation pressure.
[0078] Optionally, as an embodiment of the present invention, such as Figure 1 and 4 As shown, the process of training the first original prediction model based on the well logging pressure training set to obtain the first trained prediction model includes:
[0079] S431: Initialize the population, which includes multiple initial individuals;
[0080] S432: Train the first original prediction model based on the well logging pressure training set and each of the original individuals to obtain the initial fitness value corresponding to each of the original individuals;
[0081] S433: Filter out the maximum value of all the initial fitness values, and obtain the maximum fitness value after filtering;
[0082] S434: Determine whether the maximum fitness value is greater than or equal to a preset first fitness threshold. If not, execute S435. If yes, update the parameters of the first original prediction model according to the maximum fitness value, and use the first original prediction model with updated parameters as the first post-training prediction model.
[0083] S435: When the initial fitness value is greater than or equal to the preset second fitness threshold, the original individual corresponding to the initial fitness value is taken as the individual to be processed, thereby obtaining multiple individuals to be processed;
[0084] S436: Perform cross-processing on all the individuals to be processed to obtain multiple cross-processed individuals;
[0085] S437: Perform mutation processing on all individuals after the crossover, and return to S432 after mutation processing.
[0086] Understandably, the steps of a genetic algorithm are as follows: determine the input and output variables, and transform the optimal solution problem into a population reproduction problem through encoding (the invention adopts binary encoding); randomly generate individuals and set up the population; determine the fitness function; perform selection, crossover, and mutation operations; calculate the fitness value and eliminate individuals; if the iteration stopping condition is met, the optimal solution is obtained; otherwise, continue to perform selection, crossover, mutation, and other operations. Based on the theory of genetic algorithms, initialize the population to generate individuals.
[0087] Specifically, the genetic algorithm population is initialized to generate individuals; the well logging pressure training set is used as the input parameters for each machine learning algorithm to train the learner of each algorithm, and the error value obtained from the training is used as the initial fitness value; at this time, it is determined whether the fitness value meets the requirements for entering the next round of iteration. If it does not meet the requirements, the fitness value is adjusted in time, and the iteration continues to enter a new round of iteration of the genetic algorithm, that is, selection, crossover and mutation operations to generate new fitness values; if it meets the requirements, the optimal solution of output parameters is obtained.
[0088] In the above embodiments, the first original prediction model is trained based on the well logging pressure training set to obtain the first post-trained prediction model. Compared with traditional optimization algorithms, the present invention has strong robustness and improves global search characteristics and implicit parallelism.
[0089] Optionally, as an embodiment of the present invention, the process of scoring the plurality of first predicted values corresponding to the first post-trained prediction model, the plurality of second predicted values corresponding to the second post-trained prediction model, the plurality of third predicted values corresponding to the third post-trained prediction model, and the plurality of fourth predicted values corresponding to the fourth post-trained prediction model based on all the true values, to obtain a first score corresponding to the first post-trained prediction model, a second score corresponding to the second post-trained prediction model, a third score corresponding to the third post-trained prediction model, and a fourth score corresponding to the fourth post-trained prediction model includes:
[0090] The second formula and all the true values are used to score the multiple first predicted values corresponding to the first post-trained prediction model, the multiple second predicted values corresponding to the second post-trained prediction model, the multiple third predicted values corresponding to the third post-trained prediction model, and the multiple fourth predicted values corresponding to the fourth post-trained prediction model, respectively, to obtain the first score corresponding to the first post-trained prediction model, the second score corresponding to the second post-trained prediction model, the third score corresponding to the third post-trained prediction model, and the fourth score corresponding to the fourth post-trained prediction model. The second formula is:
[0091] M=α*R 2 +β*RMSE+γ*MAE,
[0092] in,
[0093] Where M represents the first, second, third, or fourth score, α, β, and γ are all preset weights, and R... 2 The coefficient of determination is RMSE, the root mean square error is MAE, N is the total number of the first, second, third, or fourth predicted values, or the total number of actual values, and y is the mean absolute error. b For the b-th true value, It can be the b-th first predicted value, the b-th second predicted value, the b-th third predicted value, or the b-th fourth predicted value. The average of all first-predictions, or the average of all second-predictions, or the average of all third-predictions, or the average of all fourth-predictions.
[0094] It should be understood that the coefficient of determination (R²) is used. 2 The root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate the performance of the regression model for predicting formation pressure.
[0095] Specifically, higher R 2 This means the model can explain the variability of the dependent variable well. The root mean square error and mean absolute error reflect the difference between the model's predicted values and the actual observed values. The smaller the value, the higher the model's fit. The calculation method is as follows:
[0096]
[0097]
[0098]
[0099] In the above embodiments, the first predicted value, the second predicted value, the third predicted value, and the fourth predicted value are scored based on the actual values to obtain the first score, the second score, the third score, and the fourth score. The present invention is not affected by the geological environment, does not require manual parameter adjustment, has high prediction accuracy, and can provide a new path for the prediction of carbonate rock formation pressure.
[0100] Alternatively, as another embodiment of the present invention, the prediction of formation pressure has an important impact on drilling operations and downhole equipment. However, conventional methods have low accuracy and are greatly affected by human factors. Therefore, it is necessary to use a combination of machine learning methods and logging data to predict formation pressure, so as to achieve preliminary understanding and control of it and ensure the safety and effectiveness of drilling operations.
[0101] Optionally, as another embodiment of the present invention, the purpose of the present invention is to determine the correlation and linear relationship of input parameters based on well logging data. Based on the theories of XGBoost algorithm, LightGBM algorithm, BP neural network algorithm, and LSTM algorithm, the algorithm is optimized by genetic algorithm to determine the parameters that are strongly correlated with formation pressure and have weak linear relationships with each other. An optimized machine learning algorithm is then established to predict carbonate formation pressure, providing a model reference for drilling and completion design and oil and gas development and construction measures.
[0102] Optionally, as another embodiment of the present invention, the input parameters are standardized to eliminate the influence of different dimensions. Standardization is performed using equation c, and the standardized data features conform to a normal distribution. The quantized features will be distributed in the interval [-1, 1], as shown in the following formula;
[0103]
[0104] In the formula: μ and δ represent the mean and standard deviation of each parameter, respectively;
[0105] i represents different feature parameters, and j represents a sample for each parameter.
[0106] Optionally, as another embodiment of the present invention, the present invention trains initial fitness values according to each algorithm learner; judges based on the fitness values, and obtains the optimal solution if the iteration stopping condition is met, otherwise continues to perform selection, crossover, mutation and other operations; inputs the output optimal solution as test sample data into each algorithm trainer; constructs a formation pressure prediction regression model based on the input sample data; and outputs the prediction results in the prediction model.
[0107] Optionally, as another embodiment of the present invention, the model is validated using LT-1 well data within the block. After selecting and standardizing the features of the LT-1 well data, numerical initialization is performed on the XGBoost algorithm, LightGBM algorithm, BP neural network algorithm, and LSTM algorithm, respectively, and these initialization values are encoded using a genetic algorithm binary encoding.
[0108] Optionally, as another embodiment of the present invention, after terminating the genetic operation, test sample data is input into the machine learning algorithm, and a statistical regression model of logging lithology parameters and logging pore pressure is established using machine learning methods; the determination coefficient (R²) of each algorithm model is calculated. 2The algorithm calculates the root mean square error (RMSE) and mean absolute error (MAE), outputs the calculation results, and evaluates the regression model. After terminating the algorithm, it evaluates the optimized algorithm model and obtains a model with a higher fit to the actual data. To further verify the accuracy of the intelligent algorithm model in predicting formation pressure, the model with a higher fit is used to predict the formation pressure of development well X with measured formation pressure points in the block. The prediction results are compared and analyzed with the calculation results of the commonly used Eaton method and the traditional effective stress method.
[0109] Optionally, as another embodiment of the present invention, this invention couples well logging data with machine learning algorithm theory. Specifically, it uses Spearman and Pearson functions to determine the correlation and linearity of parameters in the well logging data, and then uses a genetic algorithm to optimize XGBoost, LightGBM, BP neural network, and LSTM algorithms. Currently, there is no precedent for using a method that combines feature selection based on engineering well logging data with further optimization of machine learning algorithm theory using genetic algorithms to predict carbonate reservoir formation pressure. Only individual theories, experiments, or well logging data can be used to obtain carbonate reservoir formation pressure. This invention utilizes the above two techniques and well logging data already obtained in the field to predict and determine carbonate reservoir formation pressure. Based on existing well logging data obtained in the field, and using machine learning algorithm theory, this invention determines the input parameters and optimization algorithm, ultimately proving that the model has high accuracy in predicting carbonate reservoir formation pressure, meeting the requirements of engineering applications.
[0110] Furthermore, predicting carbonate formation pressure during drilling can help prevent complex situations such as blowouts and lost circulation. Simultaneously, considering the influence of geological structural features and formation lithology on formation pressure, feature selection allows input parameters to more accurately adapt to current formation conditions, thereby improving formation pressure accuracy and ultimately enhancing drilling efficiency and safety. Excessive or insufficient formation pressure can create difficulties and risks in drilling, potentially leading to problems such as lost circulation, well collapse, overflow, well kick, poor cementing, and even blowouts. Therefore, accurately determining the corresponding formation pressure based on geological exploration and logging data before drilling is crucial.
[0111] Secondly, this technique significantly improves prediction accuracy compared to the Eaton method and the traditional effective stress method. This method is unaffected by geological conditions, requires no manual parameter adjustment, and offers high prediction accuracy, providing a new approach for predicting pressure in carbonate formations.
[0112] Alternatively, as another embodiment of the present invention, such as Figure 3As shown, there is no significant linear relationship between the input parameters of this invention, which ensures that the model will not overfit or produce unstable estimates.
[0113] Alternatively, as another embodiment of the present invention, such as Figure 5 As shown, the present invention is specifically as follows:
[0114] (1) Raw logging data from multiple wells in the Tarim Basin's northern uplift and southern low uplift, along with formation pressure adjusted based on measured pressure points, were collected and processed. Data preprocessing mainly included outlier removal and missing value supplementation. This led to the establishment of a dataset for intelligent formation pressure prediction.
[0115] (2) Multiple rounds of training were conducted on multiple wells within the block. The correlation coefficients were calculated based on the Spearman function and the Pearson function, and the optimal correlation parameters for the formation pressure prediction model were selected. The hyperparameter selections for each model are shown in Table 1. Table 1 shows the hyperparameter selections for each model, and the parameters were standardized.
[0116] Table 1
[0117]
[0118]
[0119] (3) The model was validated using the LT-1 well in the block. Based on the correlation degree, input parameters such as “DEPTH”, “DEN”, “DT”, “SP”, “DEVI”, “AZIM”, “CAL”, “GR”, “RT”, and “RM” were selected from the well logging data. Genetic algorithms were used to optimize the XGBoost algorithm, LightGBM algorithm, BP neural network algorithm and LSTM algorithm.
[0120] (4) Evaluate the coefficient of determination (R²) of the regression model. 2 The root mean square error (RMSE) and mean absolute error (MAE) were used to evaluate the optimized regression models. It was found that the accuracy of all machine learning models using default parameters was below 92%, while the accuracy of each algorithm was significantly improved after optimization. Among them, the XGBoost model optimized by the genetic algorithm achieved the highest accuracy. 2 The accuracy was 96.99%, RMSE was 0.013%, and MAE was 0.54%, which is significantly higher than that of other model algorithms such as LightGBM, BP neural network, and LSTM.
[0121] (5) To further verify the reliability of the intelligent formation pressure prediction model constructed in this invention, the XGBoost method optimized by a genetic algorithm was used to predict the formation pressure of development well X with measured formation pressure points within the block. The prediction results were compared and analyzed with the calculation results of the commonly used Eaton method and the traditional effective stress method. The results showed that the formation pressure prediction results based on the XGBoost model optimized by the genetic algorithm were more accurate in matching the actual measurement points than those of the traditional methods.
[0122] Figure 6 This is a block diagram of a formation pressure prediction device provided in an embodiment of the present invention.
[0123] Alternatively, as another embodiment of the present invention, such as Figure 6 As shown, a formation pressure prediction device includes:
[0124] The import module is used to import multiple raw well logging data and multiple historical formation pressure values;
[0125] The preprocessing module is used to preprocess each of the original logging data to obtain preprocessed logging data corresponding to each of the original logging data.
[0126] The construction module is used to construct the original well logging pressure dataset from all the preprocessed well logging data and all the historical formation pressure values;
[0127] The filtering module is used to filter out multiple target well logging pressure data from the original well logging pressure dataset, and combine all the target well logging pressure data to obtain a target well logging pressure dataset;
[0128] The model analysis module is used to build a training model and perform model analysis on the training model using the target well logging pressure dataset to obtain a target prediction model.
[0129] The import module is also used to import well logging data to be predicted;
[0130] The prediction result acquisition module is used to predict the well logging data to be predicted using the target prediction model to obtain the formation pressure prediction result.
[0131] Optionally, as an embodiment of the present invention, the preprocessing module includes the following steps:
[0132] Outlier cleanup is performed on each of the original logging data to obtain processed logging data corresponding to each of the original logging data.
[0133] Missing values are filled in for each of the processed logging data to obtain preprocessed logging data corresponding to each of the original logging data.
[0134] Optionally, another embodiment of the present invention provides a formation pressure prediction system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the formation pressure prediction method as described above. This system can be a computer or similar system.
[0135] Optionally, another embodiment of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the formation pressure prediction method as described above.
[0136] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0137] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described apparatus and unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.
[0138] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.
[0139] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.
[0140] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0141] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0142] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for predicting formation pressure, characterized in that, Includes the following steps: Import multiple raw logging data and multiple historical formation pressure values, and preprocess each of the raw logging data to obtain preprocessed logging data corresponding to each of the raw logging data. The original logging pressure dataset is constructed using all the preprocessed logging data and all the historical formation pressure values. Multiple target well logging pressure data are selected from the original well logging pressure dataset, and all the target well logging pressure data are combined to obtain the target well logging pressure dataset; A training model is constructed, and the training model is analyzed using the target well logging pressure dataset to obtain the target prediction model; Import the logging data to be predicted, and use the target prediction model to predict the logging data to obtain the formation pressure prediction result.
2. The formation pressure prediction method according to claim 1, characterized in that, The process of preprocessing each of the original logging data to obtain preprocessed logging data corresponding to each of the original logging data includes: Outlier cleanup is performed on each of the original logging data to obtain processed logging data corresponding to each of the original logging data. Missing values are filled in for each of the processed logging data to obtain preprocessed logging data corresponding to each of the original logging data.
3. The formation pressure prediction method according to claim 1, characterized in that, The process of filtering multiple target well logging pressure data from the original well logging pressure dataset includes: Each of the preprocessed logging data in the original logging pressure dataset is grouped together with any remaining original logging pressure data in the original logging pressure dataset to obtain multiple original logging pressure data groups. By calculating the correlation coefficient for each of the original well logging pressure data sets using the first formula, the Spearman correlation coefficient corresponding to each of the original well logging pressure data sets is obtained. The first formula is: in, in, Let cov(x) be the Spearman correlation coefficient corresponding to the a-th raw well logging pressure data set. a ,y a Let be the covariance between the original well logging pressure data x and the original well logging pressure data y in the a-th original well logging pressure data set, and n be the total number of original well logging pressure data sets. Let x be the standard deviation of the original well logging pressure data in the a-th original well logging pressure data set. Let y be the standard deviation of the raw well logging pressure data in the a-th raw well logging pressure data set. Let x be the Pearson correlation coefficient corresponding to the a-th set of original well logging pressure data. a Let x and y be the raw logging pressure data in the a-th raw logging pressure data set. a Let y be the original well logging pressure data in the a-th original well logging pressure data group; When the Spearman correlation coefficient is greater than or equal to a preset correlation coefficient threshold, all the original logging pressure data in the original logging pressure data group corresponding to the Spearman correlation coefficient are used as target logging pressure data, thereby obtaining multiple target logging pressure data.
4. The formation pressure prediction method according to claim 1, characterized in that, The process of constructing a training model and performing model analysis on the training model using the target well logging pressure dataset to obtain the target prediction model includes: The target logging pressure dataset is divided into a logging pressure training set and a logging pressure test set according to a preset ratio. The first original prediction model is constructed based on the XGBoost algorithm, the second original prediction model is constructed based on the LightGBM algorithm, the third original prediction model is constructed based on the BP neural network algorithm, and the fourth original prediction model is constructed based on the LSTM algorithm. The first original prediction model is trained based on the well logging pressure training set to obtain the first trained prediction model; The second original prediction model is trained based on the well logging pressure training set to obtain the second trained prediction model; The third original prediction model is trained based on the well logging pressure training set to obtain the third trained prediction model; The fourth original prediction model is trained based on the well logging pressure training set to obtain the fourth trained prediction model; The first post-training prediction model, the second post-training prediction model, the third post-training prediction model, and the fourth post-training prediction model are tested according to the well logging pressure test set to obtain multiple first prediction values corresponding to the first post-training prediction model, multiple second prediction values corresponding to the second post-training prediction model, multiple third prediction values corresponding to the third post-training prediction model, and multiple fourth prediction values corresponding to the fourth post-training prediction model. Multiple real values are imported, and scores are assigned to multiple first predicted values corresponding to the first post-trained prediction model, multiple second predicted values corresponding to the second post-trained prediction model, multiple third predicted values corresponding to the third post-trained prediction model, and multiple fourth predicted values corresponding to the fourth post-trained prediction model based on all the real values, to obtain a first score corresponding to the first post-trained prediction model, a second score corresponding to the second post-trained prediction model, a third score corresponding to the third post-trained prediction model, and a fourth score corresponding to the fourth post-trained prediction model; The maximum value among the first score, the second score, the third score, and the fourth score is selected, and the first, second, third, or fourth post-training prediction model corresponding to the selected result is taken as the target prediction model.
5. The formation pressure prediction method according to claim 4, characterized in that, The process of training the first original prediction model based on the well logging pressure training set to obtain the first trained prediction model includes: S431: Initialize the population, which includes multiple initial individuals; S432: Train the first original prediction model based on the well logging pressure training set and each of the original individuals to obtain the initial fitness value corresponding to each of the original individuals; S433: Filter out the maximum value of all the initial fitness values, and obtain the maximum fitness value after filtering; S434: Determine whether the maximum fitness value is greater than or equal to a preset first fitness threshold. If not, execute S435. If yes, update the parameters of the first original prediction model according to the maximum fitness value, and use the first original prediction model with updated parameters as the first post-training prediction model. S435: When the initial fitness value is greater than or equal to the preset second fitness threshold, the original individual corresponding to the initial fitness value is taken as the individual to be processed, thereby obtaining multiple individuals to be processed; S436: Perform cross-processing on all the individuals to be processed to obtain multiple cross-processed individuals; S437: Perform mutation processing on all individuals after the crossover, and return to S432 after mutation processing.
6. The formation pressure prediction method according to claim 4, characterized in that, The process of scoring the multiple first predicted values corresponding to the first post-trained prediction model, the multiple second predicted values corresponding to the second post-trained prediction model, the multiple third predicted values corresponding to the third post-trained prediction model, and the multiple fourth predicted values corresponding to the fourth post-trained prediction model based on all the true values, to obtain the first score corresponding to the first post-trained prediction model, the second score corresponding to the second post-trained prediction model, the third score corresponding to the third post-trained prediction model, and the fourth score corresponding to the fourth post-trained prediction model includes: The second formula and all the true values are used to score the multiple first predicted values corresponding to the first post-trained prediction model, the multiple second predicted values corresponding to the second post-trained prediction model, the multiple third predicted values corresponding to the third post-trained prediction model, and the multiple fourth predicted values corresponding to the fourth post-trained prediction model, respectively, to obtain the first score corresponding to the first post-trained prediction model, the second score corresponding to the second post-trained prediction model, the third score corresponding to the third post-trained prediction model, and the fourth score corresponding to the fourth post-trained prediction model. The second formula is: M=α*R 2 +β*RMSE+γ*MAE, in, Where M represents the first, second, third, or fourth score, α, β, and γ are all preset weights, and R... 2 The coefficient of determination is RMSE, the root mean square error is MAE, N is the total number of the first, second, third, or fourth predicted values, or the total number of actual values, and y is the mean absolute error. b For the b-th true value, It can be the b-th first predicted value, the b-th second predicted value, the b-th third predicted value, or the b-th fourth predicted value. The average of all first-predictions, or the average of all second-predictions, or the average of all third-predictions, or the average of all fourth-predictions.
7. A formation pressure prediction device, characterized in that, include: The import module is used to import multiple raw well logging data and multiple historical formation pressure values; The preprocessing module is used to preprocess each of the original logging data to obtain preprocessed logging data corresponding to each of the original logging data. The construction module is used to construct the original well logging pressure dataset from all the preprocessed well logging data and all the historical formation pressure values; The filtering module is used to filter out multiple target well logging pressure data from the original well logging pressure dataset, and combine all the target well logging pressure data to obtain a target well logging pressure dataset. The model analysis module is used to build a training model and perform model analysis on the training model using the target well logging pressure dataset to obtain a target prediction model. The import module is also used to import well logging data to be predicted; The prediction result acquisition module is used to predict the well logging data to be predicted using the target prediction model to obtain the formation pressure prediction result.
8. The formation pressure prediction device according to claim 7, characterized in that, The preprocessing module includes the following steps: Outlier cleanup is performed on each of the original logging data to obtain processed logging data corresponding to each of the original logging data. Missing values are filled in for each of the processed logging data to obtain preprocessed logging data corresponding to each of the original logging data.
9. A formation pressure prediction system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the formation pressure prediction method as described in any one of claims 1 to 6.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the formation pressure prediction method as described in any one of claims 1 to 6.