Oil well yield intelligent prediction method and system based on integrated model
By integrating the base models of XGBoost, BPNN, and SVR, the problem of low oil well production prediction accuracy is solved, achieving higher prediction accuracy and robustness, and supporting the intelligent and digital development of oil well production.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-02
- Publication Date
- 2026-03-10
AI Technical Summary
Existing technologies for predicting oil well production have low accuracy and are prone to getting stuck in local optima, making them unsuitable for the increasing data volume and complexity required in oilfield development.
An ensemble model-based approach is adopted, combining the base models of XGBoost, BPNN, and SVR. By combining base learner layers and meta learner layers, an ensemble prediction model for oil well production is constructed. Data preprocessing and training are then performed to improve prediction accuracy and generalization performance.
It improves the accuracy and robustness of oil well production forecasting, especially under conditions of small sample size, high noise, and non-stationary sequences, and enhances the intelligence and digitalization of oil well production management.
Smart Images
Figure CN121637451A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of oil and gas development, more particularly to an intelligent oil well production prediction method and system based on an integrated model. BACKGROUND
[0002] Oil well production prediction is a key link in oilfield development, which is used to reflect the real-time dynamic changes of the reservoir, and to ensure the stable output of oil well production, accurate prediction of liquid production becomes an important task of oilfield management. In the early stage, the liquid production prediction relies on traditional empirical formula, which is convenient to calculate and easy to implement, but the prediction accuracy is low; in recent years, artificial intelligence is widely used in oilfield development and production, which is trained and analyzed through a large number of historical production sample data, and has high prediction accuracy. However, with the increase of the application range of the traditional single prediction model in the oilfield, the time span of the prediction is prolonged, and the data volume is increased, which leads to the decrease of the accuracy of the single prediction model, and the single prediction model is easy to fall into local optimal solution, which affects the prediction performance.
[0003] Therefore, how to provide an intelligent oil well production prediction method and system based on an integrated model, improve the prediction accuracy of oil well production, and provide effective assistance for realizing intelligent and digital oilfield is a problem to be solved by those skilled in the art. SUMMARY
[0004] Therefore, the present application provides an intelligent oil well production prediction method and system based on an integrated model, based on the current situation that there are too many factors affecting the accuracy of production prediction and the complexity is high, by combining the integrated model and the actual demand of oilfield development, the prediction accuracy of oil well production is improved, and the generalization is higher, which provides stronger support for oil well production management.
[0005] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows: an intelligent oil well production prediction method based on an integrated model is used to predict the production of an oil well, comprising: Collecting historical production data of the oil well; Preprocessing the historical production data of the oil well; Establishing a base model based on XGBoost algorithm, BPNN and SVR; Building an integrated model based on the base model, the integrated model comprising a base learner layer and a meta-learner layer; the base learner layer is used for learning and predicting the original features, and the meta-learner layer is used for fusing the output of the base learner layer; the integrated model is trained and tested through the preprocessed historical production data, and an integrated oil well production prediction model is obtained; After preprocessing the real-time collected oil well production data, inputting the oil well production data into the integrated oil well production prediction model, the oil well production of the next time step is predicted.
[0006] Preferably, the preprocessing includes removing outliers, imputing missing values, and normalization processing.
[0007] Preferably, the objective function of the XGBoost algorithm is as follows: ; wherein, l is a loss function; Ω( f t ) is a regularization term; and constant is a constant term. is the predicted value of the previous t-1 rounds; is the predicted value of the new tree.
[0008] Preferably, the BPNN includes a forward propagation stage and a back propagation stage. In the forward propagation stage, the network receives input data and calculates the output through multiple hidden layers layer by layer. In the back propagation stage, the error between the actual output and the target output is first calculated, and then the gradient of each layer is calculated according to the error, and then the weights and biases are updated.
[0009] Preferably, the SVR finds an optimal hyperplane in the data points and makes predictions within the allowed error range. The expression of the loss function is as follows: ; The expression of the optimization objective is as follows: ; The expression of the SVR decision function is obtained by introducing the Lagrange function as follows: ; wherein x is the model input; w is the hyperplane normal vector; b is the hyperplane intercept; and f(x) is the model output.
[0010] Preferably, a ridge regression model is used to construct the meta-learner layer, which is a linear relationship model between multiple independent variables and a dependent variable. The L2 regularization term is added to the linear regression.
[0011] Preferably, the historical production data of the oil well is divided into a training set and a test set, the integrated model is trained through the training set data, and the trained integrated model is tested by using the test set to generate an integrated prediction model of the oil well production.
[0012] Preferably, an intelligent oil well production prediction system based on an integrated model comprises: a historical data acquisition module for acquiring historical production data of an oil well; A preprocessing module is configured to preprocess historical production data of the oil well; A base model establishing module is configured to establish a base model based on an XGBoost algorithm, a BPNN and an SVR; A prediction model generating module is configured to construct an ensemble model based on the base model, the ensemble model comprising a base learner layer and a meta learner layer; the base learner layer is configured to learn and predict original features, and the meta learner layer is configured to fuse outputs of the base learner layer; the ensemble model is trained and tested based on the preprocessed historical production data to obtain an oil well production ensemble prediction model; A real-time prediction module is configured to input preprocessed real-time collected oil well production data into the oil well production ensemble prediction model to predict oil well production at a next time step.
[0013] Compared with the prior art, the oil well production intelligent prediction method and system based on an ensemble model provided by the present application combines an ensemble model and actual needs of oilfield development, has better prediction accuracy and generalization performance, provides effective assistance for realizing intelligent and digital oilfields, and makes an important contribution to sustainable development of oilfields. BRIEF DESCRIPTION OF DRAWINGS
[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of the provided drawings.
[0015] Figure 1 A flowchart of an oil well production intelligent prediction method based on an ensemble model provided by an embodiment of the present application.
[0016] Figure 2 An integrated learning model structure architecture diagram provided by an embodiment of the present application.
[0017] Figure 3 An oil well production intelligent prediction method based on an ensemble model provided by an embodiment of the present application.
[0018] Figure 4 An oil well production intelligent prediction result fitting comparison diagram provided by an embodiment of the present application.
[0019] FIG. 5(a) is a schematic diagram of a result of removing abnormal values from "stroke" data provided by an embodiment of the present application.
[0020] Fig. 5 (b) is a schematic diagram of the result of removing outliers from the "discharge rate" data according to an embodiment of the present application.
[0021] Fig. 5 (c) is a schematic diagram of the result of removing outliers from the "maximum load" data according to an embodiment of the present application.
[0022] Fig. 5 (d) is a schematic diagram of the result of removing outliers from the "minimum load" data according to an embodiment of the present application.
[0023] Fig. 5 (e) is a schematic diagram of the result of removing outliers from the "submergence" data according to an embodiment of the present application.
[0024] Fig. 5 (f) is a schematic diagram of the result of removing outliers from the "hydrodynamic surface" data according to an embodiment of the present application.
[0025] Fig. 6 (a) is a schematic diagram of the result of interpolating missing values according to an embodiment of the present application.
[0026] Fig. 6 (b) is a schematic diagram of the result of interpolating missing values according to an embodiment of the present application.
[0027] Fig. 7 (a) is a schematic diagram of the result of normalization according to an embodiment of the present application.
[0028] Fig. 7 (b) is a schematic diagram of the result of normalization according to an embodiment of the present application.
[0029] Fig. 8 (a) is a schematic diagram of the prediction process of the XGBoost algorithm according to an embodiment of the present application.
[0030] Fig. 8 (b) is a schematic diagram of the prediction process of the BPNN according to an embodiment of the present application.
[0031] Fig. 8 (c) is a schematic diagram of the prediction process of the SVR according to an embodiment of the present application. DETAILED DESCRIPTION
[0032] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.
[0033] To overcome the above problems in the prediction accuracy and generalization of the existing oil well production, the present application discloses an intelligent prediction method for oil well production based on an integrated model, which solves the problem of low prediction accuracy and the demand for efficient prediction tools in the current oil well development.
[0034] The embodiment of the present application discloses an intelligent oil well production prediction method based on an integrated model, as shown in the formula (I): Figure 1 The embodiment of the present application discloses an intelligent oil well production prediction method based on an integrated model, as shown in the formula (I): S1, oil well production data acquisition: collecting historical production data of the oil well; S2, data preprocessing: preprocessing the collected historical production data of the oil well, including removing outliers, interpolating missing values, and normalization processing; S3, base model establishment: establishing a base model based on XGBoost algorithm, BPNN and SVR; S4, integrated model establishment: constructing an integrated model based on the base model, the integrated model comprising a base learner layer and a meta learner layer; the base learner layer is used for learning and predicting original features, and the meta learner layer is used for fusing the output of the base learner layer; the integrated model is trained and tested through the preprocessed historical production data to obtain an oil well production integrated prediction model; Wherein, the first layer of the integrated model is the base learner layer, which learns and predicts the original features, and the second layer is the meta learner layer, which further learns on the basis of the first layer and fuses the advantages of multiple base models to improve the overall prediction performance; S5, integrated model prediction of oil well production: after preprocessing the real-time collected oil well production data, inputting the data into the oil well production integrated prediction model to predict the oil well production of the next time step.
[0035] The integrated model is composed of a base learner layer and a meta learner layer, which is more robust than a single prediction model, especially in the case of small sample, high noise and non-stationary sequence, and has higher robustness. The base learner layer is composed of multiple models of different types, and each model can complement the error, offset the different errors of the weak model through combination, achieve the effect of "complementing one with many", and improve the generalization ability and prediction accuracy. The method provided by the embodiment of the present application can effectively improve the accuracy and calculation efficiency of oil well production.
[0036] Specifically, in the step S1, the historical production data of the oil well is collected. The historical production data of the oil well includes stroke, stroke, suspension point displacement, suspension point load, water cut, casing pressure, dynamic liquid level, maximum load, minimum load, pump depth, pump diameter, etc., wherein the static information includes water cut, casing pressure, pump diameter, etc., and the dynamic information includes suspension point displacement, suspension point load, etc.
[0037] Specifically, the historical production data of the oil well in the step S1 is divided into static data and dynamic data, the static data includes well number, rod diameter, maximum load, pump diameter, water cut, casing pressure, oil pressure, etc., and the dynamic data includes stroke frequency, stroke, point displacement, point load, etc., the point displacement and the point load are the dynamometer card, which reflects the working condition of the oil well, in order to ensure the accuracy of the point displacement and the point load as related factors for production prediction, the dynamic data is collected once an hour.
[0038] In the step S2, the historical production data of the oil well is preprocessed, since the collected historical production data of the oil well has high-dimensional and complex nonlinear relationship, the data preprocessing is needed to avoid affecting the prediction accuracy, the operation includes removing abnormal values, interpolating missing values and data standardization.
[0039] Specifically, S2-1, for the production abnormal values, collection abnormal values and invalid data in the historical production data, the direct deletion processing is performed, and the historical production data of the oil well is re-integrated, the production abnormal values are data with too high standard deviation caused by equipment damage and production process change in the oil well production process, the collection abnormal values are abnormal data with too high standard deviation caused by sensor failure and data entry error in the collection process, and the invalid data are physically unreasonable values or data unrelated to production factors. As shown in FIGS. 5 (a)-5 (f).
[0040] S2-2, for the missing values in the historical production data, the K-nearest neighbor classification algorithm is used to interpolate the missing values, as shown in FIGS. 6 (a)-6 (b), the method calculates the distance between the adjacent parameters and the current missing value and then weights, the calculation formula is as follows: ; Wherein, is the interpolation estimate value of the jth missing attribute in the ith sample, is the observation value of the jth attribute of the kth nearest neighbor sample, is the number of selected nearest neighbor samples; The weight is usually the inverse of the distance: ; Wherein, d(x i ,x k ) is the distance between the missing sample x i and the nearest neighbor sample x k , is a very small positive number; S2-3, for the magnitude difference problem in historical production data, the data after removing outliers and imputing missing values is standardized, Min-Max normalization processing method is adopted to eliminate the influence of non-uniform dimension and time sequence data time dimension, as shown in FIG. 7(a)-7(b), the calculation formula is as follows: .
[0041] Wherein, X is the original data value, X max is the maximum value in the data set, X min is the minimum value in the data set; Specifically, in step S3, the base model is XGBoost algorithm (eXtreme Gradient Boosting), BP neural network (Backpropagation Neural Network, BPNN) and support vector regression (Support Vector Regression, SVR). XGBoost algorithm is an integrated method based on gradient boosting tree. Compared with traditional GBDT, XGBoost algorithm introduces second derivative, adds regularization term and automatically learns missing method to improve model accuracy and convergence speed. The prediction value is obtained by gradually training the additive model, as shown in FIG. 8(a).
[0042] Specifically, the objective function of XGBoost algorithm is as follows: ; Wherein, l is the loss function; Ω(f t ) is the regularization term; constant is the constant term; is the prediction value of the previous t-1 rounds; y t is the true value of the i-th sample; n is the total number of training samples; is the prediction value of the newly added tree model in the t-th round; Specifically, BPNN is a kind of multilayer feedforward neural network, which mainly includes forward propagation stage and back propagation stage. In the forward propagation stage, the network receives input data and calculates the output through multiple hidden layers layer by layer, as shown in FIG. 8(b), and the output process is expressed as follows: Hidden layer input: ; Hidden layer output: ; Output layer input: ; Network output: ; Where W and W' are the weights from the input to the hidden layer and from the hidden layer to the output layer, respectively; f(·) is the activation function; During the backpropagation phase, the error between the actual output and the target output is first calculated. Based on this error, backpropagation is performed to calculate the gradient of each layer, thereby updating the weights and biases. The output process is expressed as follows: Mean square error: ; Weights from hidden layer to input layer: ; Weights from output layer to hidden layer: .
[0043] Specifically, SVR is an algorithm based on support vector machines, mainly used for predicting continuous numerical targets. SVR finds an "optimal hyperplane," i.e., a regression function, among the data points and makes predictions within the allowable error range, as shown in Figure 8(c). The expression for the loss function is as follows: ; The expression for the optimization objective is as follows: ; in, For the true value of the i-th sample, For insensitive loss parameters, For the hyperplane normal vector, and For slack variables, This is the penalty coefficient; The SVR optimization problem is transformed into a dual problem. By introducing the Lagrangian function, the expression for the SVR decision function is obtained as follows: ; Where x is the model input; w is the hyperplane normal vector; b is the hyperplane intercept; and f(x) is the model output. and For Lagrange multipliers; Specifically, such as Figure 2 As shown in the diagram, the integrated learning model architecture of this invention, specifically step S4, establishes an oil well production prediction model based on the integrated model. The first layer is the base learner layer, comprising N base models. Its main purpose is to learn and predict from the original data, supporting further learning in the second layer. It uses the three base models from step S3 as the first layer of the integrated model. The second layer is the meta-learner layer, which further learns based on the first layer, integrating the advantages of multiple base models to improve overall prediction performance.
[0044] Specifically, the meta-learner layer more greatly affects the computational efficiency and generalization performance of the ensemble model. Since the XGBoost in the base model is a relatively complex model, a ridge regression is selected as the meta-model to reduce the computational complexity and maintain strong interpretability.
[0045] The ridge regression (RR) model is used to construct the meta-learner layer. The ridge regression model is a linear relationship model between multiple independent variables and a dependent variable, and an L2 regularization term is added on the basis of linear regression. It can effectively solve the problems of overfitting and multicollinearity, and is expressed as follows: A regularization term is added on the basis of least squares: ; The analytical solution is expressed as follows: .
[0046] Among them, is a regularization strength parameter, is an identity matrix.
[0047] Specifically, the historical production data of the oil well is divided into a training set and a test set in a ratio of 8:2. The ensemble model is trained by the training set data, and the trained ensemble model is tested by the test set to generate an integrated prediction model of the oil well production. The newly collected oil well production data is preprocessed and used as the input of the integrated model to predict the future production.
[0048] Specifically, in the step S4, as shown in Figure 3 the architecture diagram of the oil well production intelligent prediction method based on the ensemble model, the dynamic data and the static data in the step S1 are processed according to the data preprocessing method in the step S2, and then divided into a training set and a test set in a ratio of 8:2. The test set sample data is used as the input of the training model. The test set data is used to train each base model, i.e., the XGBoost, BPNN and SVR in the step S3, respectively, to obtain three training models, and then single model cross-validation is performed. Each base learner outputs a set of predicted values, which are retained as a set of new features. After rotation, each data can be used as a test set. The predicted values of the trained base learners are used as the input of the meta-learner layer, and the linear regression model is used to combine the predicted values of the base learners to find the optimal coefficient of the predicted values, thereby obtaining an intelligent prediction model of the production based on ensemble learning. Finally, the prediction model is used to predict the production data at the next time step.
[0049] As Figure 4As shown, the intelligent oil well production prediction result of the embodiment of the present application is fitted and compared after the model is trained according to the steps S1, S2, S3, S4, S5 and verified by a set of verification data sets.
[0050] The intelligent oil well production prediction method based on the integrated model provided by the embodiment of the present application can solve the problem of low accuracy caused by the influence of multiple factors on oil well production prediction. The traditional single prediction method is easy to fall into local optimum, and it is difficult to capture deep patterns in data due to the high complexity of oil well production related factors. The integrated model has a base learner layer and a meta learner layer, which realizes more accurate production prediction and provides stronger support and solutions for improving oil well production efficiency.
[0051] In one specific embodiment of the present application, an intelligent oil well production prediction system based on an integrated model comprises: a historical data acquisition module for acquiring historical production data of an oil well; a preprocessing module for preprocessing the historical production data of the oil well; a base model establishment module for establishing a base model based on XGBoost algorithm, BPNN and SVR; a prediction model generation module for constructing an integrated model based on the base model, the integrated model comprising a base learner layer and a meta learner layer; the base learner layer is used for learning and predicting original features, and the meta learner layer is used for fusing the output of the base learner layer; the integrated model is trained and tested by the preprocessed historical production data to obtain an integrated oil well production prediction model; a real-time prediction module for inputting the real-time collected oil well production data after preprocessing to the integrated oil well production prediction model to predict the oil well production of the next time step.
[0052] Each embodiment in the specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between each embodiment can be referred to each other. For the device disclosed by the embodiment, since it corresponds to the method disclosed by the embodiment, the description is relatively simple, and the related parts can be referred to the method part.
[0053] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. An intelligent prediction method of oil well production based on integrated model, characterized in that, The method comprises the following steps: collecting historical production data of an oil well; preprocessing the historical production data of the oil well; establishing base models based on XGBoost algorithm, BPNN and SVR; constructing an ensemble model based on the base models, wherein the ensemble model comprises a base learner layer and a meta-learner layer; the base learner layer is used for learning and predicting original features, and the meta-learner layer is used for fusing the output of the base learner layer; training and testing the ensemble model through the preprocessed historical production data to obtain an integrated prediction model of oil well production; inputting real-time collected oil well production data after preprocessing into the integrated prediction model of oil well production to predict the oil well production at the next time step.
2. The intelligent oil well production prediction method based on integrated model according to claim 1, characterized in that, The preprocessing comprises removing outliers, imputing missing values and normalization processing.
3. The intelligent oil well production prediction method based on integrated model according to claim 1, characterized in that, The objective function of the XGBoost algorithm is as follows: ； wherein, l is a loss function; Ω( f t ) is a regularizer; cons constant is a constant term; is the prediction value of the previous t-1 rounds; is the prediction value of the new tree.
4. The intelligent oil well production prediction method based on integrated model according to claim 1, characterized in that, The BPNN comprises a forward propagation stage and a backward propagation stage. In the forward propagation stage, the network receives input data and obtains output through layer-by-layer calculation of multiple hidden layers. In the backward propagation stage, the error between the actual output and the target output is calculated first, then the error is propagated backward to calculate the gradient of each layer, and then the weights and biases are updated.
5. The integrated model based intelligent prediction of oil well production method of claim 1, wherein, The SVR finds an optimal hyperplane in the data points and makes a prediction within the allowed error range. The expression of the loss function is as follows: ； The expression of the optimization target is as follows: ； The expression of the SVR decision function is obtained by introducing the Lagrange function as follows: ； Wherein, x is the model input; w is the hyperplane normal vector; b is the hyperplane intercept; f(x) is the model output.
6. The integrated model based intelligent prediction of oil well production method of claim 1, wherein, The meta-learner layer is constructed by using a ridge regression model, which is a linear relationship model between multiple independent variables and a dependent variable, and an L2 regularization term is added to the linear regression.
7. The integrated model based intelligent prediction of oil well production method of claim 1, wherein, The historical production data of the oil well is divided into a training set and a test set, the ensemble model is trained through the training set data, and the trained ensemble model is tested by using the test set to generate an integrated prediction model of oil well production.
8. An intelligent oil well production prediction system based on integrated model, applying the intelligent oil well production prediction method based on integrated model in any one of claims 1-7, characterized in that, The method comprises the following steps: a historical data collection module for collecting historical production data of an oil well; a preprocessing module for preprocessing the historical production data of the oil well; a base model establishment module for establishing base models based on XGBoost algorithm, BPNN and SVR; a prediction model generation module for constructing an ensemble model based on the base models, wherein the ensemble model comprises a base learner layer and a meta-learner layer; the base learner layer is used for learning and predicting original features, and the meta-learner layer is used for fusing the output of the base learner layer; training and testing the ensemble model through the preprocessed historical production data to obtain an integrated prediction model of oil well production; a real-time prediction module for inputting real-time collected oil well production data after preprocessing into the integrated prediction model of oil well production to predict the oil well production at the next time step.