Model integration prediction method for shale reservoir horizontal well post-fracturing yield
By calculating the Pearson correlation coefficient and the integrated learning algorithm, an integrated prediction model for yield after horizontal well compression in shale reservoirs is constructed, which solves the problems of low prediction accuracy and algorithm differences in traditional models, and improves the prediction accuracy and confidence.
Patent Information
- Application Number
- CN202411052966.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-02
- Publication Date
- 2024-11-12
AI Technical Summary
Traditional seepage models are difficult to accurately predict the post-pressure output of horizontal wells in the shale reservoir. There are differences and low confidence in the prediction process of machine learning algorithms, resulting in low prediction accuracy.
The evaluation index value of the Pearson correlation coefficient computer learning model was adopted, and a model with an evaluation index value greater than 1.50 was selected as the base model. The shale reservoir horizontal well pressure production ensemble prediction model was constructed through an integrated learning algorithm, and the main control production parameters were used for training. , improve prediction accuracy.
It improves the generalization performance and adaptability of machine learning prediction methods, enhances the accuracy of horizontal well pressure prediction of shale reservoirs, reduces the risk of local optimality, and improves the credibility of the prediction results.
Smart Images

Figure CN118940903A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of oil and gas reservoir development engineering, and in particular to a model integrated prediction method for the post-pressure production of horizontal wells in shale reservoirs. Background Art
[0002] Capacity prediction is the core basis for decision-making and production management in oilfield development, and is also one of the keys to improving the development benefits and resource utilization efficiency of oil reservoirs. However, shale reservoirs are relatively dense and have extremely poor physical properties, which makes the "horizontal well + volume fracturing" technology a common method for the development of such reservoirs. The transformation of the reservoir has affected the post-fracturing production of horizontal wells in shale reservoirs in many ways. The post-fracturing production refers to the initial production of the shale reservoir after fracturing transformation. The capacity calculation results based on the traditional seepage model are not ideal and it is difficult to meet the accuracy requirements. With the continuous development of machine learning algorithms, predicting production data by building intelligent prediction models has become a common capacity prediction method, but different algorithms have certain differences in the prediction process and different sensitivities to data, resulting in different prediction accuracies and reduced credibility of prediction results. Summary of the invention
[0003] In order to solve the above technical problems, the present invention provides a model integrated prediction method for the post-pressure production of horizontal wells in shale reservoirs.
[0004] The present invention is achieved through the following technical solutions:
[0005] A model integration prediction method for the post-pressure production of a horizontal well in a shale reservoir comprises the following steps:
[0006] Based on the Pearson correlation coefficient, the evaluation index values of various machine learning prediction models for the prediction of horizontal well production after pressure pumping in shale reservoirs are calculated respectively;
[0007] Select the machine learning prediction model with the evaluation index value greater than 1.50 as the base model, and use the integrated learning algorithm to construct an integrated prediction model for the post-pressure production of horizontal wells in shale reservoirs;
[0008] The integrated prediction model for the production of horizontal wells in the shale reservoir after being pressurized is trained by using the main control production parameters to obtain a trained integrated prediction model for the production of horizontal wells in the shale reservoir after being pressurized;
[0009] The trained integrated prediction model for horizontal well production after pressure reduction in shale reservoirs is used to predict the horizontal well production after pressure reduction in shale reservoirs.
[0010] Preferably, the main control production parameters include dynamic liquid level, number of segment clusters, amount of sand added per meter, amount of liquid added per meter, soaking time, and length of horizontal well segment.
[0011] The calculation formula of Pearson correlation coefficient is:
[0012]
[0013] In the formula, x i is the normalized parameter affecting the production of horizontal wells in shale reservoirs after pressure reduction; i is the measured value of the initial oil production of the horizontal well; is the average value of the initial oil production of the horizontal well; is the predicted value of the initial oil production of the horizontal well; n is the number of horizontal wells.
[0014] The calculation formula of the evaluation index value is:
[0015]
[0016] Where, d i is the evaluation index value of the i-th prediction model, dimensionless; ρ iu is the correlation between the prediction results of the ith prediction model and the actual horizontal well production after fracturing in shale reservoirs; ρ ij is the correlation coefficient between the ith prediction model and the jth prediction model, and n is the number of prediction models.
[0017] Optionally, the evaluation index values of 8 machine learning prediction models for predicting the post-pressure production of horizontal wells in shale reservoirs are calculated respectively.
[0018] Preferably, the eight machine learning prediction models are: support vector machine, generalized linear model, multilayer perceptron, random forest, decision tree, XGBoost, BP neural network, convolutional neural network. Preferably, convolutional neural network, XGBoost, decision tree, generalized linear model, multilayer perceptron model, random forest are selected to construct an integrated prediction model for the post-pressure production of horizontal wells in shale reservoirs.
[0019] Optionally, the CatBoost ensemble learning algorithm in the Boosting ensemble learning framework is used to construct an integrated prediction model for the post-pressure production of horizontal wells in shale reservoirs.
[0020] Compared with the prior art, this application has at least the following beneficial effects:
[0021] This application can avoid falling into local optimal points during the prediction process, prevent the model from losing its ability to predict unknown data, improve the generalization performance of machine learning prediction methods, improve the adaptability of the model in the actual shale reservoir horizontal well pressure production prediction, improve the accuracy of the prediction results, and help solve actual production problems. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.
[0023] Figure 1 It is a correlation diagram of main controlling factors calculated by grey correlation method in the embodiment;
[0024] Figure 2 The heat map of the correlation coefficient of the yield factor used in the embodiment;
[0025] Figure 3 It is a correlation diagram of the main controlling factors comprehensively calculated in the embodiment;
[0026] Figure 4 Correlation coefficient diagram of each machine learning algorithm model in the embodiment;
[0027] Figure 5 Average relative error graph of each machine learning algorithm model in the embodiment;
[0028] Figure 6 Correlation coefficient heat map of prediction results of each single model in the embodiment;
[0029] Figure 7 A graph showing the relationship between the predicted oil production of a horizontal well and the actual oil production in the embodiment. DETAILED DESCRIPTION
[0030] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.
[0031] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments of the present invention can be combined with each other. It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same and similar parts between the various embodiments can be referred to each other.
[0032] Example 1
[0033] like Figure 1-3 As shown, the model integration prediction method for the post-pressure production of horizontal wells in shale reservoirs disclosed in this embodiment mainly includes the following steps:
[0034] S1, collect the production well data of the study block, obtain and count the main production parameters of each production well, and construct the relevant data set.
[0035] Among them, production well data usually include: geological data, logging data, drilling data, fracturing data, production data and other related information. The main production parameters of each production well are obtained and counted, including horizontal well section length, drilling rate, number of sections, number of section clusters, amount of sand added per meter, amount of liquid added per meter, well soaking time, pumping depth, dynamic liquid level, porosity, permeability, oil saturation, etc., and relevant data sets are constructed.
[0036] Among them, the main production parameters include horizontal well section length, drilling rate, number of sections, number of section clusters, amount of sand added per meter, amount of liquid added per meter, well soaking time, pumping depth, dynamic liquid level, porosity, permeability, oil saturation, etc.
[0037] S2. Mathematical statistics methods are used to analyze the correlation between the initial production of horizontal wells and production parameters such as horizontal well section length, drilling rate, number of sections, number of section clusters, amount of sand added per meter, amount of liquid added per meter, soaking time, pumping depth, dynamic liquid level, porosity, permeability, and oil saturation. The contribution of different factors to the initial production of horizontal wells is analyzed to clarify the promoting or weakening effects of different factors on the initial production of horizontal wells.
[0038] S2, due to the different dimensions of different factors, the numerical differences are large, direct processing may affect the results of data analysis. In order to eliminate the impact of the dimensions between different factors, data normalization is required. The normalization formula is shown in formula (1):
[0039]
[0040] In the above formula, x(i)' is dimensionless and is the normalized sequence; x(i) is the original sequence; is the minimum value in the sequence; is the maximum value in the sequence.
[0041] Using the normalized data, the initial production of horizontal wells is taken as the parent sequence, and the horizontal well section length, drilling rate, number of sections, number of section clusters, amount of sand added per meter, amount of liquid added per meter, soaking time, pumping depth, dynamic liquid level, porosity, permeability, oil saturation and other factors are taken as subsequences. The grey correlation theory is used to calculate the correlation between each factor and the initial production. The calculation formulas are shown in equations (2)-(4). The calculation results are shown in Figure 1 As shown. It can be seen that the correlation between each factor and the initial production of horizontal wells is from large to small: dynamic liquid level > number of segment clusters > amount of sand added per meter > amount of liquid added per meter > soaking time > horizontal well segment length > number of segments > porosity > oil saturation > pumping depth > drilling rate > permeability.
[0042] Δ i (k)=|x0 (k)-x i (k)| (2)
[0043]
[0044] In the above formula, is the minimum absolute difference between the comparison sequence and the target sequence; is the maximum absolute difference between the comparison sequence and the target sequence; B is the resolution coefficient, B∈(0,1), the smaller B is, the greater the resolution is; ω k is the weighting coefficient, which indicates the importance of the main control factor to the target factor. In this embodiment, ω k The value of is 1.
[0045] S4, introduce the Pearson correlation analysis criterion, as shown in Formula 5; calculate the correlation coefficient of factors such as horizontal well section length, drilling rate, number of sections, number of section clusters, amount of sand added per meter, amount of liquid added per meter, soaking time, pumping depth, dynamic liquid level, porosity, permeability, oil saturation, etc. on the initial production of horizontal wells. The calculation results are as follows Figure 2 As shown in the figure, it can be seen that the correlation between each factor and the initial production of horizontal wells is from large to small: dynamic liquid level > number of segment clusters > amount of sand added per meter > number of segments > soaking time > horizontal well segment length > pumping depth > porosity > amount of fluid added per meter > drilling rate > oil saturation > permeability;
[0046]
[0047] In the above formula, y i is the measured value of the initial oil production of the horizontal well; is the average value of the initial oil production of the horizontal well; is the predicted value of the initial oil production of the horizontal well; n is the number of horizontal wells.
[0048] S5, using the calculation results of grey correlation and Pearson correlation criteria, the comprehensive correlation between the influencing factors and the initial production of horizontal wells is introduced (see formula 6). The calculation results are as follows: Figure 3 As shown in the figure, through comprehensive correlation analysis, the 12 influencing dimensions that affect the initial production of horizontal wells are reduced to 6 influencing dimensions, which reduces the amount of data calculation and improves the calculation speed. The final main controlling factors are determined as follows: dynamic liquid level, number of segment clusters, amount of sand added per meter, amount of liquid added per meter, soaking time, and horizontal well segment length;
[0049]
[0050] In the above formula, γ i is the correlation degree obtained by the grey correlation method, F i is the weight obtained by Pearson, η i is the ultimate controlling factor.
[0051] S6, taking the main controlling factors of comprehensive analysis as input parameters, using support vector machine, generalized linear model, multi-layer perceptron, random forest, decision tree, XGBoost, BP neural network, convolutional neural network and other machine learning algorithms to predict the production of horizontal wells in shale reservoirs. The correlation coefficient R 2 like Figure 4 As shown in Figure 2, the mean relative error (MRE) of each machine learning algorithm model is as follows: Figure 5 As shown; the correlation coefficient calculation formula is as shown in formula (7), and the average relative error calculation formula is as shown in formula (8):
[0052]
[0053] In the formula, y i is the actual value; It is the predicted value of the production of horizontal wells in shale reservoirs after pressure reduction; is the average of the actual values; n is the sample size.
[0054] Based on a variety of comparative indicators, it can be found that the decision tree model has the best prediction effect, and the generalized linear model has the worst prediction effect. As can be seen from the figure, the prediction accuracy of the machine learning algorithm model is relatively high, but there are also obvious differences between the post-pressure production prediction results of different machine learning algorithm models, reflecting that different machine learning algorithms have different adaptability to the post-pressure production of horizontal wells in shale reservoirs and different prediction effects.
[0055] In addition, the use of a single machine learning method in the prediction model application process is prone to fall into the local optimal point, causing the model to lose its ability to predict unknown data. In order to improve the generalization performance of the machine prediction method, the ensemble learning idea is used to integrate multiple intelligent prediction methods. The Pearson correlation criterion is used to calculate the correlation coefficient between the prediction results of each prediction model and the actual horizontal well production after pressure in the shale reservoir based on the prediction results of a single machine learning algorithm. Figure 6 As shown. Figure 6 It can be seen that there are still certain differences in the correlation coefficients between the prediction results of different machine learning algorithms and the actual post-pressure production of horizontal wells in shale reservoirs. Therefore, the selection of base models should be based on "good and different". "Good" means that the base model should have a certain prediction ability, and "different" means that there should be certain differences in the prediction ability between the base models.
[0056] To this end, based on the Pearson correlation coefficient, this embodiment proposes a prediction model evaluation index d (see Formula 9). The larger the evaluation index value of each prediction model is, the more the model conforms to the "good but different" relationship.
[0057]
[0058] Where, d i is the evaluation index value of the i-th prediction model, dimensionless; ρ iU is the correlation between the prediction results of the ith prediction model and the actual horizontal well production after fracturing in shale reservoirs; ρ ij is the correlation coefficient between the ith prediction model and the jth prediction model, and n is the number of prediction models.
[0059] The evaluation index values of each prediction model for the post-pressurization production of horizontal wells in shale reservoirs are shown in Table 1. In order to reduce the complexity of calculation and improve the prediction accuracy, the evaluation index value of 1.50 is used as the reference limit in the construction of the integrated prediction model for the post-pressurization production of shale reservoirs. When the evaluation index value of the prediction model is higher than 1.50, the prediction result of the model is more accurate and has stronger prediction ability, which is more in line with the selection requirement of "good but different" for the base model; although the prediction model with an evaluation index value less than 1.50 has good prediction accuracy, its prediction ability is relatively lower than that of the prediction model with a higher evaluation index value.
[0060] Table 1: Evaluation index values of various prediction models for the prediction of horizontal well production after pressure reduction in shale reservoirs
[0061]
[0062] Based on the model evaluation index value, the prediction model constructed by the machine learning algorithm with better prediction effect is selected, that is, the prediction model constructed by the machine learning algorithm with the prediction model evaluation index value greater than 1.50 is used as the base model of the model integration, and the CatBoost ensemble learning algorithm in the Boosting ensemble learning framework is used to build the integrated prediction model of the production of horizontal wells in shale reservoirs after pyrolysis. Subsequently, the integrated prediction model of the production of horizontal wells in shale reservoirs after pyrolysis is trained with the main control production parameters as the model input parameters and the production of horizontal wells in shale reservoirs after pyrolysis as the output.
[0063] Finally, the constructed integrated prediction model for horizontal well production after pressure reduction in shale reservoirs was used to predict the production capacity of horizontal wells in the actual study area. The prediction results are as follows: Figure 4 , Figure 5 As shown, the prediction results are Figure 7 As shown in the figure, it can be seen that the prediction effect of the integrated prediction model for the post-pressure production of shale reservoir horizontal wells is good, and the correlation coefficient R 2 The average relative error is 0.972 and 1.2301%. Compared with the prediction results of a single model, the constructed integrated prediction model has better prediction effect.
[0064] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A model integration prediction method for the post-pressure production of horizontal wells in shale reservoirs, characterized in that: The following steps are involved: Based on the Pearson correlation coefficient, the evaluation index values of various machine learning prediction models for the prediction of horizontal well production after pressure pumping in shale reservoirs are calculated respectively; The machine learning prediction model with an evaluation index value higher than 1.50 was selected as the base model, and an integrated learning algorithm was used to construct an integrated prediction model for the post-pressure production of horizontal wells in shale reservoirs. The integrated prediction model for the production of horizontal wells in the shale reservoir after being pressurized is trained by using the main control production parameters to obtain a trained integrated prediction model for the production of horizontal wells in the shale reservoir after being pressurized; The trained integrated prediction model for horizontal well production after pressure reduction in shale reservoirs is used to predict the horizontal well production after pressure reduction in shale reservoirs.
2. The model integration prediction method for the post-fracture production of horizontal wells in shale reservoirs according to claim 1 is characterized in that: The main production control parameters include dynamic liquid level, number of clusters, amount of sand added per meter, amount of liquid added per meter, well soaking time, and horizontal well section length.
3. The model integration prediction method for the post-pressure production of horizontal wells in shale reservoirs according to claim 1 is characterized in that: The calculation formula of Pearson correlation coefficient is: In the formula, x i is the normalized parameter affecting the production of horizontal wells in shale reservoirs after pressure reduction; i is the measured value of the initial oil production of the horizontal well; is the average value of the initial oil production of the horizontal well; is the predicted value of the initial oil production of the horizontal well; n is the number of horizontal wells.
4. The model integration prediction method for the post-pressure production of horizontal wells in shale reservoirs according to claim 1 is characterized in that: The calculation formula of the evaluation index value is: Where, d i is the evaluation index value of the i-th prediction model, dimensionless; ρ iU is the correlation between the prediction results of the ith prediction model and the actual horizontal well production after fracturing in shale reservoirs; ρ ij is the correlation coefficient between the ith prediction model and the jth prediction model, and n is the number of prediction models.
5. The model integration prediction method for the post-pressure production of horizontal wells in shale reservoirs according to claim 1 is characterized in that: The evaluation index values of 8 machine learning prediction models for predicting the post-pressure production of horizontal wells in shale reservoirs are calculated respectively.
6. The model integration prediction method for the post-pressure production of horizontal wells in shale reservoirs according to claim 5, characterized in that: The 8 machine learning prediction models are: support vector machine, generalized linear model, multilayer perceptron, random forest, decision tree, XGBoost, BP neural network, and convolutional neural network.
7. A model integration prediction method for shale reservoir horizontal well production after fracturing according to claim 6, characterized in that: Convolutional neural network, XGBoost, decision tree, generalized linear model, multilayer perceptron model and random forest were selected to construct an integrated prediction model for the post-pressure production of horizontal wells in shale reservoirs.
8. The model integration prediction method for the post-fracture production of horizontal wells in shale reservoirs according to claim 1 or 7, characterized in that: The CatBoost ensemble learning algorithm in the Boosting ensemble learning framework is used to build an integrated prediction model for the post-pressure production of horizontal wells in shale reservoirs.
Citation Information
Patent Citations
Non-deterministic prediction method for shale gas production capacity
CN108416475A
Complex reservoir horizontal well initial yield fusion prediction method
CN115539026A
Cited By
Method and device for intelligently predicting post-fracturing productivity of multi-layer combined producing well
CN122022024A
A method and device for intelligently predicting the deliverability of a multi-layer commingled oil well after fracturing
CN122022024B
