Fractured oil well productivity prediction method and system based on ICNN-SVR and storage medium

Abstract

The invention discloses a fractured oil well productivity prediction method and system based on ICNN-SVR, and a storage medium, relates to the technical field of petroleum engineering and measurement, and aims to solve the problems of high calculation cost, high sensitivity to hyper-parameters, large fluctuation and insufficient adaptability to missing data in an existing method. Comprising the following steps: S1, collecting fracturing production parameter data of an oil well; s2, a fractured oil well productivity prediction model is constructed and comprises a CNN module and an MLP module, the CNN module comprises two convolution blocks, each convolution block adopts a double-layer 5 * 5 convolution structure, and a layer of pooling structure is formed after two layers of convolution; each MLP module is composed of three dense layers according to an inverted pyramid structure and shares the input weight of the CNN module, and an output layer adopts a support vector regression layer; and S3, the CNN module of the fractured oil well productivity prediction model extracts oil well data features, the MLP module further performs feature association extraction on the features extracted by the CNN module, and the support vector regression layer predicts fractured oil well productivity parameters.

Description

Technical Field

[0001] The present invention relates to the field of petroleum engineering and measurement technology, and in particular to a method, system and storage medium for predicting the productivity of a fracturing oil well based on ICNN-SVR. Background Art

[0002] Fracturing technology is a commonly used oilfield development and production enhancement technology, which aims to improve the productivity and recovery rate of oil wells. Research on the prediction of production capacity after fracturing has also become a research hotspot in the field of petroleum engineering at home and abroad. The initial oil production (IOP) of an oil well can be used to reflect the production capacity of an oil well and evaluate the working status of the pumping unit, so as to achieve scientific deployment and reliable production of oil well equipment. The prediction of initial oil production can help oil companies evaluate the effect of fracturing operations and further optimize fracturing design and construction plans. The cumulative oil production (ICOP) of an oil well refers to the total amount of oil production increased by the oil well from the time of fracturing production to the present. Cumulative oil production is one of the important indicators for evaluating the productivity of an oil well and the effect of oilfield development, and can also be used to predict the remaining production and life of an oil well.

[0003] At present, the research on post-fracture production capacity prediction is mainly based on traditional physical models, numerical analysis and artificial intelligence methods. Traditional physical models and numerical simulation methods have the problem of large amount of calculation, long time consumption and need a lot of professional knowledge to support; in recent years, the research on the prediction of production capacity of fractured oil wells has mostly adopted artificial intelligence methods, including vector autoregression model, random forest and other machine learning algorithms. Among them, the research on oil production prediction based on deep learning has also made great progress, including fully connected neural network, convolutional neural network, long short-term memory neural network, etc.

[0004] Machine learning and deep learning methods have gradually become the core technologies for fracturing capacity prediction. Zhao Ming (January 2023) proposed the "Application of Neural Network Algorithms in the Prediction of Tight Oil Fracturing Capacity". Although this method has made some progress in optimizing the structure of neural networks, it still has problems such as high computational cost, sensitivity to hyperparameters, and insufficient adaptability to fluctuating and missing data. Especially in practical applications, the data of fracturing wells are often missing and volatile, which affects the stability of the model and the accuracy of prediction. During the training process, too many features were lost and the adaptability to fluctuating and missing data was insufficient, resulting in poor generalization ability of the model, which affected the effect of fracturing well capacity prediction. Summary of the invention

[0005] The technical problems to be solved by the present invention are:

[0006] Existing methods suffer from high computational cost, sensitivity to hyperparameters, and insufficient adaptability to volatile and missing data.

[0007] The present invention adopts the following technical solutions to solve the above technical problems:

[0008] The present invention provides a method for predicting the productivity of a fracturing oil well based on ICNN-SVR, comprising the following steps:

[0009] S1. Collecting the fracturing production parameter data of the oil well, and screening and preprocessing the data;

[0010] S2. Construct a fracturing oil well productivity prediction model, including a CNN module and an MLP module. The CNN module includes two convolution blocks, each of which adopts a double-layer 5×5 convolution structure, and a pooling structure after two layers of convolution; each of the MLP modules is composed of three dense layers in an inverted pyramid structure, sharing the CNN module input weights, and the output layer adopts a support vector regression layer;

[0011] S3, inputting the fracturing production parameter data of the oil well into the fracturing oil well capacity prediction model, the CNN module extracts the oil well data features, the MLP module further performs feature association extraction on the features extracted by the CNN module, and the support vector regression layer predicts the fracturing oil well capacity parameters.

[0012] Furthermore, the productivity parameters of the fracturing oil well described in S3 include: initial oil production (IOP) and cumulative increased oil production (ICOP).

[0013] Furthermore, the data screening and preprocessing described in S1 include: outlier removal, correlation analysis, missing value filling and normalization of the data.

[0014] Furthermore, Spearman correlation was used in S1 to conduct correlation analysis between the data and the cumulative oil production increase and the initial oil production, where the calculation formula of the Spearman rank correlation coefficient ρ is:

[0015]

[0016] Where, d i =x i -y i is the rank difference, x i and i is the original data, and n is the sample size.

[0017] Furthermore, the parameters in S1 with the absolute value of the Spearman correlation coefficient with the cumulative oil increase and the initial oil production greater than 0.1 are taken as the input features of the model.

[0018] Furthermore, the fracturing production parameter data of the oil wells obtained by screening and preprocessing in S1 include: single-layer oil production before fracturing, sandstone thickness ratio, sandstone thickness, single-layer effective thickness ratio, single-layer effective thickness, perforation ratio, single-well oil production before fracturing, geological reserves, oil saturation, perforation thickness, permeability, daily oil production before fracturing, recoverable volume, sand addition volume, total liquid volume, dynamic liquid level before fracturing, number of small layers and original saturation pressure.

[0019] The present invention provides a system for predicting the productivity of a fractured oil well based on ICNN-SVR. The system has a program module corresponding to the steps of the method described in any one of the above technical solutions, and executes the steps in the above-mentioned method for predicting the productivity of a fractured oil well based on ICNN-SVR during operation.

[0020] The present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps in the ICNN-SVR-based fractured oil well productivity prediction method described in any one of the above technical solutions when called by a processor.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] The present invention adopts an improved convolutional neural network-support vector regression model ICNN-SVR. By optimizing the CNN module, two convolution blocks are adopted. Each convolution block adopts a double-layer 5×5 convolution structure, and a pooling structure is formed after the two layers of convolution. The problem of excessive feature loss during training is reduced. At the same time, the MLP module is optimized and integrated with SVR to effectively process data with large fluctuations and missing data, reduce feature loss, and improve the robustness and generalization ability of the model. The model performs well in fracturing capacity prediction and has higher application value.

[0023] The present invention realizes adaptive feature extraction by improving CNN and fusing SVR method, reasonably allocates weights of effective features, so as to realize efficient prediction of fracturing capacity, has good prediction performance, and is of practical significance for evaluating the capacity of fracturing oil wells and formulating optimization plans. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a flow chart of a method for predicting the productivity of a fractured oil well based on ICNN-SVR in an embodiment of the present invention;

[0025] Figure 2 This is a Spearman correlation coefficient analysis diagram of the initial oil production and the cumulative oil increase in the embodiment of the present invention;

[0026] Figure 3 It is a structural diagram of the ICNN-SVR network model in an embodiment of the present invention;

[0027] Figure 4 It is a comparison diagram of the prediction results between different models in the embodiment of the present invention, wherein Figure (a) is the SVR prediction result, Figure (b) is the CNN prediction result, and Figure (c) is the ICNN-SVR model prediction result. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the scheme of the present invention, exemplary implementations or embodiments of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described implementations or embodiments are only implementations or embodiments of a part of the present invention, not all of them. Based on the implementations or embodiments of the present invention, all other implementations or embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0029] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0030] The present invention provides a method for predicting the productivity of a fracturing oil well based on ICNN-SVR, comprising the following steps:

[0031] S1. Collecting the fracturing production parameter data of the oil well, and screening and preprocessing the data;

[0032] S2. Construct a fracturing oil well productivity prediction model, including a CNN module and an MLP module. The CNN module includes two convolution blocks, each of which adopts a double-layer 5×5 convolution structure, and a pooling structure after two layers of convolution; each of the MLP modules is composed of three dense layers in an inverted pyramid structure, sharing the CNN module input weights, and the output layer adopts a support vector regression layer;

[0033] S3. The fracturing production parameter data of the oil well is input into the fracturing oil well capacity prediction model. The CNN module extracts the oil well data features. The MLP module further performs feature association extraction on the features extracted by the CNN module. The support vector regression layer predicts the capacity parameters of the fracturing oil well. The capacity parameters of the fracturing oil well include: initial oil production (IOP) and cumulative oil production increase (ICOP).

[0034] There are a lot of fracturing production data for oil wells, including 56 functional parameters such as well number, sandstone thickness, sand addition amount, cumulative oil increase, initial oil production, pre-fracturing oil production, effective thickness, etc. The data is screened and pre-processed, including: outlier removal, correlation analysis, missing value filling and normalization.

[0035] The Spearman correlation analysis method is more flexible in dealing with missing data, and can be calculated using ranks without complete initial data. The calculation formula for the rank correlation coefficient ρ is:

[0036]

[0037] Where, d i =x i -y i is the rank difference, x i and i is the original data, and n is the sample size.

[0038] Spearman correlation analysis was performed on the data for initial oil production and cumulative oil production, such as Figure 2 As shown;

[0039] The present invention selects parameters with an absolute value of the Spearman correlation coefficient greater than 0.1 as input features of the network model in order to improve the generalization ability of the model and obtain more accurate prediction results.

[0040] Gradient descent is usually used to reduce the numerical input variables by a certain ratio, which can effectively increase the convergence speed of the algorithm and improve its performance. Standardization and normalization can maintain the distribution rules of the original data and reduce the noise of the original data by scaling the data to the range of 0 to 1:

[0041]

[0042] Where x min is the minimum value of the data, x max is the maximum value of the data.

[0043] The screened fracturing production parameter data of the oil wells include: single-layer oil production before fracturing, sandstone thickness ratio, sandstone thickness, single-layer effective thickness ratio, single-layer effective thickness, perforation ratio, single-well oil production before fracturing, geological reserves, oil saturation, perforation thickness, permeability, daily oil production before fracturing, recoverable volume, sand addition volume, total liquid volume, dynamic liquid level before fracturing, number of small layers and original saturation pressure.

[0044] Based on the prediction of fracturing oil wells based on the fusion data set, the present invention optimizes the CNN structure, such as Figure 3 As shown in the figure. Using a large convolution kernel (kernel=5) instead of a small convolution kernel (kernel=3) has better feature extraction capabilities, higher model effects, and more stable feature expressions. At the same time, a large convolution kernel can increase the receptive field, thereby obtaining more abstract high-order features and better overall features:

[0045]

[0046] In the formula, w is the convolution kernel, y is the convolution output result, a is the feature input, and k is the convolution kernel size.

[0047] In view of the classic CNN structure of "one layer of convolution and one layer of pooling", the present invention appropriately reduces the number of maximum pooling layers and adjusts it to a "two layers of convolution and one layer of pooling" structure. That is, "convolution layer (64)-convolution layer (32)-pooling layer", two such structural blocks are connected in series, and the number of convolution kernels in the other structural block is 32 and 16. This is because although the maximum pooling layer can effectively reduce the amount of calculation and improve translation invariance, too many pooling operations will lead to the loss of detail information and may make the distribution of the perceptual field uneven, thereby affecting the model's ability to capture global features. By reducing the number of pooling layers, the model can better balance the effectiveness of feature extraction and the integrity of information.

[0048] The CNN module performs feature extraction. First, the data is convolved to obtain local features, and then global features. The maximum pooling layer is then used to maintain the significant features in the data, thereby reducing the dimensionality of the data. The CNN module can reduce the number of network parameters through parameter sharing, thereby accelerating the training of the model.

[0049] Multi-Layer Perceptron (MLP) contains multiple fully connected neural networks. Each layer of the neural network contains many neurons, and each neuron is connected to the previous layer. MLP generally consists of an input layer, multiple hidden layers, and an output layer.

[0050] MLP has some limitations, such as network structure limitations and easy overfitting. In this embodiment, the final output layer of MLP is configured with a customized support vector regression (SVR), that is, each MLP module is composed of three dense layers in an inverted pyramid structure (128, 64, 32), and the support vector machine layer is connected on the basis of the last layer. The kernel parameters in the SVR layer are continuously updated through neural network training to better meet the needs of model output. In this way, after processing the data features, small sample data can be effectively processed, and the generalization ability of the model can be improved while effectively avoiding the problem of overfitting.

[0051] The features extracted by the CNN and MLP modules are set as a i =[a1,a2,a3,…,a n ], and then assign different weights to different features, set to w i =[w1,w2,w3,…,w n], by assigning larger weights to important features, highlighting key features, and then setting the bias to b and the activation function to f, that is:

[0052]

[0053] Where w is the weight, a is the feature vector, and b is the bias.

[0054] The relationship between the input layer and the hidden layer is similar to the above. The input layer in the CNN model assigns the input optimal parameter data to the weights obtained after multiple iterations, and then outputs it through the activation function as a feature input to the next hidden layer.

[0055] The last layer is the output layer. The output tensor of the last hidden layer of the MLP module is used as the input of the custom SVR layer. The output of the custom SVR layer is calculated by matrix multiplication to calculate the product of the input and the kernel. The initial oil production prediction value Output1 and the cumulative oil increase prediction value Output2 can be associated with the features respectively, that is,

[0056] output i =output (last hidden layer) kernel, (i=1,2)

[0057] In the formula, output (last hidden layer) is the output result of the last hidden layer, and kernel is the kernel function of the custom SVR layer.

[0058] The MLP module can not only transform the feature dimension, but also facilitate data transmission; in addition, for the previously extracted features, the superimposed fully connected layer can be used to further extract the correlation between the features. The fusion effect of the CNN module and the MLP module is very good.

[0059] Each layer in the improved convolutional model will gradually extract more abstract high-order features, and through loss function evaluation and continuous updating of weights using the gradient descent algorithm, a complex nonlinear relationship will be established between the input optimal parameter values ​​and the output initial oil production and cumulative oil increase prediction values, completing the task of using existing parameters to predict unknown fracturing oil well production.

[0060] The ICNN-SVR model is a multiple-input double-output (MIDO) system. The weights learned by two modules are used to explore the relationship between input parameters and initial oil production and cumulative oil production. Two different feature extraction methods realize the comprehensive extraction of data features. Compared with a single model, there is only one model for weight learning and only a single method can be used to extract features. There are certain limitations. The features finally obtained are not rich enough and may even miss important features. The settings of the network parameters of the two modules are shown in Table 1.

[0061] Table 1 Network parameter settings

[0062]

[0063] The model training parameter settings are shown in Table 2.

[0064] Table 2. Setting of training parameters

[0065]

[0066] The ICNN-SVR-based fractured oil well productivity prediction method (algorithm) proposed in the present invention is the underlying technical core of the present invention, and various products can be derived based on the algorithm.

[0067] Based on the method proposed in the present invention, a system for predicting the productivity of a fractured oil well based on ICNN-SVR is developed using a programming language. The system has program modules corresponding to the steps of the above-mentioned technical solution, and executes the steps in the above-mentioned method for predicting the productivity of a fractured oil well based on ICNN-SVR during operation.

[0068] The computer program of the developed system (software) is stored on a computer-readable storage medium, and the computer program is configured to implement the steps of the above-mentioned ICNN-SVR-based fractured oil well productivity prediction method when called by a processor, that is, the present invention is materialized on a carrier to become a computer program product.

[0069] Various implementations of the systems and techniques described herein can be realized in digital electronic circuit systems, integrated circuit systems, dedicated ASICs (application specific integrated circuits), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0070] The computer programs (also referred to as programs, software, software applications, or codes) of the present invention include machine instructions for programmable processors, and these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, device, and / or device (e.g., disk, optical disk, memory, programmable logic device PLD) for providing machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as machine-readable signals. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0071] Example 1

[0072] This embodiment is based on Jupyter Notebook 6.3.0, using Python3.7 programming language, and the model is trained and predicted on the deep learning framework Tensorflow2.1.0. All models are based on the same data set, and the model performance is measured by three evaluation indicators: root mean square error (RMSE), mean absolute error (MAE) and prediction accuracy. For the prediction accuracy, first test the prediction accuracy of each sample in the set = [1-(true value of the sample-predicted value of the sample) / true value of the sample] × 100%, then sum and average the prediction accuracy to get the final prediction accuracy.

[0073] According to the production data of oil wells in a block of Daqing Oilfield, various oilfield data are first integrated, outliers are eliminated, parameter correlation analysis is performed, missing values ​​are filled, normalization and optimization are performed. The data set contains a total of 2196 sample data from 499 wells in different years and months. First, another 199 samples are manually selected to form a test set, which contains data from 36 wells in different years and months. Then, the training set and validation set are randomly divided according to the method that the same oil well does not appear in the training set and validation set at the same time. By observing the data, it is found that under normal circumstances, the variation range of some parameter values ​​in different layers of the same well is often not too large. If these similar samples are divided into the training set, validation set and test set at the same time, it will cause information leakage in the test set and falsely high prediction accuracy. Therefore, in order to make the experimental results more credible, the training set, validation set and test set are manually divided to ensure that the oil wells contained in each of the three are not repeated, and the data in the test set is completely not involved in the training and adjustment of the model.

[0074] Due to the difficulty of exploration, equipment and human factors, the original oil well data set is often of low quality and has certain problems, which has a certain negative impact on model training and further affects the performance of the model on the validation set and test set. Therefore, it is necessary to preprocess the data to improve the data quality as shown in Table 3:

[0075] Table 3 Problems and analysis of oil well dataset

[0076]

[0077] Because outliers can have a negative impact on data analysis and model training, and because the oil well data is huge, it is inevitable that various parameters in the production process will have recording errors, and even some format errors, so the original oil well data is manually detected and removed for outliers. If the original data is seriously missing and cannot be directly input into the network, the function in the library can be called to fill the mean. The method of deleting samples with missing values ​​will lose too much information and is not suitable for small sample data. Finally, because the measurement units of different parameters are often different, the scale differences will be large, so the minimum and maximum normalization method is used to scale the data to the range to unify the scale.

[0078] In this embodiment, the ICNN-SVR model of the present invention is compared with the SVR model and the CNN model. Each model is run 100 times, and the model prediction results are averaged for reasonable evaluation. Some specific prediction results of the three models for the test set samples are shown in Tables 4 and 5.

[0079] Table 4 Prediction results of some specific initial oil production (IOP) of the three models for the test set samples

[0080]

[0081] Table 5 Prediction results of some specific cumulative oil production (ICOP) of the three models for the test set samples

[0082]

[0083]

[0084] It can be seen from the prediction results of the three models on a single sample that the predicted value of the model proposed in the present invention is closest to the true value, and the prediction accuracy of the single sample is relatively higher.

[0085] The comparison results among the three models are shown in Table 6.

[0086] Table 6

[0087]

[0088] The ICNN-SVR model has the highest accuracy and the lowest error. The prediction results of the other two models are slightly insufficient, but the CNN model still has higher accuracy and lower error than the SVR model, proving that reasonable optimization of the model is a way to improve model performance.

[0089] Table 7 Linear regression equations of the models

[0090]

[0091] The prediction capabilities of the three models are further analyzed by using the cross plot of the predicted and true values ​​of initial oil production and cumulative oil increase. Figure 4 As shown in the figure, the ICNN-SVR model performs best among the three models; and the linear regression equation is established for the discrete points (see Table 7) to observe the deviation between the equation and y=x. The lower the deviation, the better the prediction result. In contrast, the regression equations of the SVR model (y=0.89x+0.272 (initial oil production) and y=1.197x+53.316 (cumulative oil production)) deviate greatly from the ideal y=x equation, indicating its limitation in complex feature extraction. Although the regression equations of the CNN model (y=0.94x-0.069 and y=0.906x+70.611) have improved, the deviation is still obvious. The regression equations of the ICNN-SVR model (y=0.96x+0.027 and y=0.934x+77.505) deviate the least from the ideal y=x equation, indicating that its performance in feature extraction and integration has been significantly improved. By optimizing the deficiencies of SVR and CNN, ICNN-SVR achieves more efficient feature extraction and more accurate prediction with fewer parameters, further verifying its advantages in oil well data analysis.

[0092] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A method for predicting the productivity of a fractured oil well based on ICNN-SVR, characterized in that: The steps include: S1. Collecting the fracturing production parameter data of the oil well, and screening and preprocessing the data; S2. Construct a fracturing oil well productivity prediction model, including a CNN module and an MLP module. The CNN module includes two convolution blocks, each of which adopts a double-layer 5×5 convolution structure, and a pooling structure after two layers of convolution; each of the MLP modules is composed of three dense layers in an inverted pyramid structure, sharing the CNN module input weights, and the output layer adopts a support vector regression layer; S3, inputting the fracturing production parameter data of the oil well into the fracturing oil well capacity prediction model, the CNN module extracts the oil well data features, the MLP module further performs feature association extraction on the features extracted by the CNN module, and the support vector regression layer predicts the fracturing oil well capacity parameters.

2. The method for predicting the productivity of a fractured oil well based on ICNN-SVR according to claim 1, characterized in that: The production capacity parameters of the fractured oil well described in S3 include: initial oil production (IOP) and cumulative increased oil production (ICOP).

3. The method for predicting the productivity of a fractured oil well based on ICNN-SVR according to claim 2, characterized in that: The data were screened and preprocessed as described in S1, including: outlier removal, correlation analysis, missing value filling and normalization.

4. The method for predicting the productivity of a fractured oil well based on ICNN-SVR according to claim 3, characterized in that: In S1, Spearman correlation was used to analyze the correlation between the data and the cumulative oil production and the initial oil production, where the calculation formula of the Spearman rank correlation coefficient ρ is: Where, d i =x i -y i is the rank difference, x i and i is the original data, and n is the sample size.

5. The method for predicting the productivity of a fractured oil well based on ICNN-SVR according to claim 4, characterized in that: In S1, the parameters with the absolute value of the Spearman correlation coefficient with the cumulative oil increase and the initial oil production greater than 0.1 are taken as the input features of the model.

6. The method for predicting the productivity of a fractured oil well based on ICNN-SVR according to claim 5, characterized in that: The fracturing production parameter data of the oil wells screened and pre-processed in S1 include: single-layer oil production before fracturing, sandstone thickness ratio, sandstone thickness, single-layer effective thickness ratio, single-layer effective thickness, perforation ratio, single-well oil production before fracturing, geological reserves, oil saturation, perforation thickness, permeability, daily oil production before fracturing, recoverable volume, sand addition volume, total liquid volume, dynamic liquid level before fracturing, number of small layers and original saturation pressure.

7. A fracturing oil well productivity prediction system based on ICNN-SVR, characterized in that: The system has a program module corresponding to the steps of the method described in any one of claims 1 to 6, and executes the steps of the ICNN-SVR-based fractured oil well productivity prediction method when running.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps in the ICNN-SVR-based fractured oil well productivity prediction method according to any one of claims 1 to 6 when called by a processor.

Images