Water logging level identification method based on BP neural network

Abstract

The invention provides a BP (Back Propagation) neural network-based flooding level identification method, which comprises the following steps of: 1, acquiring logging data, and selecting a logging curve which can reflect the characteristics of an oil-water layer; 2, preprocessing a logging curve, and dividing data into a training set and a test set; step 3, performing optimization to obtain a water logging level prediction identification model after BP neural network optimization; and step 4, verifying the accuracy of the flooding level prediction identification model after BP neural network optimization by calculating a root mean square error between a flooding level actual value and a predicted value in the test sample. According to the BP neural network-based flooding level identification method, the problems of poor learning stability and insufficient generalization ability of a traditional machine are solved, accurate prediction of the flooding level is realized, and a foundation is laid for exploration and development of a reservoir.

Description

A flood level identification method based on BP neural network Technical Field

[0001] This invention relates to the field of petroleum exploration and development technology, and in particular to a method for identifying flooding levels based on a BP neural network. Background Technology

[0002] Conventional well logging interpretation of water-flooded layers primarily relies on qualitative identification through well logging curve response characteristics and chart recognition methods. However, the accuracy and conclusions vary considerably, necessitating the introduction of automated computer-aided identification methods. While traditional machine learning methods have achieved good results in water-flooded layer identification, their inherent limitations often result in low accuracy. Backpropagation (BP) neural network models overcome the limitations of traditional machine learning in terms of input data and parameters. Therefore, conducting research on water-flooded layer identification based on BP neural networks is of paramount importance for improving the accuracy of water-flooded layer identification in the study area.

[0003] A backpropagation (BP) neural network is a multilayer feedforward neural network. Its main characteristic is that the signal propagates forward while the error propagates backward. In the forward propagation, the input signal is processed layer by layer from the input layer through the hidden layers until it reaches the output layer. The state of a neuron in each layer only affects the state of the neuron in the next layer. If the output layer does not produce the desired output, backpropagation begins. The network weights and thresholds are adjusted based on the prediction error, allowing the network's predicted output to continuously approach the desired output.

[0004] Chinese patent application CN201110090342.3 involves improving traditional cross-plot technology using neural network algorithms to achieve nonlinear identification and quantitative analysis of cross-plots. The proposed method utilizes a BP neural network algorithm, comprising five steps: screening object feature parameters, selecting network structure parameters, training the neural network model, testing the network model, and developing a simulated cross-plot. Based on the various properties of reservoir oil, gas, and water layers, statistical methods are used to accurately select parameter samples that best reflect the characteristics of oil, gas, and water layers from well logging calculations or well logging curves related to oil and gas interpretation. Then, the BP neural network algorithm is used to construct a network model by selecting appropriate weights and thresholds, and the model is trained and its errors are checked. Finally, the fluid type or water flooding degree of the reservoir at that depth is determined by projecting the identification vector obtained from the network output onto a plane. The key to the above technology lies in using cross plotting techniques for oil-water layer identification. The neural network model is only used to provide parameters for cross plotting, which means that the output layer of the neural network can only output two parameters. Moreover, the use of the output parameters is limited to fixed positions on the cross plotting. When judging the properties of oil-water layers, it is necessary to calculate the position of the projection point to the classification center. Therefore, its use has certain limitations.

[0005] Chinese patent application CN201811066926.5 discloses a well logging interpretation method for waterflooding in mixed-phase reservoirs based on sedimentary microfacies and lithofacies. The method includes: conducting detailed stratigraphic correlation and division; performing a classification study of sedimentary microfacies and lithofacies; establishing a four-property relationship map for each facies zone; developing a well logging interpretation model suitable for the study area based on sedimentary microfacies and lithofacies; processing actual data based on the sedimentary microfacies and lithofacies discrimination model; analyzing the differences in well logging interpretation for various reservoir types; and using an entropy-weighted quantitative evaluation method for waterflooded layers to study the waterflooding status of mixed-phase reservoirs, establishing a comprehensive evaluation standard for waterflooded layers in mixed-phase reservoirs, classifying oil-bearing layer waterflooding levels, and summarizing the waterflooding pattern of the target layer. This method improves the utilization rate and ultimate recovery rate of uncontrolled reserves in mixed-phase reservoirs, effectively reduces the rate of oilfield production decline, extends the stable production period of the oilfield, and significantly improves the ultimate recovery rate of the reservoir, achieving significant development effects and benefits. The aforementioned patented technology provides a well logging processing and interpretation method based on the concept of phase control. First, the reservoir is classified into sedimentary microfacies and lithofacies. Then, a parameter model is established based on the classification to calculate parameters such as porosity and permeability. The processing and interpretation method provided belongs to the conventional well logging processing method. Its prediction of water flooding level is based on various parameters calculated in the aforementioned steps, and the "entropy weight" method is used to assign different weights to each parameter. It does not use a BP neural network for prediction, which is completely different from the technical points of this invention.

[0006] Chinese patent application CN201911334572.2 discloses a method, apparatus, and storage medium for evaluating water-flooded layers. The method includes: acquiring measured logging curves of a target well group; preprocessing the measured logging curves to obtain standardized logging curves; wherein the target well group includes a target well and existing wells; determining the clay content, reservoir porosity, and pre-flood formation water salinity of the target well group based on the standardized logging curves; reconstructing the pre-flood spontaneous potential curve of the target well based on the clay content and pre-flood formation water salinity of the existing wells; inputting the standardized logging curve, clay content curve, and reservoir porosity curve of the target well into a reconstruction model to obtain the reconstructed pre-flood resistivity curve of the target well; and constructing a comprehensive water-flooding index to evaluate the water-flooded layer of the target well based on the reconstructed pre-flood spontaneous potential curve and the reconstructed pre-flood resistivity curve of the target well, thereby improving the accuracy of the water-flooded layer evaluation. This patent provides a technique for evaluating water-flooded layers by comparing the reconstructed curve with the measured curve of the target well. Its focus is on reconstructing the logging curves of the target well before water flooding, primarily the reconstruction of the spontaneous potential and resistivity curves. However, it does not directly use a BP neural network model for water flooding level classification, which differs from the technical focus of this invention. Furthermore, its methods for building the reconstruction model include support vector machine algorithms, gradient boosting regression tree algorithms, ensemble algorithms, regression algorithms, and Bayesian algorithms, without mentioning a neural network model. Even if it could use a BP neural network for logging curve reconstruction, it utilizes the predictive function of the neural network, while this invention uses the classification function of a BP neural network to directly classify the water flooding level; the core technologies of the two are completely different.

[0007] Chinese patent application CN202010696264.0 discloses a refined interpretation and evaluation method for water-flooded layers using a comprehensive GWO-LSSVM algorithm, relating to the field of petroleum engineering technology. This method, based on well logging curve evaluation and the theoretical foundation of the GWO-LSSVM algorithm, collects and optimizes data by analyzing the weights of different water-flooding intensity evaluation indicators, then inputs the data into a prediction module to calculate and output the water-flooding intensity result. The water-flooding level result is optimized by at least three kernel function parameters. This overcomes the limitations of traditional water-flooded layer evaluation methods that rely on a single parameter. Furthermore, this method involves fewer testing steps, lower difficulty, and higher accuracy, effectively providing reference data for practical engineering or experiments. While the aforementioned refined water-flooded layer evaluation method relates to the field of petroleum engineering technology, its model input sample values ​​include parameters from petroleum engineering fields such as mobile water saturation, water absorption profile, and production profile. However, this invention is applicable to the field of petroleum exploration and development, where all input parameters are well logging curves, without production data such as water absorption profiles or production profiles. Secondly, the above invention uses a combination of Grey Wolf Optimization (GWO) algorithm and Least Squares Support Vector Machine (LSSVM) to interpret the water-flooded layer. The main process is to first optimize the parameters according to the GWO algorithm to obtain the appropriate weights of the water flooding intensity evaluation index, and then use LSSVM to predict the final result. This is different from the algorithm and discrimination principle used in this invention, which uses the well logging curve as the input value and directly uses the BP neural network for prediction.

[0008] The existing technologies described above are significantly different from the present invention and have failed to solve the technical problem we want to address. Therefore, we have invented a new flooding level identification method based on a BP neural network. Summary of the Invention

[0009] The purpose of this invention is to provide a water inundation level identification method that optimizes the water inundation level prediction and identification model using the BP algorithm, thereby achieving high-precision prediction of water inundation level and laying the foundation for reservoir exploration and development.

[0010] The objective of this invention can be achieved through the following technical measures: a flood level identification method based on a BP neural network, which includes:

[0011] Step 1: Obtain logging data and select the logging curve that best reflects the characteristics of the oil-water reservoir;

[0012] Step 2: Preprocess the logging curves by dividing the data into training and testing sets;

[0013] Step 3: Optimize the flood level prediction and recognition model using a BP neural network.

[0014] Step 4: Calculate the root mean square error between the actual and predicted flood levels in the test samples to verify the accuracy of the flood level prediction and recognition model optimized by the BP neural network.

[0015] The objective of this invention can also be achieved through the following technical measures:

[0016] In step 1, by performing correlation analysis between the water flooding level of the water flooded layer and the conventional logging curve, the logging curve that best reflects the characteristics of the oil-water layer is selected.

[0017] In step 1, well logging data is acquired. By analyzing the correlation between conventional well logging curves and water-flooded layer levels, water-flooded layer analysis data after data standardization and dimensionality reduction is selected as model input data. These curves can reflect the characteristics of oil-water layers and can be used to predict water-flooded layer levels.

[0018] In step 2, the acquired logging curve data is preprocessed by removing outliers from each logging curve and normalizing the logging curves. The measured values ​​corresponding to each depth point on the preprocessed logging curve are used as input data.

[0019] In step 2, the calculation formula for the normalization of well logging curves is:

[0020]

[0021] In the formula, X norm X represents the measured depth value after normalization; X represents the measured depth value before normalization. max X represents the maximum measured depth value in the logging curve before normalization. min This represents the minimum measured depth value in the logging curve before normalization.

[0022] In step 2, after preprocessing the logging curve, the measured values ​​corresponding to each depth point on the logging curve are used as input data, and the input data is divided into training set and test set to obtain training samples and test samples.

[0023] In step 3, a flood level prediction and identification model is established based on the training set. A BP neural network model is set in the flood level prediction and identification model. The optimal weights and optimal biases of the extreme learning machine model are optimized by the BP neural network to obtain the flood level prediction and identification model optimized by the BP neural network.

[0024] Step 3 specifically includes:

[0025] s3.1. Construct a BP neural network model, study the network structure based on the characteristics of the system's input and output data, and reasonably set the number of nodes and activation functions of the hidden layers;

[0026] s3.2. Train the BP neural network, and continuously adjust the network weights or thresholds based on the network prediction error during the training process;

[0027] In the forward propagation process of the s3.3.BP algorithm, the input signal enters the input layer, passes through the activation function to obtain a value, enters the hidden layer, and after processing, enters the output layer.

[0028] s3.4. Calculate whether the data obtained from the output layer is within the error range of the expected output data. If not, proceed to the backpropagation process.

[0029] s3.5. During the backpropagation process, the weight values ​​between layers and the bias values ​​between neurons in each layer are updated and adjusted along the way;

[0030] s3.6. Test the neural network model. Use the established classification model to predict and back-judge the flooded layer samples, test the classification effect of the model, and obtain the flood level prediction and recognition model optimized by the BP search algorithm.

[0031] In step s3.3, the output formula of the hidden layer is shown in equation (2):

[0032]

[0033] In the formula, z k For hidden layer output; v ki Set the weights between the input layer and the hidden layer; θ is the bias value; f1 is the activation function of the hidden layer.

[0034] In step s 3.4, the backpropagation error formula is shown in equation (3):

[0035]

[0036] In the formula, p is the sample size; This indicates the output data; This is the tag value.

[0037] In step s3.5, the weight adjustment formula for the hidden layer neurons is shown in equation (4):

[0038]

[0039] In the formula, This indicates the output data; For label values; w jk f1'(S) represents the weights between the input layer and the hidden layer. k f2'(S) is the partial derivative of the hidden layer transfer function; j) is the partial derivative of the output layer transfer function.

[0040] In step 4, the test sample is input into the flood level prediction and recognition model optimized by the BP neural network. The flood level of the test sample is predicted by the BP neural network optimized flood level prediction and recognition model to obtain the predicted value of the flood level of the test sample. The accuracy of the BP neural network optimized flood level prediction and recognition model is verified by calculating the root mean square error between the measured value and the predicted value of the flood level in the test sample.

[0041] The objective of this invention can also be achieved through the following technical measures: a water flooding level identification system based on a BP neural network, which uses a BP neural network-based water flooding level identification method to predict the water flooding level of oil drilling accidents.

[0042] The water flooding level identification method based on BP neural network in this invention belongs to the category of intelligent well logging interpretation. It selects the well logging curves that best reflect the characteristics of oil-water layers as input data, preprocesses them, and divides them into training and test sets. Based on the training set, a water flooding layer level prediction and identification model with a supervised learning algorithm is established. The BP algorithm is used to find the optimal weights and thresholds, resulting in a BP-optimized water flooding layer level prediction and identification model. This model is then used to predict test set data, and the error between the predicted and measured values ​​of the water flooding layer level is analyzed to verify the accuracy of the BP-optimized water flooding layer level prediction and identification. This invention, based on a BP neural network-optimized water flooding layer level prediction and identification model, overcomes the problems of poor stability and insufficient generalization ability in traditional machine learning, achieving accurate prediction of water flooding layer levels and laying the foundation for reservoir exploration and development. Compared with existing technologies, this invention has the following beneficial effects:

[0043] This invention utilizes the BP algorithm to optimize the Extreme Learning Machine (ELM) in the prediction and identification of flooding layers, obtaining the optimal weights and biases. The BP neural network is a multi-layer feedforward neural network, characterized by forward signal propagation and backward error propagation. In forward propagation, the input signal is processed layer by layer from the input layer through hidden layers until the output layer. The state of neurons in each layer only affects the state of neurons in the next layer. If the output layer does not produce the desired output, backpropagation begins, adjusting the network weights and thresholds based on the prediction error, thereby continuously approximating the desired output.

[0044] This invention trains an extreme learning machine for predicting and identifying water-flooded layers using curve data that best reflects the characteristics of oil and water layers as a training set. It utilizes a BP neural network model to deeply explore the intrinsic correlation between various logging parameter values ​​and water-flooded layer levels, establishes a prediction and identification system for water-flooded layer levels, improves the prediction accuracy of water-flooded layer levels, and facilitates the accurate acquisition of reservoir shear wave velocities, laying the foundation for reservoir engineering evaluation and fluid identification. Attached Figure Description

[0045] Figure 1 is a flowchart of a specific embodiment of the flood level identification method based on BP neural network of the present invention;

[0046] Figure 2 is a discrimination diagram of the flooding level in a specific embodiment of the present invention;

[0047] Figure 3 is a schematic diagram of the classification error obtained by using the BP algorithm for different numbers of hidden layer nodes in a specific embodiment of the present invention. Detailed Implementation

[0048] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0049] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, and / or combinations thereof.

[0050] The flood level identification method based on BP neural network includes the following steps:

[0051] s 1. By performing correlation analysis on the logging curves of water-flooded layers and conventional logging curves, the logging curve that best reflects the characteristics of oil-water layers is selected;

[0052] s2. After preprocessing the logging curve, the measured values ​​corresponding to each depth point on the logging curve are used as input data, and the input data is divided into training set and test set to obtain training samples and test samples;

[0053] s3. Based on the training set, a flood level prediction and identification model is established. A BP neural network model is set in the flood level prediction and identification model. The optimal weights and optimal biases of the extreme learning machine model are optimized by the BP neural network to obtain the flood level prediction and identification model optimized by the BP neural network.

[0054] s4. Input the test sample into the flood level prediction and recognition model optimized by the BP neural network, use the optimized BP neural network to predict the flood level of the test sample, obtain the predicted value of the flood level of the test sample, and verify the accuracy of the optimized BP neural network flood level prediction and recognition model by calculating the root mean square error between the measured value and the predicted value of the flood level in the test sample.

[0055] In step s2, the preprocessing of the logging curves includes outlier removal and normalization. The calculation formula for the logging curve normalization is as follows:

[0056]

[0057] In the formula, X norm X represents the measured depth value after normalization; X represents the measured depth value before normalization. max X represents the maximum measured depth value in the logging curve before normalization. min This represents the minimum measured depth value in the logging curve before normalization.

[0058] Step s3 includes the following steps:

[0059] s3.1. Construct a BP neural network model and study the network structure based on the characteristics of the system's input and output data, mainly by reasonably setting the number of nodes in the hidden layer and the activation function of the hidden layer;

[0060] s3.2. Train the BP neural network, and continuously adjust the network weights or thresholds based on the network prediction error during the training process;

[0061] In the forward propagation process of the s3.3.BP algorithm, the input signal enters the input layer, passes through the activation function to obtain a value, enters the hidden layer, and after processing, enters the output layer.

[0062] s3.4. Calculate whether the data obtained from the output layer is within the error range of the expected output data. If not, proceed to the backpropagation process.

[0063] s3.5. During the backpropagation process, the weight values ​​between layers and the bias values ​​between neurons in each layer are updated and adjusted along the way;

[0064] s3.6. Test the neural network model. Use the established classification model to predict and back-judge the flooded layer samples, test the classification effect of the model, and obtain the flood level prediction and recognition model optimized by the BP search algorithm.

[0065] Preferably, in step s3.3, the output formula of the hidden layer is as shown in equation (2):

[0066]

[0067] In the formula, z k For hidden layer output; v ki Set the weights between the input layer and the hidden layer; θ is the bias value; f1 is the activation function of the hidden layer.

[0068] Preferably, in step s 3.4, the backpropagation error formula is as shown in equation (3):

[0069]

[0070] In the formula, p is the sample size; This indicates the output data; This is the tag value.

[0071] Preferably, in step s3.5, the weight adjustment formula for each neuron in the hidden layer is as shown in equation (4):

[0072]

[0073] In the formula, This indicates the output data; For label values; w jk The weights are the weights between the input layer and the hidden layer.

[0074] The following are several specific embodiments of the application of the present invention.

[0075] Example 1

[0076] In a specific embodiment 1 of the present invention, a flooding level identification method based on a BP neural network proposed in this invention is used to predict the flooding level, as shown in Figure 1, specifically including the following steps:

[0077] s1. Obtain open-hole logging data from renewal wells in water-flooded areas. By analyzing the correlation between conventional logging curves and water-flooded layer levels, select spontaneous potential, spontaneous gamma ray, deep lateral resistivity, shallow lateral resistivity, and sonic transit time curves as model input data. These curves can reflect the characteristics of oil-water layers and can be used to predict water-flooded layer levels. The water-flooding level is determined based on core analysis, oil testing data, and production data, and is divided into four levels: high water-flooded, medium water-flooded, weak water-flooded, and no water-flooded.

[0078] s2. The acquired data is preprocessed to remove outliers on each logging curve and normalize the logging curve using formula (1). The measured values ​​corresponding to each depth point on the preprocessed logging curve are used as input data. The input data is further divided into training set and test set. The training set contains 6000 training samples and the test set contains 199 test data. The test data are all taken from the reservoir block of the study area.

[0079] s3. Based on the MATLAB software platform, a flooding level prediction and identification model is trained using the training set and the BP algorithm. The flooding level prediction and identification includes an extreme learning machine model. The BP algorithm is used to optimize the optimal weights and thresholds, resulting in the optimized flooding level prediction and identification. The specific steps include:

[0080] s3.1. Establish a flood level prediction and identification model based on the training set. Construct a BP neural network model for the flood level prediction and identification model. Study the structure of the network according to the characteristics of the system input and output data, mainly by reasonably setting the number of nodes in the hidden layer and the activation function of the hidden layer.

[0081] s3.2. Train the BP neural network, and continuously adjust the network weights or thresholds based on the network prediction error during the training process;

[0082] In the forward propagation process of the s3.3.BP algorithm, the input signal enters the input layer and passes through the activation function to obtain a value, which then enters the hidden layer. After processing, it enters the output layer, as shown in equation (2):

[0083]

[0084] In the formula, z k For hidden layer output; v ki Set the weights between the input layer and the hidden layer; θ is the bias value; f1 is the activation function of the hidden layer.

[0085] s3.4. Calculate whether the data obtained from the output layer is within the error range of the expected output data. If not, proceed to the backpropagation process, as shown in equation (3):

[0086] The formula for backpropagation error is shown in equation (3):

[0087]

[0088] In the formula, p is the sample size; This indicates the output data; This is the tag value.

[0089] s3.5. During the backpropagation process, the weight values ​​between layers and the bias values ​​between neurons in each layer are updated and adjusted as shown in equation (4):

[0090]

[0091] In the formula, This indicates the output data; For label values; w jk The weights are the weights between the input layer and the hidden layer.

[0092] s4. Input the test sample into the BP algorithm-optimized flood level prediction and identification system. Use the BP algorithm-optimized flood level prediction and identification system to predict the flood level of the test sample, obtaining the predicted value of the flood level of the test sample, as shown in Figure 2. By comparing the measured and predicted values ​​of the flood level in the test set, the predicted results are found to be basically consistent with the actual flood level, verifying the accuracy of the detection model. The flood level value predicted by the method of this invention is close to the actual measured value, and the prediction effect is good, verifying the accuracy of the BP neural network-based flood level identification method in this invention.

[0093] Example 2

[0094] In a specific embodiment 2 of the present invention, a flooding level identification method based on a BP neural network proposed in this invention is used to predict the flooding level, as shown in Figure 1, specifically including the following steps:

[0095] s1. Obtain open-hole logging data of the renewal well in the water-flooded area. By analyzing the correlation between conventional logging curves and water-flooded layer levels, select spontaneous potential, spontaneous gamma, deep lateral resistivity, shallow lateral resistivity, sonic transit time curve, porosity, permeability, and oil saturation as model input data. These curves can reflect the characteristics of oil-water layers and can be used to predict the water-flooded layer level. After analyzing the water-flooded layer samples, it is shown that the water-flooded level in this embodiment is divided into five levels: no water flooding, weak water flooding, medium water flooding, medium-strong water flooding, and strong water flooding.

[0096] s2. The acquired data is preprocessed to remove outliers on each logging curve and normalize the logging curve using formula (1). The measured values ​​corresponding to each depth point on the preprocessed logging curve are used as input data. The input data is further divided into training set and test set. The training set contains 800 training samples and the test set contains 100 test data. The test data are all taken from the same research block.

[0097] s3. Based on the MATLAB software platform, a flooding level prediction and identification model is trained using the training set and the BP algorithm. The flooding level prediction and identification includes an extreme learning machine model. The BP algorithm is used to optimize the optimal weights and thresholds, resulting in the optimized flooding level prediction and identification. The specific steps include:

[0098] s3.1. A flood level prediction and identification model is established based on the training set. This model is constructed using a BP neural network. The network structure is studied based on the characteristics of the system's input and output data, primarily focusing on the appropriate setting of the number of nodes and activation functions in the hidden layers. Generally, a larger number of hidden layer nodes results in a smaller error; however, increasing the number of nodes increases the computational load and time. Therefore, a reasonable number of hidden layer nodes is necessary. As shown in Figure 3, in this embodiment, the final flood level classification error decreases with increasing hidden layer nodes. However, the error reaches the required accuracy when the number of hidden layers is 12, thus a larger number of nodes is not necessary.

[0099] s3.2. Train the BP neural network, and continuously adjust the network weights or thresholds based on the network prediction error during the training process;

[0100] In the forward propagation process of the s3.3.BP algorithm, the input signal enters the input layer and passes through the activation function to obtain a value, which then enters the hidden layer. After processing, it enters the output layer, as shown in equation (2):

[0101]

[0102] In the formula, z k For hidden layer output; v ki Set the weights between the input layer and the hidden layer; θ is the bias value; f1 is the activation function of the hidden layer.

[0103] s3.4. Calculate whether the data obtained from the output layer is within the error range of the expected output data. If not, proceed to the backpropagation process, as shown in equation (3):

[0104] The formula for backpropagation error is shown in equation (3):

[0105]

[0106] In the formula, p is the sample size; This indicates the output data; This is the tag value.

[0107] s3.5. During the backpropagation process, the weight values ​​between layers and the bias values ​​between neurons in each layer are updated and adjusted as shown in equation (4):

[0108]

[0109] In the formula, This indicates the output data; For label values; w jk The weights are the weights between the input layer and the hidden layer.

[0110] s4. Input the test sample into the BP algorithm-optimized flood level prediction and identification system. Use the BP algorithm-optimized flood level prediction and identification system to predict the flood level of the test sample, obtaining the predicted value of the flood level. By comparing the measured and predicted values ​​of the flood level in the test set, the predicted results show a good match with the actual flood level, verifying the accuracy of the detection model. The flood level values ​​predicted by the method of this invention are close to the actual measured values, demonstrating good prediction performance and verifying the accuracy of the BP neural network-based flood level identification method in this invention.

[0111] Example 3

[0112] In a specific embodiment 3 of the present invention, and in a specific embodiment 2 of the present invention, a flooding level identification method based on a BP neural network proposed in the present invention is used to predict the flooding level, as shown in Figure 1, specifically including the following steps:

[0113] s 1. Obtain logging data of old wells with casing in water-flooded areas, including natural potential, natural gamma, deep lateral resistivity, shallow lateral resistivity, sonic transit time curve, porosity, permeability, original oil saturation, and through-casing resistivity curve as model input data. These curves can reflect the characteristics of oil-water layers and can be used to predict the level of water-flooded layers. Analysis of water-flooded layer samples shows that the water-flooding level in this embodiment is divided into four levels: no water-flooded, weakly water-flooded, moderately water-flooded, and strongly water-flooded.

[0114] s2. The acquired data is preprocessed to remove outliers on each logging curve and normalize the logging curve using formula (1). The measured values ​​corresponding to each depth point on the preprocessed logging curve are used as input data. The input data is further divided into training set and test set. The training set contains 800 training samples and the test set contains 100 test data. The test data are all taken from the same research block.

[0115] s3. Based on the MATLAB software platform, a flooding level prediction and identification model is trained using the training set and the BP algorithm. The flooding level prediction and identification includes an extreme learning machine model. The BP algorithm is used to optimize the optimal weights and thresholds, resulting in the optimized flooding level prediction and identification. The specific steps include:

[0116] s3.1. Establish a flood level prediction and identification model based on the training set. Construct a BP neural network model for the flood level prediction and identification model. Study the structure of the network according to the characteristics of the system input and output data, mainly by reasonably setting the number of nodes in the hidden layer and the activation function of the hidden layer.

[0117] s3.2. Train the BP neural network, and continuously adjust the network weights or thresholds based on the network prediction error during the training process;

[0118] In the forward propagation process of the s3.3.BP algorithm, the input signal enters the input layer and passes through the activation function to obtain a value, which then enters the hidden layer. After processing, it enters the output layer, as shown in equation (2):

[0119]

[0120] In the formula, z k For hidden layer output; v ki Set the weights between the input layer and the hidden layer; θ is the bias value; f1 is the activation function of the hidden layer.

[0121] s3.4. Calculate whether the data obtained from the output layer is within the error range of the expected output data. If not, proceed to the backpropagation process, as shown in equation (3):

[0122] The formula for backpropagation error is shown in equation (3):

[0123]

[0124] In the formula, p is the sample size; This indicates the output data; This is the tag value.

[0125] s3.5. During the backpropagation process, the weight values ​​between layers and the bias values ​​between neurons in each layer are updated and adjusted as shown in equation (4):

[0126]

[0127] In the formula, This indicates the output data; For label values; w jk The weights are the weights between the input layer and the hidden layer.

[0128] s4. Input the test sample into the BP algorithm-optimized flood level prediction and identification system. Use the BP algorithm-optimized flood level prediction and identification system to predict the flood level of the test sample, obtaining the predicted value of the flood level. By comparing the measured and predicted values ​​of the flood level in the test set, the predicted results show a good match with the actual flood level, verifying the accuracy of the detection model. The flood level values ​​predicted by the method of this invention are close to the actual measured values, demonstrating good prediction performance and verifying the accuracy of the BP neural network-based flood level identification method in this invention.

[0129] This invention provides three embodiments, each with different input data. Some embodiments use open-hole logging curves, while others use cased-hole logging curves. The open-hole logging input curves also vary depending on the logging method. Water-flooded formation samples can be divided into four or five water-flood levels. Therefore, it can be summarized that the input of this invention can cover various logging curves for different well types, and the water-flood levels are also divided into different levels according to the reservoir characteristics of different blocks. The method of this invention can be applied to classify the water-flood levels in all of these cases.

[0130] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0131] Except for the technical features described in the specification, all other technologies are known to those skilled in the art.

Claims

1. A flood level identification method based on BP neural network, characterized in that, The water flooding level identification method based on BP neural network includes: Step 1, acquiring well logging data and selecting the well logging curve that best reflects the characteristics of the oil-water layer; Step 2, preprocessing the well logging curve and dividing the data into training set and test set; Step 3, optimizing to obtain the water flooding level prediction and identification model optimized by BP neural network; Step 4, verifying the accuracy of the water flooding level prediction and identification model optimized by BP neural network by calculating the root mean square error between the actual value and the predicted value of the water flooding level in the test sample.

2. The flood level identification method based on BP neural network according to claim 1, characterized in that, In step 1, by performing correlation analysis between the water flooding level of the water flooded layer and the conventional logging curve, the logging curve that best reflects the characteristics of the oil-water layer is selected.

3. The flood level identification method based on BP neural network according to claim 2, characterized in that, In step 1, well logging data of the reservoir is obtained. By analyzing the correlation between conventional well logging curves and water-flooded layer levels, water-flooded layer analysis data after data standardization and dimensionality reduction is selected as model input data. These curves can reflect the characteristics of oil-water layers and can be used to predict water-flooded layer levels.

4. The flood level identification method based on BP neural network according to claim 1, characterized in that, In step 2, the acquired logging curve data is preprocessed by removing outliers from each logging curve and normalizing the logging curves. The measured values ​​corresponding to each depth point on the preprocessed logging curve are used as input data.

5. The flood level identification method based on a BP neural network according to claim 4, characterized in that, In step 2, the calculation formula for the normalization of well logging curves is: In the formula, X norm X represents the measured depth value after normalization; X represents the measured depth value before normalization. max X represents the maximum measured depth value in the logging curve before normalization. min This represents the minimum measured depth value in the logging curve before normalization.

6. The flood level identification method based on a BP neural network according to claim 5, characterized in that, In step 2, after preprocessing the logging curve, the measured values ​​corresponding to each depth point on the logging curve are used as input data, and the input data is divided into training set and test set to obtain training samples and test samples.

7. The flood level identification method based on BP neural network according to claim 1, characterized in that, In step 3, a flood level prediction and identification model is established based on the training set. A BP neural network model is set in the flood level prediction and identification model. The optimal weights and optimal biases of the extreme learning machine model are optimized by the BP neural network to obtain the flood level prediction and identification model optimized by the BP neural network.

8. The flood level identification method based on a BP neural network according to claim 7, characterized in that, Step 3 specifically includes: s3.

1. Constructing a BP neural network model, studying the network structure based on the characteristics of the system's input and output data, and reasonably setting the number of nodes and activation functions of the hidden layers; s3.

2. Training the BP neural network, continuously adjusting the network weights or thresholds based on the network's prediction error during training; s3.

3. During the forward propagation of the BP algorithm, the input signal enters the input layer and passes through the activation function to obtain a value that enters the hidden layer, and after processing, enters the output layer; s3.

4. Calculating whether the data obtained from the output layer is within the error range of the expected output data. If not, proceeding to the back propagation process; s3.

5. Updating and adjusting the weight values ​​between layers and the bias values ​​between neurons in each layer along the back propagation process; s3.

6. Testing the neural network model, using the established classification model to predict and back-judge flooded layer samples, testing the model's classification effect, and obtaining the flood level prediction and recognition model optimized by the BP search algorithm.

9. The flood level identification method based on a BP neural network according to claim 8, characterized in that, In step s3.3, the output formula of the hidden layer is shown in equation (2): In the formula, x i For input data; z k For hidden layer output; v ki Set the weights between the input layer and the hidden layer; θ is the bias value; f1 is the activation function of the hidden layer.

10. The flood level identification method based on a BP neural network according to claim 9, characterized in that, In step s3.4, the backpropagation error formula is shown in equation (3): In the formula, p is the sample size; This indicates the output data. This is the tag value.

11. The flood level identification method based on a BP neural network according to claim 10, characterized in that, In step s3.5, the weight adjustment formula for the hidden layer neurons is shown in equation (4): In the formula, η is the learning rate; This indicates the output data; For label values; w jk f1'(S) represents the weights between the input layer and the hidden layer. k f2'(S) is the partial derivative of the hidden layer transfer function; j ) is the partial derivative of the output layer transfer function.

12. The flood level identification method based on BP neural network according to claim 1, characterized in that, In step 4, the test sample is input into the flood level prediction and recognition model optimized by the BP neural network. The flood level of the test sample is predicted by the BP neural network optimized flood level prediction and recognition model to obtain the predicted value of the flood level of the test sample. The accuracy of the BP neural network optimized flood level prediction and recognition model is verified by calculating the root mean square error between the measured value and the predicted value of the flood level in the test sample.

13. A flood level identification system based on a BP neural network, characterized in that, The BP neural network-based flooding level identification system uses the BP neural network-based flooding level identification method described in any one of claims 1-12 to predict the flooding level of oil drilling accidents.

Images