An artificial intelligence-based drilling loss identification method

Abstract

This invention discloses an artificial intelligence-based method for identifying wellbore leakage in drilling operations, comprising the following steps: S1: Collecting multi-dimensional data from drilling operations and manually labeling it; S2: Using a generative adversarial network based on quantum field optimization to generate samples and expand the data; S3: Extracting features from the expanded data and performing dimensionality reduction; S4: Using a dynamically weighted support vector machine to classify and train the dimensionality-reduced features, and optimizing the classifier model parameters based on real-time environmental factors; S5: Using the trained classifier model to identify wellbore leakage. This invention utilizes a dynamically weighted support vector machine, achieving sensitive adaptation to classification difficulty samples and environmental changes through dynamic weight adjustment based on error feedback and optimization of classifier penalty parameters combined with real-time environmental factors.

Description

Technical Field

[0001] This invention relates to the field of wellbore leakage identification technology, and in particular to a wellbore leakage identification method based on artificial intelligence. Background Technology

[0002] Well leakage is a significant issue affecting drilling efficiency and safety during operations, and its occurrence is often sudden and complex. Accurate identification of well leakage is crucial for optimizing drilling processes, reducing costs, and ensuring operational safety. However, traditional well leakage identification methods rely primarily on manual experience or single physical models, resulting in low efficiency, poor real-time performance, and insufficient accuracy. With advancements in drilling data acquisition technology, intelligent identification methods based on big data and artificial intelligence are gaining increasing attention.

[0003] Training well leakage detection models relies on high-quality sample data. However, the amount of real data collected in existing technologies is limited and unevenly distributed, which can easily lead to insufficient generalization ability of the models, especially when dealing with rare or complex samples. Furthermore, traditional classification algorithms are difficult to effectively adjust when dealing with difficult-to-classify samples, and they are slow to respond to dynamic changes in environmental conditions during drilling, resulting in insufficient classification performance and real-time capability, failing to meet the accuracy and timeliness requirements of well leakage detection. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides an artificial intelligence-based method for identifying well leakage in drilling operations.

[0005] This invention is achieved using the following technical solution: a drilling well leakage identification method based on artificial intelligence, comprising the following steps: S1: Collect multi-dimensional data during drilling operations and manually annotate it; S2: A generative adversarial network based on quantum field optimization is used to generate samples and augment the data. S3: Extract features from the expanded data and perform dimensionality reduction. S4: Use dynamic weight adjustment support vector machine to classify and train the dimensionality-reduced features, and optimize the classifier model parameters according to real-time environmental factors; S5: Use the trained classifier model to identify wellbore leakage.

[0006] Furthermore, the multi-dimensional data includes one or more of pressure, temperature, drilling speed, and drilling fluid properties, and the manually labeled categories include: no well leakage, slight well leakage, moderate well leakage, and severe well leakage.

[0007] Furthermore, the quantum field-optimized generative adversarial network optimizes the parameters of the generator and discriminator by simulating quantum behavior, and uses quantum state energy difference and system effective temperature to control the noise mode, thereby improving the quality and diversity of generated samples.

[0008] Furthermore, the specific training method for the quantum field-optimized generative adversarial network is as follows: Initialize the network parameters of the generator and discriminator of the generative adversarial network; In the quantum optimization loop iteration phase, the generator is first updated. The generator accepts random noise as input and uses the current network parameters to generate data samples. Based on the accuracy of the discriminator in distinguishing between real and fake data samples, the training strategies of the generator and the discriminator are dynamically adjusted, and the training intensity of the generator is either increased or decreased. The quality and diversity of the generated data are evaluated. Training ends if the predetermined convergence criteria are met or the quality of the generated data meets the requirements.

[0009] Furthermore, the parameter update method of the generator is as follows: ; In the formula, For the first The generator parameters for the next iteration; For the first The generator parameters for the next iteration; To generate the learning rate for the adversarial network, For the input random noise, The gradient with respect to the generator parameters; For generator functions; For the discriminator function; It is a noise vector obtained based on quantum behavior simulation.

[0010] Furthermore, the convergence criterion is evaluated by measuring the quality and diversity of the generated data, and the calculation method is expressed as follows: ; In the formula, As a quality assessment indicator; For quality evaluation functions, The number of samples generated, For generator functions; For the discriminator function, Indicates the first One real data sample; For the first The input is random noise.

[0011] Furthermore, step S3 includes the following sub-steps: The neural network algorithm based on dynamic adaptive oscillation extracts data features, optimizes the gradient problem, and improves the quality of feature extraction. By using an autoencoder based on boundary smoothing to reduce the dimensionality of the extracted features, the robustness to boundary outliers is enhanced.

[0012] Furthermore, the dynamically weighted support vector machine optimizes the classification process through a weight adjustment mechanism based on error feedback, and dynamically adjusts the penalty parameter and kernel function bandwidth according to real-time drilling data to improve the ability to identify samples with classification difficulty.

[0013] Furthermore, the specific training method for the dynamically weighted support vector machine includes: Initialize the parameters of the support vector machine classifier; A dynamic weight adjustment mechanism is set up, and the rules for weight adjustment are defined. Specifically, a weight increase / decrease strategy based on error feedback is used to ensure that misclassified samples receive more attention in subsequent iterations. During the training iterations, the weight of each sample is adjusted based on the classification results of the previous iteration; The constraints of the model are dynamically adjusted based on the current model performance and external environmental conditions. Repeat the above steps until the preset stopping iteration condition is met, which means that the model training is complete.

[0014] Furthermore, the calculation method for dynamic weight adjustment is as follows: ; In the formula, It is the first In the first iteration The weights of each sample, It is the learning rate of the support vector machine. It is an indicator function, if the first... If a sample is misclassified under the current model, the result is 1; otherwise, it is 0. Is the model in parameters The following are the prediction results; For the first The parameters of the support vector machine in the next iteration.

[0015] The beneficial effects of this invention are as follows: This invention employs a quantum field-optimized generative adversarial network (GAN) to optimize generator and discriminator parameters by simulating quantum behavior. This generates high-quality samples to expand the training data volume, alleviating the problem of insufficient training samples and enhancing the model's generalization ability. It utilizes a dynamically weighted support vector machine (SVM) to achieve sensitive adaptation to classification difficulty samples and environmental changes through dynamic weight adjustment based on error feedback and optimization of classifier penalty parameters combined with real-time environmental factors.

[0016] This invention expands the data using generative adversarial networks (GANs), enabling the model training samples to cover a wider distribution range and improving the training quality and generalization performance of the classification model. Through dynamic weight adjustment and environmental factor feedback, the classifier can more effectively identify complex samples and has better adaptability to changes in environmental conditions, thereby improving classification accuracy and real-time performance. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0018] Figure 1 This is a flowchart of the present invention; Figure 2 Based on this, a schematic diagram of the adversarial network principle is generated. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0020] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0021] The following detailed description of some embodiments of the present invention is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0022] See Figure 1 A drilling well leakage identification method based on artificial intelligence includes the following steps: S1: Collect multi-dimensional data during drilling operations and manually annotate it; S2: A generative adversarial network based on quantum field optimization is used to generate samples and augment the data. S3: Extract features from the expanded data and perform dimensionality reduction. S4: Use dynamic weight adjustment support vector machine to classify and train the dimensionality-reduced features, and optimize the classifier model parameters according to real-time environmental factors; S5: Use the trained classifier model to identify wellbore leakage.

[0023] In this embodiment, the data primarily originates from oilfield drilling operations and is collected in real-time by sensors. This data includes parameters such as pressure, temperature, drilling speed, and drilling fluid properties. The data acquisition system comprises multiple sensors installed on key parts of the drilling rig to monitor and record data in real time. The collected data is transmitted to the central processing unit via a data interface and stored in a high-performance database in a structured data table format. It should be noted that this embodiment only illustrates one data format and type; in practical applications, data typically has more than 10 attributes, and the number of attributes may reach dozens or even hundreds. The collected data is labeled manually. In one embodiment, the labeling categories include: no well leakage, slight well leakage, moderate well leakage, and severe well leakage.

[0024] The attributes of the data include: This indicates the drill bit rotation speed (rpm). Indicates the drilling depth (m). This indicates drilling fluid pressure (bar). This indicates the drilling fluid temperature (°C). This indicates the drilling speed (m / h). Density of drilling fluid (g / cm³) 3 ), This indicates the viscosity of the drilling fluid (cp). This indicates the drilling fluid flow rate (L / s). This indicates the stability index of the wellbore. This indicates the rock abrasion index.

[0025] In this embodiment, the acquisition, annotation, and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor model generalization ability and affect model accuracy. It is understood that in the task of this invention, the acquisition, annotation, and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor model generalization ability and affect model accuracy. Specifically, this invention uses a generative adversarial network based on quantum field optimization to generate samples, thereby achieving data augmentation. The generative adversarial network includes a generator and a discriminator, where the generator is responsible for generating data samples that are as realistic as possible, and the discriminator's task is to distinguish between real samples and generated samples. See also... Figure 2 Based on traditional generative adversarial networks, this invention employs a dynamic optimization method based on quantum field theory. By simulating the behavior of quantum particles, the parameters of the generator and discriminator are optimized, thereby achieving faster convergence speed and higher data generation quality, improving model stability and the diversity of generated data.

[0026] Specifically, the training process for the generative adversarial network algorithm based on quantum field optimization is as follows: S201. Initialize the network parameters of the generator and discriminator of the generative adversarial network. A quantum behavior simulation-based method is used to set the initial parameters, which helps to provide a good starting point in the early stages of training and reduces local optima problems. Specifically, the parameter initialization method is expressed as follows: ; ; In the formula, and These are the initial parameters for the generator and the discriminator, respectively. Random perturbations for parameter initialization. To control the variance of the disturbance distribution, preferably, Set to 0.01.

[0027] S202. In the quantum optimization loop iteration stage, the generator is updated first. The generator accepts random noise as input and uses the current network parameters to generate data samples. During the generator update process, quantum behavior principles are employed to simulate the uncertainty and randomness in deep learning, thereby mimicking the unpredictability and complexity of real-world data generation. Specifically, by adjusting the quantum state energy difference and the system's effective temperature, the amount and pattern of noise injection can be precisely controlled, thus enhancing the diversity and quality of generation while maintaining model stability. The generator parameter update method is expressed as follows: ; In the formula, For the first The generator parameters for the next iteration; For the first The generator parameters for the next iteration; To generate the learning rate for the adversarial network, For the input random noise, The gradient with respect to the generator parameters; For generator functions; For the discriminator function; It is a noise vector obtained based on quantum behavior simulation.

[0028] The noise vector obtained from quantum behavior simulation is represented as follows: ; In the formula, It is a random variable drawn from a standard normal distribution. It is the quantum state energy difference, the potential energy difference between the simulated sample and the real sample; It is Boltzmann's constant. It is the effective temperature of the system, used to control the intensity of noise.

[0029] The gradient of the generator parameters is calculated as follows: ; In the formula, This indicates that the generated sample corresponds to the gradient output by the discriminator.

[0030] The discriminator is updated to classify fake and real data samples received from the generator. The discriminator update method is represented as follows: ; In the formula, For the first Discriminator parameters for the next iteration; For the first Discriminator parameters for the next iteration; This is a real data sample; This represents the gradient of the discriminator parameters.

[0031] The gradient of the discriminator parameters is calculated as follows: ; In the formula, and These represent the parameter gradients of the discriminator output with respect to the real samples and the generated samples, respectively.

[0032] S203. Based on the discriminator's accuracy in distinguishing between real and fake data samples, dynamically adjust the training strategies of the generator and discriminator. If the discriminator easily identifies fake samples, increase the training intensity of the generator; otherwise, decrease it. The adaptive adjustment method of the learning rate is expressed as follows: ; In the formula, It is a coefficient for adjusting the intensity, representing the average difference in the discriminator's ability to distinguish between real and generated samples, and is used to dynamically adjust the intensity of the learning rate.

[0033] S204. Check the convergence of the algorithm and evaluate the quality and diversity of the generated data. If the predetermined convergence criterion is met or the quality of the generated data meets the requirements, then the training ends. In one embodiment, the convergence evaluation is achieved through a measure of the quality and diversity of the generated data, and the calculation method is expressed as follows: ; In the formula, As a quality assessment indicator; The quality evaluation function, in this embodiment, uses a discriminator to assess the discrimination accuracy of the generated samples.

[0034] If the quality assessment metric exceeds a preset threshold, the iteration stops, indicating that the generative adversarial network (GAN) training is complete. After the data augmentation model is trained, the number of samples is increased using the trained model. In one embodiment, if the original collected samples are 800, and the data augmentation model generates 200 additional samples, then the augmented dataset contains 1000 samples.

[0035] After data expansion, a neural network algorithm based on dynamic adaptive oscillation is used to extract data features, optimize the gradient problem, and improve the quality of feature extraction. The extracted features are then dimensionality-reduced using an autoencoder based on boundary smoothing to enhance robustness to boundary outliers. Finally, the dimensionality-reduced data is input into a classifier for model training. This invention employs a support vector machine (SVM) algorithm based on dynamic weight adjustment as the classification algorithm. By adjusting the kernel function and penalty parameters in the SVM, accurate classification of different types of well leakage is achieved. Building upon traditional SVM, this invention uses a dynamic weight adjustment mechanism, dynamically adjusting weights based on the classification difficulty of samples in each iteration. This allows the model to focus more on difficult-to-classify samples, adapting to the rapid changes in the environment and operating conditions during drilling operations.

[0036] Specifically, the training process of the support vector machine algorithm based on dynamic weight adjustment is as follows: S301. Initialize the parameters of the Support Vector Machine (SVM) classifier. Specifically, the kernel function and penalty parameters of the SVM are set using the following formulas: ; ; In the formula, The kernel function for a support vector machine; and These represent different input data samples. This is the first input sample for the support vector machine. This is the second input sample for the support vector machine; It is an L2 norm; The bandwidth parameter of the Gaussian kernel determines the amplitude of the kernel function; It is the first The label of each sample It is the average of all sample labels; It is the total number of samples input into the support vector machine; This is the penalty parameter for the support vector machine, used to adjust the model's tolerance to errors.

[0037] S302. Set up a dynamic weight adjustment mechanism and define the rules for weight adjustment. Specifically, based on the error feedback, a weight increase / decrease strategy is implemented to ensure that misclassified samples receive more attention in subsequent iterations. The dynamic weight adjustment method is expressed as follows: ; In the formula, It is the first In the first iteration The weights of each sample, It is the learning rate of the support vector machine. It is an indicator function, if the first... If a sample is misclassified under the current model, the result is 1; otherwise, it is 0. Is the model in parameters The following are the prediction results; For the first The parameters of the support vector machine in the next iteration.

[0038] The indicator function uses the marginal distance between samples to determine whether to increase the weight, and is expressed as: ; ; In the formula, It is the first The marginal distance between samples determines the accuracy of the classification; It is the first The transpose of the parameters of the support vector machine in the next iteration.

[0039] S303. During the training iteration, the weight of each sample is adjusted based on the classification results of the previous iteration. The weight of misclassified samples is increased, and the weight of correctly classified samples is decreased. The core of each training iteration is to solve the following weighted minimization problem: ; The constraints are: ; In the formula, These are model parameters. It is a function that maps input data to a higher-dimensional space. It is a bias term; These are slack variables used to handle cases where the data is not completely separable.

[0040] S304. Based on the current model performance and external environmental conditions, dynamically adjust the model's constraints. Specifically, dynamic constraint adjustment is achieved by changing the bandwidth parameter of the Gaussian kernel. The bandwidth parameter of the kernel function is adjusted according to the proportion of classification errors, making the model more sensitive or more tolerant to changes in the distribution of data points. The adjustment method is expressed as follows: ; In the formula, It is the first The bandwidth parameter of the Gaussian kernel in the next iteration; It is the first The bandwidth parameter of the Gaussian kernel in the next iteration; It is the adjustment strength parameter of the support vector machine.

[0041] Furthermore, the model parameter adaptive adjustment mechanism based on dynamic environmental feedback dynamically adjusts the classifier's penalty parameters according to real-time environmental data monitored during drilling (such as drilling depth, drilling speed, and formation pressure) to optimize the model's classification performance. Specifically, during training iterations, environmental factors are extracted from real-time data, reflecting the complexity and rate of change of the current drilling operation. These environmental factors are used to adjust the penalty parameters, making the model more sensitive to immediate environmental changes, thereby improving the accuracy and timeliness of well leakage detection. The calculation method for environmental factors is as follows: ; In the formula, For the environment factors of the support vector machine; , and These represent the measured values ​​of current drilling depth, drilling rate, and formation pressure, respectively. The first environmental weighting coefficient, This is the second environmental weighting coefficient. This is the third environmental weighting coefficient.

[0042] The penalty parameter of the support vector machine (SVM) is adjusted based on environmental factors. This increases the penalty parameter when environmental conditions change significantly, thereby enhancing the model's sensitivity to classification errors. The adjustment method is expressed as follows: ; In the formula, It is an environmental factor adjustment coefficient; For the first The penalty parameters of the support vector machine in the next iteration; For the first The penalty parameters for the support vector machine in the next iteration. Preferably, Set it to 0.2.

[0043] S305. Repeat the above steps until a preset stopping iteration condition is met, indicating that the model training is complete. In one embodiment, the preset stopping iteration condition is reaching a preset maximum number of iterations; preferably, the preset maximum number of iterations is set to 1000.

[0044] Finally, the trained model is used to identify wellbore leakage. The collected raw data is input into the trained feature extraction and feature dimensionality reduction model for feature processing. Furthermore, the processed features are input into a classifier model for classification, thus obtaining the classification results. In this embodiment, the classification categories include: no wellbore leakage, slight wellbore leakage, moderate wellbore leakage, and severe wellbore leakage.

[0045] This invention obtains drilling fluid rheological parameters without the need for any mechanized equipment or sensors, enabling real-time measurement at a frequency down to the second level. Furthermore, it can be infinitely replicated to various fluid rheological testing applications, such as engineering slurries, trenchless drilling, and tunneling, making it widely applicable. Moreover, this invention combines artificial intelligence to predict wellbore collapse pressure in multi-cleavage coal seams. The research results have significant guiding significance for ensuring wellbore stability in coal seams and achieving efficient and safe coal seam drilling.

[0046] For the foregoing embodiments, in order to simplify the description, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to this application.

[0047] The above embodiments describe the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Modifications and variations made by those skilled in the art without departing from the spirit and scope of the invention should be within the protection scope of the appended claims.

Claims

1. A drilling well leakage identification method based on artificial intelligence, characterized in that, Includes the following steps: S1: Collect multi-dimensional data during drilling operations and manually annotate it; S2: A quantum field-optimized generative adversarial network (GAN) is used for sample generation to augment the data. The GAN optimizes the generator and discriminator parameters by simulating quantum behavior, and uses quantum state energy difference and system effective temperature to control the noise mode, thereby improving the quality and diversity of the generated samples. The generator's parameter update method is as follows: ； In the formula, For the first The generator parameters for the next iteration; For the first The generator parameters for the next iteration; To generate the learning rate for the adversarial network, For the input random noise, The gradient with respect to the generator parameters; For generator functions; For the discriminator function; This is a noise vector obtained based on quantum behavior simulation, and its calculation method is expressed as follows: ； In the formula, It is a random variable drawn from a standard normal distribution. It is the quantum state energy difference, the potential energy difference between the simulated sample and the real sample; It is Boltzmann's constant. It is the effective temperature of the system, used to control the intensity of noise; Learning rate The adaptive adjustment method is expressed as: ； In the formula, It is a coefficient for adjusting the intensity, representing the average difference in the discriminator's ability to distinguish between real and generated samples, and is used to dynamically adjust the intensity of the learning rate. The number of samples generated; S3: Extract features from the expanded data and perform dimensionality reduction. S4: Use dynamic weight adjustment support vector machine to classify and train the dimensionality-reduced features, and optimize the classifier model parameters according to real-time environmental factors; The dynamically weighted support vector machine optimizes the classification process through a weight adjustment mechanism based on error feedback, and dynamically adjusts the penalty parameter and kernel function bandwidth according to real-time drilling data to improve the ability to identify samples with classification difficulty; wherein, the bandwidth adjustment method is expressed as follows: ； In the formula, It is the first The bandwidth parameter of the Gaussian kernel in the next iteration; It is the first The bandwidth parameter of the Gaussian kernel in the next iteration; These are the adjustment strength parameters of the support vector machine; It is an indicator function, if the first... If a sample is misclassified under the current model, the result is 1; otherwise, it is 0. Is the model in parameters The following are the prediction results; For the first The parameters of the support vector machine in the next iteration It is the total number of samples input into the support vector machine; The calculation method for the environmental factors is expressed as follows: ； In the formula, For the environment factors of the support vector machine; , and These represent the measured values ​​of current drilling depth, drilling rate, and formation pressure, respectively. The first environmental weighting coefficient, This is the second environmental weighting coefficient. This is the third environmental weighting coefficient; The dynamic adjustment method of the penalty parameter is expressed as follows: ； In the formula, It is the environmental factor adjustment coefficient; For the first The penalty parameters of the support vector machine in the next iteration; For the first The penalty parameters of the support vector machine in the next iteration; The specific training method for the dynamically weighted support vector machine includes: Initialize the parameters of the support vector machine classifier; Set up a dynamic weight adjustment mechanism and define the rules for weight adjustment; During the training iterations, the weight of each sample is adjusted based on the classification results of the previous iteration; The constraints of the model are dynamically adjusted based on the current model performance and external environmental conditions. The training steps of the support vector machine with dynamic weight adjustment are repeated until the preset stopping iteration condition is met, which means that the model training is complete. The dynamic weight adjustment mechanism is set up, defining the rules for weight adjustment. Specifically, it employs a weight increase / decrease strategy based on error feedback, ensuring that misclassified samples receive more attention in subsequent iterations. The calculation method for dynamic weight adjustment is as follows: ； In the formula, It is the first In the first iteration The weights of each sample; It is the learning rate of the support vector machine; It is an indicator function, if the first... If a sample is misclassified under the current model, the result is 1; otherwise, it is 0. Is the model in parameters The following are the prediction results; For the first The parameters of the support vector machine in the next iteration; S5: Use the trained classifier model to identify wellbore leakage.

2. The drilling leakage identification method based on artificial intelligence as described in claim 1, characterized in that, The multi-dimensional data includes one or more of the following: pressure, temperature, drilling speed, and drilling fluid properties. The categories of manually marked wells include: no well leakage, slight well leakage, moderate well leakage, and severe well leakage.

3. The drilling leakage identification method based on artificial intelligence as described in claim 1, characterized in that, The specific training method for the quantum field-optimized generative adversarial network is as follows: Initialize the network parameters of the generator and discriminator of the generative adversarial network; In the quantum optimization loop iteration phase, the generator is first updated. The generator accepts random noise as input and uses the current network parameters to generate data samples. Based on the accuracy of the discriminator in distinguishing between real and fake data samples, the training strategies of the generator and discriminator are dynamically adjusted, and the training intensity of the generator is either increased or decreased.

4. The drilling leakage identification method based on artificial intelligence as described in claim 1, characterized in that, The specific training method for the quantum field-optimized generative adversarial network also includes: The quality and diversity of the generated data are evaluated. Training ends if the predetermined convergence criterion is met or the quality of the generated data satisfies the requirements. The convergence criterion is evaluated by measuring the quality and diversity of the generated data, and the calculation method is as follows: ； In the formula, As a quality assessment indicator; For quality evaluation functions, The number of samples generated, For generator functions; For the discriminator function, Indicates the first One real data sample; For the first The input is random noise.

5. The drilling leakage identification method based on artificial intelligence as described in claim 1, characterized in that, Step S3 includes the following sub-steps: The neural network algorithm based on dynamic adaptive oscillation extracts data features, optimizes the gradient problem, and improves the quality of feature extraction. By using an autoencoder based on boundary smoothing to reduce the dimensionality of the extracted features, the robustness to boundary outliers is enhanced.

Images