Dynamic model detection method based on ai algorithm

Abstract

The application provides a dynamic model detection method based on an AI algorithm, and relates to the technical field of model detection. The dynamic model detection method based on the AI algorithm acquires running data of a model to be detected in real time through a multi-dimensional data acquisition module, extracts and analyzes the running data by using an improved deep learning algorithm, realizes accurate detection of an abnormal state of the model in combination with a dynamic threshold adjustment mechanism, generates a detailed detection report, issues a warning signal, feeds back abnormal information in the detection report to a model design and development team, and optimizes and improves the model according to the information. The application can effectively improve the real-time performance, accuracy and adaptability of model detection, solves the problems of low detection efficiency and high false alarm rate of a traditional model detection method in a complex dynamic scene, and can be widely applied to the detection field of various dynamic models such as artificial intelligence models and industrial control models.

Description

Technical Field

[0001] This invention relates to the field of model detection technology, specifically to a dynamic model detection method based on AI algorithms. Background Technology

[0002] With the rapid development of technologies such as artificial intelligence and industrial automation, various dynamic models have been widely applied in production, daily life, and scientific research, such as deep learning models in artificial intelligence, process control models in industrial control, and risk prediction models in the financial field. During operation, these dynamic models may be affected by factors such as changes in the external environment, fluctuations in data quality, and model parameter drift, leading to abnormal operating states and consequently affecting the accuracy of the model's output and the stability of the system. Therefore, real-time and accurate detection of dynamic models is of significant practical importance. Traditional model detection methods mainly include rule-based and statistical methods. Rule-based methods use pre-defined rules and thresholds to evaluate model performance data; when data exceeds these rules or thresholds, the model is considered anomaly-prone. However, this method is highly dependent on rules and thresholds, making it difficult to adapt to the complex and ever-changing operating scenarios of dynamic models. When the model's operating environment changes, rules and thresholds need to be manually readjusted, resulting in poor flexibility and a high risk of false positives and false negatives. Statistical methods analyze the statistical characteristics of model performance data, such as mean, variance, and probability distribution, to build a statistical model for anomaly detection. However, this method makes strict assumptions about the statistical distribution of the data, leading to poor detection performance when dealing with unstructured or non-linear data. It also struggles to capture complex relationships within the data, resulting in low detection accuracy. With the continuous development of AI algorithms, technologies such as deep learning and machine learning have provided new approaches to model detection. While some AI-based model detection methods exist, most focus on static model detection, neglecting the real-time performance of dynamic models. Furthermore, they fail to fully extract key feature information during feature extraction, resulting in lower detection efficiency and accuracy. Additionally, existing methods typically employ fixed detection thresholds, unable to dynamically adjust based on the model's operating scenario and historical data, leading to poor adaptability when facing models in different operating states. Therefore, there is an urgent need for a detection method that can improve the real-time performance, accuracy, and adaptability of dynamic model detection in order to solve the problems existing in the current technology. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a dynamic model detection method based on AI algorithms, which solves the problems of low detection efficiency, high false alarm rate, and poor adaptability of traditional model detection methods in dynamic model detection.

[0004] To achieve the above objectives, the present invention provides the following technical solution: The dynamic model detection method based on AI algorithms includes the following steps: Step S1. Multi-dimensional data collection A multi-dimensional data acquisition module is constructed to determine the key operating parameters of the model to be tested. The key operating parameter data of the model to be tested are collected in real time through this module to form the original operating dataset. Step S2. Data Preprocessing The original dataset is cleaned, standardized, and normalized. Data cleaning removes outliers and missing values ​​to ensure data quality. Standardization and normalization convert the data into a format that meets the input requirements of AI algorithms, eliminating the impact of differences in data units on the detection results. Step S3. Improve the construction of deep learning detection model Based on the convolutional neural network (CNN) framework, attention mechanism and residual connection structure are introduced. Attention mechanism highlights the importance of key features, and residual connection structure solves the gradient vanishing problem, thereby improving the model’s feature extraction capability and training stability. Step S4. Model Training The improved deep learning detection model is trained using labeled normal and abnormal operation datasets. The cross-entropy loss function and Adam optimization algorithm are used, combined with data augmentation techniques to expand the training dataset, improve the model's generalization ability and anti-interference ability, until the model converges to obtain the trained AI detection model. Step S5. Anomaly Identification The preprocessed dataset is input into the trained AI detection model, which performs feature extraction and anomaly identification on the data and outputs anomaly detection results. Step S6. Dynamic threshold adjustment A dynamic threshold adjustment mechanism is constructed to adjust the anomaly detection threshold in real time based on the model's operating scenario and historical detection data through statistical analysis and machine learning algorithms. This includes threshold initialization, updating, and verification steps to ensure the accuracy and adaptability of detection. Step S7. Anomaly Detection and Report Generation By combining the output of the AI ​​detection model with the dynamically adjusted threshold, it is determined whether the model is abnormal. If it is abnormal, a detailed detection report containing the abnormality type, occurrence time, and scope of impact is generated, and an early warning signal is issued. Step S8. Model Optimization Feedback The abnormal information in the test report is fed back to the model design and development team, who then optimize and improve the model based on the information.

[0005] Furthermore, in step S1, the multi-dimensional data acquisition module adopts a distributed data acquisition architecture, including data acquisition nodes, data transmission nodes, and data storage nodes; The data acquisition node uses multiple acquisition methods such as sensors and data interfaces to collect the running data of the model under test in real time; The data transmission node employs an encrypted transmission protocol to ensure the security of data transmission. The data storage nodes employ a distributed database to achieve efficient storage and management of large amounts of operational data.

[0006] Furthermore, in step S1, the key operating parameters cover model input data, output data, intermediate calculation results, running time, and resource utilization, comprehensively reflecting the operating status of the model.

[0007] Furthermore, in step S2, the data cleaning employs an outlier detection method based on statistical analysis and a missing value filling method based on interpolation. The outlier detection method based on statistical analysis determines the range of outliers by calculating the mean, standard deviation, and quartiles of the data and removes data that exceeds the range. The missing value filling method based on interpolation uses linear interpolation or polynomial interpolation to fill in the missing data and ensure the integrity of the data. The data standardization method uses Z-score standardization to convert the data into standard normal distribution data with a mean of 0 and a standard deviation of 1. The data normalization uses the Min-Max normalization method to map the data to the interval [0, 1].

[0008] Furthermore, in step S3, the attention mechanism adopts a combination of channel attention mechanism and spatial attention mechanism. The channel attention mechanism learns the weights of feature channels with different operating parameters to highlight the information of important feature channels, and the spatial attention mechanism learns the spatial position weights of feature maps to highlight the feature information of key spatial positions. The residual connection structure uses a shortcut connection method to directly add the output of the previous layer to the input of the next layer, reducing gradient decay during model training.

[0009] Furthermore, in step S4, when training the improved deep learning detection model, data augmentation techniques are used to expand the training dataset. The data augmentation techniques include random cropping, rotation, flipping, and adding noise. The random cropping generates new data samples by cropping the original data samples to a random size. The rotation increases the diversity of the data by rotating the original data samples at random angles; The flipping is achieved by horizontally or vertically flipping the original data sample to increase the amount of data. The addition of noise improves the model's robustness by adding Gaussian noise or salt-and-pepper noise to the original data samples.

[0010] Further, in step S6, the threshold initialization adopts a statistical method based on historical normal operation data to calculate the statistical characteristic values ​​of the historical normal operation data, including the mean, standard deviation, maximum value and minimum value, and determine the initial anomaly detection threshold based on the statistical characteristic values; The threshold update adopts a sliding window mechanism to obtain the latest running data and detection results in real time, and dynamically adjusts the threshold through machine learning algorithms, including support vector machine (SVM), decision tree or random forest. The threshold verification verifies the effectiveness of the adjusted threshold by comparing the detection accuracy, false alarm rate, and false negative rate before and after the threshold adjustment. If the verification fails, the threshold is updated again.

[0011] Furthermore, in step S7, the anomaly types include data anomaly, calculation anomaly, performance anomaly, and functional anomaly. The data anomaly refers to the input data or output data exceeding the normal range. The calculation anomaly refers to errors or deviations in intermediate calculation results. The performance anomaly refers to the model taking too long to run or having too high resource usage. The functional anomaly refers to the model output results not matching the expected function. The test report is presented in a visual format, including charts, curves, and text descriptions. The charts are used to show the changing trends of the model's operating parameters, the curves are used to compare the differences between normal operating data and abnormal operating data, and the text descriptions are used to describe the abnormal situation in detail and provide handling suggestions.

[0012] Furthermore, in step S8, the optimization and improvement of the model includes adjusting the model structure, optimizing algorithm parameters, and improving the data processing flow to form a closed-loop mechanism for model detection and optimization.

[0013] This invention provides a dynamic model detection method based on AI algorithms. It has the following beneficial effects: 1. This invention provides a dynamic model detection method based on AI algorithms. It adopts a distributed data acquisition architecture, which can collect the running data of the model to be detected in real time. At the same time, the improved deep learning detection model has high computational efficiency and can quickly process the running data and identify anomalies, thus meeting the needs of real-time detection of dynamic models.

[0014] 2. This invention provides a dynamic model detection method based on AI algorithms. By introducing an attention mechanism and a residual connection structure to improve the deep learning detection model, it can fully mine the key feature information in the running data and improve the accuracy of feature extraction. The data preprocessing step ensures the quality of the input data and reduces the interference of noise on the detection results. The dynamic threshold adjustment mechanism can adjust the threshold in real time according to the model running status, reducing the false alarm rate and false negative rate, and further improving the detection accuracy.

[0015] 3. This invention provides a dynamic model detection method based on AI algorithms. Its dynamic threshold adjustment mechanism can adapt to different model operation scenarios and data changes without the need for frequent manual threshold adjustments. At the same time, this method can be applied to the detection of various dynamic models such as artificial intelligence models, industrial control models, and financial risk prediction models, and has a wide range of applications and strong adaptability.

[0016] 4. This invention provides a dynamic model detection method based on AI algorithms. Through a model optimization feedback step, the detected anomaly information is used to optimize and improve the model, realizing a virtuous cycle of model detection and optimization, which helps to improve the overall performance and stability of the model. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the steps of the AI-based dynamic model detection method of the present invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0019] In the description of this invention, unless otherwise explicitly specified and limited, the terms "connected," "linked," and "fixed" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0020] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0021] In the description of this embodiment, the terms "upper," "lower," "right," etc., refer to the orientation or positional relationship shown in the accompanying drawings. They are used only for ease of description and simplification of operation, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention. In addition, the terms "first" and "second" are used only for distinction in description and have no special meaning.

[0022] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0023] like Figure 1 As shown, this embodiment of the invention provides a dynamic model detection method based on AI algorithms, including the following steps: Step S1. Multi-dimensional data collection A multi-dimensional data acquisition module is constructed to determine the key operating parameters of the model to be tested. The key operating parameter data of the model to be tested are collected in real time through this module to form the original operating dataset. The multi-dimensional data acquisition module adopts a distributed data acquisition architecture, including data acquisition nodes, data transmission nodes, and data storage nodes; The data acquisition node uses multiple acquisition methods such as sensors and data interfaces to collect the running data of the model under test in real time; The data transmission node employs an encrypted transmission protocol to ensure the security of data transmission. The data storage node uses a distributed database to achieve efficient storage and management of large amounts of operational data; Key operating parameters cover model input data, output data, intermediate calculation results, running time, and resource utilization, comprehensively reflecting the model's operating status.

[0024] Step S2. Data Preprocessing The original dataset is cleaned, standardized, and normalized. Data cleaning removes outliers and missing values ​​to ensure data quality. Standardization and normalization convert the data into a format that meets the input requirements of AI algorithms, eliminating the impact of differences in data units on the detection results. Data cleaning employs outlier detection based on statistical analysis and missing value imputation based on interpolation. The outlier detection method based on statistical analysis determines the range of outliers by calculating the mean, standard deviation, and quartiles of the data and removes data that exceeds the range. The missing value imputation method based on interpolation uses linear interpolation or polynomial interpolation to fill in missing data and ensure data integrity. The data standardization method uses Z-score standardization to convert the data into standard normal distribution data with a mean of 0 and a standard deviation of 1. The data normalization uses the Min-Max normalization method to map the data to the interval [0, 1].

[0025] Step S3. Improve the construction of deep learning detection model Based on the convolutional neural network (CNN) framework, attention mechanism and residual connection structure are introduced. Attention mechanism highlights the importance of key features, and residual connection structure solves the gradient vanishing problem, thereby improving the model’s feature extraction capability and training stability. The attention mechanism combines channel attention and spatial attention. The channel attention mechanism learns the weights of feature channels with different operating parameters to highlight the information of important feature channels, while the spatial attention mechanism learns the spatial position weights of the feature map to highlight the feature information of key spatial positions. The residual connection structure uses a shortcut connection method to directly add the output of the previous layer to the input of the next layer, reducing gradient decay during model training.

[0026] Step S4. Model Training The improved deep learning detection model is trained using labeled normal and abnormal operation datasets. The cross-entropy loss function and Adam optimization algorithm are used, combined with data augmentation techniques to expand the training dataset, improve the model's generalization ability and anti-interference ability, until the model converges to obtain the trained AI detection model. When training the improved deep learning detection model, data augmentation techniques are used to expand the training dataset. These data augmentation techniques include random cropping, rotation, flipping, and adding noise. The random cropping generates new data samples by cropping the original data samples to a random size. The rotation increases the diversity of the data by rotating the original data samples at random angles; The flipping is achieved by horizontally or vertically flipping the original data sample to increase the amount of data. The addition of noise improves the model's robustness by adding Gaussian noise or salt-and-pepper noise to the original data samples.

[0027] Step S5. Anomaly Identification The preprocessed dataset is input into the trained AI detection model, which performs feature extraction and anomaly identification on the data and outputs anomaly detection results. Step S6. Dynamic threshold adjustment A dynamic threshold adjustment mechanism is constructed to adjust the anomaly detection threshold in real time based on the model's operating scenario and historical detection data through statistical analysis and machine learning algorithms. This includes threshold initialization, updating, and verification steps to ensure the accuracy and adaptability of detection. The threshold initialization adopts a statistical method based on historical normal operation data to calculate the statistical characteristic values ​​of the historical normal operation data, including the mean, standard deviation, maximum value and minimum value, and determine the initial anomaly detection threshold based on the statistical characteristic values; The threshold update adopts a sliding window mechanism to obtain the latest running data and detection results in real time, and dynamically adjusts the threshold through machine learning algorithms, including support vector machine (SVM), decision tree or random forest. The threshold verification verifies the effectiveness of the adjusted threshold by comparing the detection accuracy, false alarm rate, and false negative rate before and after the threshold adjustment. If the verification fails, the threshold is updated again.

[0028] Step S7. Anomaly Detection and Report Generation By combining the output of the AI ​​detection model with the dynamically adjusted threshold, it is determined whether the model is abnormal. If it is abnormal, a detailed detection report containing the abnormality type, occurrence time, and scope of impact is generated, and an early warning signal is issued. The types of anomalies include data anomalies, computational anomalies, performance anomalies, and functional anomalies. Data anomalies refer to input or output data exceeding the normal range. Computational anomalies refer to errors or deviations in intermediate calculation results. Performance anomalies refer to excessively long model execution time or excessively high resource consumption. Functional anomalies refer to model output results that do not match the expected function. The test report is presented in a visual format, including charts, curves, and text descriptions. The charts are used to show the changing trends of the model's operating parameters, the curves are used to compare the differences between normal operating data and abnormal operating data, and the text descriptions are used to describe the abnormal situation in detail and provide handling suggestions.

[0029] Step S8. Model Optimization Feedback The abnormal information in the test report is fed back to the model design and development team. The team optimizes and improves the model based on the abnormal information, including adjusting the model structure, optimizing algorithm parameters, and improving the data processing flow, thus forming a closed-loop mechanism for model testing and optimization.

[0030] This invention is specifically applied to a dynamic detection method for industrial control models based on AI algorithms, comprising the following processes: 1) Multi-dimensional data acquisition: The model to be tested is the temperature control model of a factory. The key operating parameters to be determined include the temperature setpoint, actual temperature value, heating equipment power, heat dissipation equipment speed, running time, and CPU utilization. The multi-dimensional data acquisition module's acquisition nodes collect actual temperature values ​​through temperature sensors and parameters such as heating equipment power and heat dissipation equipment rotation speed through device interfaces. The data transmission nodes use the SSL encrypted transmission protocol, and the data storage nodes use a Hadoop distributed database to store the collected raw running datasets.

[0031] 2) Data preprocessing: An outlier detection method based on the 3σ principle (a statistical analysis-based outlier detection method) is adopted to calculate the mean and standard deviation of parameters such as actual temperature value and heating equipment power. Data exceeding the range of [mean - 3 × standard deviation, mean + 3 × standard deviation] are identified as outliers and removed; for missing values, linear interpolation is used for filling. The data is standardized using the Z-score standardization method, and then mapped to the [0, 1] interval using the Min-Max normalization method to obtain the preprocessed running dataset.

[0032] 3) Improved deep learning detection model construction: Based on the CNN framework, the channel attention mechanism is implemented through the Squeeze-and-Excitation (SE) module to learn the weights of feature channels with different operating parameters; The spatial attention mechanism generates a spatial attention weight map by adding a spatial attention module after the channel attention output feature map; The residual connection structure uses two convolutional layers plus shortcut connections to alleviate the gradient vanishing problem.

[0033] 4) Model training: Collect the normal operation data and historical abnormal operation data (such as excessive temperature fluctuations, abnormal equipment power, etc.) of the temperature control model over the past year, label the data, and form a training dataset; The training dataset was augmented using data augmentation techniques including random cropping (cropping ratio of 0.8-1.0), rotation (rotation angle of -10° to 10°), horizontal flipping, and adding Gaussian noise (noise standard deviation of 0.01). The expanded training dataset was divided into training and validation sets in an 8:2 ratio. The improved deep learning detection model was then trained using the data. The error was calculated using the cross-entropy loss function. The Adam optimization algorithm (with an initial learning rate of 0.001, which decays to 0.9 every 10 epochs) was used to adjust the model parameters. After 50 epochs of training, the model converged, resulting in the trained AI detection model.

[0034] 5) Anomaly detection: Real-time acquisition of operating data from the temperature control model, followed by preprocessing and input into the trained AI detection model. The model outputs anomaly detection results, such as "Actual temperature value deviates from the set value by 10℃, suspected anomaly". 6) Dynamic Threshold Adjustment: In the threshold initialization phase, the mean deviation of the actual temperature value from the set value in historical normal operation data is calculated to be 2℃, and the standard deviation is 1℃. The initial anomaly detection threshold is set to 5℃ (mean + 3 × standard deviation). In the threshold update phase, a sliding window (window size of 100 data points) is used to obtain the latest operating data. The threshold is dynamically adjusted using a random forest algorithm. If three consecutive data deviations are detected to be close to the threshold, the threshold is adjusted to 4.5℃. In the threshold verification phase, the detection accuracy (92% before adjustment, 95% after adjustment), false alarm rate (6% before adjustment, 3% after adjustment), and false negative rate (2% before adjustment, 1% after adjustment) are compared before adjustment to verify the effectiveness of the adjusted threshold.

[0035] 7) Anomaly Judgment and Report Generation: When the actual temperature value output by the AI ​​detection model deviates from the set value by 10℃ and exceeds the dynamically adjusted threshold of 4.5℃, the model is judged to be in an abnormal operating state, a detection report is generated, indicating that the anomaly type is "temperature control anomaly", the time of the anomaly is "date, hour and minute of 202X", the scope of the anomaly is "temperature in area A of the production workshop exceeds the standard, which may affect product quality", and an audible and visual warning signal is issued through the factory's monitoring system. 8) Model optimization feedback: The test report was fed back to the development team of the temperature control model. The team analyzed the data and found that the unstable power output was caused by the aging of the heating equipment. The heating equipment was replaced and the PID control parameters of the model were adjusted. After testing, the optimized model showed improved temperature control accuracy and reduced anomaly rate.

[0036] The following points should be noted in this article: 1. The accompanying drawings of the embodiments disclosed herein only relate to the structures involved in the embodiments disclosed herein; other structures can be referred to in a general design.

[0037] 2. Where there is no conflict, the embodiments of this disclosure and the features in the embodiments can be combined with each other to obtain new embodiments.

[0038] Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

Claims

1. A dynamic model detection method based on AI algorithms, characterized in that, Includes the following steps: Step S1. Multi-dimensional data collection A multi-dimensional data acquisition module is constructed to determine the key operating parameters of the model to be tested. This module is used to collect the key operating parameter data of the model to be tested in real time to form the original operating dataset. Step S2. Data Preprocessing The original dataset is cleaned, standardized, and normalized. Data cleaning removes outliers and missing values ​​to ensure data quality. Standardization and normalization convert the data into a format that meets the input requirements of AI algorithms, eliminating the impact of differences in data units on the detection results. Step S3. Improve the construction of deep learning detection model Based on the convolutional neural network framework, attention mechanism and residual connection structure are introduced. Attention mechanism highlights the importance of key features, and residual connection structure solves the gradient vanishing problem, thereby improving the feature extraction capability and training stability of the model. Step S4. Model Training The improved deep learning detection model is trained using labeled normal and abnormal operation datasets. The cross-entropy loss function and Adam optimization algorithm are used, combined with data augmentation techniques to expand the training dataset, improve the model's generalization ability and anti-interference ability, until the model converges to obtain the trained AI detection model. Step S5. Anomaly Identification The preprocessed dataset is input into the trained AI detection model, which performs feature extraction and anomaly identification on the data and outputs anomaly detection results. Step S6. Dynamic threshold adjustment A dynamic threshold adjustment mechanism is constructed to adjust the anomaly detection threshold in real time based on the model's operating scenario and historical detection data through statistical analysis and machine learning algorithms. This includes threshold initialization, updating, and verification steps to ensure the accuracy and adaptability of detection. Step S7. Anomaly Detection and Report Generation By combining the output of the AI ​​detection model with the dynamically adjusted threshold, it is determined whether the model is abnormal. If it is abnormal, a detailed detection report containing the abnormality type, occurrence time, and scope of impact is generated, and an early warning signal is issued. Step S8. Model Optimization Feedback The abnormal information in the test report is fed back to the model design and development team, who then optimize and improve the model based on the abnormal information.

2. The dynamic model detection method based on AI algorithm according to claim 1, characterized in that, In step S1, the multi-dimensional data acquisition module adopts a distributed data acquisition architecture, including data acquisition nodes, data transmission nodes, and data storage nodes; The data acquisition node uses multiple acquisition methods such as sensors and data interfaces to collect the running data of the model under test in real time; The data transmission node employs an encrypted transmission protocol to ensure the security of data transmission. The data storage nodes employ a distributed database to achieve efficient storage and management of large amounts of operational data.

3. The dynamic model detection method based on AI algorithm according to claim 1, characterized in that, In step S1, the key operating parameters cover model input data, output data, intermediate calculation results, running time, and resource utilization, comprehensively reflecting the operating status of the model.

4. The dynamic model detection method based on AI algorithm according to claim 1, characterized in that, In step S2, the data cleaning employs a statistical analysis-based outlier detection method and an interpolation-based missing value filling method. The statistical analysis-based outlier detection method calculates the mean, standard deviation, and quartiles of the data to determine the range of outliers and removes data that exceeds the range. The interpolation-based missing value filling method uses linear interpolation or polynomial interpolation to fill in the missing data and ensure data integrity. The data standardization method uses Z-score standardization to convert the data into standard normal distribution data with a mean of 0 and a standard deviation of 1. The data normalization uses the Min-Max normalization method to map the data to the interval [0, 1].

5. The dynamic model detection method based on AI algorithm according to claim 1, characterized in that, In step S3, the attention mechanism adopts a combination of channel attention mechanism and spatial attention mechanism. The channel attention mechanism learns the weights of feature channels with different operating parameters to highlight the information of important feature channels, and the spatial attention mechanism learns the spatial position weights of feature maps to highlight the feature information of key spatial positions. The residual connection structure uses a shortcut connection method to directly add the output of the previous layer to the input of the next layer, reducing gradient decay during model training.

6. The dynamic model detection method based on AI algorithm according to claim 1, characterized in that, In step S4, when training the improved deep learning detection model, data augmentation techniques are used to expand the training dataset. The data augmentation techniques include random cropping, rotation, flipping, and adding noise. The random cropping generates new data samples by cropping the original data samples to a random size. The rotation increases the diversity of the data by rotating the original data samples at random angles; The flipping is achieved by horizontally or vertically flipping the original data sample to increase the amount of data. The addition of noise improves the model's robustness by adding Gaussian noise or salt-and-pepper noise to the original data samples.

7. The dynamic model detection method based on AI algorithm according to claim 1, characterized in that, In step S6, the threshold initialization adopts a statistical method based on historical normal operation data to calculate the statistical feature values ​​of historical normal operation data, including mean, standard deviation, maximum value and minimum value, and determine the initial anomaly detection threshold based on the statistical feature values; The threshold update adopts a sliding window mechanism to obtain the latest running data and detection results in real time, and dynamically adjusts the threshold through machine learning algorithms, including support vector machines, decision trees, or random forests. The threshold verification verifies the effectiveness of the adjusted threshold by comparing the detection accuracy, false alarm rate, and false negative rate before and after the threshold adjustment. If the verification fails, the threshold is updated again.

8. The dynamic model detection method based on AI algorithm according to claim 1, characterized in that, In step S7, the anomaly types include data anomaly, calculation anomaly, performance anomaly, and functional anomaly. Data anomaly refers to input data or output data exceeding the normal range. Calculation anomaly refers to errors or deviations in intermediate calculation results. Performance anomaly refers to excessively long model execution time or excessively high resource utilization. Functional anomaly refers to model output results not matching the expected function. The test report is presented in a visual format, including charts, curves, and text descriptions. The charts are used to show the changing trends of the model's operating parameters, the curves are used to compare the differences between normal operating data and abnormal operating data, and the text descriptions are used to describe the abnormal situation in detail and provide handling suggestions.

9. The dynamic model detection method based on AI algorithm according to claim 1, characterized in that, In step S8, the optimization and improvement of the model includes adjusting the model structure, optimizing algorithm parameters, and improving the data processing flow to form a closed-loop mechanism for model detection and optimization.

Images