Monitoring evaluation method and system based on historical motion data

Abstract

The invention discloses a monitoring evaluation method and system based on historical motion data, and relates to the technical field of intelligent monitoring and data analysis, and the method comprises the steps: S1, extracting a plurality of motion features from historical motion data, and building a feature library; the method comprises the steps of S1, building an anomaly detection model and training the model to adapt to different motion scenes, S3, counting the occurrence frequency of false positive and analyzing the cause of the false positive according to a detection result output by the trained anomaly detection model, and S4, optimally setting a detection threshold value of an abnormal motion mode based on an analysis result to reduce the probability of the false positive. According to the monitoring evaluation method and system based on the historical motion data, the detection threshold value of the abnormal motion mode is optimally set so as to solve the problem that the false positive is too high in early warning.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent monitoring and data analysis, and particularly to a monitoring and evaluation method and system based on historical motion data. Background Art

[0002] In the prior art, the monitoring and evaluation methods based on historical motion data are widely used in the detection and early warning of abnormal motion patterns. These methods analyze the historical motion data of the target, establish a detection model based on statistics or machine learning, so as to realize the recognition and evaluation of abnormal motion patterns. Such methods have important application values in the fields of industrial production, traffic management, and public safety, and can effectively improve the safety and operation efficiency of the system.

[0003] However, when the existing monitoring and evaluation methods detect abnormal motion patterns, they usually rely on fixed detection thresholds. These fixed thresholds have poor adaptability under different environmental conditions or motion patterns, and are likely to cause two main problems: on the one hand, too loose thresholds may lead to the missed reporting of abnormal patterns, making the early warning function unable to play its role in time; on the other hand, too strict thresholds may trigger a large number of false positive alarms, increasing the workload of the system and even reducing the user's trust in the system. Especially in the early warning scenario, the problem of too high false positives is particularly prominent. This problem not only wastes limited resources but also may interfere with the subsequent decision-making process. For example, in the status monitoring of industrial production equipment, frequent false warnings will trigger unnecessary maintenance operations, resulting in a decline in production efficiency. At the same time, due to the lack of a mechanism for dynamically adjusting thresholds, the existing methods are powerless in the face of complex and variable motion patterns, thus limiting the accuracy and reliability of abnormal detection. In summary, the key defect of the prior art in the detection of abnormal motion patterns is that it fails to dynamically adjust the detection threshold according to the actual scenario requirements and historical data, resulting in a high false positive rate and affecting the effectiveness of early warning and the user experience. Summary of the Invention

[0004] The purpose of the present invention is to provide a monitoring and evaluation method and system based on historical motion data, and optimize the setting of the detection threshold for abnormal motion patterns to solve the problem of too high false positives in early warning.

[0005] To achieve the above purpose, the present invention provides the following technical solution: A monitoring and evaluation method based on historical motion data, the method includes: S1. Extract various motion features from the historical motion data and establish a feature library; S2. Based on the feature library, construct an abnormal detection model and train the model to adapt to different motion scenarios; S3. According to the detection results output by the trained anomaly detection model, count the occurrence frequency of false positives and analyze its causes, including counting the occurrence frequency of false positives and evaluating the errors in the output of the trained anomaly detection model. The specific formula is: ; where A represents the anomaly metric value, D represents the actual motion pattern frequency detected during monitoring, G represents the expected frequency, and i represents the index of each motion pattern; S4. Based on the analysis results, optimize the detection threshold for abnormal motion patterns to reduce the probability of false positives, including optimizing the detection threshold for abnormal motion patterns based on the results of analyzing the causes of false positives. The specific formula is: α 新 = ; where α 新 represents the optimized detection threshold, α represents the current monitoring threshold, a represents the learning rate, and b represents the loss function.

[0006] Preferably, S1 includes dimensionality reduction of historical motion data, reducing the data dimension and extracting features. The specific formula is: B = D × E; where B represents the transformed data, D represents the original data matrix, and E represents the feature vector matrix; The extracted features are made part of the database.

[0007] Preferably, S2 includes constructing an anomaly detection model based on the data in the feature library and making decisions by integrating multiple decision trees. The specific formula is: ; where H(x) represents the output of the final model, M represents the number of decision trees, x represents the input features, and K j (x) represents the prediction result of the j-th decision tree for the input feature x, and j represents the index of the decision tree.

[0008] Preferably, S1 also includes using an autoencoder for non-linear dimensionality reduction of historical data.

[0009] Preferably, the calculation formula for the feature vector matrix E in S1 is: E = V × ; where E represents the feature vector matrix, V represents the eigenvector of the covariance matrix, and represents the eigenvalue matrix of the covariance matrix.

[0010] Preferably, integrating multiple decision trees in S2 includes using different subsets of training data to train each decision tree.

[0011] A monitoring and evaluation system based on historical motion data, which is used to implement the steps of the monitoring and evaluation method based on historical motion data. The system includes: A data acquisition module, which is used to extract various motion features from historical motion data and establish a feature library; An anomaly detection module connected to the data acquisition module, which is used to build an anomaly detection model based on the feature library and train the model to adapt to different motion scenarios; A false positive statistical analysis module connected to the anomaly detection module, which is used to count the occurrence frequency of false positives and analyze its causes according to the detection results output by the trained anomaly detection model; A threshold optimization module connected to the anomaly detection module and the false positive statistical analysis module, which is used to optimize the detection threshold of abnormal motion patterns based on the false positive analysis results to reduce the probability of false positives.

[0012] Preferably, the data acquisition module includes a data preprocessing unit, which is used to denoise, standardize and format the collected historical motion data to ensure the accuracy and consistency of the data.

[0013] Preferably, the false positive statistical analysis module includes a statistical analysis unit, which is used to calculate the false positive rate in different motion scenarios and analyze the statistical significance of false positives by combining chi-square tests.

[0014] Preferably, the false positive statistical analysis module further includes a feedback adjustment unit, which automatically adjusts the parameters of the anomaly detection model according to the false positive analysis results to dynamically adapt to changes in motion scenarios.

[0015] As can be seen from the above technical solutions, the present invention has the following beneficial effects: The monitoring and evaluation method and system based on historical motion data extract various motion features from historical motion data, establish a feature library, build an anomaly detection model based on the feature library, train the model to adapt to different motion scenarios, count the occurrence frequency of false positives according to the detection results output by the trained anomaly detection model, analyze its causes, and optimize the detection threshold of the abnormal motion mode based on the analysis results to reduce the probability of false positives. It can effectively avoid the missed alarm and false positive problems caused by overly loose or strict fixed thresholds, improve the adaptability and accuracy of the detection model in different environments, and enhance the user's trust in the system. Especially in the early warning scenario, through more accurate threshold setting, the number of false alarms can be significantly reduced, the timeliness and accuracy of early warning can be improved, the monitoring and evaluation system can be ensured to operate efficiently in various actual application environments, false alarms or missed alarms caused by improper threshold setting can be avoided, unnecessary maintenance operations and over-responses can be avoided, and the operation efficiency of the system can be improved. In scenarios such as industrial equipment monitoring, early warning can be ensured to be timely and effective, thereby reducing production interruptions, lowering maintenance costs, and improving overall production efficiency, enhancing the user's trust and dependence on the monitoring system, improving the user's overall experience, and optimizing the detection threshold of abnormal motion modes to solve the problem of excessive false positives in early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flowchart of the method of the present invention; Figure 2 is a schematic diagram of the connection of system modules of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0018] As Figure 1 shown, the present invention provides a technical solution: a monitoring and evaluation method based on historical motion data, the method includes: S1. Extract various motion features from historical motion data and establish a feature library; S2. Build an anomaly detection model based on the feature library and train the model to adapt to different motion scenarios; S3. According to the detection results output by the trained anomaly detection model, count the occurrence frequency of false positives and analyze its causes, including counting the occurrence frequency of false positives and evaluating the error in the output of the trained anomaly detection model. The specific formula is: ; Wherein, A represents an anomaly metric, D represents the actual motion pattern frequency detected during monitoring, G represents the expected frequency, and i represents the index of each motion pattern; S4. Based on the analysis results, optimize the detection threshold for abnormal motion patterns to reduce the probability of false positives, including optimizing the detection threshold for abnormal motion patterns based on the results of analyzing the causes of false positives. The specific formula is: α 新 = ; Wherein, α 新 represents the optimized detection threshold, α represents the current monitoring threshold, a represents the learning rate, and b represents the loss function.

[0019] In an embodiment of the present invention, first, a variety of motion features are extracted from historical motion data, and a feature library is established based on these features. The feature library includes indicators such as motion frequency, intensity, and rhythm. These features provide data support for subsequent anomaly detection. Then, an anomaly detection model is constructed based on the feature library. The model is trained for different motion scenarios so that it can flexibly adapt to different motion types and motion environments. After the model training is completed, potential abnormal states are detected through real-time monitoring and analysis of new data, and detection results are output based on the comparison between historical data and real-time data. The output results of the model will be further analyzed to count the frequency of false positives and optimize the detection model in combination with the causes of false positives. By adjusting the model parameters and algorithms, it performs more accurately in a specific environment, thereby achieving the adaptability and precision of anomaly detection. The implementation scheme of the present invention has significant beneficial effects. First, based on the extraction of a variety of motion features and model training, the detection accuracy of abnormal states during the motion process can be significantly improved, thereby reducing the misdetection rate and ensuring the accuracy and reliability of the monitoring system. Second, by optimizing and adjusting the detection results, the probability of false positives can be effectively reduced, and the stability of the detection system can be improved, especially in complex motion scenarios. Third, the invention has an automatic adjustment ability. During real-time monitoring, the system can adjust the detection strategy according to different motion states and environmental changes to ensure its continuous adaptation and provide accurate feedback. In addition, the present invention also has high flexibility and can be customized and optimized according to different hardware platforms, providing targeted solutions in different application scenarios, thereby meeting various requirements and enhancing the wide applicability of the system.

[0020] S1 includes dimensionality reduction of historical motion data, reducing the data dimension and extracting features. The specific formula is: B = D × E; wherein, B represents the transformed data, D represents the original data matrix, and E represents the feature vector matrix; the extracted features are made part of the database.

[0021] First, perform dimensionality reduction on the historical motion data to reduce the data dimension in order to extract more representative features. This process is completed through matrix transformation by a mathematical model. Specifically, let the original motion data matrix be D. By multiplying it with the eigenvector matrix E, the transformed data B is obtained, that is: B = D × E, where B represents the transformed data, D is the original motion data matrix, and E is the eigenvector matrix containing the feature information related to the motion data. Through this dimensionality reduction method, not only is the data complexity reduced, but the core information of the data is also retained. On this basis, the extracted features will be added to the feature library and become the basis for subsequent anomaly detection and evaluation. These feature data can provide a rich information source for subsequent monitoring and evaluation and help train the model to judge the abnormal state during the motion process. The core advantage of this implementation is that it can effectively reduce the dimension of the original data, thereby greatly improving the data processing efficiency. Through the dimensionality reduction operation, redundant information can be removed, and the features crucial for anomaly detection are retained, making the subsequent model training and anomaly detection more efficient and accurate. In addition, the dimensionality-reduced feature library provides more concise and powerful data support for subsequent data analysis, which helps to improve the performance of the overall monitoring system. By reducing the computational amount and storage requirements, the system can still maintain a high response speed and processing capacity under large-scale data, ensuring the stability and accuracy of real-time monitoring.

[0022] S2 includes constructing an anomaly detection model based on the data in the feature library and making decisions by integrating multiple decision trees. The specific formula is: ; where H(x) represents the output of the final model, M represents the number of decision trees, x represents the input feature, and K j (x) represents the prediction result of the j-th decision tree for the input feature x, and j represents the index of the decision tree.

[0023] In this implementation, first, based on the feature library of historical motion data, multiple anomaly detection models are constructed. Each model learns through different algorithms and training methods to adapt to different types of motion data and abnormal situations. By integrating these decision tree models, a powerful anomaly detection system is finally formed. Specifically, the method of decision tree integration is used to make prediction decisions. For the input feature x, the output results of multiple decision trees M are weighted and averaged to obtain the final detection result. The formula is: , where H(x) represents the output of the final model, M is the number of decision trees, x is the input feature, and K j(x) represents the prediction result of the j-th decision tree for the input feature x, where j represents the index of the decision tree. Through this ensemble learning method, the prediction results of multiple models are comprehensively considered, effectively reducing the misjudgments that may be caused by a single model, thereby improving the accuracy of anomaly detection and the reliability of the system. This embodiment has significant beneficial effects. By integrating multiple decision tree models, the overfitting problem of a single decision tree can be effectively avoided, improving the robustness and generalization ability of the system. In addition, by performing weighted averaging on the prediction results of each decision tree, the advantages of different models can be comprehensively considered, further enhancing the accuracy of anomaly detection. In practical applications, this method can detect abnormal states during the movement in real time and accurately, avoiding unnecessary interference caused by system misjudgments. At the same time, the ensemble learning method also improves the adaptability of the model in different movement scenarios, enabling the monitoring and evaluation system to operate stably in various environments and different types of data.

[0024] S1 also includes using an autoencoder to perform non-linear dimensionality reduction on historical data. In this embodiment, in addition to performing traditional linear dimensionality reduction on historical movement data, an autoencoder is introduced for non-linear dimensionality reduction processing. An autoencoder is an artificial neural network that can map the high-dimensional features of historical data to a low-dimensional space, effectively compressing the data and preserving its main features during this process. Specifically, an autoencoder consists of an encoder and a decoder. The encoder maps the input data to a low-dimensional latent space, and the decoder then attempts to reconstruct the original data from this low-dimensional space. By training the autoencoder, dimensionality reduction can be performed in a non-linear space, extracting more representative movement features. Finally, the data after dimensionality reduction will be used as part of the feature library for subsequent anomaly detection and evaluation. The dimensionality reduction process of the autoencoder can be expressed by the following formula: ; where B is the data after non-linear dimensionality reduction, D is the original data matrix, is the encoder function. In this way, the non-linear relationships in the data are effectively retained, enhancing the accuracy of feature extraction.

[0025] Compared with traditional linear dimensionality reduction methods, this embodiment can more effectively process historical movement data with complex non-linear relationships. By learning the non-linear structure of the data, the autoencoder can retain more valuable information. Especially when the data contains complex patterns and multiple movement features, it can improve the accuracy of feature extraction. In addition, non-linear dimensionality reduction can extract the latent features of the data at a higher level, making the subsequent anomaly detection more accurate and sensitive. This method enables the monitoring and evaluation system to not only perform excellently when processing movement data with linear relationships, but also ensure high efficiency and good adaptability when processing complex and non-linear relationship movement data.

[0026] The calculation formula of the eigenvector matrix E in S1 is: E = V × ; where E represents the eigenvector matrix, V represents the eigenvectors of the covariance matrix, represents the eigenvalue matrix of the covariance matrix.

[0027] In this embodiment, the calculation of the eigenvector matrix E is achieved through eigenvalue decomposition. Specifically, assuming that V represents the eigenvectors of the covariance matrix and Λ represents the eigenvalues of the covariance matrix, then the eigenvector matrix E can be calculated by the following formula: E = V × , where E represents the eigenvector matrix, V represents the eigenvectors of the covariance matrix, Λ represents the eigenvalues of the covariance matrix, and Λ 1 / 2 represents the square root of the eigenvalue matrix. Through eigenvalue decomposition, we can extract the most representative features from the covariance matrix, thus retaining the feature information crucial for anomaly detection in the eigenvector matrix E. This step helps to further reduce the dimension and extract the most discriminative features, providing more accurate data support for subsequent anomaly detection and monitoring evaluation. Through eigenvalue decomposition and the calculation of the covariance matrix, we can more precisely extract the principal components in the data and construct the eigenvector matrix based on this. Compared with traditional linear dimensionality reduction methods, eigenvalue decomposition can better retain the variability in the data. Especially for complex motion data, it can more accurately capture the important features. The feature information extracted in this way can not only effectively reduce the data dimension but also reduce the noise in the data, ensuring that the anomaly detection model can maintain high accuracy and robustness in high-dimensional data. In addition, eigenvalue decomposition can help the model reduce the consumption of computing resources while ensuring the data fidelity during the processing, thus providing more efficient performance in real-time monitoring applications.

[0028] Integrating multiple decision trees in S2 includes using different subsets of training data to train each decision tree. In this embodiment, to improve the performance of the ensemble learning model, the random sampling technique is used to generate multiple subsets of training data, and these different subsets are used to train each decision tree. Specifically, first, different subsets are randomly selected from the entire training dataset, and the differences between each subset are ensured. Then, each subset of training data is used to train a decision tree. In this way, it can be ensured that each decision tree learns different feature patterns and decision rules, thus increasing the diversity of the model. When the model makes a prediction, the prediction results of multiple decision trees are integrated, and the final result is output through weighted voting or averaging methods. Mathematically, assuming there are M decision trees, and the subset of training data used for training each decision tree is D j , then the training process of each decision tree is: Among them, K j (x) represents the prediction result of the j-th decision tree for the input feature x, and D j represents the data subset used to train the j-th decision tree, and x is the input feature. The final integrated decision result can be obtained through weighted average: In this way, the training data of multiple subsets enhances the generalization ability of the model, avoids the overfitting problem, and improves the accuracy of anomaly detection. In this embodiment, each decision tree is trained using different training data subsets, which enhances the diversity and robustness of the ensemble learning model. Each decision tree is trained on a different data subset, and can learn different aspects of the data, thereby reducing the bias that may be brought by a single model. This method can significantly improve the generalization ability of the model on unknown data. Especially in the monitoring and evaluation of motion data, it can better adapt to changing environments and different motion patterns. In addition, the ensemble learning method can effectively reduce the impact of noise and outliers on the model by synthesizing the prediction results of multiple decision trees, and improve the stability and accuracy of anomaly detection.

[0029] As Figure 2 shown, a monitoring and evaluation system based on historical motion data is also provided, which is used to implement the steps of the monitoring and evaluation method based on historical motion data. The system includes: A data acquisition module, which is used to extract various motion features from historical motion data and establish a feature library; An anomaly detection module connected to the data acquisition module, which is used to build an anomaly detection model based on the feature library and train the model to adapt to different motion scenarios; A false positive statistical analysis module connected to the anomaly detection module, which is used to count the occurrence frequency of false positives and analyze its causes according to the detection results output by the trained anomaly detection model; A threshold optimization module connected to the anomaly detection module and the false positive statistical analysis module, which is used to optimize the detection threshold of abnormal motion patterns based on the false positive analysis results to reduce the probability of false positives.

[0030] The data acquisition module is responsible for extracting various motion features from historical motion data, such as key indicators like motion frequency, intensity, rhythm, etc. This module preprocesses the data, removes redundant information, and further extracts important features through dimensionality reduction (including methods such as linear and non - linear dimensionality reduction), thereby constructing a feature library. The feature library provides the necessary data support for subsequent anomaly detection and evaluation. The anomaly detection module constructs an anomaly detection model based on the feature library and trains the model to adapt to different motion scenarios. During the model construction process, ensemble learning methods (such as decision tree ensembles or random forests, etc.) can be adopted to enhance the diversity and robustness of the model by using different subsets of training data. The model conducts in - depth learning on data patterns during training, can accurately identify abnormal motion patterns, and ensure the stability and adaptability of the detection system. The false positive statistical analysis module is connected to the anomaly detection module. By analyzing the output results of the model, it statistically calculates the occurrence frequency of false positives and conducts in - depth analysis in combination with the reasons for false positives. This module uses false positive metrics (such as false positive rate and error metrics) to evaluate the model performance and provides the analysis results to the subsequent threshold optimization module as a reference. The threshold optimization module adjusts the detection threshold for abnormal motion patterns based on the false positive statistical analysis results to reduce the probability of false positives. This module precisely adjusts the threshold through optimization algorithms (such as gradient descent or Bayesian optimization) to ensure that the model can accurately detect anomalies while minimizing misjudgment cases to improve the overall performance of the system. The collaboration among the above - mentioned modules enables the system to automatically and efficiently complete the entire process of data acquisition, anomaly detection, and optimization analysis, and is applicable to different motion scenarios and complex data environments. Through the collaboration of the data acquisition module, anomaly detection module, false positive statistical analysis module, and threshold optimization module, the system can achieve full - process automation from data acquisition to anomaly detection and optimization analysis, greatly reducing the need for manual intervention. The system uses the feature library combined with various detection algorithms (such as ensemble learning and non - linear dimensionality reduction) to quickly and accurately identify abnormal states during motion, improving the accuracy and real - time performance of detection. The role of the false positive statistical analysis module combined with the threshold optimization module effectively reduces the false positive rate of the model, thereby improving the reliability of the detection results. The system supports multiple motion scenarios and can adapt to changes in different environments and data characteristics by dynamically optimizing the detection model, ensuring its stability in complex application scenarios.

[0031] The data acquisition module includes a data preprocessing unit, which is used to denoise, standardize, and format the collected historical motion data to ensure the accuracy and consistency of the data. In this embodiment, the data preprocessing unit in the data acquisition module undertakes the task of comprehensively cleaning and processing the historical motion data to ensure that the data meets the requirements of subsequent analysis. Specifically, it includes the following steps: The data preprocessing unit denoises the collected historical motion data. By applying filtering algorithms (such as low-pass filtering, mean filtering, or wavelet denoising), the noise interference in the data is effectively removed, and the real motion characteristics are retained. For example, in the motion data collected by an acceleration sensor, denoising can reduce the data deviation caused by device vibration or environmental noise. The denoised data will be standardized, that is, the numerical range of the data is adjusted to a certain standard range (such as normalizing the data to [0, 1] or a distribution with a mean of 0 and a standard deviation of 1). The purpose of standardization is to eliminate the dimensionality differences between different feature values to avoid bias in model training caused by differences in data scale. The data preprocessing unit formats the historical motion data to unify the storage and expression forms of the data and ensure data consistency. For example, converting data from different sources into a unified time series format or JSON / XML format for the subsequent use of the feature extraction and model training modules. After the above processing, the high-quality data generated by the data preprocessing unit is stored in the feature library, providing reliable input data support for the subsequent training and optimization of the anomaly detection model. Through denoising, standardization, and formatting processing, noise and redundant information are effectively removed, improving the accuracy of the data and providing a reliable basis for the subsequent training and detection of the model. The standardized data reduces the scale differences between features and avoids training bias in the model due to different numerical ranges of features, thereby improving the performance and stability of the anomaly detection model. The formatting process unifies the data structure, enabling the system to be compatible with multiple data sources and enhancing the scalability and applicability of the system. High-quality data input can improve the adaptability of the system in different motion scenarios and ensure the stable performance of the anomaly detection and optimization modules in complex environments.

[0032] The false positive statistical analysis module includes a statistical analysis unit for calculating the false positive rate in different motion scenarios and analyzing the statistical significance of false positives by combining the chi-square test. The false positive statistical analysis module further evaluates the statistical significance of false positives in different motion scenarios through the chi-square test. Specifically, the statistical analysis unit first constructs a hypothesis testing framework, where the null hypothesis is that the probability of false positives has no significant difference in different motion scenarios. The statistical analysis unit outputs the calculated false positive rate and the chi-square test results and generates an analysis report, including the false positive rate, significance level, and conclusion of the statistical analysis for each motion scenario. The results provide a reliable basis for optimization for the subsequent threshold optimization module. By classifying and counting the false positive rates in different motion scenarios, the system can more accurately identify the specific sources of misjudgment and provide clear targets for improving the anomaly detection model. Analyzing the statistical significance of false positives by using the chi-square test helps to determine whether the false positive differences are statistically significant, thus more scientifically guiding the optimization direction of the detection model. Combining the statistical analysis results can focus on optimizing the scenarios with higher false positives, thus effectively reducing the overall misjudgment rate of the system and improving the accuracy and reliability of anomaly detection. The system supports false positive analysis in multiple motion scenarios, enhancing its ability to adapt to complex environments and diverse data.

[0033] The false positive statistical analysis module further includes a feedback adjustment unit, which automatically adjusts the parameters of the anomaly detection model according to the false positive analysis results to dynamically adapt to the changes in the motion scenario. The false positive statistical analysis module calculates the false positive rates under different motion scenarios, and analyzes the causes and significance of false positives through the statistical analysis unit. The feedback adjustment unit receives these analysis results and uses them as the basis for adjusting the parameters of the anomaly detection model. For example, when the false positive rate of a certain motion scenario is significantly high, the feedback adjustment unit can identify the unique data features and model performance differences in that scenario. According to the false positive analysis results, the feedback adjustment unit dynamically optimizes the key parameters of the anomaly detection model. For example: Threshold adjustment: Adjust the detection threshold of the anomaly detection model to adapt to the false positive characteristics under different scenarios. For example, increasing the detection threshold can reduce the over-sensitive false positive rate, while decreasing the threshold helps to improve the sensitivity of detection; Weight update: For models using ensemble learning methods (such as decision tree ensembles), the feedback adjustment unit can reassign the weights of different decision trees to optimize the model's response ability to different features; Learning rate optimization: Adjust the learning rate or the number of training iterations of the model to avoid overfitting or underfitting and further improve the accuracy of the detection results. The feedback adjustment unit adopts a closed-loop feedback control mechanism, enabling the anomaly detection model to dynamically adapt to the changes in the motion scenario. For example, when the motion data in the monitored environment changes significantly, the feedback adjustment unit dynamically adjusts the model by receiving new false positive analysis results in real time to ensure its effectiveness in the new scenario. The adjusted model will be reapplied to the false positive statistical analysis module for verification to confirm the optimization effect. If the adjusted false positive rate reaches the expected target, the system will save the current parameters; otherwise, the feedback adjustment unit will adjust the parameters again, forming a cyclic optimization process. Through the feedback adjustment unit, the system can dynamically adjust the model parameters according to the real-time analysis results of the false positive rate, enabling the anomaly detection model to adapt to different motion scenarios and data feature changes. Automatically adjusting the model parameters effectively reduces the probability of false positives, especially in complex or changing environments, improving the accuracy and stability of anomaly detection. The adjusted model will go through a verification process to ensure that the optimized parameters can significantly improve the model performance, forming an efficient and stable closed-loop optimization process. The feedback adjustment unit provides an automated model optimization tool, enabling the system to operate efficiently in different application scenarios without manual intervention and adapt to complex dynamic environments.

[0034] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made therein without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A monitoring and evaluation method based on historical motion data, characterized in that, The method includes: S1. Extract multiple motion features from historical motion data and establish a feature library; S2. Based on the feature library, construct an anomaly detection model and train the model to adapt to different motion scenarios; S3. According to the detection results output by the trained anomaly detection model, count the occurrence frequency of false positives and analyze its causes, including counting the occurrence frequency of false positives and evaluating the errors in the output of the trained anomaly detection model. The specific formula is: ; where A represents the anomaly metric value, D represents the actual motion pattern frequency detected during monitoring, G represents the expected frequency, and i represents the index of each motion pattern; S4. Based on the analysis results, optimize the detection threshold of the abnormal motion pattern to reduce the probability of false positives, including optimizing the detection threshold of the abnormal motion pattern based on the results of analyzing the causes of false positives. The specific formula is: α 新 = ; Among them, α 新 represents the optimized detection threshold, α represents the current monitoring threshold, a represents the learning rate, and b represents the loss function.

2. The monitoring and evaluation method based on historical motion data according to claim 1, wherein: S1 includes dimensionality reduction of the historical motion data to reduce the data dimension and extract features. The specific formula is: B = D × E; where B represents the transformed data, D represents the original data matrix, and E represents the feature vector matrix; The extracted features are made part of the database.

3. The monitoring and evaluation method based on historical motion data according to claim 2, characterized in that: The above-mentioned S2 includes constructing an anomaly detection model based on the data in the feature library and making decisions by integrating multiple decision trees. The specific formula is as follows: ; Among them, H(x) represents the output of the final model, M represents the number of decision trees, x represents the input feature, and K j (x) represents the prediction result of the j-th decision tree for the input feature x, and j represents the index of the decision tree.

4. The monitoring and evaluation method based on historical motion data according to claim 2, wherein: S1 also includes using an autoencoder for non-linear dimensionality reduction of the historical data.

5. The monitoring and evaluation method based on historical motion data according to claim 2, characterized in that: The calculation formula for the eigenvector matrix E in S1 is: E = V × ; Among them, E represents the eigenvector matrix, and V represents the eigenvector of the covariance matrix, which represents the eigenvalue matrix of the covariance matrix.

6. The monitoring and evaluation method based on historical motion data according to claim 3, wherein: Integrating multiple decision trees in S2 includes using different subsets of training data to train each decision tree.

7. A monitoring and evaluation system based on historical motion data, which is used to implement the steps of the monitoring and evaluation method based on historical motion data according to any one of claims 1-6, characterized in that, The system includes: A data acquisition module for extracting multiple motion features from historical motion data and establishing a feature library; An anomaly detection module connected to the data acquisition module for constructing an anomaly detection model based on the feature library and training the model to adapt to different motion scenarios; A false positive statistical analysis module connected to the anomaly detection module for statistically analyzing the occurrence frequency of false positives and analyzing their causes according to the detection results output by the trained anomaly detection model; A threshold optimization module connected to the anomaly detection module and the false positive statistical analysis module for optimizing the detection threshold of abnormal motion patterns based on the false positive analysis results to reduce the probability of false positives.

8. A monitoring and evaluation system based on historical motion data according to claim 7, characterized in that: The data acquisition module includes a data preprocessing unit for denoising, normalizing, and formatting the collected historical motion data to ensure the accuracy and consistency of the data.

9. The monitoring and evaluation system based on historical motion data according to claim 7, wherein: The false positive statistical analysis module includes a statistical analysis unit for calculating the false positive rate in different motion scenarios and analyzing the statistical significance of the occurrence of false positives in combination with the chi-square test.

10. A monitoring and evaluation system based on historical movement data according to claim 7, characterized in that: The false positive statistical analysis module also includes a feedback adjustment unit for automatically adjusting the parameters of the anomaly detection model according to the false positive analysis results to dynamically adapt to changes in the motion scenario.

Images