Sound barrier anomaly identification method and device based on anomaly detection model

Abstract

The invention discloses a sound barrier anomaly recognition method and device based on an anomaly detection model, and relates to the field of anomaly detection. In the method, a spatio-temporal data matrix is constructed according to a timestamp, spatial coordinates of a sensor and multi-dimensional data, and a preset state feature vector is constructed according to the timestamp, environmental data and current traffic flow data; constructing a sensor topological graph according to the spatio-temporal data matrix, obtaining a spatio-temporal feature vector of each time step according to the sensor topological graph, and determining an abnormal candidate event according to the target spatio-temporal feature vector; determining the confidence of the abnormal candidate event; and determining the sensor corresponding to the abnormal candidate event of which the confidence coefficient is greater than a second threshold value as a target sensor, determining an abnormal position in the sound barrier according to a space coordinate of the target sensor, and determining an abnormal type according to a target spatial-temporal feature vector corresponding to the target sensor. By implementing the technical scheme provided by the invention, the effects of accurately identifying the abnormal condition of the sound barrier and improving the detection efficiency and accuracy are achieved.

Description

Technical Field

[0001] The present application relates to the technical field of anomaly detection, and specifically to a method and device for identifying anomalies in a sound barrier based on an anomaly detection model. Background Art

[0002] Sound barriers play a vital role in modern transportation infrastructure. Their primary function is to reduce the impact of traffic noise on the surrounding environment, providing people with a quieter and more harmonious living space. With the acceleration of urbanization and the continued growth of traffic volume, the widespread use of sound barriers has not only improved the quality of the urban environment but also played a positive role in protecting the ecological environment. However, sound barriers are exposed to complex and changing external environments for a long time, making them susceptible to various factors such as wind loads, temperature fluctuations, and rainfall erosion. This can lead to gradual deterioration of their structural performance and even abnormalities.

[0003] To ensure the safety and reliability of sound barriers, traditional inspection methods primarily rely on manual inspections and regular maintenance. Manual inspections rely on the operator's experience, using visual inspections or simple tool measurements to determine if there are any anomalies in the sound barrier. While these methods can detect potential problems to a certain extent, they are generally limited by inefficiency, lack of accuracy, and a lack of real-time feedback. These traditional methods are inadequate for complex working conditions, especially when it is necessary to quickly and accurately locate and identify anomalies.

[0004] Therefore, how to efficiently and reliably identify abnormal conditions in sound barriers has become a key technical issue that needs to be solved urgently. Summary of the Invention

[0005] The present application provides a method and device for identifying sound barrier anomalies based on an anomaly detection model, which accurately identifies abnormal conditions of the sound barrier and improves detection efficiency and accuracy.

[0006] In a first aspect of the present application, a method for identifying sound barrier anomalies based on an anomaly detection model is provided, which is applied to a sound barrier detection platform. The method comprises: Acquiring multidimensional data, current traffic flow data, and environmental data from each sensor, constructing a spatiotemporal data matrix based on a timestamp, the spatial coordinates of the sensor, and the multidimensional data, and constructing a preset state feature vector based on the timestamp, the environmental data, and the current traffic flow data, wherein the multidimensional data includes noise sound pressure level, vibration spectrum, and structural stress value, and the environmental data includes wind speed, rainfall, and temperature; constructing a sensor topology map according to the spatiotemporal data matrix, obtaining a spatiotemporal feature vector for each time step according to the sensor topology map, screening out a target spatiotemporal feature vector having a deviation from the preset state feature vector greater than a first threshold, and determining an abnormal candidate event according to the target spatiotemporal feature vector; Reconstructing the target spatiotemporal feature vector of the abnormal candidate event to obtain a reconstructed vector, calculating a first difference between the reconstructed vector and the target spatiotemporal feature vector, calculating the KL divergence of the abnormal candidate event to obtain a second difference, and performing a weighted summation of the first difference and the second difference to obtain a confidence score; The sensor corresponding to the abnormal candidate event with a confidence greater than the second threshold is determined as the target sensor, the abnormal position in the sound barrier is determined according to the spatial coordinates of the target sensor, and the abnormal type is determined according to the target spatiotemporal feature vector corresponding to the target sensor.

[0007] Optionally, constructing a sensor topology map according to the spatiotemporal data matrix includes: Taking each sensor as a graph node, constructing a node feature vector of the graph node according to the multidimensional data in the spatiotemporal data matrix, wherein the node feature vector includes noise sound pressure level, vibration spectrum energy distribution, and structural stress gradient; Constructing a graph edge according to the adjacent relationship of the sensors, and constructing an edge weight matrix according to the distance between the two sensors corresponding to the graph edge and the attenuation coefficient of the material of the sound barrier; A sensor topology graph is constructed according to the node feature vectors and the edge weight matrix.

[0008] Optionally, obtaining the spatiotemporal feature vector of each time step according to the sensor topology graph includes: A spatial feature matrix is ​​obtained according to the sensor topology map and the node feature matrix corresponding to each time step, the spatial feature matrix is ​​arranged according to the time step to form a time series, and a gated recurrent unit is used to obtain a time feature matrix according to the time series; the spatial feature matrix and the time feature matrix are spliced ​​according to the node dimension to obtain intermediate features, and the intermediate features are reduced to the target dimension to obtain a spatiotemporal feature vector.

[0009] Optionally, screening out a target spatiotemporal feature vector whose deviation from the preset state feature vector is greater than a first threshold includes: Dimensionally aligning the spatiotemporal eigenvector with the preset state eigenvector, and calculating the vibration frequency deviation and the stress gradient deviation respectively; Performing dynamic time warping alignment on the harmonic component quantity change rate and energy attenuation slope in the time series, and then calculating the difference to obtain a time series dynamic deviation, and performing weighted summation on the vibration frequency deviation, the stress gradient deviation, and the time series dynamic deviation to obtain a target deviation; The spatiotemporal feature vector obtained when the target deviation is greater than the first threshold is determined as the target spatiotemporal feature vector.

[0010] Optionally, reconstructing the target spatiotemporal feature vector of the abnormal candidate event to obtain a reconstructed vector, calculating a first difference between the reconstructed vector and the target spatiotemporal feature vector, and calculating the KL divergence of the abnormal candidate event to obtain a second difference includes: Inputting the target spatiotemporal feature vector into a deep variational autoencoder to obtain a latent space distribution parameter, obtaining a latent vector according to the latent space distribution parameter, and inputting the latent vector into a decoder to obtain the reconstructed vector; Calculating a square error for each element of the reconstructed vector and the target spatiotemporal feature vector, and averaging the square errors of all elements to obtain a first difference; Determine the probability distribution of each element of the target spatiotemporal feature vector, determine the corresponding preset probability distribution, and sum the KL divergence between the probability distribution of all elements and the preset probability distribution to obtain a second difference.

[0011] Optionally, determining the abnormal position in the sound barrier according to the spatial coordinates of the target sensor includes: A propagation matrix is ​​constructed according to the spatial coordinates of the target sensor and the attenuation coefficient of the material of the sound barrier, and an inverse solution is performed based on the propagation matrix to obtain the abnormal position.

[0012] Optionally, determining the abnormality type based on the target spatiotemporal feature vector corresponding to the target sensor includes: extracting key spectrum features from the target spatiotemporal feature vector, the key spectrum features including the main resonance peak frequency, the number of harmonic components and the energy attenuation slope, matching the key spectrum features with the fault modes in a preset knowledge base to determine the abnormality type.

[0013] In a second aspect of the present application, a sound barrier anomaly recognition system based on an anomaly detection model is provided, comprising an acquisition module, a screening module, a calculation module, and an execution module, wherein: an acquisition module configured to acquire multidimensional data, current traffic flow data, and environmental data from each sensor, construct a spatiotemporal data matrix based on a timestamp, spatial coordinates of the sensor, and the multidimensional data, and construct a preset state feature vector based on the timestamp, the environmental data, and the current traffic flow data, wherein the multidimensional data includes noise sound pressure level, vibration spectrum, and structural stress value, and the environmental data includes wind speed, rainfall, and temperature; a screening module configured to construct a sensor topology map based on the spatiotemporal data matrix, obtain a spatiotemporal feature vector for each time step based on the sensor topology map, screen out a target spatiotemporal feature vector whose deviation from the preset state feature vector is greater than a first threshold, and determine an abnormal candidate event based on the target spatiotemporal feature vector; a calculation module configured to reconstruct the target spatiotemporal feature vector of the abnormal candidate event to obtain a reconstructed vector, calculate a first difference between the reconstructed vector and the target spatiotemporal feature vector, calculate the KL divergence of the abnormal candidate event to obtain a second difference, and perform a weighted summation of the first difference and the second difference to obtain a confidence level; An execution module is configured to determine the sensor corresponding to the abnormal candidate event with a confidence level greater than a second threshold as a target sensor, determine the abnormal position in the sound barrier according to the spatial coordinates of the target sensor, and determine the abnormal type according to the target spatiotemporal feature vector corresponding to the target sensor.

[0014] In the third aspect of the present application, an electronic device is provided, including a processor, a memory, a user interface and a network interface, the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device performs any of the methods described above.

[0015] In a fourth aspect of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores instructions. When the instructions are executed, any one of the methods described above is executed.

[0016] In summary, one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: By integrating multi-dimensional data, including noise sound pressure level, vibration spectrum and structural stress value, as well as comprehensive consideration of environmental data and vehicle flow data, the actual operating status of the sound barrier can be more comprehensively reflected, reducing the errors and uncertainties brought by a single data source; by constructing a sensor topology map using a spatiotemporal data matrix and obtaining spatiotemporal feature vectors, the complex spatial and temporal change patterns of the sound barrier can be captured, improving the sensitivity and recognition ability to abnormal events; by reconstructing the target spatiotemporal feature vectors of abnormal candidate events and combining them with KL divergence calculations, the degree of abnormality can be more accurately measured, thereby improving the accuracy of anomaly detection; by determining the abnormal position in the sound barrier through the spatial coordinates of the target sensor, and combining the reasonable layout of the sensor network with a high-precision positioning algorithm, the abnormal position can be accurately identified, providing clear guidance for subsequent repairs and maintenance; accurately identifying the type of abnormality can help maintenance personnel quickly determine the cause of the fault, take targeted repair measures, and reduce repair time and costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of a method for identifying anomalies in a sound barrier based on anomaly detection models disclosed in an embodiment of the present application; Figure 2 This is a module diagram of a sound barrier anomaly recognition system based on anomaly detection model disclosed in an embodiment of the present application; Figure 3 This is a structural diagram of an electronic device disclosed in an embodiment of the present application.

[0018] Explanation of the accompanying drawings: 201, acquisition module; 202, screening module; 203, calculation module; 204, execution module; 301, processor; 302, communication bus; 303, user interface; 304, network interface; 305, memory. DETAILED DESCRIPTION

[0019] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0020] In the description of the embodiments of this application, words such as "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "for example" or "for instance" in the embodiments of this application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "for example" or "for instance" is intended to present the relevant concepts in a concrete manner.

[0021] In the description of the embodiments of the present application, the term "multiple" means two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0022] This embodiment discloses a sound barrier anomaly recognition method based on an anomaly detection model, which is applied to a sound barrier detection platform. Figure 1 This is a flow chart of the method for identifying anomalies in a sound barrier based on anomaly detection model disclosed in the embodiment of the present application. Figure 1 As shown, the method includes the following steps: S101. Acquire multidimensional data, current traffic flow data, and environmental data from each sensor, construct a spatiotemporal data matrix based on a timestamp, the spatial coordinates of the sensor, and the multidimensional data, and construct a preset state feature vector based on the timestamp, the environmental data, and the current traffic flow data, wherein the multidimensional data includes noise sound pressure level, vibration spectrum, and structural stress value, and the environmental data includes wind speed, rainfall, and temperature. S102: construct a sensor topology map based on the spatiotemporal data matrix, obtain a spatiotemporal feature vector for each time step based on the sensor topology map, screen out target spatiotemporal feature vectors whose deviation from the preset state feature vector is greater than a first threshold, and determine abnormal candidate events based on the target spatiotemporal feature vectors; S103: reconstruct the target spatiotemporal feature vector of the abnormal candidate event to obtain a reconstructed vector, calculate a first difference between the reconstructed vector and the target spatiotemporal feature vector, calculate the KL divergence of the abnormal candidate event to obtain a second difference, and perform a weighted sum of the first difference and the second difference to obtain a confidence level; S104. Determine the sensor corresponding to the abnormal candidate event with a confidence level greater than a second threshold as the target sensor, determine the abnormal position in the sound barrier according to the spatial coordinates of the target sensor, and determine the abnormal type according to the target spatiotemporal feature vector corresponding to the target sensor.

[0023] Multidimensional data from each sensor is collected, including noise sound pressure level, vibration spectrum, and structural stress values. Current traffic flow data and environmental data, including wind speed, rainfall, and temperature, are also obtained. A spatiotemporal data matrix is ​​constructed based on the timestamp, the sensor's spatial coordinates, and this multidimensional data. Furthermore, a preset state feature vector is constructed based on the timestamp, environmental data, and current traffic flow data. This preset state feature vector can be considered a reference feature under normal conditions. A sensor topology map is constructed based on this constructed spatiotemporal data matrix. Each sensor is considered a node in the topology map, and its feature vector contains information such as noise sound pressure level, vibration spectrum energy distribution, and structural stress gradient. Graph edges are constructed based on the proximity relationships between sensors, and the edge weight matrix is ​​determined based on the distance between sensors and the attenuation coefficient of the sound barrier material. This sensor topology map, generated in this manner, reflects the spatial relationships between sensors and their data characteristics. A spatiotemporal feature vector is obtained for each time step based on the sensor topology map. This vector integrates the sensor's spatial characteristics and their temporal characteristics. These spatiotemporal feature vectors are compared with the preset state feature vectors, and the deviation between them is calculated. If the deviation is greater than a set first threshold, the corresponding spatiotemporal feature vectors are selected as target spatiotemporal feature vectors. The events corresponding to these target spatiotemporal feature vectors are then identified as candidate anomaly events. For each identified candidate anomaly event, the corresponding target spatiotemporal feature vector is reconstructed to obtain a reconstructed vector. The reconstruction error is measured by calculating the first difference between the reconstructed vector and the original target spatiotemporal feature vector, which reflects the degree of data change during the reconstruction process. Simultaneously, the KL divergence of the candidate anomaly event is calculated to obtain a second difference. The KL divergence measures the difference between the probability distribution of the candidate anomaly event and the normal distribution. The first and second differences are weighted and summed to obtain a comprehensive confidence value that more comprehensively reflects the degree of anomaly for the candidate anomaly event. The calculated confidence value is compared with a set second threshold. If the confidence value is greater than the second threshold, the sensor associated with the candidate anomaly event is identified as the target sensor. Based on the spatial coordinates of these target sensors, the location of the anomaly within the sound barrier can be precisely located. Furthermore, based on the target spatiotemporal feature vector corresponding to the target sensor, the specific abnormality type is determined using the pre-established mapping relationship between the feature vector and the abnormality type, thereby completing the identification of the sound barrier abnormality.

[0024] Optionally, constructing a sensor topology map according to the spatiotemporal data matrix includes: Taking each sensor as a graph node, constructing a node feature vector of the graph node according to the multidimensional data in the spatiotemporal data matrix, wherein the node feature vector includes noise sound pressure level, vibration spectrum energy distribution, and structural stress gradient; Constructing a graph edge according to the adjacent relationship of the sensors, and constructing an edge weight matrix according to the distance between the two sensors corresponding to the graph edge and the attenuation coefficient of the material of the sound barrier; A sensor topology graph is constructed according to the node feature vectors and the edge weight matrix.

[0025] When constructing the sensor topology graph, each sensor is considered a node in the graph. These sensors collect relevant data about the sound barrier, such as noise pressure level, vibration spectrum, and structural stress. A node feature vector is constructed for each graph node (i.e., sensor) based on the multidimensional data in the spatiotemporal data matrix. This node feature vector includes the noise pressure level, vibration spectrum energy distribution, and structural stress gradient. The noise pressure level reflects the noise intensity at the sensor's location and is an important indicator of the sound barrier's acoustic performance. The vibration spectrum energy distribution analyzes the vibration signal's spectrum to determine the energy distribution at different frequency components, helping to identify the structure's vibration characteristics and potential abnormal vibration modes. The structural stress gradient indicates the degree of stress variation between different locations on the structure. It can indicate whether the sound barrier structure is subjected to uneven forces and is important for detecting structural damage or stress concentration. The physical positional relationships between sensors are analyzed to identify adjacent sensors. Adjacent sensors have a certain degree of spatial correlation, and their monitored data will normally be correlated. For each graph edge (i.e., the connection between two adjacent sensors), an edge weight is calculated based on the actual distance between the two sensors and the attenuation coefficient of the sound barrier material. The calculation formula of edge weight can be expressed as: Among them, W ij represents the weight of the edge between sensor i and sensor j, r ij is the distance between sensor i and sensor j, where i and j are both positive integers. α is the attenuation coefficient of the sound barrier material, which reflects the energy attenuation characteristics of sound waves as they propagate through the barrier material. The edge weight reflects the strength of the data correlation between the two sensors. The closer the distance and the smaller the material attenuation coefficient, the larger the edge weight, indicating a higher correlation. Based on the node feature vectors and edge weight matrix constructed above, all sensor nodes and their connections are integrated to form a complete sensor topology map. This topology map not only reflects the spatial distribution of sensors but also integrates the data characteristics monitored by the sensors and the correlations between them through the node feature vectors and edge weight matrix, providing the infrastructure for subsequent spatiotemporal feature extraction and anomaly detection.

[0026] Each sensor is treated as a node in the graph, and node feature vectors are constructed using multi-dimensional data such as noise pressure level, vibration spectrum energy distribution, and structural stress gradients. This method comprehensively describes the sound barrier status information perceived by each sensor, providing a rich data foundation for subsequent feature extraction and anomaly detection. The graph edges are constructed based on the proximity of sensors, and the edge weight matrix is ​​constructed based on the actual distance between sensors and the attenuation coefficient of the sound barrier material. This approach not only considers the physical location of the sensors but also incorporates the influence of the sound barrier material properties on sound wave propagation, making the constructed sensor topology map more accurately reflect the actual structure and sound wave propagation characteristics of the sound barrier. By combining node feature vectors and edge weight matrices to construct a sensor topology map, it realistically simulates the layout and data interaction of the sound barrier sensor network, providing an accurate topological foundation for subsequent spatiotemporal feature extraction and anomaly detection algorithms, helping to improve the accuracy and reliability of anomaly detection. This method can flexibly adapt to sound barrier sensor networks of varying layouts and sizes. Whether the sound barrier structure is linear, curved, or other complex shapes, the rationally constructed sensor topology map can effectively monitor the sound barrier status and detect anomalies, demonstrating its versatility and adaptability. By constructing a sensor topology map, sensor data processing can be elevated from the individual level to the network level. Leveraging the advantages of the graph structure for parallel computing and feature propagation, this helps improve the computational efficiency and real-time performance of the anomaly detection algorithm, meeting the needs of real-time monitoring of sound barriers. Physical model parameters such as the attenuation coefficient of the sound barrier material are introduced when constructing the edge weight matrix, achieving an organic combination of physical models and data-driven methods. This not only fully utilizes prior physical knowledge, but also enables continuous optimization and adjustment of the model through a data-driven approach, improving the interpretability and reliability of the anomaly detection model. An accurate sensor topology map provides a reliable basis for anomaly location and tracing. When an abnormal event is detected, the specific location of the anomaly can be quickly and accurately located by analyzing the feature changes of the relevant nodes and edges in the topology map, and the path of the anomaly propagation can be traced, providing strong support for the maintenance and repair of sound barriers.

[0027] Optionally, obtaining the spatiotemporal feature vector of each time step according to the sensor topology graph includes: A spatial feature matrix is ​​obtained according to the sensor topology map and the node feature matrix corresponding to each time step, the spatial feature matrix is ​​arranged according to the time step to form a time series, and a gated recurrent unit is used to obtain a time feature matrix according to the time series; the spatial feature matrix and the time feature matrix are spliced ​​according to the node dimension to obtain intermediate features, and the intermediate features are reduced to the target dimension to obtain a spatiotemporal feature vector.

[0028] Based on the constructed sensor topology, each time step has a corresponding node feature matrix. This matrix contains the node feature vectors of all sensors at that time step. These feature vectors include information such as noise sound pressure level, vibration spectrum energy distribution, and structural stress gradients. Using graph neural network technologies such as graph convolutional networks (GCNs), spatial features are extracted from the sensor topology and node feature matrix to generate a spatial feature matrix. This process, through convolution operations on the graph structure, captures spatial correlations and dependencies between sensors, fusing the features of each sensor with those of its neighbors to obtain more representative spatial features. The resulting spatial feature matrix is ​​arranged in the order of time steps to form a time series. This time series reflects the changes in the spatial features of the sound barrier at different time steps. This time series is processed using a gated recurrent unit (GRU). The GRU is a recurrent neural network structure that effectively processes sequential data and can capture the dependencies and dynamic patterns between previous and subsequent time steps in the time series. Through GRU processing, a temporal feature matrix is ​​generated. This matrix contains feature information in the temporal dimension and reflects the changing trends of the sound barrier status over time. The spatial and temporal feature matrices obtained above are concatenated according to the node dimension to produce an intermediate feature matrix. This intermediate feature matrix incorporates feature information from both spatial and temporal dimensions, but its dimensionality may be relatively high. To reduce the feature dimensionality, improve computational efficiency, and improve model generalization, dimensionality reduction techniques are used to reduce the intermediate feature matrix to the target dimension, ultimately yielding a spatiotemporal feature vector. This spatiotemporal feature vector incorporates both spatial correlation features and temporal dynamic change characteristics, comprehensively describing the state characteristics of the sound barrier in the spatiotemporal domain and providing strong support for subsequent anomaly detection and analysis.

[0029] The spatial feature matrix is ​​derived from the sensor topology map and the node feature matrix corresponding to each time step. This fully utilizes the spatial relationships between sensors and the multidimensional feature information of each sensor to accurately extract the spatial features of the sensor network at each time step, providing a solid foundation for subsequent spatiotemporal analysis. The spatial feature matrix is ​​arranged by time step to form a time series, and this time series is processed using a gated recurrent unit (GRU) to generate a temporal feature matrix. This effectively captures the dynamic changes and dependencies of sensor data in the temporal dimension, facilitating the discovery of anomalous patterns and trends that evolve over time. The spatial and temporal feature matrices are concatenated along the node dimension to generate intermediate features, which are then reduced to the target dimension to generate a spatiotemporal feature vector. This fusion method organically combines spatial and temporal features to form a comprehensive, low-dimensional spatiotemporal feature vector, preserving critical spatiotemporal information while reducing data dimensionality, improving computational efficiency and model tractability. The resulting spatiotemporal feature vector contains rich spatiotemporal information and can more comprehensively reflect the operational status of the sound barrier. This helps improve the accuracy and sensitivity of anomaly detection, promptly identifying potential abnormal events, and ensuring the safe operation of the sound barrier. By processing data in time steps and capturing temporal features, this method can adapt to dynamic changes in the environment in which the sound barrier is located, such as fluctuations in traffic volume and changes in weather conditions. This makes the anomaly detection model more robust and adaptable, and can maintain stable performance under different operating conditions.

[0030] Optionally, screening out a target spatiotemporal feature vector whose deviation from the preset state feature vector is greater than a first threshold includes: Dimensionally aligning the spatiotemporal eigenvector with the preset state eigenvector, and calculating the vibration frequency deviation and the stress gradient deviation respectively; Performing dynamic time warping alignment on the harmonic component quantity change rate and energy attenuation slope in the time series, and then calculating the difference to obtain a time series dynamic deviation, and performing weighted summation on the vibration frequency deviation, the stress gradient deviation, and the time series dynamic deviation to obtain a target deviation; The spatiotemporal feature vector obtained when the target deviation is greater than the first threshold is determined as the target spatiotemporal feature vector.

[0031] Align the spatiotemporal eigenvector with the preset state eigenvector. This means ensuring consistency across all characteristic dimensions. For example, dimensions such as vibration frequency, stress gradient, number of harmonic components, and energy decay slope must correspond one-to-one. Only when these dimensions are aligned can subsequent deviation calculations be performed accurately. Calculate the deviation between the spatiotemporal eigenvector and the preset state eigenvector in the vibration frequency dimension. This can be achieved through a simple difference calculation: subtract the corresponding value in the preset state eigenvector from the vibration frequency value in the spatiotemporal eigenvector to obtain the vibration frequency deviation. Similarly, calculate the deviation between the two in the stress gradient dimension: subtract the corresponding value in the preset state eigenvector from the stress gradient value in the spatiotemporal eigenvector to obtain the stress gradient deviation. For the number of harmonic components in the spatiotemporal eigenvector and the preset state eigenvector, calculate their rate of change over time. This can be achieved by performing operations such as differencing or differentiation on the time series data to obtain the rate of change of the number of harmonic components over time. Calculate the energy decay slope for the spatiotemporal eigenvector and the preset state eigenvector. This can be achieved by linearly fitting the change in energy value over time to obtain a slope value that reflects the speed of energy decay. Since time series data may have time delays or distortions, in order to accurately compare the difference between the rate of change of the number of harmonic components and the energy decay slope, it is necessary to use the dynamic time warping (DTW) algorithm to align the two on the time axis. DTW can find the optimal matching path between the two time series, so that the comparison of corresponding features is more comparable at the aligned time point. At the aligned time point, the difference between the rate of change of the number of harmonic components and the energy decay slope is calculated respectively to obtain the time series dynamic deviation. This reflects the dynamic difference between the spatiotemporal feature vector and the preset state feature vector in terms of harmonic components and energy decay in the time series. The vibration frequency deviation, stress gradient deviation and time series dynamic deviation are weighted and summed to obtain the target deviation. The weight setting can be determined based on the importance of each deviation in reflecting the abnormal characteristics of the sound barrier. For example, if the vibration frequency deviation is more sensitive to the sound barrier abnormality, a larger weight can be given. The calculated target deviation is compared with the preset first threshold. If the target deviation is greater than the first threshold, it is considered that the deviation between the space-time feature vector and the preset state feature vector is significant, which may indicate that there is an abnormality in the sound barrier. Therefore, the space-time feature vector is determined as the target space-time feature vector and as the feature vector of the abnormal candidate event for further analysis and processing.

[0032] By dimensionally aligning the spatiotemporal feature vector with the preset state feature vector and calculating the vibration frequency deviation and stress gradient deviation, the differences between the two can be comprehensively measured across multiple key feature dimensions, ensuring comprehensive and accurate anomaly detection. Dynamic time warping and alignment are performed on the harmonic component count change rate and energy decay slope in the time series, followed by subtraction to obtain the time series dynamic deviation. This method effectively captures dynamic feature differences over time and adapts to the time series feature changes under different operating conditions of the sound barrier. The target deviation is obtained by weighted summation of the vibration frequency deviation, stress gradient deviation, and time series dynamic deviation. This is used to screen the target spatiotemporal feature vector. This method comprehensively considers the contributions of different feature dimensions and, by setting a first threshold, effectively selects feature vectors that significantly deviate from the normal state as candidate anomaly events. This comprehensive assessment of multi-dimensional feature deviations allows for more accurate identification of truly anomalous feature vectors, reducing false positives and false negatives, and improving the accuracy and reliability of anomaly detection. Dynamic time warping and alignment can adapt to feature changes in the time series caused by factors such as environmental changes and traffic volume fluctuations, ensuring that the anomaly detection model maintains stable performance across different time periods and operating conditions.

[0033] Optionally, reconstructing the target spatiotemporal feature vector of the abnormal candidate event to obtain a reconstructed vector, calculating a first difference between the reconstructed vector and the target spatiotemporal feature vector, and calculating the KL divergence of the abnormal candidate event to obtain a second difference includes: Inputting the target spatiotemporal feature vector into a deep variational autoencoder to obtain a latent space distribution parameter, obtaining a latent vector according to the latent space distribution parameter, and inputting the latent vector into a decoder to obtain the reconstructed vector; Calculating a square error for each element of the reconstructed vector and the target spatiotemporal feature vector, and averaging the square errors of all elements to obtain a first difference; Determine the probability distribution of each element of the target spatiotemporal feature vector, determine the corresponding preset probability distribution, and sum the KL divergence between the probability distribution of all elements and the preset probability distribution to obtain a second difference.

[0034] The target spatiotemporal feature vector is input into a deep variational autoencoder (D-VAE). D-VAE is a generative model consisting of an encoder and a decoder. The encoder encodes the input data into distribution parameters in a latent space, typically including parameters such as the mean and variance, which describe the probability distribution of the data in the latent space. Based on the obtained latent space distribution parameters, a latent vector is obtained through sampling and other methods. This latent vector represents the data in the latent space and captures the data's main features and intrinsic structure. The latent vector is input into the decoder, which decodes the latent vector into a reconstructed vector. The reconstructed vector is the model's output of restoring the original input data and should be as close as possible to the original input data in the feature space. The squared error is calculated for each corresponding element of the reconstructed vector and the target spatiotemporal feature vector. This is the squared difference between the reconstructed value and the original value of each element. The squared error measures the reconstruction error of a single element. The squared errors of all elements are averaged to obtain the first difference. This first difference reflects the average error between the reconstructed vector and the target spatiotemporal feature vector. The smaller the first difference, the more faithful the reconstructed vector is to the original feature vector. Determine the probability distribution of each element of the target spatiotemporal feature vector. This can be obtained by statistically analyzing the data or using the output distribution of the model. The probability distribution describes the uncertainty of the value of each element. Determine the corresponding preset probability distribution. The preset probability distribution is usually set based on the data distribution characteristics under normal conditions and represents the pattern of element values ​​under normal conditions. Sum the KL divergence between the probability distribution of all elements and the preset probability distribution to obtain the second difference. The KL divergence is used to measure the degree of difference between two probability distributions. The larger the second difference, the greater the difference between the distribution of the target spatiotemporal feature vector and the normal distribution, that is, the more likely there is an anomaly.

[0035] Inputting the target spatiotemporal feature vector into a deep variational autoencoder learns the data's deep feature representation and latent space distribution parameters. The decoder then maps the latent vector back to the original data space to produce a reconstructed vector. This process captures complex patterns and inherent structure in the data, providing richer information for anomaly detection. The squared error between each element of the reconstructed vector and the target spatiotemporal feature vector is calculated and averaged to obtain the first difference. This method is simple, direct, and effective in quantifying the reconstruction error, reflecting the degree of difference between the reconstructed vector and the original vector. A large first difference may indicate that the original vector contains anomalous features. The probability distribution of each element of the target spatiotemporal feature vector is determined and compared with a pre-determined probability distribution. The KL divergence between them is calculated to obtain the second difference. The KL divergence measures the difference between two probability distributions, capturing data distributional variations and is important for detecting anomalies or distribution shifts in the data generation process. The first and second differences reflect the degree of anomaly of the candidate anomaly from different perspectives. The first difference focuses on the reconstruction error of the data, while the second difference focuses on differences in the probability distribution of the data. Combining the two provides a more comprehensive assessment of the degree of anomaly of the candidate anomaly, improving the accuracy and reliability of anomaly detection. The deep variational autoencoder has a certain degree of generalization ability, can adapt to different types of spatiotemporal feature vectors, and can still effectively reconstruct and detect anomalies in the presence of noisy or incomplete data through the learned latent space distribution, thus enhancing the robustness of the model. By analyzing the differences between the reconstructed vector and the target spatiotemporal feature vector, as well as the changes in the probability distribution, the characteristics and regularities of the anomaly pattern can be discovered, which helps to further understand the types and causes of sound barrier anomalies and provide a basis for subsequent fault diagnosis and maintenance strategy formulation.

[0036] Optionally, determining the abnormal position in the sound barrier according to the spatial coordinates of the target sensor includes: A propagation matrix is ​​constructed according to the spatial coordinates of the target sensor and the attenuation coefficient of the material of the sound barrier, and an inverse solution is performed based on the propagation matrix to obtain the abnormal position.

[0037] During the sound barrier anomaly detection process, after the target sensors have been identified through the aforementioned steps, a propagation matrix is ​​constructed based on the spatial coordinates of these target sensors and the attenuation coefficient of the barrier material to accurately locate the anomaly within the barrier. The spatial coordinates of the target sensors provide information about their specific location within the barrier structure, while the attenuation coefficient of the barrier material reflects the energy loss characteristics of sound waves propagating through the barrier material. The propagation matrix aims to establish a mathematical relationship between the signals received by the sensors and the possible anomaly locations within the barrier. It comprehensively considers the physical properties and geometric structure of the barrier, as well as the propagation patterns of sound waves within it. Based on the constructed propagation matrix, the anomaly location is determined through an inverse solution method. This inverse solution process typically involves mathematical optimization algorithms or matrix inversion operations. Its goal is to find a point within the barrier such that the acoustic signal emitted from this point, after passing through the propagation characteristics described by the propagation matrix, optimally matches the actual signal received by the target sensor. This method leverages multi-point measurement data from the sensor network, combined with the characteristics of the barrier material, to infer the most likely source of the anomaly signal, thereby accurately locating the anomaly within the barrier.

[0038] By constructing a propagation matrix and combining it with the spatial coordinates of the target sensor and the attenuation coefficient of the sound barrier material, anomalies in the sound barrier can be accurately located. This method leverages the propagation characteristics of sound waves in the sound barrier and the sensor's location information to improve positioning accuracy. Combining the physical properties of the sound barrier (such as the material's attenuation coefficient) with sensor data, it achieves a fusion of physical models and data-driven methods. This not only improves positioning reliability but also better interprets the physical meaning and actual location of the anomaly. An inverse solution method based on the propagation matrix can effectively infer the location of the anomaly source from the sensor data. This method is computationally feasible and efficient, and can obtain relatively accurate anomaly locations in a short time. It is adaptable to sound barrier structures of varying shapes and complexity. Whether the sound barrier is linear, curved, or otherwise complex, as long as the spatial coordinates of the target sensor and the attenuation coefficient of the sound barrier material are accurately obtained, the corresponding propagation matrix can be constructed and the anomaly location can be located.

[0039] Optionally, determining the abnormality type based on the target spatiotemporal feature vector corresponding to the target sensor includes: extracting key spectrum features from the target spatiotemporal feature vector, the key spectrum features including the main resonance peak frequency, the number of harmonic components and the energy attenuation slope, matching the key spectrum features with the fault modes in a preset knowledge base to determine the abnormality type.

[0040] Spectral analysis is performed on the target's spatiotemporal feature vector to extract key spectral features. These key spectral features, including the main resonant peak frequency, the number of harmonic components, and the energy decay slope, reflect the vibration and acoustic characteristics of the sound barrier from different perspectives and are important indicators for identifying anomaly types. The main resonant peak frequency refers to the frequency component with the highest energy concentration in the spectrum. It is typically related to factors such as the sound barrier's structural characteristics, material properties, and the excitation force it is subject to. When an anomaly occurs in a sound barrier, such as structural damage or loose bolts, the main resonant peak frequency may shift or split. Therefore, detecting changes in the main resonant peak frequency can provide a preliminary assessment of the presence of an anomaly. The number of harmonic components reflects the amount of harmonic components in the spectrum besides the fundamental wave. Under normal circumstances, the vibration spectrum of a sound barrier may contain some harmonic components, but when an anomaly occurs, the number of harmonic components may increase or decrease. For example, certain types of faults may cause an increase in high-frequency harmonic components, while others may cause the harmonic components to become irregular or decrease. Therefore, changes in the number of harmonic components are also an important characteristic for identifying anomaly types. The energy decay slope describes the tendency of spectral energy to decay as frequency increases. Different abnormal conditions can cause variations in the energy decay slope. For example, material aging may accelerate energy decay in high-frequency bands, while structural looseness may slow energy decay in certain frequency bands. By analyzing the energy decay slope, different abnormality types can be further distinguished. A pre-set knowledge base stores spectral feature templates corresponding to various known fault modes. These templates are derived from analyzing extensive historical data and actual fault cases. By comparing and matching the extracted key spectral features with the fault mode templates in the knowledge base, the fault mode most similar to the current key spectral features can be found, thereby determining the type of abnormality. For example, if the extracted main resonance peak frequency is within the 4-6kHz range, the number of harmonic components is 1-2, and the energy decay slope conforms to a specific pattern, the knowledge base matching results indicate a loose bolt fault. Based on this matching result, the type of abnormality occurring in the sound barrier is ultimately determined. Different abnormality types correspond to different fault causes and repair strategies. Accurately identifying the abnormality type helps maintenance personnel quickly and effectively implement appropriate repair measures to restore the sound barrier to normal operation.

[0041] Key spectral features, including the main resonance peak frequency, number of harmonic components, and energy decay slope, are extracted from the target's spatiotemporal feature vector. These features are crucial indicators of sound barrier anomalies and accurately reflect the unique spectral manifestations of different anomaly types. By matching these features with fault patterns in a pre-defined knowledge base, the anomaly type can be accurately determined, providing clear guidance for subsequent maintenance and repair. This method can quickly extract the most critical spectral features from the complex target's spatiotemporal feature vector and efficiently match them with the fault patterns in the knowledge base, significantly improving the efficiency of anomaly identification and meeting the requirements of real-time sound barrier monitoring and rapid response. The pre-defined knowledge base includes a variety of common sound barrier failure modes and their corresponding spectral features, such as loose bolts, panel cracking, and foundation settlement. This ensures that the method covers a wide range of anomaly types and adapts to the needs of sound barrier anomaly detection in various scenarios. Matching key spectral features with fault patterns provides a reliable diagnostic basis for determining anomaly types, reducing the possibility of misjudgment. Furthermore, the clear feature matching relationships help maintenance personnel better understand the causes and mechanisms of anomalies, enabling more effective maintenance measures. The fault modes and corresponding spectral feature ranges in the preset knowledge base can be flexibly adjusted according to the structure and material characteristics of different sound barriers. It has strong adaptability and versatility and is suitable for anomaly detection of various types of sound barriers.

[0042] This embodiment also discloses a sound barrier anomaly recognition system based on an anomaly detection model. Figure 2 This is a module diagram of the sound barrier anomaly recognition system based on the anomaly detection model disclosed in the embodiment of the present application, such as Figure 2 As shown, the system includes a collection module 201, a screening module 202, a calculation module 203 and an execution module 204, wherein: An acquisition module 201 is configured to acquire multidimensional data, current traffic flow data, and environmental data from each sensor, construct a spatiotemporal data matrix based on a timestamp, the spatial coordinates of the sensor, and the multidimensional data, and construct a preset state feature vector based on the timestamp, the environmental data, and the current traffic flow data, wherein the multidimensional data includes noise sound pressure level, vibration spectrum, and structural stress value, and the environmental data includes wind speed, rainfall, and temperature; A screening module 202 is configured to construct a sensor topology map based on the spatiotemporal data matrix, obtain a spatiotemporal feature vector for each time step based on the sensor topology map, screen out target spatiotemporal feature vectors whose deviation from the preset state feature vector is greater than a first threshold, and determine abnormal candidate events based on the target spatiotemporal feature vectors; A calculation module 203 is configured to reconstruct the target spatiotemporal feature vector of the abnormal candidate event to obtain a reconstructed vector, calculate a first difference between the reconstructed vector and the target spatiotemporal feature vector, calculate the KL divergence of the abnormal candidate event to obtain a second difference, and perform a weighted sum of the first difference and the second difference to obtain a confidence level; The execution module 204 is configured to determine the sensor corresponding to the abnormal candidate event with a confidence level greater than a second threshold as a target sensor, determine the abnormal position in the sound barrier according to the spatial coordinates of the target sensor, and determine the abnormal type according to the target spatiotemporal feature vector corresponding to the target sensor.

[0043] Optionally, the screening module 202 is configured to: Taking each sensor as a graph node, constructing a node feature vector of the graph node according to the multidimensional data in the spatiotemporal data matrix, wherein the node feature vector includes noise sound pressure level, vibration spectrum energy distribution, and structural stress gradient; Constructing a graph edge according to the adjacent relationship of the sensors, and constructing an edge weight matrix according to the distance between the two sensors corresponding to the graph edge and the attenuation coefficient of the material of the sound barrier; A sensor topology graph is constructed according to the node feature vectors and the edge weight matrix.

[0044] Optionally, the screening module 202 is configured to: A spatial feature matrix is ​​obtained according to the sensor topology map and the node feature matrix corresponding to each time step, the spatial feature matrix is ​​arranged according to the time step to form a time series, and a gated recurrent unit is used to obtain a time feature matrix according to the time series; the spatial feature matrix and the time feature matrix are spliced ​​according to the node dimension to obtain intermediate features, and the intermediate features are reduced to the target dimension to obtain a spatiotemporal feature vector.

[0045] Optionally, the screening module 202 is configured to: Dimensionally aligning the spatiotemporal eigenvector with the preset state eigenvector, and calculating the vibration frequency deviation and the stress gradient deviation respectively; Performing dynamic time warping alignment on the harmonic component quantity change rate and energy attenuation slope in the time series, and then calculating the difference to obtain a time series dynamic deviation, and performing weighted summation on the vibration frequency deviation, the stress gradient deviation, and the time series dynamic deviation to obtain a target deviation; The spatiotemporal feature vector obtained when the target deviation is greater than the first threshold is determined as the target spatiotemporal feature vector.

[0046] Optionally, the calculation module 203 is configured to: Inputting the target spatiotemporal feature vector into a deep variational autoencoder to obtain a latent space distribution parameter, obtaining a latent vector according to the latent space distribution parameter, and inputting the latent vector into a decoder to obtain the reconstructed vector; Calculating a square error for each element of the reconstructed vector and the target spatiotemporal feature vector, and averaging the square errors of all elements to obtain a first difference; Determine the probability distribution of each element of the target spatiotemporal feature vector, determine the corresponding preset probability distribution, and sum the KL divergence between the probability distribution of all elements and the preset probability distribution to obtain a second difference.

[0047] Optionally, the execution module 204 is configured to: A propagation matrix is ​​constructed according to the spatial coordinates of the target sensor and the attenuation coefficient of the material of the sound barrier, and an inverse solution is performed based on the propagation matrix to obtain the abnormal position.

[0048] Optionally, the execution module 204 is configured to: Key spectral features are extracted from the target spatiotemporal feature vector. The key spectral features include the main resonance peak frequency, the number of harmonic components, and the energy attenuation slope. The key spectral features are matched with the fault modes in the preset knowledge base to determine the abnormality type.

[0049] It should be noted that the above embodiments provide devices that implement their functions using only the division of the above functional modules as examples. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiments are based on the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0050] This embodiment also discloses an electronic device, referring to Figure 3 The electronic device may include: at least one processor 301 , at least one communication bus 302 , a user interface 303 , a network interface 304 , and at least one memory 305 .

[0051] The communication bus 302 is used to implement the connection and communication between these components.

[0052] The user interface 303 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 303 may also include a standard wired interface and a wireless interface.

[0053] The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0054] The processor 301 may include one or more processing cores. The processor 301 utilizes various interfaces and lines to connect various parts of the entire server. By running or executing instructions, programs, code sets, or instruction sets stored in the memory 305, and calling data stored in the memory 305, the processor 301 performs various server functions and processes data. Optionally, the processor 301 may be implemented in at least one hardware form of a digital signal processing (DSP), a field-programmable gate array (FPGA), or a programmable logic array (PLA). The processor 301 may integrate one or a combination of a central processing unit (CPU), a graphics processing unit (GPU), and a modem. The CPU primarily processes the operating system, user interface, and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display; and the modem is used to handle wireless communications. It is understood that the modem may not be integrated into the processor 301 and may be implemented separately on a single chip.

[0055] Among them, the memory 305 may include a random access memory (RAM) or a read-only memory (Read-Only Memory). Optionally, the memory 305 includes a non-transitory computer-readable storage medium. The memory 305 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 305 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store data involved in the above-mentioned various method embodiments, etc. The memory 305 may also be optionally at least one storage device located away from the aforementioned processor 301. As Figure 3 As shown, the memory 305 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an application program of a sound barrier anomaly recognition method based on an anomaly detection model.

[0056] exist Figure 3In the electronic device shown, the user interface 303 is mainly used to provide an input interface for the user and obtain data input by the user; and the processor 301 can be used to call the application program of the sound barrier anomaly recognition method based on the anomaly detection model stored in the memory 305. When executed by one or more processors 301, the electronic device executes one or more methods as in the above embodiments.

[0057] It should be noted that for the aforementioned method embodiments, for simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that this application is not limited by the order of the actions described, because according to this application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for this application.

[0058] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0059] In the several embodiments provided in this application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are merely schematic, such as the division of units, which is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interface, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0060] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0061] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0062] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a memory 305 and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned memory 305 includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a magnetic disk, or an optical disk.

[0063] The above is only an exemplary embodiment of the present disclosure and cannot be used to limit the scope of the present disclosure. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification, those skilled in the art will easily think of other embodiments of the present disclosure. This application is intended to cover any variations, uses or adaptive changes of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the technical field that are not recorded in the present disclosure. The description and examples are to be regarded as exemplary only, and the scope and spirit of the present disclosure are defined by the claims.

Claims

1. A method for identifying anomalies in a sound barrier based on anomaly detection model, characterized in that: Applied to a sound barrier detection platform, the method includes: Acquiring multidimensional data, current traffic flow data, and environmental data from each sensor, constructing a spatiotemporal data matrix based on a timestamp, the spatial coordinates of the sensor, and the multidimensional data, and constructing a preset state feature vector based on the timestamp, the environmental data, and the current traffic flow data, wherein the multidimensional data includes noise sound pressure level, vibration spectrum, and structural stress value, and the environmental data includes wind speed, rainfall, and temperature; constructing a sensor topology map according to the spatiotemporal data matrix, obtaining a spatiotemporal feature vector for each time step according to the sensor topology map, screening out a target spatiotemporal feature vector having a deviation from the preset state feature vector greater than a first threshold, and determining an abnormal candidate event according to the target spatiotemporal feature vector; Reconstructing the target spatiotemporal feature vector of the abnormal candidate event to obtain a reconstructed vector, calculating a first difference between the reconstructed vector and the target spatiotemporal feature vector, calculating the KL divergence of the abnormal candidate event to obtain a second difference, and performing a weighted summation of the first difference and the second difference to obtain a confidence score; The sensor corresponding to the abnormal candidate event with a confidence greater than the second threshold is determined as the target sensor, the abnormal position in the sound barrier is determined according to the spatial coordinates of the target sensor, and the abnormal type is determined according to the target spatiotemporal feature vector corresponding to the target sensor.

2. The method for identifying anomalies in a sound barrier based on anomaly detection model according to claim 1, characterized in that: The constructing of a sensor topology map according to the spatiotemporal data matrix comprises: Taking each sensor as a graph node, constructing a node feature vector of the graph node according to the multidimensional data in the spatiotemporal data matrix, wherein the node feature vector includes noise sound pressure level, vibration spectrum energy distribution, and structural stress gradient; Constructing a graph edge according to the adjacent relationship of the sensors, and constructing an edge weight matrix according to the distance between the two sensors corresponding to the graph edge and the attenuation coefficient of the material of the sound barrier; A sensor topology graph is constructed according to the node feature vectors and the edge weight matrix.

3. The method for identifying anomalies in a sound barrier based on anomaly detection model according to claim 1, characterized in that: The step of obtaining the spatiotemporal feature vector of each time step according to the sensor topology graph includes: A spatial feature matrix is ​​obtained according to the sensor topology map and the node feature matrix corresponding to each time step, the spatial feature matrix is ​​arranged according to the time step to form a time series, and a time feature matrix is ​​obtained according to the time series using a gated recurrent unit; The spatial feature matrix and the temporal feature matrix are concatenated according to the node dimension to obtain intermediate features, and the intermediate features are reduced to the target dimension to obtain a spatiotemporal feature vector.

4. The method for identifying anomalies in a sound barrier based on anomaly detection model according to claim 1, characterized in that: The step of screening out a target spatiotemporal feature vector whose deviation from the preset state feature vector is greater than a first threshold comprises: Aligning the spatiotemporal eigenvector with the preset state eigenvector, and calculating the vibration frequency deviation and the stress gradient deviation respectively; Performing dynamic time warping alignment on the harmonic component quantity change rate and energy attenuation slope in the time series, and then calculating the difference to obtain a time series dynamic deviation, and performing weighted summation on the vibration frequency deviation, the stress gradient deviation, and the time series dynamic deviation to obtain a target deviation; The spatiotemporal feature vector obtained when the target deviation is greater than the first threshold is determined as the target spatiotemporal feature vector.

5. The method for identifying anomalies in a sound barrier based on anomaly detection model according to claim 1, characterized in that: The reconstructing the target spatiotemporal feature vector of the abnormal candidate event to obtain a reconstructed vector, calculating a first difference between the reconstructed vector and the target spatiotemporal feature vector, and calculating the KL divergence of the abnormal candidate event to obtain a second difference comprises: Inputting the target spatiotemporal feature vector into a deep variational autoencoder to obtain a latent space distribution parameter, obtaining a latent vector according to the latent space distribution parameter, and inputting the latent vector into a decoder to obtain the reconstructed vector; Calculating a square error for each element of the reconstructed vector and the target spatiotemporal feature vector, and averaging the square errors of all elements to obtain a first difference; Determine the probability distribution of each element of the target spatiotemporal feature vector, determine the corresponding preset probability distribution, and sum the KL divergence between the probability distribution of all elements and the preset probability distribution to obtain a second difference.

6. The method for identifying anomalies in a sound barrier based on anomaly detection model according to claim 1, characterized in that: Determining the abnormal position in the sound barrier according to the spatial coordinates of the target sensor includes: A propagation matrix is ​​constructed according to the spatial coordinates of the target sensor and the attenuation coefficient of the material of the sound barrier, and an inverse solution is performed based on the propagation matrix to obtain the abnormal position.

7. The method for identifying anomalies in a sound barrier based on anomaly detection model according to claim 1, characterized in that: Determining the abnormality type according to the target spatiotemporal feature vector corresponding to the target sensor includes: Key spectral features are extracted from the target spatiotemporal feature vector. The key spectral features include the main resonance peak frequency, the number of harmonic components, and the energy attenuation slope. The key spectral features are matched with the fault modes in the preset knowledge base to determine the abnormality type.

8. A sound barrier anomaly recognition system based on anomaly detection model, characterized in that: It includes acquisition module, screening module, calculation module and execution module, among which: an acquisition module configured to acquire multidimensional data, current traffic flow data, and environmental data from each sensor, construct a spatiotemporal data matrix based on a timestamp, spatial coordinates of the sensor, and the multidimensional data, and construct a preset state feature vector based on the timestamp, the environmental data, and the current traffic flow data, wherein the multidimensional data includes noise sound pressure level, vibration spectrum, and structural stress value, and the environmental data includes wind speed, rainfall, and temperature; a screening module configured to construct a sensor topology map based on the spatiotemporal data matrix, obtain a spatiotemporal feature vector for each time step based on the sensor topology map, screen out a target spatiotemporal feature vector whose deviation from the preset state feature vector is greater than a first threshold, and determine an abnormal candidate event based on the target spatiotemporal feature vector; a calculation module configured to reconstruct the target spatiotemporal feature vector of the abnormal candidate event to obtain a reconstructed vector, calculate a first difference between the reconstructed vector and the target spatiotemporal feature vector, calculate the KL divergence of the abnormal candidate event to obtain a second difference, and perform a weighted summation of the first difference and the second difference to obtain a confidence level; An execution module is configured to determine the sensor corresponding to the abnormal candidate event with a confidence level greater than a second threshold as a target sensor, determine the abnormal position in the sound barrier according to the spatial coordinates of the target sensor, and determine the abnormal type according to the target spatiotemporal feature vector corresponding to the target sensor.

9. An electronic device, characterized in that: The electronic device comprises a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are both used to communicate with other devices, and the processor is used to execute the instructions stored in the memory so that the electronic device executes the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores instructions, and when the instructions are executed, the method according to any one of claims 1 to 7 is executed.

Images