Structural acceleration data anomaly detection method based on two-dimensional convolutional neural network

Abstract

The application relates to a structure acceleration data anomaly detection method based on a two-dimensional convolutional neural network, which comprises the following steps: acquiring an original acceleration response data image of a target structure; dividing the original acceleration response data image into multiple data segment images according to fixed time intervals, pre-processing each data segment image, setting a plurality of acceleration data anomaly types, adding data anomaly labels to the data segment images according to the acceleration data anomaly types, and forming a training sample set; constructing a two-dimensional convolutional neural network, training the two-dimensional convolutional neural network by using the training sample set, and optimizing and adjusting network layer parameters and hyperparameters to obtain a trained and optimized structure acceleration data anomaly detection model; acquiring a current acceleration response data image of the target structure, inputting the pre-processed current acceleration response data image into the structure acceleration data anomaly detection model, and obtaining a data anomaly detection result.

Description

Technical Field

[0001] This invention relates to a method for detecting structural acceleration data anomalies based on a two-dimensional convolutional neural network, belonging to the field of civil engineering structural health monitoring technology. Background Technology

[0002] Civil engineering structural health monitoring systems acquire structural response data through the deployment of numerous sensors, thereby enabling structural health diagnosis. During the service life of a structure, sensor failures are inevitable, and data collected by faulty sensors can easily lead to misinterpretations in subsequent analysis. Therefore, real-time and efficient identification of anomalies in civil engineering structural monitoring data caused by sensor failures is crucial for the effective operation of the sensor network.

[0003] Machine learning-based structural response anomaly detection methods do not require the establishment of corresponding mechanistic models, and therefore have been widely used. Existing technologies, such as "Data Anomaly Identification in Structural Health Monitoring Systems [J]. Computer-Aided Engineering, 2016, 25(05):60-65," reduce false negatives and false negatives of anomaly data by combining the advantages of different support vector machines. Existing technologies, such as "Anomaly Detection Method for Wireless Sensor Networks Based on Neural Networks [J]. Computer Science, 2014, 41(S2):208-211," apply the BP neural network model to anomaly data diagnosis in wireless sensor networks, achieving self-diagnosis of data anomalies. The above studies mainly extract feature parameters from structural response data as anomaly detection indicators; most of these features are manually defined or sensitive indicators within a specific range. Furthermore, traditional machine learning methods involve complex data preprocessing, slow network convergence, and are prone to overfitting. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention proposes a method for detecting structural acceleration data anomalies based on a two-dimensional convolutional neural network.

[0005] The technical solution of the present invention is as follows:

[0006] On the one hand, this invention proposes a method for detecting structural acceleration data anomalies based on a two-dimensional convolutional neural network, comprising the following steps:

[0007] Obtain the raw acceleration response data image of the target structural component;

[0008] The original acceleration response data image is divided into multiple data segment images at fixed time intervals. Each data segment image is preprocessed, several acceleration data anomaly types are set, and data anomaly labels are added to the data segment images according to the acceleration data anomaly types to form a training sample set.

[0009] A two-dimensional convolutional neural network is constructed, trained using a training sample set, and the network layer parameters and hyperparameters are optimized and adjusted to obtain a trained and optimized structural acceleration data anomaly detection model.

[0010] The current acceleration response data image of the target structural component is obtained, preprocessed, and then input into the structural acceleration data anomaly detection model to obtain the data anomaly detection results.

[0011] In a preferred embodiment, the preprocessing method for each data segment image is as follows:

[0012] For each data segment image, a weighted average method is used for grayscale processing, as shown in the following formula:

[0013] Gray(i,j)=0.3R(i,j)+0.59G(i,j)+0.11B(i,j)

[0014] Where Gray(i,j) is the gray level of the pixel in the i-th row and j-th column of the corresponding data segment image after grayscale conversion; R(i,j), G(i,j), and B(i,j) are the brightness of the red, green, and blue corresponding to the pixel in the i-th row and j-th column of the corresponding data segment image, respectively.

[0015] In a preferred embodiment, the process of training and optimizing the two-dimensional convolutional neural network using the training sample set specifically includes:

[0016] For any input sample, obtain the gray level of its preprocessed data segment image, multiply the gray level by a weight coefficient and add a bias term, then add a nonlinear factor through the ReLU activation function to obtain the feature value c of each type of anomalous acceleration image. i As shown in the following formula:

[0017] c i =K(ωx(i,j)+β)

[0018] Where: ω is the weight parameter; is the bias parameter; K is the ReLU activation function; and c is the feature value based on each type of anomalous acceleration image. i The probability P of the structural acceleration data anomaly type to which the current input sample belongs is calculated as follows:

[0019]

[0020] Where q is the number of data anomaly types; e is the natural constant;

[0021] Based on the calculated probability P and the true values ​​corresponding to the outlier data labels of the current sample, the loss value is calculated as follows:

[0022]

[0023] Where D is the loss value; y i is the true value corresponding to the outlier data label of the sample; N is the total number of input samples;

[0024] The network layer parameters and hyperparameters of the two-dimensional convolutional neural network are optimized and adjusted based on the calculated loss values.

[0025] The above steps are executed iteratively until the iteration termination condition is met. The current network layer parameters and hyperparameters are saved to obtain the trained and optimized structural acceleration data anomaly detection model.

[0026] In a preferred embodiment, the acceleration data anomaly types include offset, drift, accuracy degradation, gain, missing data, and noise.

[0027] On the other hand, the present invention also proposes a structural acceleration data anomaly detection system based on a two-dimensional convolutional neural network, comprising:

[0028] The raw data acquisition module is used to acquire raw acceleration response data images of the target structural component;

[0029] The sample processing module is used to divide the original acceleration response data image into multiple data segment images at fixed time intervals, preprocess each data segment image, set several acceleration data anomaly types, add data anomaly labels to the data segment images according to the acceleration data anomaly types, and form a training sample set.

[0030] The model training module is used to construct a two-dimensional convolutional neural network. It trains the two-dimensional convolutional neural network using a training sample set and optimizes and adjusts the network layer parameters and hyperparameters to obtain a trained and optimized structural acceleration data anomaly detection model.

[0031] The detection module is used to acquire the current acceleration response data image of the target structural component. After preprocessing, the data is input into the structural acceleration data anomaly detection model to obtain the data anomaly detection results.

[0032] In a preferred embodiment, the preprocessing method for each data segment image is as follows:

[0033] For each data segment image, a weighted average method is used for grayscale processing, as shown in the following formula:

[0034] Gray(i,j)=0.3R(i,j)+0.59G(i,j)+0.11B(i,j)

[0035] Where Gray(i,j) is the gray level of the pixel in the i-th row and j-th column of the corresponding data segment image after grayscale conversion; R(i,j), G(i,j), and B(i,j) are the brightness of the red, green, and blue corresponding to the pixel in the i-th row and j-th column of the corresponding data segment image, respectively.

[0036] In a preferred embodiment, the process of training and optimizing the two-dimensional convolutional neural network using the training sample set in the model training module specifically involves:

[0037] For any input sample, obtain the gray level of its preprocessed data segment image, multiply the gray level by a weight coefficient and add a bias term, then add a nonlinear factor through the ReLU activation function to obtain the feature value c of each type of anomalous acceleration image. i As shown in the following formula:

[0038] c i =K(ωx(i,j)+β)

[0039] Where: ω is the weight parameter; β is the bias parameter; K is the ReLU activation function;

[0040] Based on the feature value c of each type of anomalous acceleration image i The probability P of the structural acceleration data anomaly type to which the current input sample belongs is calculated as follows:

[0041]

[0042] Where q is the number of data anomaly types; e is the natural constant;

[0043] Based on the calculated probability P and the true values ​​corresponding to the outlier data labels of the current sample, the loss value is calculated as follows:

[0044]

[0045] Where D is the loss value; y i is the true value corresponding to the outlier data label of the sample; N is the total number of input samples;

[0046] The network layer parameters and hyperparameters of the two-dimensional convolutional neural network are optimized and adjusted based on the calculated loss values.

[0047] The above steps are executed iteratively until the iteration termination condition is met. The current network layer parameters and hyperparameters are saved to obtain the trained and optimized structural acceleration data anomaly detection model.

[0048] In a preferred embodiment, the acceleration data anomaly types include offset, drift, accuracy degradation, gain, missing data, and noise.

[0049] The present invention has the following beneficial effects:

[0050] 1. This invention provides a structural acceleration data anomaly detection method based on a two-dimensional convolutional neural network. It utilizes the advantages of convolutional neural networks in autonomous learning and automatic data feature extraction. Combined with structural response anomaly data features, it can automatically extract structural response features and distinguish common structural response anomaly data types. The model has a fast convergence speed, short training and optimization time, and high detection accuracy. It avoids the large amount of manual feature extraction operations of traditional methods, reducing the workload and application cost of data anomaly detection.

[0051] 2. The present invention provides a method for detecting structural acceleration data anomalies based on a two-dimensional convolutional neural network, which performs grayscale processing on the image to eliminate the influence of color correlation on the accuracy of data anomaly detection. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention;

[0053] Figure 2 A flowchart illustrating a workflow provided in an embodiment of the present invention;

[0054] Figure 3 A numerical model diagram of the truss provided in an embodiment of the present invention;

[0055] Figure 4 Example image of acceleration response data of truss model node provided in an embodiment of the present invention;

[0056] Figure 5 This is an example diagram of data anomaly feature visualization provided in an embodiment of the present invention;

[0057] Figure 6 Example diagram of the optimized two-dimensional convolutional neural network model structure provided in the embodiments of the present invention;

[0058] Figure 7 An example diagram of data anomaly detection results provided in an embodiment of the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0061] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0062] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0063] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0064] Implementation Case:

[0065] To address the shortcomings of existing machine learning-based structural response anomaly detection methods, such as susceptibility to environmental factors, the potential failure of manually defined feature parameters in complex environments, the underutilization of many effective feature information during extraction, and the limitations of widespread application due to high computational costs, this paper proposes a structural response anomaly detection method capable of real-time and rapid processing of massive datasets and automatic feature extraction. Combining structural response anomaly data features, a structural acceleration data anomaly detection method based on a 2D Convolutional Neural Network (2D-CNN) is proposed, integrating data preprocessing, preliminary convolutional neural network design, network training, and optimization. This method can automatically extract structural anomaly response features, enabling rapid and accurate differentiation of acceleration data anomaly types.

[0066] See Figure 1 and Figure 2 The method provided in this embodiment specifically includes:

[0067] S100: Obtain the original acceleration response data image of the target structural component.

[0068] S200. Divide the original acceleration response data image into multiple data segment images at fixed time intervals, and perform grayscale processing on each data segment image using a weighted average method, as shown in the following formula:

[0069] Gray(i,j)=0.3R(i,j)+0.59G(i,j)+0.11B(i,j)

[0070] Where Gray(i,j) is the gray level of the pixel in the i-th row and j-th column of the corresponding data segment image after grayscale conversion; R(i,j), G(i,j), and B(i,j) are the brightness of the red, green, and blue corresponding to the pixel in the i-th row and j-th column of the corresponding data segment image, respectively.

[0071] Several acceleration data anomaly types are defined, and data anomaly labels are added to the data segment images according to the acceleration data anomaly types to form a training sample set.

[0072] S300. The process of constructing a two-dimensional convolutional neural network and training and optimizing the two-dimensional convolutional neural network using a training sample set. The specific method is as follows:

[0073] For any input sample, obtain the gray level of its preprocessed data segment image, multiply the gray level by a weight coefficient and add a bias term, then add a nonlinear factor through the ReLU activation function to obtain the feature value c of each type of anomalous acceleration image. i As shown in the following formula:

[0074] c i =K(ωx(i,j)+β)

[0075] Where: ω is the weight parameter; β is the bias parameter; K is the ReLU activation function;

[0076] Based on the feature value c of each type of anomalous acceleration image i The probability P of the structural acceleration data anomaly type to which the current input sample belongs is calculated as follows:

[0077]

[0078] Where q is the number of data anomaly types; e is the natural constant;

[0079] Based on the calculated probability P and the true values ​​corresponding to the outlier data labels of the current sample, the loss value is calculated as follows:

[0080]

[0081] Where D is the loss value; y i is the true value corresponding to the outlier data label of the sample; N is the total number of input samples;

[0082] The network layer parameters and hyperparameters of the two-dimensional convolutional neural network are optimized and adjusted based on the calculated loss value. The above steps are executed iteratively until the iteration termination condition is reached. In this embodiment, the iteration termination condition is D < 0.1. When the calculated loss value D ≥ 0.1, the network layer parameters and hyperparameters of the two-dimensional convolutional neural network are adjusted until D < 0.1. The current network layer parameters and hyperparameters are then saved to obtain the trained and optimized structural acceleration data anomaly detection model.

[0083] S400: Obtain the current acceleration response data image of the target structural component, perform preprocessing, and input it into the structural acceleration data anomaly detection model to obtain the data anomaly detection result.

[0084] To verify the effectiveness and superiority of the system and method provided in this embodiment, a specific experimental case is provided below:

[0085] This experimental case established the following: Figure 3 The numerical model of the truss shown consists of 10 nodes and 21 members, with the cross-sectional area of ​​each member uniformly set to A = 1 × 10⁻⁶. -4 m 2 The elastic modulus is E = 2.1 × 10⁻⁶. 11 Pa, with a mass density of ρ = 7300 kg / m³ 3 A random excitation is applied to node 4 in the vertically downward direction. The acceleration response of the truss node is calculated using the Newmark method, with parameters γ = 0.5 and β = 0.25.

[0086] The experimental platform used in the case study is configured as follows: Experimental system: Windows 10; Hardware configuration: CPU is Intel Core i5-1050, system memory is 16G; Experimental framework: MATLAB 2020 Deep Learning Toolbox.

[0087] First, according to step S100, the acceleration response data image of node 6 is collected as follows: Figure 4 As shown, the sampling frequency is 1000Hz and the sampling duration is 2000s.

[0088] Then, according to step S200, the original acceleration response data images are divided into images every 1 second, and the images are processed into grayscale. Six types of data anomalies were simulated: offset, drift, precision degradation, gain, missing data, and noise. A total of 12,000 data segment images were obtained. Data anomaly labels were added to the data segment images according to the acceleration data anomaly type to form a training sample set.

[0089] Next, following step S300, a two-dimensional convolutional neural network model is initially established, consisting of one input layer, one convolutional layer, one pooling layer, one fully connected layer, and one output layer. The convolutional layer uses eight 3×3 convolutional kernels, the pooling layer is 2×2, and the output layer contains label distributions for six types of structural acceleration data anomalies. Hyperparameter settings: initial learning rate 0.1, 500 training iterations, and a validation frequency of 50. The training sample set is input into the two-dimensional convolutional neural network for training. The feature values ​​of each type of anomalous acceleration image are calculated, and the extracted data anomaly features are visualized as follows: Figure 5 As shown, the calculated training loss value D is 0.32, indicating that the detection result has a large error compared to the actual result, and further optimization and adjustment of the network layer parameters and hyperparameters are needed.

[0090] After seven rounds of hyperparameter optimization, the following network layer parameters were obtained: 1 input layer, 4 convolutional layers, 4 pooling layers, 1 fully connected layer, and 1 output layer. The first convolutional layer used 8 3×3 convolutional kernels, while the 2nd, 3rd, and 4th convolutional layers used 16, 32, and 64 3×3 convolutional kernels, respectively. Hyperparameters: learning rate 0.001, 290 training iterations, and validation frequency 10. After 290 iterations, the training loss value D was 0.09, which is less than the threshold of 0.1, resulting in the following... Figure 6 The trained and optimized structural acceleration data anomaly detection model is shown.

[0091] Finally, according to step S400, the current acceleration response data image of node 6 is obtained, preprocessed, and then input into the trained and optimized structural acceleration data anomaly detection model for detection. The data anomaly detection results are as follows: Figure 7 As shown in the figure, the prediction probability of the input acceleration image is above 90%, and it is predicted as offset, drift, accuracy degradation, gain, missing, and noise, respectively. The recognition results are consistent with the actual assumed situation. The system and method proposed in this embodiment can effectively detect anomalies in the acceleration data of the structure.

[0092] In one embodiment, a structural acceleration data anomaly detection system based on a two-dimensional convolutional neural network is also provided, comprising:

[0093] The raw data acquisition module is used to acquire the raw acceleration response data image of the target structural component; this module is used to implement the function of step S100 above, and will not be described in detail here.

[0094] The sample processing module is used to divide the original acceleration response data image into multiple data segment images at fixed time intervals, preprocess each data segment image, set several acceleration data anomaly types, add data anomaly labels to the data segment images according to the acceleration data anomaly types, and form a training sample set. This module is used to implement the functions of step S200 above, which will not be described in detail here.

[0095] The model training module is used to construct a two-dimensional convolutional neural network. It trains the two-dimensional convolutional neural network using a training sample set and optimizes and adjusts the network layer parameters and hyperparameters to obtain a trained and optimized structural acceleration data anomaly detection model. This module is used to implement the functions of step S300 above, which will not be described in detail here.

[0096] The detection module is used to acquire the current acceleration response data image of the target structural component, and after preprocessing, input it into the structural acceleration data anomaly detection model to obtain the data anomaly detection result. This module is used to implement the function of step S400 above, which will not be described in detail here.

[0097] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0098] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0099] In the embodiments provided in this application, any system module, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0100] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A method for detecting abnormality of structural acceleration data based on a two-dimensional convolutional neural network, characterized in that, The method comprises the following steps: obtaining an original acceleration response data image of a target structure; dividing the original acceleration response data image into a plurality of data segment images according to fixed time intervals, pre-processing each data segment image, setting a plurality of acceleration data anomaly types, adding data anomaly labels to the data segment images according to the acceleration data anomaly types, and forming a training sample set; constructing a two-dimensional convolutional neural network, training the two-dimensional convolutional neural network using the training sample set, and optimizing and adjusting network layer parameters and hyperparameters to obtain a trained and optimized structure acceleration data anomaly detection model; obtaining a current acceleration response data image of the target structure, pre-processing the current acceleration response data image, inputting the pre-processed current acceleration response data image into the structure acceleration data anomaly detection model, and obtaining a data anomaly detection result; wherein the pre-processing method of each data segment image is as follows: each data segment image is subjected to gray processing by using a weighted average method, and the formula is as follows: Gray(i,j) = 0.3R(i,j) + 0.59G(i,j) + 0.11B(i,j) wherein Gray(i,j) is the gray level of the i-th row and j-th column pixel of the corresponding data segment image after gray processing; R(i,j), G(i,j), and B(i,j) are the red, green, and blue brightness of the i-th row and j-th column pixel of the corresponding data segment image, respectively; wherein the process of training and optimizing the two-dimensional convolutional neural network using the training sample set is as follows: For each input sample, the gray level of the preprocessed data segment image thereof is obtained, the gray level is multiplied by a weight coefficient and superimposed with a bias term, a non-linear factor is added through a ReLU activation function, and a feature value c of each type of abnormal acceleration image is obtained i As shown in the following formula: c i = K(ωx(i,j) + β) wherein ω is a weight parameter, β is a bias parameter, and K is a ReLU activation function; a feature value c based on each type of abnormal acceleration image i a probability P that the current input sample belongs to a structure acceleration data abnormal type, as shown in the following formula: wherein q is the number of data anomaly types, and e is a natural constant; based on the calculated probability P and the true value corresponding to the anomaly data label of the training sample, a loss value is calculated, and the formula is as follows: wherein D is a loss value; y i is a true value corresponding to an abnormal data label of the sample; N is a total number of input samples; based on the calculated loss value, the network layer parameters and hyperparameters of the two-dimensional convolutional neural network are optimized and adjusted; the above steps are iteratively executed until the iteration termination condition is reached, the current network layer parameters and hyperparameters are saved, and the trained and optimized structure acceleration data anomaly detection model is obtained.

2. The structure acceleration data anomaly detection method based on a two-dimensional convolutional neural network according to claim 1, characterized in that: the acceleration data anomaly types include offset, drift, precision degradation, gain, loss, and noise.

3. A two-dimensional convolutional neural network-based structural acceleration data anomaly detection system, characterized by, comprising: an original data acquisition module for acquiring an original acceleration response data image of a target structure; a sample processing module for dividing the original acceleration response data image into a plurality of data segment images according to fixed time intervals, pre-processing each data segment image, setting a plurality of acceleration data anomaly types, adding data anomaly labels to the data segment images according to the acceleration data anomaly types, and forming a training sample set; a model training module for constructing a two-dimensional convolutional neural network, training the two-dimensional convolutional neural network using the training sample set, and optimizing and adjusting network layer parameters and hyperparameters to obtain a trained and optimized structure acceleration data anomaly detection model; The detection module is configured to acquire a current acceleration response data image of a target structure, pre-process the current acceleration response data image, and input the pre-processed current acceleration response data image into a structure acceleration data anomaly detection model to obtain a data anomaly detection result. The pre-processing method of each data segment image is as follows: The weighted average method is used for gray processing of each data segment image, and the gray processing is as shown in the following formula: Gray(i,j) = 0.3R(i,j) + 0.59G(i,j) + 0.11B(i,j) Gray(i,j) is the gray level of the i-th row and j-th column pixel of the corresponding data segment image after gray processing; R(i,j), G(i,j), and B(i,j) are the red, green, and blue brightness of the i-th row and j-th column pixel of the corresponding data segment image, respectively. In the model training module, the process of training and optimizing the two-dimensional convolutional neural network by using the training sample set is as follows: For each input sample, the gray level of the preprocessed data segment image thereof is obtained, the gray level is multiplied by a weight coefficient and superimposed with a bias term, a non-linear factor is added through a ReLU activation function, and a feature value c of each type of abnormal acceleration image is obtained i As shown in the following formula: c i = K(ωx(i,j) + β) ω is a weight parameter; β is a bias parameter; and K is a ReLU activation function. a feature value c based on each type of abnormal acceleration image i a probability P that the current input sample belongs to a structure acceleration data abnormal type, as shown in the following formula: q is the number of data anomaly types; and e is a natural constant. Based on the calculated probability P and the real value corresponding to the anomaly data label of the current sample, a loss value is calculated, as shown in the following formula: wherein D is a loss value; y i is a true value corresponding to an abnormal data label of the sample; N is a total number of input samples; Based on the calculated loss value, the network layer parameters and hyperparameters of the two-dimensional convolutional neural network are optimized and adjusted. The above steps are iteratively executed until the iteration termination condition is reached, the current network layer parameters and hyperparameters are saved, and the trained and optimized structure acceleration data anomaly detection model is obtained.

4. The structure acceleration data anomaly detection system based on a two-dimensional convolutional neural network according to claim 3, characterized in that: The acceleration data anomaly types include offset, drift, precision degradation, gain, missing, and noise.

Images