Visual real-time monitoring method and system for unbalanced bending moment of bridge rotation

Abstract

The present invention relates to the field of data processing, and specifically provides a visualized real-time monitoring method and monitoring system for unbalanced bending moment of bridge rotation, by obtaining the bridge rotation sensor data stream to be identified, inputting the bridge rotation sensor data stream into the pre-trained target recognition algorithm, and obtaining the recognition result data stream, the target recognition algorithm is a recognition algorithm obtained by inputting knowledge samples into the basic recognition algorithm to be optimized, and the knowledge samples are samples constructed by extracting multiple sample characterization vectors from multiple bridge rotation sensor data set samples and the unbalanced bending moment recognition results and feedback information corresponding to each sample characterization vector. Unbalanced bending moment classification is performed based on past data and implementation data, combining the before-after correlation and change characteristics of the data, thereby improving the recognition accuracy without the need for other calculation processes.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a visualized real-time monitoring method and a monitoring system for unbalanced bending moments of bridge rotations. Background Art

[0002] With the rapid development of infrastructure construction, bridge rotation construction technology has been widely used due to its high efficiency and safety. However, the unbalanced bending moment problem that may occur during the bridge rotation process has always been an important challenge in construction. Unbalanced bending moment may cause structural deformation or even damage, seriously affecting the safety and service life of the bridge. Therefore, real-time monitoring of unbalanced bending moment during the bridge rotation process is crucial to ensure construction quality and safety. Traditional unbalanced bending moment monitoring methods mainly rely on manual observation and empirical judgment. This method is not only time-consuming and labor-intensive, but also easily affected by subjective factors, resulting in inaccurate monitoring results. In recent years, with the advancement of sensor technology and data processing technology, automatic monitoring methods based on sensor data have gradually become a research hotspot. These methods usually involve multiple links such as sensor deployment, data acquisition, feature extraction and pattern recognition, which can realize real-time monitoring of unbalanced bending moment during bridge rotation. Existing automatic monitoring methods often only identify unbalanced bending moments based on real-time data, ignoring the changes in data, and the accuracy of identification needs to be improved. Summary of the invention

[0003] In view of this, an embodiment of the present invention provides a visual real-time monitoring method and monitoring system for unbalanced bending moment of bridge rotation. The technical solution of the embodiment of the present invention is implemented as follows:

[0004] On the one hand, an embodiment of the present invention provides a visualized real-time monitoring method for unbalanced bending moment of a bridge rotation, comprising: obtaining a bridge rotation sensor data stream to be identified, wherein the bridge rotation sensor data stream includes target data to be identified, and the bridge rotation sensor data stream is sequence data composed of multiple bridge rotation sensor data sets; inputting the bridge rotation sensor data stream into a pre-trained target recognition algorithm to obtain a recognition result data stream, wherein the target recognition algorithm is a recognition algorithm obtained by optimizing a basic recognition algorithm to be optimized by inputting knowledge samples, and the knowledge samples are samples constructed by extracting multiple sample representation vectors from samples of multiple bridge rotation sensor data sets and unbalanced bending moment recognition results and feedback information corresponding to each sample representation vector, and the unbalanced bending moment The balanced bending moment identification result represents an unbalanced bending moment classification of a sample characterization vector, the feedback information represents the benefit corresponding to the unbalanced bending moment classification, the multiple bridge rotation sensor data set samples are multiple data samples with a continuous time series relationship, and each sample characterization vector includes features collaboratively extracted based on a bridge rotation sensor data set sample and a previous bridge rotation sensor data set sample of the bridge rotation sensor data set sample; based on the identification result data stream, the identification result corresponding to each bridge rotation sensor data set in the bridge rotation sensor data stream is determined, wherein the identification result represents the unbalanced bending moment classification of the target data; according to the unbalanced bending moment classification of the target data, a corresponding visualization display is performed in a preset bridge rotation digital twin.

[0005] On the other hand, an embodiment of the present invention provides a monitoring system, including a memory and a processor, wherein the memory stores a computer program executable on the processor, and the processor implements the steps in the above method when executing the program.

[0006] The beneficial effects of the present invention include: the present invention obtains a bridge rotation sensor data stream to be identified, the bridge rotation sensor data stream includes target data to be identified, and inputs the bridge rotation sensor data stream into a pre-trained target recognition algorithm to obtain a recognition result data stream, the target recognition algorithm is a recognition algorithm obtained by inputting a knowledge sample into a basic recognition algorithm to be optimized, the knowledge sample is a sample constructed by extracting multiple sample characterization vectors from multiple bridge rotation sensor data set samples and an unbalanced moment recognition result and feedback information corresponding to each sample characterization vector, the unbalanced moment recognition result represents an unbalanced moment classification of a sample characterization vector, the feedback information represents the benefit corresponding to the unbalanced moment classification, and multiple bridge The beam rotation sensor data set samples are multiple data samples with a continuous time series relationship. Each sample representation vector includes features extracted collaboratively based on a bridge rotation sensor data set sample and a bridge rotation sensor data set sample of the previous bridge rotation sensor data set sample. The recognition result corresponding to each bridge rotation sensor data set in the bridge rotation sensor data stream is determined based on the recognition result data stream. The recognition result represents the unbalanced moment classification of the target data. Based on the past data in the bridge rotation sensor data stream, the representation information of the past data and the real-time data is fused and fed back to complete the algorithm optimization. The optimized algorithm obtained in this way has a more accurate recognition result for the input bridge rotation sensor data stream when identifying the unbalanced moment. The unbalanced moment classification based on the past data and the real-time data combines the previous and next correlation and change characteristics of the data, thereby improving the recognition accuracy without the need for other calculation processes.

[0007] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and are not intended to limit the technical solutions of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The accompanying drawings herein are incorporated into and constitute a part of the specification. These drawings illustrate embodiments consistent with the present invention and, together with the specification, are used to explain the technical solutions of the present invention.

[0009] Figure 1 A schematic diagram of the implementation flow of a visualized real-time monitoring method for unbalanced bending moment of a bridge rotation provided by an embodiment of the present invention.

[0010] Figure 2 A schematic diagram of a hardware entity of a monitoring system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0011] In order to make the purpose, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention are further elaborated in detail below in conjunction with the drawings and embodiments. The described embodiments should not be regarded as limiting the present invention. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present invention.

[0012] The embodiment of the present invention provides a visual real-time monitoring method for unbalanced bending moment of a bridge rotation, which can be executed by a processor of a monitoring system. The monitoring system can refer to a server, a laptop, a tablet computer, a desktop computer, a mobile device (such as a mobile phone, a portable video player, a personal digital assistant, a dedicated messaging device, a portable gaming device) and other devices with data processing capabilities.

[0013] Figure 1 A schematic diagram of the implementation flow of a visual bridge rotation unbalanced bending moment real-time monitoring method provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the method includes:

[0014] Step S110: Acquire a bridge rotation sensor data stream to be identified, wherein the bridge rotation sensor data stream includes target data to be identified, and the bridge rotation sensor data stream is sequence data composed of multiple bridge rotation sensor data sets.

[0015] In step S110, the monitoring system is responsible for acquiring the sensor data stream of the bridge rotation to be identified. These data streams are continuously collected by multiple sensors to form a set of ordered bridge rotation sensor data sets, which together constitute a continuous time series. For example, suppose a bridge is undergoing rotation construction, one end of the bridge body is fixed to the shore, and the other end gradually docks with the other shore through rotation. In this process, the balance of the bridge body is very important, and any unbalanced bending moment may cause structural damage or even accidents. Therefore, the sensors installed on the bridge body will continuously monitor the state of the bridge body, such as the weight distribution of various parts of the bridge body, the inclination angle of the bridge body, the rotation speed, etc. The monitoring system will continue to receive data streams from these sensors. These data streams are in the form of time series, that is, a series of data points arranged in time order. For example, assuming that data is collected once per second, the data stream is represented by a series of data sets arranged by seconds, and each data set is a snapshot of the bridge state collected at a certain point in time. Each item in these data sets can contain multiple features, such as weight distribution, inclination angle, etc.

[0016] Specifically, a data set can be as follows: Dataset = [weight distribution, tilt angle, rotation speed]. Among them, the weight distribution is, for example, a vector, such as [1200, 1100, 1250, 1300] kilograms, which represents the weight of the bridge at different positions. The tilt angle is, for example, (-3°) or (+2°), which represents the degree of inclination of the bridge relative to the horizontal plane. The rotation speed is, for example, (0.5° / s), which represents the angle at which the bridge rotates per second. Over time, the monitoring system will receive one such data set after another, forming a continuous data stream. For example, in the previous minute, the system may receive a data stream as follows:

[0017] ;

[0018] In this example, the monitoring system identifies potential unbalanced bending moment issues from these data sets. For example, if the weight distribution of the bridge begins to change so that one side is much heavier than the other, this could be a sign of unbalanced bending moment. Therefore, each data set in the data stream may contain target data to be identified - information that may indicate unbalanced bending moment.

[0019] Step S120: Input the bridge rotation sensor data stream into the pre-trained target recognition algorithm to obtain a recognition result data stream, wherein the target recognition algorithm is a recognition algorithm obtained by inputting knowledge samples into the basic recognition algorithm to be optimized for optimization, and the knowledge samples are samples constructed by extracting multiple sample characterization vectors from multiple bridge rotation sensor data set samples and the unbalanced bending moment recognition result and feedback information corresponding to each sample characterization vector, the unbalanced bending moment recognition result represents the unbalanced bending moment classification of a sample characterization vector, and the feedback information represents the benefit corresponding to the unbalanced bending moment classification, the multiple bridge rotation sensor data set samples are multiple data samples with a continuous time series relationship, and each sample characterization vector includes features extracted collaboratively based on a bridge rotation sensor data set sample and a previous bridge rotation sensor data set sample of the bridge rotation sensor data set sample.

[0020] In step S120, the monitoring system inputs the received bridge rotation sensor data stream into a pre-trained target recognition algorithm to obtain the recognition result data stream of the unbalanced bending moment. This process involves the reinforcement learning technology and data processing methods in machine learning. At this stage, the bridge rotation sensor data stream is a sequence data composed of multiple bridge rotation sensor data sets, each of which is a snapshot of sensor data collected at a specific time point, reflecting various state information during the bridge rotation process.

[0021] The monitoring system processes a series of such data sets to form a time series data stream. For example, assuming that data is collected once every second, the data stream can be expressed as:

[0022] ;in, Represents data flow, Represents a dataset.

[0023] In order to train the target recognition algorithm, the monitoring system constructs knowledge samples. These knowledge samples include sample representation vectors, unbalanced bending moment recognition results, and feedback information. The sample representation vector is extracted from multiple bridge rotation sensor data set samples. It not only contains the information of the current data set, but also considers the correlation with the previous data set. For example, a sample representation vector can be expressed as: .in, is the feature vector of the current data set, is the feature change vector between the current dataset and the previous dataset. For example, if the current dataset is , the previous data set is , then the change vector can be expressed as: The unbalanced bending moment recognition result is a classification of the sample characterization vector, such as "normal" or "abnormal". The feedback information is the benefit corresponding to the unbalanced bending moment classification, which is similar to the reward mechanism in reinforcement learning. For example, if the recognition result is correct, the feedback information is positive; if the recognition result is wrong, the feedback information is negative.

[0024] The monitoring system uses these knowledge samples to optimize the basic recognition algorithm to obtain the target recognition algorithm. This process can use reinforcement learning methods, such as the Q-learning algorithm. Assume that in the initial state, the parameters of the algorithm are , for each sample characterization vector s and the corresponding unbalanced bending moment identification result a, the goal of the algorithm is to minimize a loss function L, which can be expressed as:

[0025] ;

[0026] in, The algorithm is in parameters is the estimate of the expected return of state s and action a, r is the immediate feedback information, is the discount factor, s' is the next state, and a' is the action in the next state. The monitoring system will iteratively update the algorithm parameters , until the convergence condition is met. For example, the gradient descent method can be used to update the parameters, the formula is as follows:

[0027] ;

[0028] Among them, t represents the current time, and t+1 represents the next time after the current time; is the learning rate.

[0029] Once the target recognition algorithm is trained, the monitoring system can input the real-time bridge rotation sensor data stream into the algorithm to obtain a recognition result data stream. For example, for each data set in the data stream, the algorithm can output a classification of unbalanced bending moment, such as "normal" or "abnormal."

[0030] Step S130: Determine the recognition result corresponding to each bridge rotation sensor data set in the bridge rotation sensor data stream based on the recognition result data stream, wherein the recognition result represents the unbalanced bending moment classification of the target data.

[0031] In step S130, the monitoring system determines the recognition result of each bridge rotation sensor data set in the bridge rotation sensor data stream based on the recognition result data stream obtained from the target recognition algorithm. The recognition result here represents the unbalanced bending moment classification of the target data, that is, it indicates whether there is an unbalanced bending moment problem during the bridge rotation process. At this stage, the monitoring system has obtained the recognition result data stream from the target recognition algorithm, and this data stream contains the unbalanced bending moment classification results for each bridge rotation sensor data set. Each element in the recognition result data stream corresponds to a bridge rotation sensor data set and indicates whether the data set contains an unbalanced bending moment.

[0032] Suppose there is a bridge rotation sensor data stream, which contains a series of bridge rotation sensor data sets. These data sets are arranged in chronological order, and each data set represents the data collected from the sensor at a certain point in time. Suppose there is such a data stream:

[0033] , the monitoring system obtains the recognition result data stream from the target recognition algorithm, and each element in this data stream corresponds to a data set in the above data stream. For example, the recognition result data stream can be:

[0034] ; Represents the recognition result data stream, where each express For example, "normal" means that no unbalanced bending moment is detected in the dataset, and "abnormal" means that unbalanced bending moment is detected.

[0035] In actual applications, the monitoring system continuously collects data from sensors and inputs this data into the target recognition algorithm for processing. Once the recognition result data stream is generated, the system can immediately determine the recognition result of each bridge rotation sensor data set. For example, if at a certain moment, an "abnormal" appears in the recognition result data stream, it means that there is an unbalanced bending moment problem in the data set at that moment. In this way, the monitoring system can quickly identify whether an unbalanced bending moment problem occurs during the bridge rotation process during real-time monitoring. This capability is crucial to ensuring the safety and efficiency of bridge rotation construction. Through timely feedback and visual display, monitoring personnel can quickly take measures to deal with possible problems and avoid potential risks.

[0036] Step S140: According to the unbalanced bending moment classification of the target data, a corresponding visual display is performed in a preset bridge rotation digital twin.

[0037] In step S140, the monitoring system performs corresponding visualization in the preset bridge rotation digital twin based on the unbalanced bending moment classification results of the target data. This step is a key step in converting complex data analysis into intuitive and easy-to-understand information.

[0038] At this stage, the monitoring system has obtained the classification results of the unbalanced bending moment and is ready to be visualized in the bridge rotation digital twin. A digital twin is a virtual model that simulates a physical object in the real world and reflects its real-time status in real time. In this case, the digital twin simulates the rotation process of the bridge under construction.

[0039] In order to display the classification results of the unbalanced bending moment, a digital twin of the bridge swivel is first created. This digital twin should be able to accurately simulate the physical structure and dynamic behavior of the bridge swivel. For example, the digital twin can include the following elements: a three-dimensional model of the bridge, motion simulation of the swivel mechanism, real-time display of sensor data, and visualization. Once the digital twin is built, the monitoring system can visualize the classification results of the unbalanced bending moment in the digital twin. For example, the following steps are included:

[0040] 1. Determine the classification result: The monitoring system determines the classification result of the unbalanced bending moment of each bridge rotation sensor data set based on the output of the target recognition algorithm. For example, for each data set, the classification result can be "normal" or "abnormal".

[0041] 2. Mapping to the digital twin: Based on the classification results, the monitoring system will color or otherwise annotate different parts of the bridge in the digital twin to visually show which areas have unbalanced bending moment problems. For example, if an area is identified to have unbalanced bending moment, the area can be marked red in the digital twin.

[0042] In one example, assume that the surveillance system has obtained the following classification results from the target recognition algorithm:

[0043] ;

[0044] For the bridge model in the digital twin, the monitoring system can be visualized based on these classification results. For example, suppose the bridge in the digital twin is divided into four parts, corresponding to four bridge rotation sensor data sets. Based on the above classification results, it can be displayed as follows:

[0045] The first part (corresponding to ) is displayed in green, indicating that the unbalanced bending moment classification result in this area is “Normal”.

[0046] The second part (corresponding to ) is displayed in red, indicating that the classification result of the unbalanced bending moment in this area is “abnormal”.

[0047] Other parts (corresponding to ) are displayed in green, indicating that they are also classified as Normal.

[0048] In this way, monitoring personnel can clearly see in the digital twin whether there is an unbalanced bending moment problem during the bridge rotation process. For example, if a red-marked part appears in the digital twin, it indicates that there is a potential risk of unbalanced bending moment in the area, which needs further inspection and processing.

[0049] Through step S140, the monitoring system can not only effectively identify the unbalanced bending moment problem during the bridge rotation process, but also intuitively display these problems through the bridge rotation digital twin. This visualization method greatly improves the work efficiency of monitoring personnel, enabling them to quickly identify potential risk points and take corresponding measures to ensure construction safety.

[0050] In some feasible implementation schemes, the process of obtaining the above sample representation vector can be implemented by the following steps:

[0051] Step S1201: extract first characterization information from the plurality of bridge rotation sensor data set samples to obtain a first characterization vector cluster, wherein a characterization vector in the first characterization vector cluster corresponds to a bridge rotation sensor data set sample among the plurality of bridge rotation sensor data set samples.

[0052] In step S1201, the monitoring system extracts the first characterization information from multiple bridge rotation sensor data set samples to obtain a first characterization vector cluster. Each characterization vector in these characterization vector clusters corresponds to a specific bridge rotation sensor data set sample and is used to characterize the key features of the data set sample. In step S1201, the monitoring system processes multiple bridge rotation sensor data set samples and extracts the first characterization information from each sample to obtain a first characterization vector cluster. Each characterization vector in these characterization vector clusters corresponds to a specific bridge rotation sensor data set sample and is used to characterize the key features of the data set sample.

[0053] The monitoring system will extract the first characterization information for each data set sample. For example, for the data set , the monitoring system will extract the first representation information and obtain a representation vector. These representation vectors will be combined into a first representation vector cluster.

[0054] In order to extract the first characterization information, the monitoring system can use a variety of feature extraction methods. For example, principal component analysis (PCA), autoencoder, or deep learning models such as convolutional neural networks (CNN) can be used. These methods can help the monitoring system extract the most meaningful features from the raw data. Continuing with the previous example, assume that PCA is used to extract features. PCA is a common dimensionality reduction technique that can help extract the principal components that best represent the direction of data change from the data set. For each data set sample, the monitoring system can use PCA to extract the two most important principal components as characterization vectors. For example, for the data set , the monitoring system can extract the two most important principal components as the characterization vector. Assume that the first principal component extracted is closely related to the change of weight distribution, and the second principal component is closely related to the change of tilt angle. Then, for , the extracted first representation vector can be expressed as: ; This means The first characterization vector cluster of is: .in, represents the first characterization vector cluster; Represents the first characterization vector.

[0055] The process of PCA calculation can be as follows:

[0056] 1. Data standardization: First, the data is standardized to eliminate the dimensional effects between features. The standardization formula is:

[0057] ;

[0058] in, is the standardized data, X is the original data, is the mean, is the standard deviation.

[0059] 2. Covariance matrix calculation: Next, calculate the covariance matrix of the standardized data. The covariance matrix can help understand the correlation between features. The calculation formula of the covariance matrix is: ;

[0060] Where C is the covariance matrix, N is the number of samples, is the i-th sample and x is the sample mean.

[0061] 3. Calculate eigenvalues ​​and eigenvectors: Then, calculate the eigenvalues ​​and eigenvectors of the covariance matrix. Eigenvalues ​​and eigenvectors can help determine the main direction of change in the data.

[0062] 4. Select principal components: Finally, select the eigenvectors corresponding to the largest eigenvalues ​​as principal components. These principal components constitute the characterization vector of . Assume that the calculated covariance matrix is: ;

[0063] The calculated eigenvalues ​​and corresponding eigenvectors are:

[0064] Eigenvalue , the eigenvector ;

[0065] Eigenvalue , the eigenvector ;

[0066] Eigenvalue , the eigenvector ;

[0067] If the eigenvectors corresponding to the first two eigenvalues ​​are selected as principal components, then for the data set samples , the extracted first representation vector can be expressed as: .

[0068] Through step S1201, the monitoring system can effectively extract the first characterization information from multiple bridge rotation sensor data set samples, and integrate the information into a first characterization vector cluster. Each characterization vector in these characterization vector clusters corresponds to a specific bridge rotation sensor data set sample, and is used to characterize the key features of the data set sample. In this way, the monitoring system can understand and process the data more comprehensively, thereby improving the accuracy of unbalanced bending moment identification.

[0069] Step S1202: extract second characterization information from the multiple bridge rotation sensor data set samples to obtain a second characterization vector cluster, wherein a first characterization vector in the second characterization vector cluster corresponds to a first bridge rotation sensor data set sample in the bridge rotation sensor data set samples, the previous bridge rotation sensor data set of the first bridge rotation sensor data set sample is a second bridge rotation sensor data set sample, and the first characterization vector characterizes the association between the first bridge rotation sensor data set sample and the second bridge rotation sensor data set sample.

[0070] In this step, the monitoring system considers the association between adjacent dataset samples to extract the second representation information, thereby obtaining a second representation vector cluster. Each first representation vector in these vector clusters corresponds to a specific bridge rotation sensor dataset sample and considers the changes between the sample and its previous dataset sample.

[0071] Step S1203: merging the first characterization vector cluster and the second characterization vector cluster respectively to obtain the multiple sample characterization vectors.

[0072] In step S1203, the monitoring system merges the first characterization vector cluster and the second characterization vector cluster obtained previously to obtain the final sample characterization vector. These sample characterization vectors will be used for subsequent unbalanced bending moment identification tasks. In this step, the monitoring system merges the first characterization vector cluster and the second characterization vector cluster obtained previously to obtain the final sample characterization vector. These sample characterization vectors contain the key features of each bridge rotation sensor data set sample and the correlation with the previous data set sample. The monitoring system has extracted the first characterization vector cluster and the second characterization vector cluster from these data set samples.

[0073] Assume that the first characterization vector cluster and the second characterization vector cluster have been obtained. For example, for the data set , the characterization vector in the first characterization vector cluster is: ; The characterization vectors in the second characterization vector cluster are: To obtain the final sample representation vector, the monitoring system merges the two representation vectors. Usually, this can be achieved through a simple vector concatenation operation.

[0074] Through step S1203, the monitoring system can effectively merge the first characterization vector cluster and the second characterization vector cluster to obtain the final sample characterization vector. These sample characterization vectors contain the key features of each bridge rotation sensor data set sample and the correlation with the previous data set sample. In this way, the monitoring system can understand and process the data more comprehensively, thereby improving the accuracy of unbalanced bending moment identification.

[0075] In some feasible implementation schemes, the step S1202, extracting the second characterization information from the plurality of bridge rotation sensor data set samples to obtain the second characterization vector cluster, can be implemented in the following manner:

[0076] The second characterization information is extracted from the multiple bridge rotation sensor data set samples according to the following steps to obtain a second characterization vector cluster. Each time the bridge rotation sensor data set sample that undergoes the second characterization information extraction is regarded as a real-time bridge rotation sensor data set sample, the obtained characterization vector is regarded as a real-time characterization vector, and the second characterization vector cluster includes the real-time characterization vector.

[0077] Step S12021: Perform unbalanced bending moment identification on the real-time bridge rotation sensor data set sample to determine a first main data item cluster, where the first main data item cluster is used to locate the position of the target data in the real-time bridge rotation sensor data set sample.

[0078] In step S12021, the monitoring system identifies the unbalanced bending moment of the real-time bridge rotation sensor data set sample and determines the first main data item cluster. This cluster is used to locate the target data, that is, the specific position of the key information of the unbalanced bending moment in the real-time bridge rotation sensor data set sample. In order to better understand this process, let us explain it in detail through a specific scenario. In order to identify the unbalanced bending moment, the monitoring system will first analyze the key data points in the real-time bridge rotation sensor data set sample. The key data points here are those physical quantities that are closely related to the unbalanced bending moment. For example, the displacement difference at both ends of the bridge body is caused by the unbalanced bending moment, so these displacement data become the focus of attention.

[0079] The monitoring system then uses a recognition algorithm to analyze these key data points. The algorithm can be a pre-trained machine learning model, such as a support vector machine (SVM), random forest, or a deep learning model such as a convolutional neural network (CNN). In the case of a convolutional neural network, the model may have been trained with a large amount of historical data to identify unbalanced bending moments in a specific mode.

[0080] During the identification process, the monitoring system extracts features that are closely related to the unbalanced bending moment. For example, assuming that the displacement difference at both ends of the bridge is one of the key features, the monitoring system can calculate the displacement difference at both ends of the bridge and use it as one of the input features. These features will be organized into a feature vector, such as in Represents the difference in displacement between the two ends of the bridge. is the inclination angle of the bridge relative to the horizontal plane, is the pressure measurement at the center of the bridge deck.

[0081] Next, the monitoring system inputs the feature vector into the unbalanced bending moment identification algorithm. For example, for a convolutional neural network model, it may contain multiple convolutional layers, pooling layers, and fully connected layers. The model processes the feature vector through these layers and finally outputs the classification result of the unbalanced bending moment. The output of the model can be a vector representing the probability distribution of different unbalanced bending moment levels. For example, the output vector may be [0.1, 0.6, 0.3], which means that the probability of slight imbalance is 10%, the probability of moderate imbalance is 60%, and the probability of severe imbalance is 30%. The monitoring system determines the level of unbalanced bending moment based on these probability distributions. For example, if the probability of moderate imbalance is the highest, the system determines that the current unbalanced bending moment level is moderate.

[0082] Finally, the monitoring system determines the first main data item cluster, which contains the data items most related to the unbalanced bending moment. For example, if the displacement difference between the two ends of the bridge is the main factor of the unbalanced bending moment, then this difference value will become part of the first main data item cluster. Assuming that in the current rotation operation, the difference in displacement between the two ends of the bridge is 5cm, and the inclination angle of the bridge relative to the horizontal plane is 2°, then the first main data item cluster can be expressed as In this way, the monitoring system not only identifies the level of unbalanced bending moment, but also determines the most relevant key data item clusters, which will be used in the subsequent unbalanced bending moment classification and algorithm optimization process.

[0083] Step S12022: Perform unbalanced bending moment identification on the previous bridge rotation sensor data set sample of the real-time bridge rotation sensor data set sample to determine a second main data item cluster, wherein the second main data item cluster is used to locate the position of the target data in the previous bridge rotation sensor data set sample.

[0084] In step S12022, the monitoring system performs unbalanced bending moment identification on the previous sample of the real-time bridge rotation sensor data set sample and determines the second main data item cluster. This cluster is used to locate the key data related to the unbalanced bending moment in the previous sensor data set sample. Assume that a bridge is undergoing rotation construction, and sensors are installed at different parts of the bridge to continuously monitor various physical parameters during the rotation process, such as stress, displacement, tilt angle, etc. After receiving these sensor data, the monitoring system organizes them into real-time bridge rotation sensor data set samples. For example, assume that the real-time sensor data set sample of the current rotation operation contains the following groups of data points: pressure measurement values ​​at the center of the bridge deck, displacement measurement values ​​at both ends of the bridge body, and the tilt angle of the bridge body relative to the horizontal plane. In order to identify the unbalanced bending moment, the monitoring system first analyzes the previous sample of the real-time sensor data set sample. For example, assume that the real-time sensor data set sample of the current rotation operation is the nth sample, then the n-1th sample is the previous sample. This previous sample contains physical parameters similar to the current sample, but is collected at an earlier time point.

[0085] The monitoring system then performs unbalanced bending moment identification on the n-1th sample. This identification process is similar to the identification of samples in the real-time sensing data set, but it is based on the data in the previous sample. For example, the monitoring system will focus on key data points such as the displacement difference between the two ends of the bridge and the inclination angle of the bridge relative to the horizontal plane.

[0086] During this process, the monitoring system uses an unbalanced bending moment identification algorithm to analyze these key data points. The algorithm can be a pre-trained machine learning model, such as a support vector machine (SVM), random forest (RandomForest), or a deep learning model such as a convolutional neural network (CNN). Taking a convolutional neural network as an example, the model may have been trained with a large amount of historical data and can identify unbalanced bending moments in specific patterns. During the identification process, the monitoring system extracts features that are closely related to the unbalanced bending moment. For example, assuming that the displacement difference at both ends of the bridge is one of the key features, the monitoring system can calculate the displacement difference at both ends of the bridge and use it as one of the input features. These features are organized into a feature vector, such as ,in Represents the difference in displacement between the two ends of the bridge. is the inclination angle of the bridge relative to the horizontal plane, is the pressure measurement at the center of the bridge deck.

[0087] Next, the monitoring system inputs the feature vector into the unbalanced bending moment identification algorithm. For example, for a convolutional neural network model, it may contain multiple convolutional layers, pooling layers, and fully connected layers. The model processes the feature vector through these layers and finally outputs the classification result of the unbalanced bending moment. The output of the model can be a vector representing the probability distribution of different unbalanced bending moment levels. For example, the output vector may be [0.1, 0.6, 0.3], which means that the probability of slight unbalance is 10%, the probability of moderate unbalance is 60%, and the probability of severe unbalance is 30%. The monitoring system determines the level of unbalanced bending moment based on these probability distributions. For example, if the probability of moderate unbalance is the highest, then the system determines that the current level of unbalanced bending moment is moderate. Finally, the monitoring system determines the second main data item cluster, which contains the data items most related to the unbalanced bending moment. For example, if the displacement difference between the two ends of the bridge is the main factor of the unbalanced bending moment, then this difference value will become part of the second main data item cluster. Assuming that in the previous rotation operation, the difference in displacement between the two ends of the bridge is 4 cm, and the inclination angle of the bridge relative to the horizontal plane is 1.5°, then the second main data item cluster can be expressed as In this way, the monitoring system not only identifies the level of unbalanced bending moment in the previous sample, but also determines the key data item clusters that are most relevant to it. This information will be used in the subsequent unbalanced bending moment classification and algorithm optimization process, helping the system to better understand the changing trend of unbalanced bending moment and make more accurate predictions in real-time monitoring.

[0088] Step S12023: Determine the real-time characterization vector based on the first main data item cluster and the second main data item cluster, wherein the real-time characterization vector characterizes the degree of evolution between corresponding main data items in the first main data item cluster and the second main data item cluster.

[0089] In step S12023, the monitoring system determines a real-time characterization vector based on the first main data item cluster and the second main data item cluster. This real-time characterization vector reflects the degree of evolution between the corresponding main data items in the first main data item cluster and the second main data item cluster. Assume that a bridge is undergoing rotation construction, and sensors are installed at different parts of the bridge to continuously monitor various physical parameters during the rotation process, such as stress, displacement, tilt angle, etc. After receiving these sensor data, the monitoring system organizes them into real-time bridge rotation sensor data set samples. For example, assume that the real-time sensor data set sample of the current rotation operation contains the following groups of data points: pressure measurement values ​​at the center of the bridge deck, displacement measurement values ​​at both ends of the bridge body, and the tilt angle of the bridge body relative to the horizontal plane. In order to identify unbalanced bending moments, the monitoring system first analyzes the previous sample of the real-time sensor data set sample. For example, assume that the real-time sensor data set sample of the current rotation operation is the nth sample, then the n-1th sample is the previous sample. This previous sample contains physical parameters similar to the current sample, but is collected at an earlier time point.

[0090] The monitoring system has identified two main clusters of data items: the first main cluster of data items comes from the real-time sensor data set sample, and the second main cluster of data items comes from the previous sensor data set sample. These two clusters contain the data items that are most relevant to the unbalanced bending moment. For example, the first main cluster of data items can be expressed as , where 5cm is the difference in displacement between the two ends of the bridge, and 2° is the inclination angle of the bridge relative to the horizontal plane; the second main data item cluster can be expressed as , which also includes the difference in displacement between the two ends of the bridge and the inclination angle of the bridge relative to the horizontal plane.

[0091] Next, the monitoring system determines a real-time characterization vector based on the two main data item clusters. This vector is used to describe the changes in the corresponding data items in the first main data item cluster and the second main data item cluster. For example, for the displacement difference, the change is , for the tilt angle, the change is 2° - 1.5° = 0.5°. To construct a real-time characterization vector, the monitoring system can use a simple mathematical formula to quantify this change. For example, the Euclidean distance can be used to measure the difference between two main data item clusters. Assume and are the first and second main data item clusters respectively, then the real-time representation vector It can be calculated by the following formula:

[0092] ;

[0093] Taking displacement difference and tilt angle as examples, real-time representation of vector v real-time It can be expressed as: ;

[0094] This vector shows that in the most recent rotation operation, the displacement difference at both ends of the bridge increased by 1 cm, and the inclination angle of the bridge relative to the horizontal plane increased by 0.5°. Such change information is very important for evaluating the changing trend of the unbalanced bending moment. In this way, the monitoring system can not only identify the level of unbalanced bending moment, but also capture the changing trend of the unbalanced bending moment over time, which is crucial for real-time monitoring and prediction of unbalanced bending moments during bridge rotation. The real-time characterization vector will be used as input to the target recognition algorithm in subsequent steps to help optimize the algorithm and improve the accuracy of unbalanced bending moment identification.

[0095] In some feasible implementation schemes, the unbalanced bending moment identification result and feedback information corresponding to each sample characterization vector in the multiple sample characterization vectors may be determined respectively according to the following steps, wherein the sample characterization vector for determining the unbalanced bending moment identification result and feedback information each time is regarded as a real-time sample characterization vector, and the unbalanced bending moment identification result corresponding to the real-time sample characterization vector is regarded as a real-time unbalanced bending moment identification result and real-time feedback information:

[0096] Step S11: inputting the real-time sample characterization vector into the basic recognition algorithm to obtain the real-time unbalanced bending moment recognition result, wherein the basic recognition algorithm is a recognition algorithm that has completed initialization;

[0097] Step S12: Determine the real-time feedback information based on the real-time unbalanced bending moment identification result and the supervisory information corresponding to the real-time sample characterization vector, the supervisory information includes the actual identification information in the bridge rotation sensor data set sample corresponding to the real-time sample characterization vector, the actual identification information includes the target unbalanced bending moment classification of the target data, and the feedback information is used to reflect whether the real-time unbalanced bending moment identification result is the same or different from the target unbalanced bending moment classification.

[0098] In this embodiment, the monitoring system determines the unbalanced bending moment identification result and feedback information corresponding to each sample characterization vector in the plurality of sample characterization vectors according to specific steps. These steps are intended to help the monitoring system generate the unbalanced bending moment identification result based on the input sample characterization vector, and determine the feedback information by comparing the identification result with the actual label, so as to evaluate the accuracy of the identification result.

[0099] In this step, the monitoring system inputs the real-time sample representation vector into the initialized basic recognition algorithm to obtain the real-time unbalanced bending moment recognition result. The basic recognition algorithm here can be a pre-trained machine learning model, such as support vector machine (SVM), decision tree, random forest or deep learning model (such as convolutional neural network CNN).

[0100] Assume that a sample representation vector has been extracted from the bridge rotation sensor data set sample, and the currently processed sample representation vector is regarded as a real-time sample representation vector. The monitoring system inputs this real-time sample representation vector into the initialized basic recognition algorithm to obtain the real-time unbalanced bending moment recognition result. As a specific example, assume that an unbalanced bending moment recognition model based on a convolutional neural network (CNN) is used. The output of the model is a vector representing the probability distribution of each unbalanced bending moment category. For example, for the real-time sample representation vector , the output of the model can be expressed as: This means that the real-time unbalanced bending moment identification result is “normal” (corresponding to a probability of 0.8), and the probabilities of the other categories are “slightly abnormal” (0.15) and “severely abnormal” (0.05), respectively.

[0101] In step S12, the monitoring system determines the real-time feedback information based on the real-time unbalanced bending moment recognition result and the supervision information corresponding to the real-time sample characterization vector. The supervision information here refers to the accurate label in advance, that is, the known actual recognition information, including the target unbalanced bending moment classification of the target data. By comparing the recognition result with the actual label, the real-time feedback information can be determined to reflect whether the recognition result is correct.

[0102] Assume that there is a real-time sample characterization vector, and the monitoring system has obtained the real-time unbalanced bending moment recognition result. At the same time, there is a prior accurate label, namely the supervision information, which represents the actual recognition information in the data set. For example, assuming that the actual recognition information is "slightly abnormal", the monitoring system will compare the real-time unbalanced bending moment recognition result with the actual recognition information to determine the feedback information. For example, assume that the real-time unbalanced bending moment recognition result is "normal", and the actual recognition information is "slightly abnormal". In this case, the feedback information will indicate that the recognition result is incorrect. The feedback information can be expressed by a simple logical expression. If the recognition result is the same as the actual label, the feedback information is "correct"; if not, the feedback information is "wrong". For example, for the above example, the feedback information can be expressed as: ;

[0103] To better understand how to determine the feedback information, a simple logical expression can be used. Assume that R represents the real-time unbalanced bending moment identification result and L represents the actual identification information (supervision information). The feedback information F can be determined by the following logical expression:

[0104] ;

[0105] Through steps S11 and S12, the monitoring system can effectively use the initialized basic recognition algorithm to generate unbalanced bending moment recognition results, and determine feedback information by comparing the recognition results with the actual labels. These steps ensure the accuracy and reliability of the recognition results, and also provide important feedback for subsequent adjustment and optimization of the recognition algorithm. In this way, the monitoring system can more accurately identify the unbalanced bending moment problem during the bridge rotation process, thereby improving the performance and safety of the monitoring system.

[0106] In some feasible implementation schemes, the above step S11, inputting the real-time sample characterization vector into the basic recognition algorithm to obtain the real-time unbalanced bending moment recognition result, can be implemented by the following steps:

[0107] Step S111: inputting each unbalanced bending moment classification in the classification set into a first classification association function in sequence together with the real-time sample characterization vector to obtain a first association result cluster, wherein the first association result cluster includes association results corresponding to each unbalanced bending moment classification;

[0108] Step S112: classifying the candidate unbalanced bending moment corresponding to the first correlation result corresponding to the maximum result in the first correlation result cluster as the real-time unbalanced bending moment identification result.

[0109] In some feasible implementations, step S11 is subdivided into two sub-steps S111 and S112. These two steps together constitute a process of inputting the real-time sample characterization vector into the basic recognition algorithm and obtaining the real-time unbalanced bending moment recognition result.

[0110] In step S111, the monitoring system first inputs each unbalanced bending moment classification in the classification set into the first classification association function in sequence together with the real-time sample characterization vector. The first classification association function is used to calculate the correlation between each unbalanced bending moment classification and the real-time sample characterization vector, and the output result is a group of correlation results, namely, the first correlation result cluster. Each correlation result corresponds to an unbalanced bending moment classification.

[0111] Assume that the classification set contains three unbalanced bending moment classifications: "normal", "slightly abnormal", and "severely abnormal". The real-time sample representation vector is a vector representation extracted from the bridge rotation sensor data set. The monitoring system inputs each unbalanced bending moment classification in the classification set in sequence together with the real-time sample representation vector into the first classification association function. The first classification association function can be a pre-trained machine learning model, such as a support vector machine (SVM), a decision tree, a random forest, or a deep learning model (such as a convolutional neural network CNN).

[0112] As a specific example, assume that the first classification association function is a trained multi-layer perceptron (MLP), whose output is a vector representing the probability distribution of each unbalanced bending moment category. , the output of the model can be expressed as: This means that for the three unbalanced bending moment categories in the classification set, “normal” has the highest association (corresponding to a probability of 0.8), followed by “slightly abnormal” (0.15), and finally “severely abnormal” (0.05).

[0113] In step S112, the monitoring system selects the maximum result in the first association result cluster, and uses the unbalanced bending moment classification corresponding to the first association result corresponding to the maximum result as the real-time unbalanced bending moment identification result. Continuing with the above example, the monitoring system selects the unbalanced bending moment classification corresponding to the maximum result from the first association result cluster. In this example, the unbalanced bending moment classification corresponding to the maximum result is "normal".

[0114] The monitoring system selects the largest result by comparing all the results in the first associated result cluster. For example, in the output vector Among them, the “Normal” category has the highest correlation and is therefore selected as the real-time unbalanced bending moment identification result.

[0115] To better understand how to determine the real-time unbalanced bending moment identification results, a simple logical expression can be used. Assume that O represents the first associated result cluster ,in The correlation results corresponding to the three unbalanced bending moment classifications of "normal", "slightly abnormal" and "severely abnormal" respectively. The real-time unbalanced bending moment identification result R can be determined by the following logical expression:

[0116] ;

[0117] To further illustrate the calculation process, a specific calculation formula can be used to show how to determine the maximum result in the first association result cluster. Assume that O is a vector containing three elements, representing the association results of "normal", "slightly abnormal" and "severely abnormal". Then, the maximum result can be expressed as:

[0118] ;

[0119] in In the example above, O = [0.8, 0.15, 0.05], so the maximum result is: max(O) = max(0.8, 0.15, 0.05) = 0.8.

[0120] Through steps S111 and S112, the monitoring system can effectively use the first classification association function to calculate the correlation between each unbalanced bending moment classification and the real-time sample characterization vector, and determine the real-time unbalanced bending moment identification result based on these correlations. This method not only ensures the accuracy and reliability of the identification result, but also provides an important basis for determining the feedback information in the subsequent steps. In this way, the monitoring system can more accurately identify the unbalanced bending moment problem during the bridge rotation process, thereby improving the performance and safety of the monitoring system.

[0121] In some feasible implementation schemes, in step S111, that is, inputting each unbalanced bending moment classification in the classification set in sequence together with the real-time sample characterization vector into the first classification association function to obtain the first association result cluster, the method further includes the following steps:

[0122] Step S101: Determine target attempt probability;

[0123] Step S102: determining whether the target attempt probability exceeds a probability threshold;

[0124] Step S103: when the target attempt probability is greater than the probability threshold, randomly obtaining an alternative unbalanced bending moment classification in the classification set as the real-time unbalanced bending moment identification result;

[0125] Step S104: When the target attempt probability is not greater than the probability threshold, the real-time sample characterization vector is input into the basic recognition algorithm to obtain the real-time unbalanced bending moment recognition result.

[0126] In some feasible implementations, the method before step S111 further includes additional steps S101 to S104. These steps introduce a probability-based mechanism to determine whether to directly use the basic identification algorithm for unbalanced bending moment identification or to randomly select an unbalanced bending moment classification as the identification result. This mechanism increases the flexibility of the system and helps to reduce the computational burden in some cases.

[0127] In step S101, the monitoring system first determines a target attempt probability. This probability is a threshold used to determine whether to directly use the basic recognition algorithm for unbalanced bending moment recognition. If the target attempt probability exceeds the set probability threshold, the system will randomly select an unbalanced bending moment classification as the recognition result; otherwise, the basic recognition algorithm will be used for recognition.

[0128] The purpose of setting the trial probability is to introduce a probabilistic decision-making mechanism to enhance the adaptability and efficiency of the system. In the scenario of real-time monitoring of unbalanced bending moments, this mechanism can help the system make more reasonable decisions under different conditions. In some cases, directly using the basic recognition algorithm can consume a lot of computing resources. This consumption is more significant when the algorithm is complex and processes a large amount of data. By introducing probabilistic decision-making, complex recognition algorithms can be avoided in some cases, thereby saving computing resources. Real-time monitoring systems usually maintain a high processing speed while ensuring accuracy. Setting the trial probability can be used as a trade-off strategy, allowing the system to adopt a simpler decision path under high load or low risk conditions, and a more accurate algorithm in other cases. During the rotation of the bridge, changes in external environmental factors (such as weather conditions, traffic conditions, etc.) can cause changes in the probability of occurrence of unbalanced bending moments. Setting the trial probability can adjust the recognition strategy according to the current environmental conditions to adapt to these changes. If the basic recognition algorithm is overfitted during the training process, then incorrect recognition may occur when processing new data. By randomly selecting classifications, the impact of overfitting can be alleviated to a certain extent.

[0129] The acquisition of the attempt probability can be based on a variety of different strategies and methods. Here are a few possible ways:

[0130] 1. Static setting. The simplest way is to artificially set a fixed trial probability value during the system design phase. For example, if it is found through analysis that the probability of occurrence of unbalanced bending moment is low in most cases, the trial probability can be set to a smaller value.

[0131] 2. Dynamic adjustment. Based on real-time monitoring data and historical records, the monitoring system can dynamically adjust the trial probability. For example, if there are many unbalanced bending moment events detected recently, the system can automatically increase the trial probability to increase the chance of directly using the basic identification algorithm.

[0132] 3. Prediction based on machine learning. The monitoring system can use historical data to train a machine learning model to predict the probability of an attempt. This model can consider multiple features, such as time, weather conditions, past test results, etc., to predict the probability of unbalanced bending moment at the current moment and determine the probability of an attempt accordingly.

[0133] In one example, assume that a machine learning-based method is used to predict the probability of an attempt. To this end, the monitoring system collects a series of data related to the bridge rotation, including but not limited to: time features, environmental features, traffic conditions, historical detection results, etc. Assume that a machine learning-based model has been trained, which can receive the above features as input and output a probability value, namely the target attempt probability. This probability value represents the necessity of directly using the basic recognition algorithm for recognition under given conditions.

[0134] In an example of a training process, suppose there is a feature vector Represents the current observation value, such as: ;

[0135] Among them: 14 represents the number of hours in a day; 3 represents the number of days in a week (Monday); 20 represents temperature; 45 represents humidity; 10 represents wind speed; 15 represents traffic volume; 0.2 represents the probability of sunny days; 0.8 represents the probability of cloudy days; 0 represents the probability of rainy days; 1 represents the probability of morning days; 0 represents the probability of afternoon days; 0 represents the probability of evening days; 2 represents the number of minor abnormal events detected in the past week; 5 represents the number of major abnormal events detected in the past week.

[0136] Based on the above feature vector x, the hypothetical machine learning model outputs a trial probability value of P = 0.3. This means that under this specific condition, the monitoring system has a 30% probability of directly using the basic identification algorithm to identify the unbalanced bending moment, and a 70% probability of randomly selecting an unbalanced bending moment classification.

[0137] Assume that the monitoring system receives a real-time sample characterization vector and predicts that the attempt probability is 0.3 based on the current environmental conditions. If the current target attempt probability is exactly 0.35, then according to the logic of step S102, the monitoring system will enter step S103 and randomly select an unbalanced bending moment classification as the real-time unbalanced bending moment identification result. In this way, the monitoring system can flexibly adjust its behavior according to the actual situation, thereby achieving the purpose of ensuring both accuracy and efficiency.

[0138] In step S102, the monitoring system determines whether the target attempt probability exceeds the set probability threshold. If the target attempt probability is greater than the probability threshold, it proceeds to step S103; otherwise, it proceeds to step S104. Continuing with the above example, assume that the probability threshold is 0.3 and the target attempt probability is 0.25. Because 0.25 is not greater than 0.3, the monitoring system will proceed to step S104.

[0139] In step S103, if the target attempt probability is greater than the probability threshold, the monitoring system will randomly select an unbalanced bending moment classification from the classification set as the real-time unbalanced bending moment identification result. Assume that the probability threshold is 0.3 and the target attempt probability is 0.35. Since 0.35 is greater than 0.3, the monitoring system will randomly select an unbalanced bending moment classification from the classification set as the real-time unbalanced bending moment identification result. Assume that the classification set contains three unbalanced bending moment classifications: "normal", "slightly abnormal", and "severely abnormal".

[0140] The monitoring system can use a random number generator to select an unbalanced bending moment classification from the classification set. For example, assuming that the result of the random selection is "slightly abnormal", "slightly abnormal" will become the real-time unbalanced bending moment identification result.

[0141] In step S104, if the target attempt probability is not greater than the probability threshold, the monitoring system will use the basic recognition algorithm to obtain the real-time unbalanced bending moment recognition result. This step is similar to the previous steps S111 to S112, but it is emphasized here that the basic recognition algorithm will only be used when the target attempt probability is not greater than the probability threshold.

[0142] Assume that the probability threshold is 0.3 and the target attempt probability is 0.25. Since 0.25 is not greater than 0.3, the monitoring system will use the basic recognition algorithm to obtain the real-time unbalanced bending moment recognition result. For example, assuming that the real-time sample characterization vector for: The monitoring system inputs this real-time sample characterization vector into the initialized basic identification algorithm to obtain the real-time unbalanced bending moment identification result.

[0143] As a specific example, assume that a convolutional neural network (CNN)-based unbalanced bending moment recognition model is used. The output of the model is a vector representing the probability distribution of each unbalanced bending moment category. For example, for the real-time sample representation vector, the output of the model can be expressed as: [0.8, 0.15, 0.05]. This means that the real-time unbalanced bending moment recognition result is "normal" (the corresponding probability is 0.8), and the probabilities of the other categories are "slightly abnormal" (0.15) and "severely abnormal" (0.05).

[0144] To better understand how to determine the recognition path, a simple logical expression can be used. Assume that P represents the target attempt probability and T represents the probability threshold. The recognition path R can be determined by the following logical expression:

[0145] ;

[0146] Through steps S101 to S104, the monitoring system can flexibly decide to a certain extent whether to directly use the basic recognition algorithm for unbalanced bending moment recognition, or to randomly select an unbalanced bending moment classification as the recognition result. This method not only ensures the accuracy and reliability of the recognition result, but also provides an important basis for determining the feedback information in the subsequent steps. In this way, the monitoring system can more accurately identify the unbalanced bending moment problem during the bridge rotation process, thereby improving the performance and safety of the monitoring system.

[0147] In some feasible implementation schemes, the step S112, i.e., classifying the candidate unbalanced bending moment corresponding to the first correlation result corresponding to the maximum result in the first correlation result cluster as the real-time unbalanced bending moment identification result, can be implemented by the following steps:

[0148] Step S1121: if the candidate unbalanced bending moment classification corresponding to the first association result is the same as the target unbalanced bending moment classification, the real-time feedback information is determined as the first feedback information, and the first feedback information represents that the basic recognition algorithm has correctly identified the problem;

[0149] Step S1122: If the candidate unbalanced bending moment classification corresponding to the first association result is different from the target unbalanced bending moment classification, the real-time feedback information is determined as second feedback information, and the second feedback information represents that the basic recognition algorithm has made an error in recognition.

[0150] In some feasible implementations, step S112 is subdivided into two sub-steps S1121 and S1122. These two steps together constitute a process of determining real-time feedback information to reflect the accuracy of the recognition result of the basic recognition algorithm.

[0151] In step S1121, if the candidate unbalanced bending moment classification corresponding to the first association result is the same as the target unbalanced bending moment classification, the monitoring system determines the real-time feedback information as the first feedback information. The target unbalanced bending moment classification here refers to the accurate label in advance, that is, the actual identification information contained in the supervision information. The first feedback information represents that the basic recognition algorithm has correctly identified.

[0152] Assume that the classification set contains three unbalanced bending moment classifications: "normal", "slightly abnormal", and "severely abnormal". The real-time sample representation vector is a vector representation extracted from the bridge rotation sensor dataset, for example: The monitoring system uses the first classification association function (such as a trained multi-layer perceptron MLP) to process the real-time sample representation vector, and the output result is a vector representing the probability distribution of each unbalanced bending moment category. For example, for the real-time sample representation vector , the output of the model can be expressed as: . This means that for the three unbalanced bending moment classifications in the classification set, "normal" has the highest correlation (corresponding to a probability of 0.8), followed by "slightly abnormal" (0.15), and finally "severely abnormal" (0.05). Assume that the target unbalanced bending moment classification provided in the supervision information is "normal". In this case, since the alternative unbalanced bending moment classification corresponding to the first correlation result is also "normal", the monitoring system determines the real-time feedback information as the first feedback information, that is, the basic recognition algorithm recognizes it correctly.

[0153] In step S1122, if the candidate unbalanced bending moment classification corresponding to the first association result is different from the target unbalanced bending moment classification, the monitoring system determines the real-time feedback information as the second feedback information. The target unbalanced bending moment classification here refers to the accurate label in advance, that is, the actual identification information contained in the supervision information. The second feedback information represents that the basic recognition algorithm has an error in recognition.

[0154] Continuing with the previous example, assume that the target unbalanced bending moment classification provided in the supervision information is "slightly abnormal". In this case, since the alternative unbalanced bending moment classification corresponding to the first association result is "normal", which is different from the target unbalanced bending moment classification, the monitoring system determines the real-time feedback information as the second feedback information, that is, the basic recognition algorithm is incorrectly identified.

[0155] To better understand how to determine the feedback information, a simple logical expression can be used. Assume that R represents the real-time unbalanced bending moment identification result and L represents the target unbalanced bending moment classification. The feedback information F can be determined by the following logical expression:

[0156] ;

[0157] Assume that the first association result cluster has been obtained from the real-time sample characterization vector, wherein the first association result corresponding to the maximum result is "normal", that is, R = normal. Assume that the target unbalanced bending moment is classified as "normal", that is, L = normal.

[0158] Since R=L, the monitoring system determines the real-time feedback information as the first feedback information, that is, the basic recognition algorithm is correct. Assume that the target unbalanced bending moment classification is changed to "slightly abnormal", that is, L="slightly abnormal". R≠L, the monitoring system determines the real-time feedback information as the second feedback information, that is, the basic recognition algorithm is incorrect.

[0159] Through steps S1121 and S1122, the monitoring system can effectively determine the real-time feedback information based on the consistency between the real-time unbalanced bending moment identification result and the target unbalanced bending moment classification. This method not only ensures the accuracy and reliability of the identification result, but also provides an important basis for the determination of feedback information in subsequent steps. In this way, the monitoring system can more accurately identify the unbalanced bending moment problem during the bridge rotation process, thereby improving the performance and safety of the monitoring system.

[0160] In some feasible implementations, the method further includes:

[0161] Step S210: determining a plurality of data component sets from the knowledge sample, wherein an m-th data component set in the plurality of data component sets includes an m-th sample characterization vector, an m-th unbalanced bending moment identification result corresponding to the m-th sample characterization vector, an m-th feedback information, and an m+1-th sample characterization vector, where m is a natural number greater than 0;

[0162] Step S220: Optimize the basic recognition algorithm to be optimized based on the multiple data component sets to obtain the target recognition algorithm. When the optimized state of the basic recognition algorithm meets the preset convergence condition, determine the basic recognition algorithm as the target recognition algorithm. When the optimized state of the basic recognition algorithm does not meet the preset convergence condition, optimize the algorithm parameters of the basic recognition algorithm based on a preset cost function. The execution data of each optimization is a data component set among the multiple data component sets.

[0163] In some feasible implementations, in addition to the above steps, steps S210 and S220 are also included. These two steps together constitute a process of optimizing the basic recognition algorithm based on the collected knowledge samples to improve the accuracy and stability of the algorithm.

[0164] In step S210, the monitoring system determines multiple data component sets from the knowledge sample. Each data component set includes four components: a sample characterization vector, a corresponding unbalanced bending moment identification result, feedback information, and a next sample characterization vector. These data component sets are used in the subsequent algorithm optimization process.

[0165] Assume that data has been collected for a period of time, including multiple sample characterization vectors and their corresponding unbalanced bending moment identification results and feedback information. For example, the following data sets are composed:

[0166] The first data set consists of:

[0167] The first sample representation vector: final_representation_vector1=[0.8, -0.2, 0.5, -0.3,-0.2];

[0168] The first unbalanced bending moment identification result: "Normal";

[0169] The first feedback information: "First feedback information";

[0170] The second sample representation vector: final_representation_vector2=[0.7, -0.1, 0.6, -0.2,-0.1];

[0171] The second data set consists of:

[0172] The second sample representation vector: final_representation_vector2=[0.7, -0.1, 0.6, -0.2, -0.1];

[0173] The second unbalanced bending moment identification result: "slightly abnormal";

[0174] The second feedback information: "Second feedback information";

[0175] The third sample representation vector: final_representation_vector3=[0.6, -0.2, 0.7, -0.1,-0.2];

[0176] In step S220, the monitoring system optimizes the basic recognition algorithm to be optimized based on multiple data component sets. The goal of the optimization is to improve the accuracy and stability of the algorithm. If the optimization process meets the preset convergence condition, the optimized basic recognition algorithm is determined as the target recognition algorithm; otherwise, the optimization is continued until the convergence condition is met. It is assumed that a series of data component sets have been collected and it is desired to use these data to optimize the basic recognition algorithm. The optimization process can use a gradient descent method or other optimization algorithms, such as an Adam optimizer.

[0177] For example, the optimization process could be:

[0178] 1. The monitoring system initializes the parameters of the basic recognition algorithm and sets the initial learning rate and other hyperparameters of the optimization algorithm.

[0179] 2. For each data set, the monitoring system uses the basic recognition algorithm to predict the unbalanced bending moment recognition result and compares it with the actual unbalanced bending moment recognition result. The loss function (such as the cross entropy loss function) is calculated based on the feedback information, and the algorithm parameters are updated to minimize the loss function.

[0180] 3. At the end of each optimization cycle, check whether the convergence condition is met (for example, the change in the loss function is less than a certain threshold or the predetermined number of iterations is reached). If the convergence condition is met, the currently optimized basic recognition algorithm is determined as the target recognition algorithm. If the convergence condition is not met, continue to optimize.

[0181] 4. If the convergence conditions are not met, the monitoring system can adjust the parameters of the optimization algorithm (such as learning rate) and continue the optimization.

[0182] To better understand the optimization process, we can use the gradient descent method as an example of an optimization algorithm. The goal of the gradient descent method is to minimize the loss function, which can be a cross entropy loss function. ,in is the mth sample representation vector, is the corresponding unbalanced bending moment identification result, and the loss function It can be expressed as: ;

[0183] in, Indicates that the basic recognition algorithm is in parameters Next prediction The probability of is a binary variable (0 or 1) indicating the actual unbalanced bending moment classification.

[0184] Assume that the basic identification algorithm is optimized using the gradient descent method. During the optimization process, the monitoring system continuously adjusts the algorithm parameters to minimize the loss function. For example, for the first data component set, if the prediction result of the basic identification algorithm is "normal" and the actual unbalanced bending moment identification result is also "normal", then based on the first feedback information, the monitoring system will adjust the algorithm parameters to maintain the accuracy of the prediction result. On the contrary, if the prediction result is different from the actual result, then based on the second feedback information, the monitoring system will adjust the algorithm parameters to correct the prediction error.

[0185] Through steps S210 and S220, the monitoring system can effectively use the collected knowledge samples to optimize the basic recognition algorithm to improve the accuracy and stability of the algorithm. This method not only ensures the accuracy and reliability of the recognition results, but also provides an important basis for determining the feedback information in the subsequent steps. In this way, the monitoring system can more accurately identify the unbalanced bending moment problem during the bridge rotation process, thereby improving the performance and safety of the monitoring system.

[0186] In some feasible implementation schemes, the target recognition algorithm can be obtained by optimizing the basic recognition algorithm to be optimized based on multiple data component sets according to the following steps. The data component set input into the basic recognition algorithm is the mth data component set:

[0187] Step S221: determining whether the bridge rotation sensor data set sample corresponding to the m+1th sample characterization vector is the penultimate bridge rotation sensor data set sample in the sample bridge rotation sensor data stream;

[0188] Step S222: if the bridge rotation sensor data set sample corresponding to the m+1th sample characterization vector is the penultimate bridge rotation sensor data set sample, determining the setting parameter output by the basic recognition algorithm based on the mth feedback information;

[0189] Step S223: if the bridge rotation sensor data set sample corresponding to the m+1th sample characterization vector is not the penultimate bridge rotation sensor data set sample, then determining the set parameter based on the mth feedback information and the output of the comparison recognition algorithm, the comparison recognition algorithm is an initialization recognition algorithm, and the algorithm parameters of the comparison recognition algorithm and the basic recognition algorithm are different;

[0190] Step S224: determining a cost value of the cost function based on the set parameters and the output of the basic recognition algorithm, and optimizing algorithm parameters of the basic recognition algorithm based on the cost value of the cost function;

[0191] Step S225: If the basic recognition algorithm reaches a convergence state, the basic recognition algorithm is used as the target recognition algorithm.

[0192] In some feasible implementations, the monitoring system optimizes the basic recognition algorithm to be optimized based on multiple data sets in accordance with specific steps to obtain a target recognition algorithm. These steps are intended to ensure that the algorithm can be dynamically adjusted according to real-time data to improve the accuracy and stability of unbalanced bending moment recognition.

[0193] In step S221, the monitoring system first determines whether the bridge rotation sensor data set sample corresponding to the m+1th sample characterization vector is the last bridge rotation sensor data set sample in the sample bridge rotation sensor data stream, in order to determine whether to use the contrast recognition algorithm for optimization.

[0194] Assume there are three sample representation vectors , and , and the corresponding unbalanced bending moment identification results and feedback information. Among them, The corresponding bridge rotation sensor data set sample will be the penultimate sample in the sample bridge rotation sensor data stream.

[0195] In step S222, if the bridge rotation sensor data set sample corresponding to the m+1th sample characterization vector is the penultimate bridge rotation sensor data set sample, then the monitoring system will determine the setting parameters of the basic recognition algorithm output based on the mth feedback information.

[0196] Assume that the bridge rotation sensor data set sample corresponding to the m+1th sample representation vector is the last bridge rotation sensor data set sample, and the monitoring system only relies on the mth feedback information to determine the setting parameters of the basic recognition algorithm output. For example, if the mth feedback information is the first feedback information, it means that the basic recognition algorithm recognizes correctly, then the setting parameters may remain unchanged or slightly adjusted to maintain the current good performance.

[0197] In step S223, if the bridge rotation sensor data set sample corresponding to the m+1th sample characterization vector is not the last bridge rotation sensor data set sample, then the monitoring system will determine the setting parameters based on the mth feedback information and the output of the comparison recognition algorithm. The comparison recognition algorithm here is another recognition algorithm that has completed initialization, and the algorithm parameters of the basic recognition algorithm are different.

[0198] Assuming that the bridge rotation sensor data set sample corresponding to the m+1th sample representation vector is not the last bridge rotation sensor data set sample, the monitoring system will consider the mth feedback information and the output of the comparative recognition algorithm. For example, if the mth feedback information is the second feedback information, it means that the basic recognition algorithm has an error in recognition, while the comparative recognition algorithm gives a correct recognition result, then the set parameters will be adjusted to be closer to the output of the comparative recognition algorithm.

[0199] In step S224 , the monitoring system determines a cost value of the cost function based on the set parameters and the output of the basic recognition algorithm, and optimizes the algorithm parameters of the basic recognition algorithm based on the cost value of the cost function.

[0200] Assume that the set parameters have been determined and the output of the basic recognition algorithm has been obtained. In order to optimize the basic recognition algorithm, a cost function is defined, such as a cross entropy loss function. Assume that the unbalanced bending moment recognition result corresponding to the mth sample representation vector is , the probability output by the basic recognition algorithm is ,in is the mth sample representation vector, is the parameter of the basic recognition algorithm. Cost function It can be expressed as: ;

[0201] In order to minimize the cost function , the monitoring system can use gradient descent to update parameters .

[0202] In step S225, the monitoring system checks whether the basic recognition algorithm has reached a convergence state. If so, the current basic recognition algorithm is determined as the target recognition algorithm.

[0203] Assuming that the monitoring system has been optimized for multiple iterations, and has detected that the loss function of the basic recognition algorithm has changed very little and has reached a convergence state, the monitoring system will determine the current basic recognition algorithm as the target recognition algorithm.

[0204] Assume that the basic identification algorithm is optimized using the gradient descent method. During the optimization process, the monitoring system continuously adjusts the algorithm parameters to minimize the loss function. For example, for the first data component set, if the prediction result of the basic identification algorithm is "normal" and the actual unbalanced bending moment identification result is also "normal", then based on the first feedback information, the monitoring system will adjust the algorithm parameters to maintain the accuracy of the prediction result. On the contrary, if the prediction result is different from the actual result, then based on the second feedback information, the monitoring system will adjust the algorithm parameters to correct the prediction error.

[0205] Through steps S221 to S225, the monitoring system can effectively use the collected knowledge samples to optimize the basic recognition algorithm to improve the accuracy and stability of the algorithm. This method not only ensures the accuracy and reliability of the recognition results, but also provides an important basis for determining the feedback information in the subsequent steps. In this way, the monitoring system can more accurately identify the unbalanced bending moment problem during the bridge rotation process, thereby improving the performance and safety of the monitoring system.

[0206] In some feasible implementation schemes, in step S223, determining the setting parameter based on the mth feedback information and the output of the comparison recognition algorithm can be implemented by the following steps:

[0207] Step S2231: inputting each unbalanced bending moment classification in the classification set and the m+1th sample characterization vector into a second classification association function in sequence to obtain a second association result cluster, wherein the second association result cluster includes association results corresponding to each unbalanced bending moment classification;

[0208] Step S2232: taking the sum of the second association result corresponding to the maximum result in the second association result cluster and the mth feedback information as the set parameter.

[0209] In some feasible implementations, step S223 is subdivided into two sub-steps S2231 and S2232. These two steps together constitute a process of determining the setting parameter based on the mth feedback information and the output of the comparison recognition algorithm.

[0210] In step S2231, the monitoring system sequentially inputs each unbalanced bending moment classification in the classification set and the m+1th sample characterization vector into a second classification association function to obtain a second association result cluster. The second classification association function can be a pre-trained machine learning model for calculating the degree of association between each unbalanced bending moment classification and the m+1th sample characterization vector.

[0211] Assume that the classification set contains three unbalanced bending moment classifications: “normal”, “slightly abnormal”, and “severely abnormal”. The m+1th sample representation vector is a vector representation extracted from the bridge rotation sensor dataset, for example: .

[0212] The monitoring system inputs each unbalanced bending moment classification in the classification set in sequence together with the m+1th sample representation vector into the second classification association function. The second classification association function can be a pre-trained machine learning model, such as a support vector machine (SVM), a decision tree, a random forest, or a deep learning model (such as a convolutional neural network CNN).

[0213] As a specific example, assume that the second classification association function is a trained multi-layer perceptron (MLP), whose output is a vector representing the probability distribution of each unbalanced bending moment category. , the output of the model can be expressed as: This means that for the three unbalanced bending moment categories in the classification set, “normal” has the highest association (corresponding to a probability of 0.7), followed by “slightly abnormal” (0.2), and finally “severely abnormal” (0.1).

[0214] In step S2232, the monitoring system adds the second association result corresponding to the maximum result in the second association result cluster and the mth feedback information, and uses this as a set parameter. The set parameter is used to subsequently optimize the algorithm parameters of the basic recognition algorithm.

[0215] Continuing with the previous example, assume that the second association result corresponding to the maximum result of the second association result cluster is "normal", and the corresponding probability is 0.7. Assume that the mth feedback information is the first feedback information, that is, the basic recognition algorithm recognizes it correctly. To simplify the example, assume that the value of the first feedback information is represented by 1 (representing correct feedback) and the value of the second feedback information is represented by -1 (representing incorrect feedback).

[0216] The monitoring system adds the second correlation result (0.7) corresponding to the maximum result in the second correlation result cluster and the mth feedback information (1) to obtain a set parameter of 1.7.

[0217] To better understand how to determine the setting parameter, a simple logical expression can be used. Assume that A represents the second associated result corresponding to the maximum result in the second associated result cluster, and F represents the mth feedback information. The setting parameter S can be determined by the following logical expression: S = A + F.

[0218] Through steps S2231 and S2232, the monitoring system can effectively determine the set parameters based on the mth feedback information and the output of the comparison recognition algorithm. This method not only ensures the accuracy and reliability of the recognition results, but also provides an important basis for the determination of feedback information in subsequent steps. In this way, the monitoring system can more accurately identify the unbalanced bending moment problem during the bridge rotation process, thereby improving the performance and safety of the monitoring system.

[0219] An embodiment of the present invention provides a monitoring system, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and the processor implements part or all of the steps in the above method when executing the program.

[0220] The embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, some or all of the steps in the above method are implemented. The computer-readable storage medium may be transient or non-transient.

[0221] An embodiment of the present invention provides a computer program, including computer-readable codes. When the computer-readable codes are run in a computer device, a processor in the computer device executes some or all of the steps for implementing the above method.

[0222] An embodiment of the present invention provides a computer program product, the computer program product includes a non-transitory computer-readable storage medium storing a computer program, and when the computer program is read and executed by a computer, some or all of the steps in the above method are implemented. The computer program product can be implemented specifically by hardware, software or a combination thereof. In some embodiments, the computer program product is specifically embodied as a computer storage medium, and in other embodiments, the computer program product is specifically embodied as a software product, such as a software development kit (SDK), etc.

[0223] It should be noted here that the description of the various embodiments above tends to emphasize the differences between the various embodiments, and the same or similar aspects can be referenced to each other. The description of the above device, storage medium, computer program and computer program product embodiments is similar to the description of the above method embodiment, and has similar beneficial effects as the method embodiment. For technical details not disclosed in the embodiments of the device, storage medium, computer program and computer program product of the present invention, please refer to the description of the method embodiment of the present invention for understanding.

[0224] Figure 2 A hardware entity diagram of a monitoring system provided by an embodiment of the present invention, such as Figure 2 As shown, the hardware entity of the monitoring system 1000 includes: a processor 1001 and a memory 1002, wherein the memory 1002 stores a computer program that can be run on the processor 1001, and the processor 1001 implements the steps in the method of any of the above embodiments when executing the program.

[0225] The memory 1002 stores a computer program that can be run on the processor. The memory 1002 is configured to store instructions and applications executable by the processor 1001. It can also cache the data to be processed or processed by the processor 1001 and each module in the monitoring system 1000 (for example, image data, audio data, voice communication data, and video communication data). This can be achieved through flash memory (FLASH) or random access memory (Random Access Memory, RAM).

[0226] When the processor 1001 executes the program, the steps of any of the above-mentioned visualized bridge rotation unbalanced bending moment real-time monitoring methods are implemented. The processor 1001 generally controls the overall operation of the monitoring system 1000.

[0227] An embodiment of the present invention provides a computer storage medium storing one or more programs, which can be executed by one or more processors to implement the steps of the visual real-time monitoring method of unbalanced bending moment of a bridge rotation as in any of the above embodiments.

[0228] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, etc., various media that can store program codes.

[0229] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the relevant technology. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0230] The above description is only an implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A visual real-time monitoring method for unbalanced bending moment of bridge rotation, characterized in that: include: Acquire a bridge rotation sensor data stream to be identified, wherein the bridge rotation sensor data stream includes target data to be identified, and the bridge rotation sensor data stream is sequence data composed of multiple bridge rotation sensor data sets; Input the bridge rotation sensor data stream into a pre-trained target recognition algorithm to obtain a recognition result data stream, wherein the target recognition algorithm is a recognition algorithm obtained by optimizing the basic recognition algorithm to be optimized by inputting knowledge samples, the knowledge samples are samples constructed by extracting multiple sample characterization vectors from multiple bridge rotation sensor data set samples and unbalanced bending moment recognition results and feedback information corresponding to each sample characterization vector, the unbalanced bending moment recognition result represents an unbalanced bending moment classification of a sample characterization vector, the feedback information represents the benefit corresponding to the unbalanced bending moment classification, the multiple bridge rotation sensor data set samples are multiple data samples with a continuous time series relationship, and each sample characterization vector includes features extracted collaboratively based on a bridge rotation sensor data set sample and a previous bridge rotation sensor data set sample of the bridge rotation sensor data set sample; Determine, based on the recognition result data stream, a recognition result corresponding to each bridge rotation sensor data set in the bridge rotation sensor data stream, wherein the recognition result represents an unbalanced bending moment classification of the target data; According to the unbalanced bending moment classification of the target data, a corresponding visualization is performed in the preset bridge rotation digital twin.

2. The method according to claim 1, characterized in that The method further comprises: Performing first characterization information extraction on the plurality of bridge rotation sensor data set samples to obtain a first characterization vector cluster, wherein a characterization vector in the first characterization vector cluster corresponds to a bridge rotation sensor data set sample among the plurality of bridge rotation sensor data set samples; Performing second characterization information extraction on the multiple bridge rotation sensor data set samples to obtain a second characterization vector cluster, wherein a first characterization vector in the second characterization vector cluster corresponds to a first bridge rotation sensor data set sample in the bridge rotation sensor data set samples, a previous bridge rotation sensor data set of the first bridge rotation sensor data set sample is a second bridge rotation sensor data set sample, and the first characterization vector characterizes the association between the first bridge rotation sensor data set sample and the second bridge rotation sensor data set sample; The first characterization vector cluster and the second characterization vector cluster are respectively merged to obtain the multiple sample characterization vectors.

3. The method according to claim 2, characterized in that The step of extracting second characterization information from the plurality of bridge rotation sensor data set samples to obtain a second characterization vector cluster includes: Performing second characterization information extraction on the plurality of bridge rotation sensor data set samples according to the following steps to obtain a second characterization vector cluster, wherein each bridge rotation sensor data set sample subjected to the second characterization information extraction is regarded as a real-time bridge rotation sensor data set sample, and the obtained characterization vector is regarded as a real-time characterization vector, and the second characterization vector cluster includes the real-time characterization vector; Perform unbalanced bending moment identification on the real-time bridge rotation sensor data set sample to determine a first main data item cluster, wherein the first main data item cluster is used to locate the position of the target data in the real-time bridge rotation sensor data set sample; Perform unbalanced bending moment identification on a previous bridge rotation sensor data set sample of the real-time bridge rotation sensor data set sample to determine a second main data item cluster, wherein the second main data item cluster is used to locate the position of the target data in the previous bridge rotation sensor data set sample; The real-time characterization vector is determined based on the first main data item cluster and the second main data item cluster, and the real-time characterization vector characterizes the evolution degree between the corresponding main data items in the first main data item cluster and the second main data item cluster.

4. The method according to claim 1, characterized in that The method further comprises: Determine the unbalanced bending moment identification result and feedback information corresponding to each sample characterization vector in the plurality of sample characterization vectors respectively according to the following steps, wherein the sample characterization vector for determining the unbalanced bending moment identification result and feedback information each time is regarded as a real-time sample characterization vector, and the unbalanced bending moment identification result corresponding to the real-time sample characterization vector is regarded as a real-time unbalanced bending moment identification result and real-time feedback information: Inputting the real-time sample characterization vector into the basic recognition algorithm to obtain the real-time unbalanced bending moment recognition result, wherein the basic recognition algorithm is a recognition algorithm that completes initialization; The real-time feedback information is determined based on the real-time unbalanced bending moment identification result and the supervisory information corresponding to the real-time sample characterization vector, the supervisory information includes the actual identification information in the bridge rotation sensor data set sample corresponding to the real-time sample characterization vector, the actual identification information includes the target unbalanced bending moment classification of the target data, and the feedback information is used to reflect whether the real-time unbalanced bending moment identification result is the same or different from the target unbalanced bending moment classification.

5. The method according to claim 4, characterized in that The step of inputting the real-time sample characterization vector into the basic recognition algorithm to obtain the real-time unbalanced bending moment recognition result includes: Inputting each unbalanced bending moment classification in the classification set in sequence together with the real-time sample characterization vector into a first classification association function to obtain a first association result cluster, wherein the first association result cluster includes association results corresponding to each unbalanced bending moment classification; The candidate unbalanced bending moment classification corresponding to the first correlation result corresponding to the maximum result in the first correlation result cluster is used as the real-time unbalanced bending moment identification result.

6. The method according to claim 5, characterized in that Before inputting each unbalanced bending moment classification in the classification set in sequence together with the real-time sample characterization vector into the first classification association function to obtain the first association result cluster, it includes: Determine target attempt probability; Determining whether the target attempt probability exceeds a probability threshold; When the target attempt probability is greater than the probability threshold, randomly obtaining an alternative unbalanced bending moment classification in the classification set as the real-time unbalanced bending moment identification result; When the target attempt probability is not greater than the probability threshold, inputting the real-time sample characterization vector into the basic recognition algorithm to obtain the real-time unbalanced bending moment recognition result; The classifying the candidate unbalanced bending moment corresponding to the first correlation result corresponding to the maximum result in the first correlation result cluster as the real-time unbalanced bending moment identification result includes: If the candidate unbalanced bending moment classification corresponding to the first association result is the same as the target unbalanced bending moment classification, the real-time feedback information is determined as the first feedback information, and the first feedback information represents that the basic recognition algorithm has correctly recognized the information; If the candidate unbalanced bending moment classification corresponding to the first association result is different from the target unbalanced bending moment classification, the real-time feedback information is determined as second feedback information, and the second feedback information represents that the basic recognition algorithm has made an error in recognition.

7. The method according to claim 1, characterized in that The method further comprises: Determine a plurality of data component sets from the knowledge sample, wherein an m-th data component set in the plurality of data component sets includes an m-th sample characterization vector, an m-th unbalanced bending moment identification result corresponding to the m-th sample characterization vector, an m-th feedback information, and an m+1-th sample characterization vector, where m is a natural number greater than 0; The basic recognition algorithm to be optimized is optimized based on the multiple data component sets to obtain the target recognition algorithm. When the optimized state of the basic recognition algorithm meets the preset convergence condition, the basic recognition algorithm is determined as the target recognition algorithm. When the optimized state of the basic recognition algorithm does not meet the preset convergence condition, the algorithm parameters of the basic recognition algorithm are optimized based on a preset cost function, and the execution data of each optimization is a data component set among the multiple data component sets.

8. The method according to claim 7, characterized in that The method further comprises: According to the following steps, the basic recognition algorithm to be optimized is optimized based on multiple data composition sets to obtain the target recognition algorithm, wherein the data composition set input to the basic recognition algorithm is the mth data composition set: Determine whether the bridge rotation sensor data set sample corresponding to the m+1th sample characterization vector is the last bridge rotation sensor data set sample in the sample bridge rotation sensor data stream; If the bridge rotation sensor data set sample corresponding to the m+1th sample characterization vector is the penultimate bridge rotation sensor data set sample, determining the setting parameter output by the basic recognition algorithm based on the mth feedback information; If the bridge rotation sensor data set sample corresponding to the m+1th sample characterization vector is not the penultimate bridge rotation sensor data set sample, then determining the set parameter based on the mth feedback information and the output of the comparison recognition algorithm, the comparison recognition algorithm is an initialization recognition algorithm, and the algorithm parameters of the comparison recognition algorithm and the basic recognition algorithm are different; Determining a cost value of the cost function based on the set parameters and the output of the basic recognition algorithm, and optimizing algorithm parameters of the basic recognition algorithm based on the cost value of the cost function; If the basic recognition algorithm reaches a convergence state, the basic recognition algorithm is used as the target recognition algorithm.

9. The method according to claim 8, characterized in that The determining the setting parameter based on the m-th feedback information and the output of the comparison recognition algorithm includes: Inputting each unbalanced bending moment classification in the classification set in sequence and the m+1th sample characterization vector into a second classification association function simultaneously to obtain a second association result cluster, wherein the second association result cluster includes association results corresponding to each unbalanced bending moment classification; The sum of the second association result corresponding to the maximum result in the second association result cluster and the mth feedback information is used as the set parameter.

10. A monitoring system, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the program, the steps in the method according to any one of claims 1 to 9 are implemented.

Images