Bridge monitoring data missing complementation method based on deep learning LSTM model

Abstract

The invention discloses a bridge monitoring data missing complementing method based on a deep learning LSTM model, relates to the technical field of bridge monitoring, and solves the technical problem that the bridge data analysis effect is limited due to the fact that the existing bridge health monitoring data missing complementing method is low in data complementing precision. Comprising the following steps: step 1, acquiring bridge health monitoring data in real time; 2, generating a data state according to the bridge health monitoring data; step 3, when the data state is a missing state, generating preliminary complementation data; otherwise, acquiring bridge health monitoring data of the next period; entering the step 2; 4, verifying the preliminary completion data according to the bridge health monitoring data to obtain verified completion data when the preliminary completion data is generated in the current set number of collection periods; 5, taking the verification completion data as completion data; data completion is carried out; the bridge health monitoring data can be accurately and efficiently supplemented; and the bridge data analysis effect is improved.

Description

Technical Field

[0001] This application belongs to the field of bridge monitoring technology and relates to bridge monitoring technology, specifically a method for completing missing bridge monitoring data based on a deep learning LSTM model. Background Art

[0002] The Bridge Health Monitoring System (BHMS) is a crucial component of modern bridge structure monitoring. It installs various sensors on the bridge structure to monitor and collect real-time data on the bridge's status, such as temperature, humidity, deflection, stress, strain, and traffic volume, thereby determining the health of the bridge structure. Typically, a BHMS consists of sensors, a data collector, a data transmission module, and a data analysis module: Sensors: These devices are directly attached to key locations on the bridge structure to monitor physical changes in the structure. Common sensors include strain sensors, accelerometers, and displacement sensors. Data collector: This device is responsible for collecting the monitoring results from each sensor into the system, providing a foundation for centralized data processing. Data transmission module: This module transmits the collected data to a data analysis platform or remote server via wireless networks, wired networks, or IoT devices. Data analysis module: Within the data analysis module, various algorithms are used to analyze the collected data to identify damage, fatigue, deformation, and other phenomena on the bridge, thereby providing a basis for bridge maintenance and decision-making.

[0003] In the process of acquiring bridge health data, equipment failure may occur: for example, sensor failure or damage, data collector abnormality or data transmission network interruption, which will lead to the loss of monitoring data; environmental factors: extreme weather conditions such as heavy rain, extreme cold, strong wind, etc. will affect the working stability of monitoring equipment, resulting in difficulty in data acquisition; data transmission delay and packet loss: the data transmission of the bridge monitoring system relies on wireless network or wired network. When the network signal is unstable or congested, it may cause packet loss during the data transmission process, resulting in data loss; human factors: human negligence or operational errors in data management may also lead to data loss or erroneous recording; the loss of monitoring data caused by these reasons not only affects the real-time perception of the structural status of the bridge health monitoring system, but also affects the accuracy of subsequent analysis and prediction models;

[0004] Current methods for completing bridge health monitoring data primarily include simple interpolation methods such as mean interpolation, linear interpolation, and polynomial interpolation. These methods are somewhat effective for addressing small-scale data gaps. However, for long-term, large-scale missing data, traditional interpolation methods struggle to meet the high accuracy and continuity requirements of bridge health monitoring data due to their overly simplistic models and inability to capture the complex temporal characteristics of the data. Furthermore, bridge monitoring data typically exhibits complex temporal dependencies and is affected by factors such as temperature, humidity, and traffic load. Traditional interpolation methods are unable to accurately capture the influence of these time-varying factors, often resulting in low accuracy in the completed data and limiting the effectiveness of data analysis. Therefore, a method for completing missing bridge monitoring data based on a deep learning LSTM model is needed. Summary of the Invention

[0005] This application aims to solve at least one of the technical problems existing in the prior art; to this end, this application proposes a bridge monitoring data missing completion method based on a deep learning LSTM model, which is used to solve the technical problem that the existing bridge health monitoring data missing completion method has low accuracy in completing the data, resulting in limiting the effect of bridge data analysis.

[0006] To achieve the above objectives, the first aspect of the present application provides a method for completing missing bridge monitoring data based on a deep learning LSTM model, comprising the following steps:

[0007] Step 1: Obtain bridge health monitoring data in real time according to the set collection cycle;

[0008] Step 2: generating a data state according to the bridge health monitoring data;

[0009] Step 3: When the data status is missing, generate several project completion input data based on the bridge health monitoring data; input the project completion input data into the corresponding bridge monitoring data missing completion model to obtain predicted completion data, and mark the predicted completion data as preliminary completion data; otherwise, obtain the bridge health monitoring data of the next collection cycle and proceed to step 2;

[0010] Step 4: When the preliminary supplementary data is generated in the current set number of collection cycles, the preliminary supplementary data is verified according to the bridge health monitoring data to obtain the verified supplementary data;

[0011] Step 5: Use the verified completed data as the completed data to complete the missing data.

[0012] In this application, bridge health monitoring data is acquired in real time according to a set collection cycle, and a data state is generated according to the bridge health monitoring data; when the data state is a missing state, a number of project completion input data are generated according to the bridge health monitoring data; the project completion input data is input into the corresponding bridge monitoring data missing completion model to obtain predicted completion data, and the predicted completion data is marked as preliminary completion data; when preliminary completion data is generated in the current set number of collection cycles, the preliminary completion data is verified according to the bridge health monitoring data to obtain verified completion data; the verified completion data is used as completion data; and is used to complete the corresponding missing data; wherein the bridge monitoring data missing completion model is obtained by training the LSTM model; the missing data is inferred from the known data through the bridge monitoring data missing completion model, thereby achieving accurate completion, and the completion data is subsequently verified to achieve accurate and efficient completion of the bridge health monitoring data, thereby improving the bridge data analysis effect.

[0013] Preferably, generating data status according to bridge health monitoring data includes:

[0014] S21: extracting the collected values ​​of several monitoring items corresponding to each collection moment in the collection period;

[0015] S22: determining in sequence whether the collected values ​​of the monitoring items corresponding to each collection time are empty; if so, marking the collected values ​​of the monitoring items corresponding to the collection time as missing values; if not, marking the collected values ​​of the monitoring items corresponding to the collection time as normal values;

[0016] S23: sequentially determining whether there are missing values ​​in each monitoring item during the collection period; if yes, marking the monitoring item as a to-be-completed item; if no, marking the monitoring item as a normal item;

[0017] S24: Determine whether there are items to be completed within the collection period; if yes, set the data status to a missing state; if not, set the data status to a normal state; the data status includes a missing state or a normal state.

[0018] Preferably, the step of generating a plurality of items of complementary input data based on the bridge health monitoring data includes:

[0019] Obtain the missing values ​​and normal values ​​corresponding to each item to be completed during the collection period; generate the missing impact coefficient based on the missing values ​​and normal values, and generate the project status based on the missing impact coefficient;

[0020] When the project status is complex and missing, the completion cycle is set to long completion cycle;

[0021] When the item status is a simple missing state, the completion cycle is set to a short completion cycle; the number of collection cycles of the long completion cycle is greater than the number of collection cycles of the short completion cycle;

[0022] Acquire a number of project completion input data according to the completion cycle.

[0023] Preferably, generating the missing impact coefficient according to the missing values ​​and the normal values ​​includes:

[0024] Obtain the number of missing values ​​and normal values ​​corresponding to the monitoring items, and mark them as QS and ZS respectively; number the normal values ​​according to their corresponding collection time and mark them as XS t ; t is the number of the collection time;

[0025] By formula The missing influence coefficient QY is calculated; wherein α1 and α2 are proportional coefficients; ZB is the average change rate corresponding to the monitoring item; XS0 is the normal value or the completed value corresponding to the last acquisition moment of the previous acquisition cycle of the monitoring item; DT is the time length between the vector acquisition moments; t=1, 2,…, T; T is the total number of acquisition moments in the acquisition cycle.

[0026] Preferably, generating the project status according to the missing impact coefficient includes:

[0027] Obtain the missing impact coefficient corresponding to each monitoring item and determine whether the missing impact coefficient is greater than a set threshold; if yes, set the item status to a complex missing state; if not, set the item status to a simple missing state.

[0028] Preferably, the step of obtaining the input data for a plurality of items to be completed according to the completion cycle includes:

[0029] Get the completion cycle of each item to be completed,

[0030] When the completion cycle is a long completion cycle, obtaining the number of collection cycles corresponding to the long completion cycle, obtaining the collected values ​​of each collection moment of the monitoring item corresponding to the item to be completed in the collection cycles of the number between the current collection cycle; integrating the collected values ​​into the item completion input data of the item to be completed in the time sequence corresponding to the collection moments;

[0031] When the completion cycle is a short completion cycle, obtain the number of collection cycles corresponding to the short completion cycle, obtain the collected values ​​of each collection moment of the monitoring item corresponding to the item to be completed in the collection cycles of the number between the current collection cycle; integrate the collected values ​​into the item completion input data of the item to be completed in the time sequence corresponding to each collection moment;

[0032] The item completion input data of each item to be completed is obtained in sequence.

[0033] This application sets the project completion input data for predicting missing data based on the amount of missing data and the fluctuation of the monitoring items corresponding to the missing data; and generates project completion input data in a targeted manner according to the specific circumstances of the missing data, thereby ensuring the accuracy of missing data prediction while improving processing efficiency.

[0034] Preferably, the bridge monitoring data missing completion model is obtained by training an LSTM model, including:

[0035] Obtain a number of historically collected values ​​and their corresponding collection times for each monitoring item, integrate a set number of historically collected values ​​into a project time feature vector in the order of collection times; obtain a number of time feature vectors in sequence; and divide the number of time feature vectors into a training set and a validation set;

[0036] Use several time feature vectors in the training set to train the LSTM model to obtain a trained LSTM model;

[0037] The trained LSTM model was tested using the time feature vectors in the validation set, and finally a bridge monitoring data missing completion model for the corresponding monitoring project was obtained.

[0038] Preferably, the verifying the preliminary supplemented data according to the bridge health monitoring data to obtain the verified supplemented data includes:

[0039] Obtaining an initial collection period of the project completion input data corresponding to the preliminary completion data, marking the collection period and the current collection period, as well as all collection periods between the collection period and the current collection period, as verification collection periods;

[0040] Use the preliminary completed data to replace the corresponding missing data as the collected values ​​at the collection time in the corresponding collection period;

[0041] Obtaining a number of collected values ​​corresponding to the monitoring items within the verification collection period, fitting the collected data into a verification curve in the order of their corresponding collection moments; generating a completed data state based on the verification curve;

[0042] When the status of the completion data is normal, the preliminary completion data is set as the verification completion data; otherwise, the completion cycle corresponding to the preliminary completion data is obtained, and the completion cycle is updated to obtain the verification completion cycle; the project completion input data is obtained according to the verification completion cycle, and the project completion input data is input into the corresponding bridge monitoring data missing completion model to obtain the predicted completion data, and the predicted completion data is marked as the verification completion data.

[0043] Preferably, generating the completed data state according to the verification curve includes:

[0044] Obtain the slope of the verification curve at each acquisition moment; obtain the slope change value between each adjacent acquisition moment;

[0045] Determine whether the variance of the slope change value is less than the set stable variance threshold; if yes, set the completion data status to normal; if not, obtain the collection time corresponding to the slope change value that is greater than the average slope change value and mark it as an abnormal time;

[0046] It is determined whether there are any abnormal moments in the collection moments corresponding to the initial supplemented data. If yes, the supplemented data status is set to abnormal; if not, the supplemented data status is set to normal.

[0047] Preferably, updating the completion period to obtain the verification completion period includes:

[0048] Obtain the number of abnormal moments and the acquisition time data in the preliminary completed data, marked as YS and CS respectively;

[0049] YS

[0050] By the formula YBS = β × e CS ×CBS is used to calculate the number of cycles in the verification and completion cycle, YBS. β is a proportional coefficient used to adjust the impact of abnormal moments on the verification and completion cycle. The specific value is set based on experience. CBS is the number of acquisition cycles in the completion cycle.

[0051] Compared with the prior art, the present invention has the following advantages:

[0052] 1. In this application, bridge health monitoring data is acquired in real time according to a set collection cycle, and a data state is generated according to the bridge health monitoring data; when the data state is a missing state, a number of project completion input data are generated according to the bridge health monitoring data; the project completion input data is input into the corresponding bridge monitoring data missing completion model to obtain predicted completion data, and the predicted completion data is marked as preliminary completion data; when preliminary completion data is generated in the current set number of collection cycles, the preliminary completion data is verified according to the bridge health monitoring data to obtain verified completion data; the verified completion data is used as completion data; and is used to complete the corresponding missing data; wherein the bridge monitoring data missing completion model is obtained by training the LSTM model; the missing data is inferred from the known data through the bridge monitoring data missing completion model, thereby achieving accurate completion, and the completion data is subsequently verified to achieve accurate and efficient completion of the bridge health monitoring data, thereby improving the bridge data analysis effect.

[0053] 2. This application sets the project completion input data for predicting missing data based on the amount of missing data and the fluctuation of the monitoring items corresponding to the missing data; and generates project completion input data in a targeted manner according to the specific circumstances of the missing data, thereby ensuring the accuracy of missing data prediction while improving processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0055] Figure 1 This is a flow chart of the method for completing missing bridge monitoring data in this application;

[0056] Figure 2 A flow chart for generating the data status in this application;

[0057] Figure 3 The detailed drawings of the repair of missing data of expansion joint displacement of a real bridge obtained by this method for this application. DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions of this application in conjunction with the embodiments. Obviously, the embodiments described are only a part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0059] See also Figure 1 The first aspect of the present application provides a method for completing missing bridge monitoring data based on a deep learning LSTM model, comprising the following steps:

[0060] Step 1: Acquire bridge health monitoring data in real time according to a set collection cycle. Bridge health monitoring data refers to the monitoring data of various monitoring items required for bridge health monitoring, such as expansion joint displacement. The collection cycle is the length of a set period. Several collection moments are set at fixed intervals within a collection cycle. When a collection moment arrives, each collection device is controlled to collect data for each monitoring item. The collection cycle can be one month, one quarter, one day, or 12 hours.

[0061] Step 2: Generate a data status based on the bridge health monitoring data; the data status indicates whether the bridge health monitoring data in the corresponding collection period is missing; it includes a missing state and a normal state, and the missing state indicates that the bridge health monitoring data in the corresponding collection period is missing;

[0062] Step 3: When the data status is missing, generate several project completion input data based on the bridge health monitoring data; input the project completion input data into the corresponding bridge monitoring data missing completion model to obtain predicted completion data, and mark the predicted completion data as preliminary completion data; otherwise, obtain the bridge health monitoring data of the next collection cycle and proceed to step 2;

[0063] The project completion input data in step 3 is a time feature sequence generated based on the missing bridge health monitoring data within the corresponding collection period for predicting missing data; the predicted completion data is the predicted data corresponding to the collection time of the missing data generated by the bridge monitoring data missing completion model based on the change characteristics of the project completion input data;

[0064] Step 4: When preliminary completion data is generated for the currently set number of collection cycles, the preliminary completion data is verified based on the bridge health monitoring data to obtain verified completion data; the set data is set based on experience; the verified completion data is the data ultimately used for data completion; if data is missing in a certain collection cycle and preliminary completion data is generated, the preliminary completion data generated in the previous corresponding collection cycle will be verified in the collection cycle with the set data interval thereafter;

[0065] Step 5: Use the verified completed data as the completed data to complete the missing data.

[0066] In this embodiment, bridge health monitoring data is acquired in real time according to a set collection cycle, and a data state is generated according to the bridge health monitoring data; when the data state is a missing state, a number of project completion input data are generated according to the bridge health monitoring data; the project completion input data are input into the corresponding bridge monitoring data missing completion model to obtain predicted completion data, and the predicted completion data are marked as preliminary completion data; when preliminary completion data is generated in the current set number of collection cycles, the preliminary completion data is verified according to the bridge health monitoring data to obtain verified completion data; the verified completion data is used as completion data; and is used to complete the corresponding missing data; wherein the bridge monitoring data missing completion model is obtained by training the LSTM model; the missing data is inferred from the known data through the bridge monitoring data missing completion model, thereby achieving accurate completion, and the completion data is subsequently verified to achieve accurate and efficient completion of the bridge health monitoring data, thereby improving the bridge data analysis effect.

[0067] See also Figure 2 , generate data status based on bridge health monitoring data, including:

[0068] S21: extracting the collected values ​​of several monitoring items corresponding to each collection moment in the collection period;

[0069] S22: determining in sequence whether the collected values ​​of the monitoring items corresponding to each collection moment are empty; if so, marking the collected values ​​of the monitoring items corresponding to the collection moment as missing values; if not, marking the collected values ​​of the monitoring items corresponding to the collection moment as normal values; wherein, a collected value being empty indicates that no collected data is recorded at the collection moment, which may be due to a malfunction of the collection device or other reasons;

[0070] S23: sequentially determine whether there are missing values ​​in each monitoring item within the collection period; if yes, mark the monitoring item as a to-be-completed item; if no, mark the monitoring item as a normal item; wherein the to-be-completed item indicates an item with missing data within the period that requires data completion;

[0071] S24: Determine whether there are items to be completed within the collection period; if yes, set the data status to a missing state; if not, set the data status to a normal state; the data status includes a missing state or a normal state.

[0072] Generate several project-completing input data based on bridge health monitoring data, including:

[0073] Obtain the missing values ​​and normal values ​​corresponding to each item to be completed during the collection period; generate the missing impact coefficient based on the missing values ​​and normal values, which can be understood as the difficulty of completing the missing data; the larger the value of the missing impact coefficient, the greater the corresponding completion difficulty; generate the project status based on the missing impact coefficient; the project status can be understood as the difficulty of completing the corresponding project, including complex missing status and simple missing status; complex missing status indicates that the amount of data required to complete the data missing is large; simple missing status indicates that the amount of data required to complete the data missing is small;

[0074] When the project status is complex and missing, the completion cycle is set to long completion cycle;

[0075] When the item status is a simple missing state, the completion cycle is set to a short completion cycle; the number of collection cycles of the long completion cycle is greater than the number of collection cycles of the short completion cycle; it can be understood that the long completion cycle and the short completion cycle are both manually set numbers of collection cycles;

[0076] Acquire a number of project completion input data according to the completion cycle.

[0077] Generate missing impact coefficients based on missing values ​​and normal values, including: obtaining the number of missing values ​​and normal values ​​corresponding to the monitoring items, and marking them as QS and ZS respectively; numbering the normal values ​​according to their corresponding collection time, and marking them as XS t ; t is the number of the collection time;

[0078] By formula The missing influence coefficient QY is calculated; wherein, α1 and α2 are proportional coefficients, which are used to adjust the proportional relationship between the two terms in the above formula, and the specific values ​​are set according to experience; ZB is the average change rate corresponding to the monitoring item; XS0 is the normal value or the completed value corresponding to the last acquisition moment of the previous acquisition cycle of the monitoring item; DT is the time length between the vector acquisition moments; t=1, 2,…, T; T is the total number of acquisition moments in the acquisition cycle.

[0079] This embodiment uses the above formula to comprehensively consider the impact of the number of missing values ​​of a certain monitoring item and the recent degree of change of the item on the difficulty of completing the missing data, and generates a missing impact coefficient; when the number of missing values ​​is greater, the difficulty in completing the missing data is greater; when the change trend of the corresponding monitoring item is more complex, the difficulty in completing the missing data is greater.

[0080] Generate the project status according to the missing impact coefficient, including: obtaining the missing impact coefficient corresponding to each monitoring project, and judging whether the missing impact coefficient is greater than the set threshold; if yes, set the project status to a complex missing state; if not, set the project status to a simple missing state. The complex missing state indicates that the monitoring project has more missing data in this period, or the corresponding project has a large fluctuation in changes, resulting in more data that needs to be referenced when completing the project; the simple missing state indicates that the monitoring project has less missing data in this period, or the corresponding project has a small fluctuation in changes, resulting in a corresponding reduction in the data that needs to be referenced when completing the project.

[0081] Acquire several items of completion input data according to the completion cycle, including: acquiring the completion cycle of each item to be completed,

[0082] When the completion cycle is a long completion cycle, obtaining the number of collection cycles corresponding to the long completion cycle, obtaining the collected values ​​of each collection moment of the monitoring item corresponding to the item to be completed in the collection cycles of the number between the current collection cycle; integrating the collected values ​​into the item completion input data of the item to be completed in the time sequence corresponding to the collection moments;

[0083] When the completion cycle is a short completion cycle, obtain the number of collection cycles corresponding to the short completion cycle, obtain the collected values ​​of each collection moment of the monitoring item corresponding to the item to be completed in the collection cycles of the number between the current collection cycle; integrate the collected values ​​into the item completion input data of the item to be completed in the time sequence corresponding to each collection moment;

[0084] The item completion input data of each item to be completed is obtained in sequence.

[0085] This embodiment sets reference data for predicting missing data, i.e., project completion input data, based on the amount of missing data and the fluctuation of the monitoring items corresponding to the missing data. When there are many missing data, or the fluctuation of the monitoring items corresponding to the missing data is more complex, monitoring data within more collection cycles are used as reference data and integrated into project completion input data. When there are few missing data, or the fluctuation of the monitoring items corresponding to the missing data is relatively simple, monitoring data within fewer collection cycles are used as reference data and integrated into project completion input data. Project completion input data is generated in a targeted manner according to the specific circumstances of the missing data, thereby ensuring the accuracy of missing data prediction while improving processing efficiency.

[0086] The bridge monitoring data missing completion model is trained through the LSTM model, including:

[0087] Obtain a number of historically collected values ​​and their corresponding collection times for each monitoring item, integrate a set number of historically collected values ​​into a project time feature vector in the order of collection times; obtain a number of time feature vectors in sequence; and divide the number of time feature vectors into a training set and a validation set;

[0088] Use several time feature vectors in the training set to train the LSTM model to obtain a trained LSTM model;

[0089] The trained LSTM model was tested using the time feature vectors in the validation set, ultimately resulting in a model for completing missing bridge monitoring data for the corresponding monitoring project. It is understood that after model training is complete, the trained model is evaluated using the validation set to calculate the error between the predicted and actual values. Common error metrics include mean squared error (MSE) and mean absolute error (MAE). If the error is large, optimization can be achieved by adjusting LSTM hyperparameters such as the number of network layers, number of neurons, and learning rate.

[0090] Verifying the preliminary supplementary data according to the bridge health monitoring data to obtain the verified supplementary data, including: obtaining an initial collection period of the project supplementary input data corresponding to the preliminary supplementary data, marking the collection period and the current collection period, as well as all collection periods between the collection period and the current collection period, as verification collection periods; the verification collection period is data used to verify the accuracy of the supplementary data, including a set number of collection periods before the collection period corresponding to the supplementary data, and monitoring values ​​of the corresponding project within a set number of collection periods thereafter;

[0091] Use the preliminary completed data to replace the corresponding missing data as the collected values ​​at the collection time in the corresponding collection period;

[0092] Obtain several collected values ​​of corresponding monitoring items within the verification collection cycle, and fit the collected data into a verification curve according to the order of their corresponding collection moments; generate a supplementary data status based on the verification curve; the supplementary data status indicates the accuracy of the supplementary data, including normal and abnormal. A normal state indicates that the accuracy of the supplementary data meets the requirements; an abnormal state indicates that the accuracy of the supplementary data does not meet the requirements;

[0093] When the status of the completion data is normal, the preliminary completion data is set as the verification completion data; otherwise, the completion cycle corresponding to the preliminary completion data is obtained, and the completion cycle is updated to obtain the verification completion cycle; the project completion input data is obtained according to the verification completion cycle, and the project completion input data is input into the corresponding bridge monitoring data missing completion model to obtain the predicted completion data, and the predicted completion data is marked as the verification completion data.

[0094] Generate a complete data state according to the verification curve, including: obtaining the slope corresponding to each acquisition moment of the verification curve; obtaining the slope change value between each adjacent acquisition moment;

[0095] Determine whether the variance of the slope change value is less than the set stable variance threshold; the stable variance threshold is set based on expert experience; if yes, set the completed data status to normal; if not, obtain the collection time corresponding to the slope change value that is greater than the average slope change value and mark it as an abnormal time;

[0096] It is determined whether there are any abnormal moments in the collection moments corresponding to the initial supplemented data. If yes, the supplemented data status is set to abnormal; if not, the supplemented data status is set to normal.

[0097] In this embodiment, it is determined whether the fluctuation of the supplemented data is consistent or close to the fluctuation of other data during its verification and supplementation cycle. If they are consistent or close, it means that the supplemented data is relatively accurate and within the requirements; otherwise, it means that the supplemented data needs to be regenerated.

[0098] The verification completion cycle is updated to obtain the verification completion cycle, including: obtaining the number of abnormal moments and the acquisition time data in the preliminary completion data, marked as YS and CS respectively; The number of cycles in the verification and completion cycle, YBS, is calculated. β is a proportional coefficient used to adjust the impact of abnormal moments on the verification and completion cycle. The specific value is set based on experience. CBS is the number of acquisition cycles in the completion cycle.

[0099] This embodiment calculates the number of reference collection cycles of data required for regenerating the supplementary data through the above formula, that is, the number of cycles for verifying the supplementary cycle; when the supplementary data corresponds to a large number of abnormal moments, it means that the degree of deviation of the supplementary data is large, and the supplementary data needs to be regenerated; since the input data for generating the supplementary data is a time series, the longer the length of the time series, the higher the accuracy of the predicted value output by the model, so this embodiment increases the periodic data of the verification supplementary cycle according to the degree of deviation of the supplementary data; when the degree of deviation of the supplementary data is greater, that is, the number of abnormal moments is greater, the corresponding number of cycles for verifying the supplementary cycle is greater; thereby ensuring the accuracy of the supplementary data.

[0100] See also Figure 3 , which is a detailed diagram of using this method to repair the missing data of expansion joint displacement of a real bridge; the horizontal axis in the figure is the time corresponding to the collection moment, and the vertical axis is the degree of expansion joint displacement.

[0101] This embodiment can accurately capture the complex relationships between various factors in bridge monitoring data, thereby achieving more accurate data completion. The completion results can effectively restore the true data trend in both short-term and long-term data loss scenarios. Unlike the fixed models in the prior art, the present invention can dynamically adjust the completion strategy based on characteristics such as the duration and pattern of the data loss, ensuring data completion effectiveness under different loss patterns. Even in scenarios with complex data loss patterns, the technology of the present invention can provide good completion results. It also enhances data reliability: By verifying the completed data, the present invention can provide confidence intervals for the completion results, ensure the reliability of the completed data, and provide more accurate data support for subsequent bridge health status assessments. Real-time and efficient processing: The optimized LSTM network can more efficiently process large-scale bridge health monitoring data, reduce computing resource requirements, adapt to the needs of real-time monitoring systems, and improve the system's computational efficiency and real-time performance. In summary, this embodiment can overcome the shortcomings of the prior art and significantly improve the completion accuracy, reliability, and computational efficiency of bridge health monitoring data through innovations such as multi-feature input, dynamic completion strategy, and error control, providing more accurate and efficient technical support for bridge health monitoring.

[0102] Some of the data in the above formula are calculated by removing the dimensions and taking their numerical values. The formula is a formula that is closest to the actual situation obtained by software simulation of a large amount of collected data; the preset parameters and preset thresholds in the formula are set by technical personnel in this field according to actual conditions or obtained through simulation of a large amount of data.

[0103] How this application works:

[0104] In this application, bridge health monitoring data is acquired in real time according to a set collection cycle, and a data state is generated according to the bridge health monitoring data; when the data state is a missing state, a number of project completion input data are generated according to the bridge health monitoring data; the project completion input data is input into the corresponding bridge monitoring data missing completion model to obtain predicted completion data, and the predicted completion data is marked as preliminary completion data; when preliminary completion data is generated in the current set number of collection cycles, the preliminary completion data is verified according to the bridge health monitoring data to obtain verified completion data; the verified completion data is used as completion data; and is used to complete the corresponding missing data; wherein the bridge monitoring data missing completion model is obtained by training the LSTM model; the missing data is inferred from the known data through the bridge monitoring data missing completion model, thereby achieving accurate completion, and the completion data is subsequently verified to achieve accurate and efficient completion of the bridge health monitoring data, thereby improving the bridge data analysis effect.

[0105] The above embodiments are only used to illustrate the technical method of the present application and are not intended to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present application.

Claims

1. A method for completing missing bridge monitoring data based on a deep learning LSTM model, characterized in that: The following steps are involved: Step 1: Obtain bridge health monitoring data in real time according to the set collection cycle; Step 2: generating a data state according to the bridge health monitoring data; Step 3: When the data status is missing, several items are generated based on the bridge health monitoring data to complete the input data; Input the project completion input data into the corresponding bridge monitoring data missing completion model to obtain the predicted completion data, and mark the predicted completion data as preliminary completion data; otherwise, obtain the bridge health monitoring data of the next collection cycle; and proceed to step 2; Step 4: When the initial supplementary data is generated by the current set number of collection cycles, the initial supplementary data is verified according to the bridge health monitoring data to obtain the verified supplementary data; Step 5: Use the verified completed data as the completed data to complete the missing data.

2. The method for completing missing bridge monitoring data based on a deep learning LSTM model according to claim 1 is characterized in that: The generating of data status according to the bridge health monitoring data includes: S21: extracting the collected values ​​of several monitoring items corresponding to each collection moment in the collection cycle; S22: determining in sequence whether the collected values ​​of the monitoring items corresponding to each collection time are empty; if yes, marking the collected values ​​of the monitoring items corresponding to the collection time as missing values; if no, marking the collected values ​​of the monitoring items corresponding to the collection time as normal values; S23: sequentially determine whether there are missing values ​​in each monitoring item within the collection period; if yes, mark the monitoring item as a to-be-completed item; if no, mark the monitoring item as a normal item; S24: Determine whether there are items to be completed within the collection period; if yes, set the data status to a missing status; if no, set the data status to a normal status; the data status includes a missing status or a normal status.

3. The method for completing missing bridge monitoring data based on a deep learning LSTM model according to claim 2 is characterized in that: The method of generating a plurality of items to complete the input data according to the bridge health monitoring data includes: Obtain the missing values ​​and normal values ​​corresponding to each item to be completed during the collection period; generate the missing impact coefficient according to the missing values ​​and the normal values, and generate the project status according to the missing impact coefficient; When the project status is complex and missing, the completion cycle is set to a long completion cycle; When the item status is a simple missing status, the completion cycle is set to a short completion cycle; the number of collection cycles of the long completion cycle is greater than the number of collection cycles of the short completion cycle; According to the completion cycle, a number of items of completion input data are obtained.

4. The method for completing missing bridge monitoring data based on a deep learning LSTM model according to claim 3 is characterized in that: The generating of missing impact coefficients according to missing values ​​and normal values ​​includes: Obtain the number of missing values ​​and normal values ​​corresponding to the monitoring items, and mark them as QS and ZS respectively; number the normal values ​​according to their corresponding collection time and mark them as XS t ; t is the number of the collection time; By formula The missing influence coefficient QY is calculated; wherein α1 and α2 are proportional coefficients; ZB is the average change rate corresponding to the monitoring item; XS0 is the normal value corresponding to the last acquisition moment of the previous acquisition cycle of the monitoring item; DT is the time length between the vector acquisition moments; t=1, 2,…, T; T is the total number of acquisition moments in the acquisition cycle.

5. The method for completing missing bridge monitoring data based on deep learning LSTM model according to claim 4 is characterized in that: The generating of the project status according to the missing impact coefficient includes: Obtain the missing impact coefficient corresponding to each monitoring item, and determine whether the missing impact coefficient is greater than a set threshold; if yes, set the item status to a complex missing status; if no, set the item status to a simple missing status.

6. The method for completing missing bridge monitoring data based on deep learning LSTM model according to claim 3 is characterized in that: The step of obtaining a plurality of items of completion input data according to the completion cycle includes: Get the completion period of each item to be completed. When the completion cycle is a long completion cycle, the number of collection cycles corresponding to the long completion cycle is obtained, and the collection values ​​of each collection time of the monitoring project corresponding to the project to be completed in the collection cycles of the number between the current collection cycle are obtained; the collection values ​​are integrated into the project completion input data of the project to be completed according to the time sequence corresponding to each collection time; When the completion cycle is a short completion cycle, the number of collection cycles corresponding to the short completion cycle is obtained, and the collection values ​​of each collection time of the monitoring project corresponding to the project to be completed in the number of collection cycles between the current collection cycle are obtained; the collection values ​​are integrated into the project completion input data of the project to be completed according to the time sequence corresponding to each collection time; The item completion input data of each item to be completed is obtained in sequence.

7. The method for completing missing bridge monitoring data based on deep learning LSTM model according to claim 1 is characterized in that: The bridge monitoring data missing completion model is obtained by training the LSTM model, including: Obtain a number of historical collection values ​​and their corresponding collection times for each monitoring project, and integrate a set number of historical collection values ​​into a project time feature vector in the order of collection time; obtain a number of time feature vectors in sequence; and divide the number of time feature vectors into a training set and a validation set; Use several time feature vectors in the training set to train the LSTM model to obtain a trained LSTM model; The trained LSTM model was tested using the time feature vector in the validation set, and finally a bridge monitoring data missing completion model for the corresponding monitoring project was obtained.

8. The method for completing missing bridge monitoring data based on deep learning LSTM model according to claim 1 is characterized in that: The verifying of the preliminary supplemented data according to the bridge health monitoring data to obtain the verified supplemented data includes: Obtaining an initial collection cycle of the project completion input data corresponding to the preliminary completion data, marking the collection cycle and the current collection cycle, as well as all collection cycles between the collection cycle and the current collection cycle as verification collection cycles; Replace the corresponding missing data with the preliminary completed data as the collected values ​​at the collection time in the corresponding collection period; Obtaining a number of collected values ​​corresponding to the monitoring items within the verification collection period, fitting the collected data into a verification curve in the order of their corresponding collection moments; generating a completed data state according to the verification curve; When the completion data status is normal, the preliminary completion data is set as the verification completion data; otherwise, the completion cycle corresponding to the preliminary completion data is obtained, and the completion cycle is updated to obtain the verification completion cycle; the project completion input data is obtained according to the verification completion cycle, and the project completion input data is input into the corresponding bridge monitoring data missing completion model to obtain the predicted completion data, and the predicted completion data is marked as the verification completion data.

9. The method for completing missing bridge monitoring data based on deep learning LSTM model according to claim 8, characterized in that: The generating of the completion data state according to the verification curve includes: Obtain the slope of the verification curve corresponding to each acquisition time; obtain the slope change value between each adjacent acquisition time; Determine whether the variance of the slope change value is less than the set stable variance threshold; if yes, set the completion data status to normal; if no, obtain the collection time corresponding to the slope change value greater than the average slope change value and mark it as an abnormal time; It is determined whether there are abnormal moments at each collection moment corresponding to the initial supplementary data. If yes, the supplementary data state is set to abnormal; if no, the supplementary data state is set to normal.

10. The method for completing missing bridge monitoring data based on deep learning LSTM model according to claim 9, characterized in that: The completion cycle is updated to obtain the verification completion cycle, including: Obtain the number of abnormal moments and the data of the acquisition moment in the preliminary completed data, marked as YS and CS respectively; By formula The number of cycles YBS in the verification and completion cycle is calculated; wherein β is the proportionality coefficient; and CBS is the number of acquisition cycles in the completion cycle.

Images