Urban multi-source data element quality perception fusion method and system

Abstract

The application discloses a city multi-source data element quality perception fusion method and system, belongs to the technical field of city water management network data fusion monitoring, and comprises a data fusion pairing module, a data acquisition module, a preventive missed detection module and a pipeline abnormality auxiliary judgment module. The data fusion pairing module collects city subordinate bus line maps and water management network line maps, divides the detection radius of the water management network line, screens bus line sections around the detection radius, records the pairing relationship between the water management network line and the bus line section, establishes a city water management network missed detection database, establishes a file and records the pairing relationship of different water management network lines in the city, and combines the continuously matched bus line sections as missed detection paths. The city multi-source data element quality perception fusion method and system rely on city public transportation to realize multi-source data fusion, differentiated missed detection and early warning, realize the continuous scanning and preventive missed detection of the city water management network with low cost, high frequency and wide coverage.

Description

Technical Field

[0001] This invention belongs to the field of urban water supply network data fusion and monitoring technology, specifically a method and system for quality perception and fusion of multi-source data elements in cities. Background Technology

[0002] Urban water supply networks are an important part of urban infrastructure, responsible for transporting domestic water and industrial water, as well as discharging rainwater and sewage. As cities continue to expand, the coverage of urban water supply networks is also becoming increasingly extensive.

[0003] Currently, leak detection in water pipe networks mainly relies on two methods: traditional manual inspection and fixed sensor monitoring. Dedicated sensors are costly to deploy and difficult to maintain, making it difficult to achieve full coverage of the pipe network. Manual inspections are inefficient and have limited frequency, often only detecting leaks after they have developed into visible leaks or pipe bursts, making it difficult to provide early warnings of leaks. The existing single-data monitoring model fails to effectively integrate multi-source data and cannot combine existing urban infrastructure with water pipe network monitoring, resulting in wasted monitoring resources and difficulty in improving monitoring efficiency and accuracy. Furthermore, for detected pipe network anomalies, there is a lack of effective auxiliary judgment methods, often requiring further manual verification. Pipe network leak detection monitoring cannot form a three-dimensional monitoring network, making it difficult to quickly locate the specific location of abnormal pipes and failing to meet the needs of refined and intelligent monitoring of urban water pipe networks, thus exhibiting low practicality. Summary of the Invention

[0004] This invention aims to address at least one of the technical problems existing in the prior art. To this end, this invention proposes a method and system for quality perception and fusion of multi-source data elements in urban areas, which can effectively solve the problems of high cost, narrow coverage, low efficiency, and limited functionality in existing water pipe network monitoring.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: The urban multi-source data element quality perception and fusion system includes a data fusion and matching module, a data acquisition module, a preventive missed detection module, and a pipeline anomaly auxiliary judgment module. The data fusion and matching module collects bus route maps and water pipeline network route maps under the city's jurisdiction, divides the detection radius of the water pipeline network route, filters bus route segments around the detection radius, records the matching relationship between water pipeline network route and bus route segment, establishes a city water pipeline network missed detection database, establishes archives and records the matching relationship of different water pipeline network routes in the city, and merges continuously matched bus route segments as missed detection paths. The data acquisition module establishes a file in the urban water supply network leak detection database using bus routes as the index, and establishes sub-files in the file using the license plates of buses under the bus routes as the index. It collects bus operation data through inertial measurement units and pressure sensors deployed on buses, and collects surrounding environmental data during bus operation through high-definition cameras deployed on buses, and records the collected data in the corresponding bus sub-files. The preventive omission detection module, based on the collected data of different omission paths in the urban water pipe network omission detection database, combines the collected data under the same omission path with the pairing relationship, and performs preventive omission detection marking for abnormal pipe section locations under the omission path; The pipeline anomaly auxiliary judgment module, for pipelines with abnormal monitoring data in the water system, queries the corresponding missed detection path in the urban water network missed detection database, and uses the preventive missed detection module to assist in the judgment of suspicious abnormal road sections, and outputs the location coordinates of the abnormal road sections.

[0006] Furthermore, the data fusion and pairing module includes the following steps: The basic parameters of the current urban water supply network are collected through the urban water supply network GIS database, including the network line number, type, spatial coordinates, pipe diameter, material, burial depth, and service life. The current urban bus line data is obtained through the urban bus line operation map, including the bus line number, trajectory coordinates, stops, and information on vehicles under the line. The detection radius of different pipeline lines is divided according to the type of pipeline network. Combined with urban bus route data, the pipeline lines are matched with bus routes, including direct matching relationships and indirect matching relationships. A database of missed detections of urban water pipeline networks is established through MySQL to create files for urban pipeline lines and record the matching relationships of pipeline lines.

[0007] Furthermore, the detection radius of different pipeline routes is divided according to the type of pipeline network. Combined with urban bus route data, the pipeline routes are paired with bus routes, including direct and indirect matching relationships. A database of missed detections in urban water supply networks is established using MySQL to create files for urban pipeline routes and record their pairing relationships. Specifically, this includes the following steps: The detection radius *r* of different pipeline network types is determined based on the network type, which includes main pipelines, secondary pipelines, and branch pipelines. For each pipeline segment *e*, its buffer zone is calculated using a GIS buffer zone generator operator. L e The geometric line string representing pipe segment e, p is any point on the two-dimensional plane, R 2 Represents the two-dimensional Cartesian plane. Representative point p to pipe segment line L e The shortest vertical distance; For each bus route segment Where i is the bus route number and u is the bus route segment number, calculate its path string. The spatial relationship between each pipe segment e and the bus route segment is used to distinguish between direct and indirect matching relationships. For each pipe segment e, all bus routes are traversed sequentially to calculate direct and indirect matching relationships. When the pipe segment and the bus route segment meet the direct matching threshold condition, they are directly determined to be in direct matching relationships, and indirect matching calculations in the buffer zone are no longer performed. If the direct matching standard is not met, the indirect matching determination is completed by analyzing the spatial intersection between the bus route segment and the pipe segment buffer zone. A leak detection database for urban water pipe network is established using MySQL. An archive is created for each pipe segment, and all matched bus route segments and their matching types are recorded. Continuous bus route segments are merged as the matching leak detection path for each pipe segment. An archive is created separately, and each leak detection path is divided into pipe segment positions according to the length of a single detection unit. The direct matching relationship is calculated as the Hausdorf distance between the bus route and the segment, i.e.: ; in, The binary result represents a direct match between the representative pipe section and the bus route section. These represent the pipe segment and the bus route segment path, respectively, and ε represents the linear overlap distance threshold. Hausdorf distance representing the section of the pipeline and the bus route; Indirect matching is calculated by measuring the area of ​​the intersection between the bus route and the buffer zone of the management segment, i.e.: ; in, The binary result represents the indirect matching between the pipeline segment and the bus route, with Area representing the area of ​​the intersecting region.

[0008] Furthermore, the data acquisition module includes an inertial measurement unit, a high-definition camera, and a pressure sensor. It establishes a file in the urban water supply network leak detection database using bus routes as an index, and creates sub-files within the file using the license plates of buses belonging to those bus routes as indexes. The inertial measurement unit deployed on the bus collects bus operation data, and the high-definition camera deployed on the bus collects surrounding environmental data during bus operation. The collected data is recorded in the corresponding bus sub-files. Specifically, this includes the following steps: An inertial measurement unit is installed in the chassis or middle of the bus to collect vehicle operating status data. High-definition cameras are installed on both sides of the top of the bus to collect image data of the surrounding environment during vehicle operation. The operating status data includes the vehicle's three-axis acceleration, angular velocity, and attitude angle. The real-time load of the vehicle during operation is collected through the suspension system pressure sensor. In the urban water supply network leak detection database, a primary archive index is established by bus route ID. The archive stores the basic information of the route, including route trajectory coordinates and stops. Under each bus route archive, the bus license plate is used as a secondary index to create an independent sub-archive for each bus, recording the license plate ID, the route it belongs to, the equipment number, the data collection timestamp, the inertial measurement unit data record, and the camera image record.

[0009] Furthermore, the preventive missed detection module specifically includes the following steps: By using the urban waterworks pipeline network leak detection database, the matching relationship between different pipeline segments is extracted. For leak detection paths that are directly matched with the current pipeline segment, bus vehicle data on the current leak detection path under normal pipeline network conditions is collected. Vehicle-leak detection path baseline models are constructed for different leak detection paths. By monitoring the vibration residuals of different bus routes on the same leak detection path, preventive leak detection marking is carried out on pipeline segment location anomalies under the leak detection path. For missed detection paths that are indirectly matched with the current pipe segment, the images of the missed detection path are extracted and compared horizontally. The shadow changes of the continuous images are judged, and the abnormal positions of the pipe segments under the missed detection path are marked for preventive missed detection.

[0010] Furthermore, by extracting the matching relationships between different pipe segments from the urban waterworks network leak detection database, and collecting bus data on the current leak detection path under normal network conditions for leak detection paths that are directly matched with the current pipe segment, a vehicle-leak detection path baseline model is constructed for each different leak detection path. By monitoring the vibration residuals of different bus routes on the same leak detection path, preventative leak detection marking is performed on pipe segment location anomalies under the leak detection path. Specifically, this includes the following steps: The original triaxial acceleration signal is subjected to gravity component removal, and a pure vibration acceleration signal is obtained by bandpass filtering. Time-frequency analysis is performed on the acceleration signal of a single detection unit with a length Δl, including short-time Fourier transform and frequency band energy calculation of characteristic frequency bands, to form an output feature vector. The specific steps include: Short-time Fourier transform is performed on the acceleration signal within the length Δl of a single detection unit: ; in, This represents the time-frequency complex amplitude of the acceleration signal along the k-th axis at frequency f and time τ. This represents the time-domain acceleration signal after single-axis filtering. τ represents the Hamming window function, j represents the center time of the sliding time window, and k represents the imaginary unit. Bandwidth energy ,in Represents the vibrational energy of the characteristic frequency band of the k-th axis, and the lower limit of the characteristic frequency band. Upper limit The values ​​are 10 and 30 respectively, to obtain the output feature vector. : ,in Represents the energy of the three-axis frequency band. Represents the time-domain standard deviation of triaxial acceleration. , , Represents the time-domain kurtosis of triaxial acceleration; Collect bus data on currently missed detection paths under normal pipeline network conditions, and construct model inputs: ; in, This represents the multidimensional input feature vector of the i-th sample. These represent the vehicle's real-time estimated load, the vehicle's real-time speed, hourly rainfall, time of day, and pipe section location, respectively. A Transformer encoder is used as the model architecture to train a vehicle-missed detection path baseline model, and the matched output feature vectors are then used. Input feature vector The training set, validation set, and test set are divided into a ratio of 7:1.5:1.5. The AdamW optimizer is used to complete parameter iteration. In each round of training, forward propagation of samples, loss calculation, back propagation, and network parameter update are completed. After each round of iteration, the model prediction error is quantified on the validation set. When the validation loss does not decrease significantly in N rounds, the training is automatically terminated to obtain the vehicle-missed detection path baseline model. The input features are fed into the vehicle-missed detection path baseline model to obtain the conditional predicted distribution of the pipe segment vibration characteristics. The model output follows a Gaussian predicted distribution. ,in The input feature vector representing the new observed sample, The measured vibration feature vector corresponding to this sample The mean vector of vibration characteristics predicted by the model. The covariance matrix representing the prediction is normalized to the residuals using the squared Mahalanobis distance, under the null hypothesis and steady-state conditions. Satisfying the chi-square distribution q represents the characteristic dimension of pipe segment vibration; For the same missed detection path pipe segment location within day t, n t Mean residuals of each shift ,in Let represent the standardized residual of the i-th observation on day t. Divide the vehicles into K groups according to their load, and calculate the average residual for each group to verify consistency. ; in, The function is a 0-1 indicator function, where α is the significance level. This indicates that different load-bearing vehicles simultaneously exceed the residual limit in the same pipe section. When the residual limit exceeds the limit for a cumulative period of 7 days, the current pipe section location is marked as the location to be prevented for leak detection in the leak detection path file of the urban water supply network leak detection database. The daily residual series is smoothed using a 7-day moving average, and the moving average is calculated as follows: τ is the number of backtracking days, used to smooth the sequence. For the object, the Mann-Kendall nonparametric test is used to identify long-term degradation trends, where the trend statistic is... ), where T is the total number of days in the time series, sgn(x) is the sign function, taking the value 1 for x > 0, 0 for x = 0, and -1 for x < 0, and calculating the corrected variance. Where g is the number of sequence nodes, t p Let Z represent the number of samples in the p-th node. Calculate the standardized test statistic Z: ; Under a one-sided significance test, when Furthermore, if the conditions persist for 14 days, it is determined that the residual of the pipe segment corresponding to the missed detection path shows an upward deterioration trend. The current pipe segment location is marked as the location to be prevented for missed detection in the missed detection path file of the urban water pipe network missed detection database.

[0011] Furthermore, for the missed detection path that has an indirect matching relationship with the current pipe segment, the method involves extracting and horizontally comparing the continuous timestamp images of the missed detection path, determining shadow changes in the continuous images, and marking the pipe segment with abnormal position under the missed detection path for preventive missed detection. Specifically, this includes the following steps: Images of pipe segments with different missed detection paths under indirect matching relationships in the urban water supply network leak detection database are extracted, sorted by timestamp, and obtained as a continuous image string. The image string is converted to HSV color space, the shadow area in the image is calculated, and based on the rate of change of shadow area in consecutive images and the rate of change of shadow area in interval images, the abnormal pipe segment location under the missed detection path is determined, and preventive missed detection marking is performed. The specific steps include: The brightness and saturation in the image are: Where V(x,y) represents the brightness of pixel (x,y), S(x,y) represents the saturation of pixel (x,y), and H(x,y) represents the hue of pixel (x,y). and represent the mean and standard deviation of the hue components of the shadow area, respectively, where the shadow area A is: ; in, The threshold for determining the shadow index is represented by Ω, which represents the set of all pixels in the image. When the rate of change of the shadow area of ​​a continuous image Rate of change of shadow area in interval images At that time, the current pipe section location is recorded as the location for pending preventative missed detection. Let represent the shadow area in frame t and the shadow area in frame (t-1), respectively, where t represents the time index of the current frame and k is the interval number of frames. Represents a minimal constant. , These represent the threshold for determining continuous shadow anomalies and the threshold for determining interval shadow anomalies, respectively. Sensor data for potential preventative leak detection locations are extracted from the water system. Anomalies in sensor data for consecutive frames and interval frames corresponding to these locations are statistically analyzed. Locations where both sensor data anomalies and abnormal shadow area change rates in consecutive or interval images are identified are marked as potential preventative leak detection locations in the leak detection path archive of the urban water network leak detection database.

[0012] Furthermore, the pipeline anomaly auxiliary judgment module, for pipelines with abnormal monitoring data in the water system, queries the corresponding missed detection path in the urban water network missed detection database, and uses the preventive missed detection module to assist in the judgment of suspected abnormal sections, specifically including the following steps: In response to anomalies in the network data monitoring sensors of the water system, the pipeline numbers involved in the sensor anomalies are extracted, and the missing detection path combinations under the corresponding pipeline numbers are extracted from the urban water network missing detection database to assist in the anomaly determination of the missing detection paths.

[0013] Furthermore, regarding the anomalies in the network data monitoring sensors of the water system, the method involves extracting the pipeline numbers involved in the sensor anomalies, retrieving the corresponding missed detection path combinations under the urban water network missed detection database, and performing anomaly auxiliary judgment on the missed detection paths. This specifically includes the following steps: For the pipeline numbers involved in the obtained anomalies, all missed detection paths under the corresponding pipeline numbers are extracted from the urban water supply network missed detection database. Data collected from buses with different license plates in the bus routes involved in the missed detection paths are extracted. Through the direct matching relationship judgment process and the indirect matching relationship judgment process in the preventive missed detection module, the locations of the pipe segments under all missed detection paths are judged as anomalies. The locations to be prevented and detected are upgraded to abnormal missed detection locations, and the coordinates of the abnormal pipe segment locations are obtained.

[0014] A method for quality perception and fusion of multi-source data elements in urban areas includes the following steps: S1. Collect bus route maps and water pipeline network maps under the jurisdiction of the city, divide the detection radius of the water pipeline network, screen the bus route segments around the detection radius, record the pairing relationship between water pipeline network and bus route segments, establish a city water pipeline network missed detection database, establish files and record the pairing relationship of different water pipeline network lines in the city, and merge continuously matched bus route segments as missed detection paths. S2. Establish archives in the urban water supply network leak detection database using bus routes as the index, and establish sub-archives in the archives using the license plates of buses under the bus routes as the index. Collect bus operation data through inertial measurement units and pressure sensors deployed on buses, and collect surrounding environmental data during bus operation through high-definition cameras deployed on buses. Record the collected data in the corresponding bus sub-archives. S3. Through the urban water supply network leak detection database, extract the matching relationship of different network segments. For leak detection paths that are directly matched with the current segment, collect bus data on the current leak detection path under normal network conditions. Construct vehicle-leak detection path baseline models for different leak detection paths. By monitoring the vibration residuals of different bus routes on the same leak detection path, preventive leak detection marking is carried out for abnormal pipe segment positions under the leak detection path. S4. Through the urban water supply network leak detection database, extract the matching relationship of different network segments. For leak detection paths that are indirectly matched with the current segment, extract and compare the continuous timestamp images of the leak detection path, determine the shadow changes of the continuous images, and mark the abnormal position of the segment under the leak detection path for preventive leak detection. S5. For pipelines with abnormal monitoring data in the water system, query the corresponding missed detection path in the urban water network missed detection database, and perform auxiliary judgment on suspicious abnormal road sections based on the matching relationship of missed detection paths, and output the location coordinates of abnormal road sections. Compared with the prior art, the beneficial effects of the present invention are: In this invention, by relying on urban public transportation, multi-source data fusion of urban water information is carried out, and different preventive leakage detection modes are adopted for water pipe networks and bus routes with different pairing relationships. The operation trajectory of buses is used to achieve large-scale coverage of water pipe networks to assist in leakage monitoring, thereby realizing low-cost, high-frequency, and wide-coverage continuous scanning of urban water pipe networks and enhancing practicality. In this invention, different analysis modes are used to preventively detect leaks in water pipe networks for different pairing relationships, so as to discover early minor leaks and issue warnings. By combining cross-sectional comparison of multi-shift data and historical baseline models, the accuracy of leak detection is improved and misjudgments caused by environmental interference are reduced. Relying on the high-frequency operation characteristics of buses, real-time data collection and leak detection are achieved, thereby improving leak detection efficiency. In this invention, by combining abnormal sensor data from the water system, abnormal pipelines are located through sensor anomalies. The abnormal pipelines are then matched with the missed detection paths in the data fusion and pairing module to form an auxiliary judgment network. Based on the missed detection judgment results of the pipeline segment locations under the missed detection paths, the abnormal location coordinates of the current abnormal pipeline are determined in reverse, forming a three-dimensional water pipeline monitoring network that combines points and surfaces, thus enhancing functionality. Attached Figure Description

[0015] Figure 1 This is a block diagram of the urban multi-source data element quality perception and fusion system of the present invention; Figure 2 This is a flowchart of the urban multi-source data element quality perception and fusion method of the present invention; Figure 3 This is a block diagram of the operation of the preventive missed detection module of the urban multi-source data element quality perception and fusion system of the present invention. Detailed Implementation

[0016] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1: like Figure 1 As shown, the urban multi-source data element quality perception and fusion system includes a data fusion and matching module, a data acquisition module, a preventive missed detection module, and a pipeline anomaly auxiliary judgment module. The data fusion and matching module collects bus route maps and water pipeline network route maps under the city's jurisdiction, divides the detection radius of the water pipeline network route, filters the bus route segments around the detection radius, records the matching relationship between the water pipeline network route and the bus route segment, establishes a city water pipeline network missed detection database, establishes archives and records the matching relationship of different water pipeline network routes in the city, and merges continuously matched bus route segments as missed detection paths. The basic parameters of the current urban water supply network are collected through the urban water supply network GIS database, including the network line number, type, spatial coordinates, pipe diameter, material, burial depth, and service life. The current urban bus line data is obtained through the urban bus line operation map, including the bus line number, trajectory coordinates, stops, and information on vehicles under the line. It should be noted that urban water supply network data and urban bus route data can be obtained through official channels such as the Urban Public Information Data Service Center, which divides the stops of the same route into bus route segments.

[0018] The detection radius of different pipeline routes is divided according to the type of pipeline network. Combined with urban bus route data, pipeline routes are paired with bus routes, including direct and indirect matching relationships. A database of missed detections in urban water supply networks is established using MySQL, creating files for urban pipeline routes and recording their pairing relationships. The specific steps include: The detection radius *r* of different pipeline network types is determined based on the network type, which includes main pipelines, secondary pipelines, and branch pipelines. For each pipeline segment *e*, its buffer zone is calculated using a GIS buffer zone generator operator. L e The geometric line string representing pipe segment e, p is any point on the two-dimensional plane, R 2 Represents the two-dimensional Cartesian plane. Representative point p to pipe segment line L e The shortest vertical distance; It should be noted that the detection radius of the pipeline network is preset based on the main pipe, secondary pipe, and branch pipe types. The specific preset value needs to be set by consulting experts in the field, taking into account the parameters of the main pipe, secondary pipe, and branch pipe in the actual pipeline network. 2 Representing the two-dimensional Cartesian plane, i.e., the geographic projection plane, buffer zone A e That is, the tubular neighborhood range, L e The geometric line string representing pipe segment e is a continuous sequence of geographic coordinates, where e represents a single pipe segment.

[0019] For each bus route segment Where i is the bus route number and u is the bus route segment number, calculate its path string. The spatial relationship between each pipe segment e and the bus route segment is used to distinguish between direct and indirect matching relationships. For each pipe segment e, all bus routes are traversed sequentially to calculate direct and indirect matching relationships. When the pipe segment and the bus route segment meet the direct matching threshold condition, they are directly determined to be in direct matching relationships, and indirect matching calculations in the buffer zone are no longer performed. If the direct matching standard is not met, the indirect matching determination is completed by analyzing the spatial intersection between the bus route segment and the pipe segment buffer zone. A leak detection database for urban water pipe network is established using MySQL. An archive is created for each pipe segment, and all matched bus route segments and their matching types are recorded. Continuous bus route segments are merged as the matching leak detection path for each pipe segment. An archive is created separately, and each leak detection path is divided into pipe segment positions according to the length of a single detection unit. The direct matching relationship is calculated as the Hausdorf distance between the bus route and the segment, i.e.: ; in, The binary result represents a direct match between the representative pipe section and the bus route section. These represent the pipe segment and the bus route segment path, respectively, and ε represents the linear overlap distance threshold. Hausdorf distance representing the section of the pipeline and the bus route; It should be noted that the linear overlap distance threshold is usually set to 1.5m, but it can also be preset according to the actual situation. The bus route is a geometric line string based on the road centerline by default. Hausdorff distance is used to measure the overall geometric closeness of two lines. In the binary matching result, 1 represents a successful match and 0 represents a failed match.

[0020] Indirect matching is calculated by measuring the area of ​​the intersection between the bus route and the buffer zone of the management segment, i.e.: ; in, The binary result represents the indirect matching between the pipeline segment and the bus route, with Area representing the area of ​​the intersecting region.

[0021] It should be noted that, This indicates that there is a valid overlap between the two, as shown in the water supply network map. In the vertex set Represents network nodes, edge sets It is a collection of pipeline segments, each segment By recording all matching bus route segments and their matching types, and merging consecutive bus route segments, the missed detection path can be obtained.

[0022] Example 2: The data acquisition module, including an inertial measurement unit, a high-definition camera, and a pressure sensor, establishes a file in the urban water supply network leak detection database using bus routes as an index, and creates sub-files within the file using the license plates of buses belonging to the bus routes as an index. The inertial measurement unit deployed on the bus collects bus operation data, and the high-definition camera deployed on the bus collects data on the surrounding environment during bus operation. The collected data is recorded in the corresponding bus sub-file. Specifically, the module includes the following steps: An inertial measurement unit is installed in the chassis or middle of the bus to collect vehicle operating status data. High-definition cameras are installed on both sides of the top of the bus to collect image data of the surrounding environment during vehicle operation. The operating status data includes the vehicle's three-axis acceleration, angular velocity, and attitude angle. The real-time load of the vehicle during operation is collected through the suspension system pressure sensor. It should be noted that the vehicle's BeiDou positioning coordinates are uploaded simultaneously when collecting data, forming an integrated record of timestamp, location, and data. The sampling frequency of the inertial measurement unit (IMU) is usually set to 100Hz, but it can be adjusted according to actual needs. Attitude angles include pitch, roll, and yaw. High-definition cameras typically collect images of the surrounding environment at 20fps. By recording the pressure values ​​of each sensor when the vehicle has no passengers and no additional load, an empty reference pressure vector is established. Then, standard weights or objects of known weight are loaded, and the pressure values ​​under different load levels are recorded to establish a pressure-load mapping model. This enables the collection of real-time load data of the vehicle during operation through the suspension system pressure sensors. During operation, the sensors continuously collect suspension system pressure data, and the current total load of the vehicle is calculated using the calibrated pressure-load model.

[0023] In the urban water supply network leak detection database, a primary archive index is established by bus route ID. The archive stores the basic information of the route, including route trajectory coordinates and stops. Under each bus route archive, the bus license plate is used as a secondary index to create an independent sub-archive for each bus, recording the license plate ID, the route it belongs to, the equipment number, the data collection timestamp, the inertial measurement unit data record, and the camera image record.

[0024] It should be noted that the data undergoes preliminary processing at the in-vehicle terminal, such as noise reduction and compression, before being uploaded to the cloud database via 5G / 4G network.

[0025] Example 3: like Figure 3 As shown, the preventive missed detection module, based on the collected data from different missed detection paths in the urban water pipe network missed detection database, combines the collected data under the same missed detection path with pairing relationships to preventively mark abnormal pipe section locations under the missed detection path. Specifically, it includes the following steps: By utilizing the urban waterworks network leak detection database, matching relationships between different network segments are extracted. For leak detection paths that are directly matched with the current segment, bus vehicle data on the current leak detection path under normal network conditions is collected. Vehicle-leak detection path baseline models are constructed for different leak detection paths. By monitoring the vibration residuals of different bus routes on the same leak detection path, preventative leak detection markings are performed on pipe segment locations with abnormalities under the leak detection path. The specific steps include: The original triaxial acceleration signal is subjected to gravity component removal, and a pure vibration acceleration signal is obtained by bandpass filtering. Time-frequency analysis is performed on the acceleration signal of a single detection unit with a length Δl, including short-time Fourier transform and frequency band energy calculation of characteristic frequency bands, to form an output feature vector. The specific steps include: Short-time Fourier transform is performed on the acceleration signal within the length Δl of a single detection unit: ; in, This represents the time-frequency complex amplitude of the acceleration signal along the k-th axis at frequency f and time τ. This represents the time-domain acceleration signal after single-axis filtering. τ represents the Hamming window function, j represents the center time of the sliding time window, and k represents the imaginary unit. It should be noted that when removing the gravity component from the original triaxial acceleration signal: ; in It is the three-axis acceleration vector of the body coordinate system after eliminating the coupling between gravity and attitude. This is the raw triaxial acceleration observation vector acquired by the IMU, where g is the constant vector of gravitational acceleration. It is a real-time attitude rotation matrix used for coordinate decoupling. The lower limit of the bandpass filter frequency is 5Hz, and the upper limit of the frequency is 50Hz. The length of a single detection unit needs to be set according to the route conditions and by consulting experts in the field.

[0026] Bandwidth energy ,in Represents the vibrational energy of the characteristic frequency band of the k-th axis, and the lower limit of the characteristic frequency band. Upper limit The values ​​are 10 and 30 respectively, to obtain the output feature vector. : ,in Represents the energy of the three-axis frequency band. Represents the time-domain standard deviation of triaxial acceleration. , , Represents the time-domain kurtosis of triaxial acceleration; It should be noted that feature vectors As the output feature value for subsequent model construction, it is the core predicted variable. The purpose is to isolate the vibration anomaly caused purely by changes in the pipeline structure state by eliminating the influence of vehicle and environmental changes. The frequency domain features are obtained through short-time Fourier transform. The standard deviation in the time domain represents the three-axis acceleration. It is calculated directly on the filtered acceleration signal. The time domain kurtosis is obtained by first calculating the standardized signal and then calculating the excess kurtosis for each axis.

[0027] Collect bus data on currently missed detection paths under normal pipeline network conditions, and construct model inputs: ; in, This represents the multidimensional input feature vector of the i-th sample. These represent the vehicle's real-time estimated load, the vehicle's real-time speed, hourly rainfall, time of day, and pipe section location, respectively. It should be noted that hourly rainfall can be obtained from real-time information released by the meteorological station on the day of data collection. The pipe segment location refers to different pipe segments of the same pipeline that are pre-divided according to a fixed interval. The fixed interval needs to be consistent with the length of a single detection unit. The historical data collection period is usually set to three months. It is necessary to collect data under different conditions as comprehensively as possible to train the adaptation model. In practice, the collection period can be extended or shortened as needed.

[0028] A Transformer encoder is used as the model architecture to train a vehicle-missed detection path baseline model, and the matched output feature vectors are then used. Input feature vector The training set, validation set, and test set are divided into a ratio of 7:1.5:1.5. The AdamW optimizer is used to complete parameter iteration. In each round of training, forward propagation of samples, loss calculation, back propagation, and network parameter update are completed. After each round of iteration, the model prediction error is quantified on the validation set. When the validation loss does not decrease significantly in N rounds, the training is automatically terminated to obtain the vehicle-missed detection path baseline model. It should be noted that during model training, the 9-dimensional vibration statistical features obtained from the vehicle-mounted inertial measurement unit were used as supervision labels. Training was conducted using only normal operating data from the pipeline network. A multi-layer Transformer encoder structure was built, with a multi-head self-attention mechanism and a residual normalization module. The feature dimension was unified through the embedding layer, and the final output layer was mapped to a 9-dimensional vibration prediction vector using a fully connected layer. The training loss employed joint constraint loss, combining weighted mean square error regression loss and spatial smoothing regularization loss. The AdamW optimizer was used for parameter iteration, with a fixed batch size, initial learning rate, and weight decay coefficient. Cosine annealing was used for learning rate scheduling. The model underwent iterative training in rounds, with each round including sample forward propagation, loss calculation, backpropagation, and network parameter updates. After each iteration, the model was switched to evaluation mode, and the prediction error was quantified on the validation set for continuous monitoring. The validation set loss is varied, and training automatically terminates when the validation loss does not decrease significantly in N rounds. The model weights with the optimal validation loss are retained as the final inference model. The number of rounds needs to be preset based on the length of the missed detection path using an empirical method. The Transformer encoder contains F encoder layers connected in series, where F is an integer greater than 1, with a preferred value of 6. Each encoder layer contains a multi-head self-attention sublayer and a feedforward neural network sublayer. The number of attention heads in the multi-head self-attention sublayer is preferably 8. Each attention head maps the input sequence to a different representation subspace and computes self-attention in parallel. Its output is concatenated and linearly transformed to generate the output of the corresponding sublayer. The feedforward neural network sublayer is a two-layer fully connected network with a preferred hidden layer dimension of 1024 and an input-output dimension of 256. Residual connections and layer normalization operations are applied after each sublayer to ensure the stability of the training process.

[0029] The input features are fed into the vehicle-missed detection path baseline model to obtain the conditional predicted distribution of the pipe segment vibration characteristics. The model output follows a Gaussian predicted distribution. ,in The input feature vector representing the new observed sample, The measured vibration feature vector corresponding to this sample The mean vector of vibration characteristics predicted by the model. The covariance matrix representing the prediction is normalized to the residuals using the squared Mahalanobis distance, under the null hypothesis and steady-state conditions. Satisfying the chi-square distribution q represents the characteristic dimension of pipe segment vibration; It should be noted that q represents the characteristic dimension of the pipe segment vibration, which is 9 here, and the squared residual of the Mahalanobis distance. .

[0030] For the same missed detection path pipe segment location within day t, n t Mean residuals of each shift ,in Let represent the standardized residual of the i-th observation on day t. Divide the vehicles into K groups according to their load, and calculate the average residual for each group to verify consistency. ; in, The function is a 0-1 indicator function, where α is the significance level. This indicates that different load-bearing vehicles simultaneously exceed the residual limit in the same pipe section. When the residual limit exceeds the limit for a cumulative period of 7 days, the current pipe section location is marked as the location to be prevented for leak detection in the leak detection path file of the urban water supply network leak detection database. It should be noted that α is the significance level, which is usually taken as 0.05. The value of K, representing the upper quantile of the chi-square distribution (1-α), is typically preferred to be 3. This setting divides vehicles into three representative working condition groups: "light load," "medium load," and "heavy load." This effectively characterizes the main influence mode of load on vehicle-road vibration response and ensures that each group has a sufficient number of vehicle samples to meet the basic requirements of statistical testing, given the typical number of daily shifts. The specific grouping rules can be based on the distribution of historical load data, divided according to percentiles. That is, Group 1 consists of vehicles with loads below the 33rd percentile, Group 2 consists of vehicles with loads between the 33rd and 67th percentiles, and Group 3 consists of vehicles with loads above the 67th percentile. For routes with complex vehicle types and a wide range of loads, the value of K can be increased to 4-6 to improve the sensitivity of condition discrimination and the ability to perform hierarchical evaluation.

[0031] The daily residual series is smoothed using a 7-day moving average, and the moving average is calculated as follows: τ is the number of backtracking days, used to smooth the sequence. For the object, the Mann-Kendall nonparametric test is used to identify long-term degradation trends, where the trend statistic is... Where T is the total number of days in the time series, sgn(x) is the sign function, taking the value 1 for x > 0, 0 for x = 0, and -1 for x < 0. Calculate the corrected variance. Where g is the number of sequence nodes, t p Let Z represent the number of samples in the p-th node. Calculate the standardized test statistic Z: ; Under a one-sided significance test, when Furthermore, if the conditions persist for 14 days, it is determined that the residual of the pipe segment corresponding to the missed detection path shows an upward deterioration trend. The current pipe segment location is marked as the location to be prevented for missed detection in the missed detection path file of the urban water pipe network missed detection database.

[0032] It should be noted that for the marked locations to be prevented from being detected, on-site inspections are conducted by dispatching personnel to prevent leaks in the pipeline network. By analyzing residuals and long-term trends, the system can capture signal drift caused by the slow deterioration of underground structures from complex urban noise, thus enabling preventive judgment.

[0033] For missed detection paths that are indirectly matched with the current pipe segment, the following steps are taken: Images of the missed detection path with consecutive timestamps are extracted and compared horizontally. Shadow changes in the consecutive images are determined, and preventative missed detection markers are used to identify abnormal pipe segment locations along the missed detection path. Images of pipe segments with different missed detection paths under indirect matching relationships in the urban water supply network leak detection database are extracted, sorted by timestamp, and obtained as a continuous image string. It should be noted that during the image selection process, GPS positioning constraints are used to ensure that images from different shifts are collected from the same spatial range. ,in These represent the coordinates of the two images taken. The coordinate range threshold needs to be determined by combining the vehicle's speed and the number of images captured per second by the high-definition camera when acquiring images of the surrounding environment. At the same time, the camera needs to be calibrated and distortion corrected before image acquisition.

[0034] The image string is converted to HSV color space, the shadow area in the image is calculated, and based on the rate of change of shadow area in consecutive images and the rate of change of shadow area in interval images, the abnormal pipe segment location under the missed detection path is determined, and preventive missed detection marking is performed. The specific steps include: The brightness and saturation in the image are: Where V(x,y) represents the brightness of pixel (x,y), S(x,y) represents the saturation of pixel (x,y), and H(x,y) represents the hue of pixel (x,y). and represent the mean and standard deviation of the hue components of the shadow area, respectively, where the shadow area A is: ; in, The threshold for determining the shadow index is represented by Ω, which represents the set of all pixels in the image. It should be noted that the threshold for determining the shadow index is automatically calculated using the Otsu method, which is the threshold that maximizes the variance of the shadow class and the non-shadow class. It is a 0-1 indicator function.

[0035] When the rate of change of the shadow area of ​​a continuous image Rate of change of shadow area in interval images At that time, the current pipe section location is recorded as the location for pending preventative missed detection. Let represent the shadow area in frame t and the shadow area in frame (t-1), respectively, where t represents the time index of the current frame and k is the interval number of frames. Represents a minimal constant. , These represent the threshold for determining continuous shadow anomalies and the threshold for determining interval shadow anomalies, respectively. Sensor data for potential preventative leak detection locations are extracted from the water system. Anomalies in sensor data for consecutive frames and interval frames corresponding to these locations are statistically analyzed. Locations where both sensor data anomalies and abnormal shadow area change rates in consecutive or interval images are identified are marked as potential preventative leak detection locations in the leak detection path archive of the urban water network leak detection database.

[0036] It should be noted that k represents the frame interval, which needs to be set based on the number of images taken daily at the same pipe section location. Typically, the interval should be between 1 and 2 days to identify potential water seepage within the area. , These represent the threshold values ​​for continuous shadow anomalies and interval shadow anomalies, respectively, used to pre-determine rapid and slow leakage in the pipeline network. The specific thresholds can be determined using an labeled leakage sample set and the ROC curve method to find the optimal value, or set based on engineering experience. Here, the area change rate is a positive rate of change, meaning it only applies when... The judgment is triggered only at certain times, and the range of shadows is uncertain. It is necessary to combine sensor data from continuous frames and interval frames of the pending preventive leak detection location in the water system for targeted joint judgment. By analyzing sensor abrupt changes caused by sudden leakage in the pipeline network, such as a sudden drop in pressure or a sudden increase in flow, and combining them with the pending preventive leak detection location, the specific location of the actual pipeline network leakage can be determined. For the sensor data in the interval frames, by capturing long-term sensor offsets caused by slow leakage, such as a continuous drop in pressure or a slow increase in flow, and combining them with the pending preventive leak detection location, the specific location of the actual pipeline network leakage can be determined.

[0037] The pipeline anomaly auxiliary judgment module targets pipelines with abnormal monitoring data in the water system. It queries the urban water network missed detection database for the corresponding missed detection path, and uses a preventative missed detection module to assist in the judgment of suspected abnormal sections, outputting the location coordinates of the abnormal sections. The specific steps include: In response to the anomalies of network data monitoring sensors in the water system, the pipeline numbers involved in the sensor anomalies are extracted, and the missing detection path combinations under the corresponding pipeline numbers are extracted from the urban water network missing detection database to assist in the anomaly judgment of the missing detection paths. For the pipeline numbers involved in the obtained anomalies, all missed detection paths under the corresponding pipeline numbers are extracted from the urban water supply network missed detection database. Data collected from buses with different license plates in the bus routes involved in the missed detection paths are extracted. Through the direct matching relationship judgment process and the indirect matching relationship judgment process in the preventive missed detection module, the locations of the pipe segments under all missed detection paths are judged as anomalies. The locations to be prevented and detected are upgraded to abnormal missed detection locations, and the coordinates of the abnormal pipe segment locations are obtained.

[0038] It should be noted that, for abnormal sensor signal pipelines in the water system, the data fusion and matching module obtains the missed detection path corresponding to the currently existing abnormal sensor signal pipeline. That is, the current abnormal sensor signal pipeline is obtained by combining the missed detection paths. The collected data under the missed detection path is used to determine the corresponding missed detection path pairing relationship through the preventive missed detection module. Through the direct matching and indirect matching judgment process, the pipe segment positions included in each missed detection path are screened for anomalies. The locations to be preventive missed detection are directly upgraded to abnormal missed detection locations, and the location coordinates of the abnormal pipe segments are output to assist in subsequent manual pipeline maintenance.

[0039] Example 4: like Figure 2 As shown, the method for quality perception fusion of multi-source data elements in cities includes the following steps: S1. Collect bus route maps and water pipeline network maps under the jurisdiction of the city, divide the detection radius of the water pipeline network, screen the bus route segments around the detection radius, record the pairing relationship between water pipeline network and bus route segments, establish a city water pipeline network missed detection database, establish files and record the pairing relationship of different water pipeline network lines in the city, and merge continuously matched bus route segments as missed detection paths. S2. Establish archives in the urban water supply network leak detection database using bus routes as the index, and establish sub-archives in the archives using the license plates of buses under the bus routes as the index. Collect bus operation data through inertial measurement units and pressure sensors deployed on buses, and collect surrounding environmental data during bus operation through high-definition cameras deployed on buses. Record the collected data in the corresponding bus sub-archives. S3. Through the urban water supply network leak detection database, extract the matching relationship of different network segments. For leak detection paths that are directly matched with the current segment, collect bus data on the current leak detection path under normal network conditions. Construct vehicle-leak detection path baseline models for different leak detection paths. By monitoring the vibration residuals of different bus routes on the same leak detection path, preventive leak detection marking is carried out for abnormal pipe segment positions under the leak detection path. S4. Through the urban water supply network leak detection database, extract the matching relationship of different network segments. For leak detection paths that are indirectly matched with the current segment, extract and compare the continuous timestamp images of the leak detection path, determine the shadow changes of the continuous images, and mark the abnormal position of the segment under the leak detection path for preventive leak detection. S5. For abnormal pipelines in the water system, query the corresponding missed detection path in the urban water network missed detection database, and make auxiliary judgments on the suspicious abnormal road sections based on the matching relationship of the missed detection paths, and output the location coordinates of the abnormal road sections.

[0040] The present invention relates to a method and system for quality perception and fusion of urban multi-source data elements. When in use, it relies on urban public transportation to perform multi-source data fusion of urban water information. Different preventive leakage detection modes are adopted for water pipe networks and bus routes with different pairing relationships. The operation trajectory of buses is used to achieve large-scale coverage of water pipe networks to assist in leakage monitoring, thereby realizing low-cost, high-frequency, and wide-coverage continuous scanning of urban water pipe networks. Different analysis modes are used to preventively detect leaks in the water supply network under different pairing relationships, in order to discover and warn of early minor leaks. By combining cross-shift data comparison and historical baseline models, the accuracy of leak detection is improved and misjudgments caused by environmental interference are reduced. Relying on the high-frequency operation characteristics of buses, real-time data collection and leak detection are achieved, improving the efficiency of leak detection. By combining abnormal sensor data in the water supply system, abnormal pipelines are located through sensor anomalies. The abnormal pipelines are matched with the leak detection paths in the data fusion pairing module to form an auxiliary judgment network. Based on the leak detection judgment results of the pipeline segment location under the leak detection path, the abnormal location coordinates of the current abnormal pipeline are determined in reverse, forming a three-dimensional water supply network monitoring network that combines points and surfaces, thus enhancing functionality.

[0041] In the embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the method in this embodiment according to actual needs.

[0042] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A city multi-source data element quality perception and fusion system, characterized in that: It includes a data fusion and pairing module, a data acquisition module, a preventative missed detection module, and a pipeline anomaly auxiliary judgment module; The data fusion and matching module collects bus route maps and water pipeline network route maps under the city's jurisdiction, divides the detection radius of the water pipeline network route, filters bus route segments around the detection radius, records the matching relationship between water pipeline network route and bus route segment, establishes a city water pipeline network missed detection database, establishes archives and records the matching relationship of different water pipeline network routes in the city, and merges continuously matched bus route segments as missed detection paths. The data acquisition module establishes a file in the urban water supply network leak detection database using bus routes as the index, and establishes sub-files in the file using the license plates of buses under the bus routes as the index. It collects bus operation data through inertial measurement units and pressure sensors deployed on buses, and collects surrounding environmental data during bus operation through high-definition cameras deployed on buses, and records the collected data in the corresponding bus sub-files. The preventive omission detection module, based on the collected data of different omission paths in the urban water pipe network omission detection database, combines the collected data under the same omission path with the pairing relationship, and performs preventive omission detection marking for abnormal pipe section locations under the omission path; The pipeline anomaly auxiliary judgment module, for pipelines with abnormal monitoring data in the water system, queries the corresponding missed detection path in the urban water network missed detection database, and uses the preventive missed detection module to assist in the judgment of suspicious abnormal road sections, and outputs the location coordinates of the abnormal road sections.

2. The urban multi-source data element quality perception and fusion system according to claim 1, characterized in that: The data fusion and pairing module includes the following steps: The basic parameters of the current urban water supply network are collected through the urban water supply network GIS database, including the network line number, type, spatial coordinates, pipe diameter, material, burial depth, and service life. The current urban bus line data is obtained through the urban bus line operation map, including the bus line number, trajectory coordinates, stops, and information on vehicles under the line. The detection radius of different pipeline lines is divided according to the type of pipeline network. Combined with urban bus route data, the pipeline lines are matched with bus routes, including direct matching relationships and indirect matching relationships. A database of missed detections of urban water pipeline networks is established through MySQL to create files for urban pipeline lines and record the matching relationships of pipeline lines.

3. The urban multi-source data element quality perception and fusion system according to claim 2, characterized in that: The method involves classifying the detection radius of different pipeline routes based on their type, combining urban bus route data, and pairing pipeline routes with bus routes, including direct and indirect matching relationships. A database of missed detections in urban water supply networks is established using MySQL to create files for urban pipeline routes and record their pairing relationships. The specific steps include: The detection radius of different pipelines is divided according to the type of pipeline, which includes main pipelines, secondary pipelines and branch pipelines. For each pipeline segment, the buffer is calculated by generating an operator through the GIS buffer. For each bus route segment, the spatial relationship between its path string and each pipe segment is calculated to distinguish between direct and indirect matching relationships. For pipe e, all bus routes are traversed sequentially to calculate direct and indirect matching relationships. When the pipe segment and the bus route segment meet the direct matching threshold condition, they are directly determined to be in a direct matching relationship, and no further buffer indirect matching calculation is performed. If the direct matching standard is not met, the indirect matching determination is completed through spatial intersection analysis between the bus route segment and the pipe segment buffer. A city waterworks network leak detection database is established using MySQL, and an archive is created for each pipe segment, recording all matched bus route segments and their matching types. Consecutive bus route segments are merged as the matching leak detection path for each pipe segment, and an archive is created separately. Each leak detection path is divided into pipe segment locations according to the length of a single detection unit.

4. The urban multi-source data element quality perception and fusion system according to claim 2, characterized in that: The data acquisition module specifically includes the following steps: An inertial measurement unit is installed in the chassis or middle of the bus to collect vehicle operating status data. High-definition cameras are installed on both sides of the top of the bus to collect image data of the surrounding environment during vehicle operation. The operating status data includes the vehicle's three-axis acceleration, angular velocity, and attitude angle. The real-time load of the vehicle during operation is collected through the suspension system pressure sensor. In the urban water supply network leak detection database, a primary archive index is established by bus route ID. The archive stores the basic information of the route, including route trajectory coordinates and stops. Under each bus route archive, the bus license plate is used as a secondary index to create an independent sub-archive for each bus, recording the license plate ID, the route it belongs to, the equipment number, the data collection timestamp, the inertial measurement unit data record, and the camera image record.

5. The urban multi-source data element quality perception and fusion system according to claim 4, characterized in that: The preventive missed detection module specifically includes the following steps: By using the urban waterworks pipeline network leak detection database, the matching relationship between different pipeline segments is extracted. For leak detection paths that are directly matched with the current pipeline segment, bus vehicle data on the current leak detection path under normal pipeline network conditions is collected. Vehicle-leak detection path baseline models are constructed for different leak detection paths. By monitoring the vibration residuals of different bus routes on the same leak detection path, preventive leak detection marking is carried out on pipeline segment location anomalies under the leak detection path. For missed detection paths that are indirectly matched with the current pipe segment, the images of the missed detection path are extracted and compared horizontally. The shadow changes of the continuous images are judged, and the abnormal positions of the pipe segments under the missed detection path are marked for preventive missed detection.

6. The urban multi-source data element quality perception and fusion system according to claim 5, characterized in that: The process involves extracting matching relationships between different pipe segments from the urban waterworks network leak detection database. For leak detection paths that are directly matched with the current pipe segment, bus data on the current leak detection path under normal network conditions is collected. Vehicle-leak detection path baseline models are constructed for different leak detection paths. By monitoring the vibration residuals of different bus routes on the same leak detection path, preventative leak detection markings are performed on pipe segment locations with abnormalities under the leak detection path. The specific steps include: Gravity components are removed from the original triaxial acceleration signal, and bandpass filtering is used to obtain a pure vibration acceleration signal. Time-frequency analysis is performed on the acceleration signal of a single detection unit with a length Δl, including short-time Fourier transform and frequency band energy calculation of characteristic frequency bands, to form an output feature vector. Data on buses on currently missed detection routes under normal pipeline network conditions is collected. A Transformer encoder is used as the model architecture to train a vehicle-missed detection route baseline model. Input features are fed into the vehicle-missed detection route baseline model to obtain the conditional prediction distribution of pipeline segment vibration characteristics. The residuals are standardized using Mahalanobis distance squared, and the results are used to predict the vibration characteristics of the same missed detection route pipeline segment within n days t. t The average residual of each shift is used to divide the vehicles into K groups according to their load. The average residual of each group is calculated to verify consistency. When the residual exceeds the standard for a cumulative period of 7 days, the current pipe section is marked as the location to be prevented for leak detection in the leak detection path file of the urban water supply network leak detection database. The daily residual series was smoothed using a 7-day moving average. Long-term deterioration trends were identified using the Mann-Kendall nonparametric test. The standardized test statistic Z was calculated. Under a one-sided significance test, when... Furthermore, if the conditions persist for 14 days, it is determined that the residual of the pipe segment corresponding to the missed detection path shows an upward deterioration trend. The current pipe segment location is marked as the location to be prevented for missed detection in the missed detection path file of the urban water pipe network missed detection database.

7. The urban multi-source data element quality perception and fusion system according to claim 6, characterized in that: For missed detection paths that are indirectly matched with the current pipe segment, the following steps are taken: Images of consecutive timestamps taken along the missed detection path are extracted and compared horizontally; shadow changes in the consecutive images are determined; and preventative missed detection markers are used to identify abnormal pipe segment locations along the missed detection path. Images of pipe segments with different missed detection paths under indirect matching relationships in the urban water supply network leak detection database are extracted, sorted by timestamp, and obtained as a continuous image string. HSV color space conversion is performed on the image string, the shadow area in the image is calculated, and the abnormal pipe segment location under the missed detection path is determined based on the shadow area change rate of continuous images and the shadow area change rate of interval images, and preventive missed detection marking is performed. When the rate of change of shadow area in continuous images or the rate of change of shadow area in interval images exceeds a set threshold, the current pipe segment location is recorded as a pending preventive leak detection location. Sensor data for the pending preventive leak detection location is extracted through the water system, and abnormal sensor data in continuous frames and interval frames corresponding to the pending preventive leak detection location are statistically analyzed. Pending preventive leak detection locations that simultaneously exhibit abnormal sensor data and abnormal shadow area change rate in either continuous or interval images are marked as pending preventive leak detection locations in the leak detection path file of the urban water pipe network leak detection database.

8. The urban multi-source data element quality perception and fusion system according to claim 7, characterized in that: The pipeline anomaly auxiliary judgment module specifically includes the following steps: In response to anomalies in the network data monitoring sensors of the water system, the pipeline numbers involved in the sensor anomalies are extracted, and the missing detection path combinations under the corresponding pipeline numbers are extracted from the urban water network missing detection database to assist in the anomaly determination of the missing detection paths.

9. The urban multi-source data element quality perception and fusion system according to claim 7, characterized in that, The method for addressing anomalies in network data monitoring sensors within the water system involves extracting the pipeline numbers associated with the sensor anomalies, retrieving the corresponding missed detection path combinations from the urban water network missed detection database, and then performing anomaly assessment on the missed detection paths. This process specifically includes the following steps: For the pipeline numbers involved in the obtained anomalies, all missed detection paths under the corresponding pipeline numbers are extracted from the urban water supply network missed detection database. Data collected from buses with different license plates in the bus routes involved in the missed detection paths are extracted. Through the direct matching relationship judgment process and the indirect matching relationship judgment process in the preventive missed detection module, the locations of the pipe segments under all missed detection paths are judged as anomalies. The locations to be prevented and detected are upgraded to abnormal missed detection locations, and the coordinates of the abnormal pipe segment locations are obtained.

10. A method for quality perception and fusion of multi-source data elements in urban areas, characterized in that, The method employs the urban multi-source data element quality perception fusion system as described in any one of claims 1-9.

Images