Intelligent signal lamp fault diagnosis method and system based on internet of things

Abstract

The application provides a signal lamp fault intelligent diagnosis method and system based on the Internet of Things, relates to the technical field of data processing, and comprises the following steps: 1, collecting electrical parameters, environmental data, working time sequence information and video image data of a signal lamp unit in real time, and constructing a multi-modal data set of the signal lamp operation state; 2, transmitting the multi-modal data set to a central diagnosis platform, performing feature extraction and preliminary fault identification by using a pre-trained fault identification model, and generating fault type and fault position information; and 3, based on the fault position information, selecting a plurality of monitoring devices associated with time and space to form a diagnosis group, and performing fusion analysis and collaborative diagnosis on the multi-source data of the diagnosis group to generate a fault determination result. The application realizes intelligent diagnosis and operation and maintenance management of the signal lamp operation state, and improves the accuracy and disposal efficiency of fault identification.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for intelligent diagnosis of traffic light faults based on the Internet of Things. Background Technology

[0002] In the field of intelligent traffic management, monitoring and diagnosing the operational status of traffic lights is a crucial aspect of ensuring road traffic safety and efficiency. Currently, most existing monitoring solutions rely on the collection and analysis of electrical parameters of traffic light units to identify faults. However, in practical applications, the data analysis dimensions and synergy of these methods may be relatively limited. For example, when a system judges a traffic light fault solely based on abnormal current, it may be difficult to effectively distinguish whether the fault is due to damage to the light source itself or to instantaneous fluctuations caused by poor wiring contact, environmental interference, or other factors. This single-dimensional judgment criterion may sometimes affect the accuracy of fault location and may also limit the assessment of the impact of the fault.

[0003] In addition, in the management and handling of fault information, some existing systems have limitations in the level of intelligence in terms of the structured presentation of alarm information and the automated scheduling of operation and maintenance resources. Sometimes, operation and maintenance personnel need to manually compare information from different systems to make a comprehensive judgment, which may affect the timeliness of operation and maintenance response to some extent. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for intelligent diagnosis of traffic light faults based on the Internet of Things, so as to realize intelligent diagnosis and operation and maintenance management of traffic light operating status, and improve the accuracy of fault identification and handling efficiency.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] Firstly, a smart fault diagnosis method for traffic lights based on the Internet of Things (IoT) includes:

[0007] Step 1: Collect electrical parameters, environmental data, working timing information and video image data of the traffic light unit in real time to construct a multimodal dataset of traffic light operating status;

[0008] Step 2: Transmit the multimodal dataset to the central diagnostic platform, use the pre-trained fault identification model to perform feature extraction and preliminary fault identification, and generate fault type and fault location information;

[0009] Step 3: Based on the fault location information, select multiple monitoring devices that are spatially and temporally correlated to form a diagnostic group, and perform fusion analysis and collaborative diagnosis on the multi-source data of the diagnostic group to generate fault judgment results;

[0010] Step 4: Based on the fault determination results and combined with the fault level rule library preset in the platform knowledge base, the impact range is quantified by calculating the area enclosed by the boundary coordinates of the traffic impact area. The fault level and traffic impact range are automatically determined, structured fault alarm information is generated, and pushed to the visual decision-making dashboard of the command center for dynamic display in order to obtain intelligent handling strategies.

[0011] Step 5: Based on the intelligent handling strategy, generate a standardized electronic operation and maintenance work order, and assign the standardized electronic operation and maintenance work order to the mobile terminal of the corresponding operation and maintenance personnel to start the fault handling process.

[0012] Step 6: During the fault handling process, record diagnostic data and operation and maintenance response information, and use the operation and maintenance response information to adjust the diagnostic logic and handling strategy to realize intelligent diagnosis and operation and maintenance management of traffic light faults.

[0013] Secondly, the IoT-based intelligent fault diagnosis system for traffic lights includes:

[0014] The acquisition module is used to collect electrical parameters, environmental data, working timing information and video image data of the traffic light unit in real time, and to build a multimodal dataset of the traffic light's operating status.

[0015] The identification module is used to transmit multimodal datasets to the central diagnostic platform, use a pre-trained fault identification model to perform feature extraction and preliminary fault identification, and generate fault type and fault location information.

[0016] The diagnostic module is used to select multiple spatiotemporally correlated monitoring devices to form a diagnostic group based on fault location information, and to perform fusion analysis and collaborative diagnosis on the multi-source data of the diagnostic group to generate fault judgment results.

[0017] The judgment module is used to quantify the impact range by calculating the area enclosed by the boundary coordinates of the traffic impact area based on the fault judgment result and the fault level rule library preset in the platform knowledge base. It automatically determines the fault level and traffic impact range, generates structured fault alarm information, and pushes it to the visual decision-making dashboard of the command center for dynamic display in order to obtain intelligent handling strategies.

[0018] The assignment module is used to generate standardized electronic operation and maintenance work orders based on intelligent handling strategies, and assign the standardized electronic operation and maintenance work orders to the mobile terminals of the corresponding operation and maintenance personnel to start the fault handling process.

[0019] The management module is used to record diagnostic data and operation and maintenance response information during the fault handling process, and to adjust the diagnostic logic and handling strategy using the operation and maintenance response information to realize intelligent diagnosis and operation and maintenance management of traffic light faults.

[0020] Thirdly, a computing device, comprising:

[0021] One or more processors;

[0022] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0023] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0024] The above-described solution of the present invention has at least the following beneficial effects:

[0025] By integrating multimodal data such as electrical parameters, environmental data, time-series information, and video images, and combining this with the feature extraction capabilities of pre-trained models, the operational status of traffic lights can be analyzed more comprehensively, improving the accuracy of fault identification and location. Constructing diagnostic groups based on spatiotemporal correlation and performing multi-source data fusion analysis allows for verification of fault information from multiple dimensions, reducing the probability of misjudgment and improving the reliability of diagnostic results. Calculating the traffic impact area and combining it with a fault level rule base allows for a quantitative assessment of the degree of traffic impact of faults, providing a basis for the rational allocation of operation and maintenance resources.

[0026] By automatically generating standardized electronic work orders and assigning them to mobile terminals, the system achieves standardization and automation of fault handling processes, shortens response time, and improves operation and maintenance efficiency. By recording operation and maintenance response information and dynamically adjusting diagnostic logic and handling strategies, the system has self-learning capabilities and can continuously optimize diagnostic accuracy and operation and maintenance management level. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating the intelligent diagnostic method for traffic light faults based on the Internet of Things provided in an embodiment of the present invention.

[0028] Figure 2 This is a schematic diagram of an IoT-based intelligent diagnostic system for traffic light faults provided in an embodiment of the present invention. Detailed Implementation

[0029] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0030] like Figure 1As shown, embodiments of the present invention propose an intelligent diagnostic method for traffic light faults based on the Internet of Things, the method comprising the following steps:

[0031] Step 1: Collect electrical parameters, environmental data, working timing information and video image data of the traffic light unit in real time to construct a multimodal dataset of traffic light operating status;

[0032] Step 2: Transmit the multimodal dataset to the central diagnostic platform, use the pre-trained fault identification model to perform feature extraction and preliminary fault identification, and generate fault type and fault location information;

[0033] Step 3: Based on the fault location information, select multiple monitoring devices that are spatially and temporally related to form a diagnostic group, and perform fusion analysis and collaborative diagnosis on the multi-source data of the diagnostic group to generate fault judgment results;

[0034] Step 4: Based on the fault determination results and combined with the fault level rule library preset in the platform knowledge base, the impact range is quantified by calculating the area enclosed by the boundary coordinates of the traffic impact area. The fault level and traffic impact range are automatically determined, structured fault alarm information is generated, and pushed to the visual decision-making dashboard of the command center for dynamic display in order to obtain intelligent handling strategies.

[0035] Step 5: Based on the intelligent handling strategy, generate a standardized electronic operation and maintenance work order, and assign the standardized electronic operation and maintenance work order to the mobile terminal of the corresponding operation and maintenance personnel to start the fault handling process.

[0036] Step 6: During the fault handling process, record diagnostic data and operation and maintenance response information, and use the operation and maintenance response information to adjust the diagnostic logic and handling strategy to realize intelligent diagnosis and operation and maintenance management of traffic light faults.

[0037] In this embodiment of the invention, by fusing multimodal data such as electrical parameters, environmental data, time-series information, and video images, and combining the feature extraction capabilities of a pre-trained model, the operating status of traffic lights can be analyzed more comprehensively, improving the accuracy of fault identification and location. Constructing diagnostic groups based on spatiotemporal correlation and performing multi-source data fusion analysis can verify fault information from multiple dimensions, reducing the probability of misjudgment and improving the reliability of diagnostic results. By calculating the area of ​​the traffic impact zone and combining it with a fault level rule base, the degree of traffic impact of the fault can be quantitatively assessed, providing a basis for the rational allocation of operation and maintenance resources.

[0038] By automatically generating standardized electronic work orders and assigning them to mobile terminals, the system achieves standardization and automation of fault handling processes, shortens response time, and improves operation and maintenance efficiency. By recording operation and maintenance response information and dynamically adjusting diagnostic logic and handling strategies, the system has self-learning capabilities and can continuously optimize diagnostic accuracy and operation and maintenance management level.

[0039] In a preferred embodiment of the present invention, step 1 above, which involves real-time acquisition of electrical parameters, environmental data, operating timing information, and video image data of the traffic light unit to construct a multimodal dataset of the traffic light's operating status, may include:

[0040] In this embodiment of the invention, traffic light monitoring hardware is first deployed, including a traffic light monitoring host, a sub-module, and an active device monitoring module. The traffic light monitoring host is wall-mounted or rack-mounted, and the sub-module is mounted on a DIN rail. These hardware devices are connected to the traffic light power supply line to collect the voltage and current values ​​of each traffic light in real time, while simultaneously monitoring electrical conditions such as leakage current, overvoltage, undervoltage, and overcurrent. The data collection process is continuous to ensure real-time data. Temperature and humidity sensors, water level sensors, and door status sensors are installed inside the traffic light control box and pole box corresponding to the traffic light. The temperature and humidity sensors detect the temperature and humidity inside the box in real time, the water level sensors detect whether there is water ingress, and the door status sensors detect whether the door is abnormally opened. These sensors upload the collected environmental data in real time through a data interface or a wireless communication method, and the upload time is synchronized with the electrical parameter acquisition time to ensure spatiotemporal consistency.

[0041] By monitoring the control signals of the traffic lights through the traffic light monitoring host, the system can acquire the phase switching sequence of the traffic lights in real time, including the switching time of each color (red, green, and yellow) and the duration of each phase. Simultaneously, it captures the operating mode of the traffic lights, such as normal operation or flashing yellow. When the traffic light timing changes, the system automatically identifies and records the changed timing data, ensuring that the working timing information is updated synchronously with the actual operating status of the traffic lights. Video equipment such as IPC cameras are deployed at each intersection. The cameras are connected to a video aggregation server, transmitting real-time video streams to the backend via the network. While collecting electrical parameters, environmental data, and working timing information, the system simultaneously retrieves video images from the corresponding intersections in real time, focusing on capturing the actual on / off status and switching of the traffic lights, and adding timestamps to the video images to ensure consistency with the time dimension of other data types.

[0042] The above four types of data are transmitted to the integrated management and control platform via a private network or 4G network. The platform first timestamps all the data to link electrical parameters, environmental data, working sequence information and corresponding video frames at the same time point. Then, it unifies the data format, standardizes the units of electrical parameters such as volts, amperes, and milliamperes, the units of environmental data such as degrees Celsius and percentage humidity, the time format of working sequence information and the resolution and frame format of video images. Finally, the data is classified according to the intersection and traffic light number, and a dedicated data group is established for each traffic light at each intersection. The corresponding four types of data are integrated to form a multimodal dataset covering all dimensions of traffic light operation.

[0043] In a preferred embodiment of the present invention, step 2 above, which involves transmitting the multimodal dataset to the central diagnostic platform and using a pre-trained fault identification model for feature extraction and preliminary fault identification to generate fault type and fault location information, may include:

[0044] In this embodiment of the invention, step 220 involves preprocessing the multimodal dataset to generate regularized multimodal data, and then inputting the regularized multimodal data into a pre-trained fault identification model. Specifically, this includes: first, determining the four core data categories contained in the multimodal dataset: electrical parameter data of the traffic lights (voltage, current, power), environmental data (temperature, humidity, water immersion status, and door opening / closing status within the chassis), operating sequence information (duration of red light on / off, green light on / off, yellow light on / off, and switching interval), and video image data (real-time view of the traffic light panel). For the electrical parameter data, the system automatically filters out abnormal values ​​exceeding the equipment's rated range. Taking the common rated voltage of a traffic light (220V) as an example, values ​​exceeding... Values ​​outside the range of 176V to 264V (plus or minus 20% of the rated voltage) are considered outliers and replaced with the median of all valid values ​​for that parameter within the past 10 minutes. For missing values, if the missing duration is no more than 5 minutes, the average of the two consecutive valid values ​​before and after the missing period is used to fill the gap; if the missing duration is more than 5 minutes, the data for that period is marked as invalid, and the start and end times of the missing data are recorded. For environmental data, temperature units are uniformly converted to degrees Celsius, and humidity units are uniformly converted to percentages. Sudden jumps caused by sensor malfunctions are removed; for example, if the difference between a temperature value at a certain moment and the average of all valid values ​​within one minute before and after it exceeds 30%, the average value within that minute is used to replace the jump value. For water immersion... The status and door open / close status are converted from / no Boolean values ​​to 1 / 0 numerical forms. For the timing information, the timing data of all traffic light units are first uniformly adjusted to a sampling frequency of once per second. For periods with insufficient sampling frequency, missing values ​​are supplemented using linear interpolation. For example, if only the green light durations of the 1st and 3rd seconds are recorded for a certain period, the duration of the 2nd second is calculated by dividing (the duration of the 1st second + the duration of the 3rd second) by 2. Then, key timing features such as the on / off cycle and switching interval of each traffic light are extracted to ensure that the timing feature format is completely consistent across different intersections and different traffic lights. For video image data, the system first automatically locates the position of the traffic lights in the image using image recognition technology, using the outline of the traffic light panel as the boundary. The system crops out irrelevant background areas, such as surrounding buildings and trees, retaining only the valid area containing traffic lights. Then, it uniformly adjusts the resolution of all cropped images to 128×128 pixels to ensure consistent image size. Finally, it divides the values ​​of the red, green, and blue channels of each pixel in the image by 255 to normalize the values ​​to between 0 and 1, eliminating the impact of pixel value differences on the fault identification model's analysis. After completing all single-type data processing, the system aligns the data according to timestamps accurate to milliseconds, matching electrical parameters, environmental data, working sequence information, and video image data from the same moment to form regularized multimodal data. This regularized data is then batch-input into the pre-trained fault identification model.

[0045] Step 221: Utilize the feature extraction network in the fault identification model to perform deep feature learning on the input normalized multimodal data, generating multidimensional feature vectors representing the signal light status. Specifically, the feature extraction network adopts a multi-subnetwork parallel extraction + feature fusion architecture. Dedicated subnetworks are designed for the characteristics of different modal data to ensure that features in each dimension are fully mined. For electrical parameters and environmental data, a total of 6 dimensions—voltage, current, power, temperature, humidity, and immersion / door status values—a fully connected subnetwork is used for processing. This subnetwork contains 3 fully connected layers. The first layer maps the original features of the 6 dimensions to a 64-dimensional feature space, and the second layer... The first layer maps the features to 32 dimensions, and the third layer maps these 32-dimensional features to 16 dimensions. After processing at each layer, a ReLU activation function is used to filter linearly irrelevant features, retaining effective nonlinear information. For the time-series information, four dimensions—the duration of red, yellow, and green lights on / off and the switching interval—an LSTM sub-network is used. This sub-network contains two LSTM layers. The first LSTM layer has 32 hidden units to capture short-term temporal dependencies, and the second LSTM layer has 16 hidden units to uncover potential patterns in long-term sequences, ultimately outputting a 16-dimensional temporal feature vector. For video image data, a CNN sub-network is used. The network consists of three convolutional layers, two pooling layers, and two fully connected layers. The first convolutional layer uses 32 3×3 convolutional kernels to scan the image with a stride of 1, extracting the edge contour features of the traffic lights and outputting 32 64×64 feature maps. Then, a max-pooling layer (2×2 pooling kernels, stride of 2) downsamples the feature maps, preserving key features and reducing computation. The second convolutional layer uses 64 3×3 convolutional kernels to extract the local texture features of the traffic lights, such as the brightness distribution when the lights are on, resulting in 64 32×32 feature maps, which are then downsampled again by a max-pooling layer. The third convolutional layer uses 128 3×3 convolutional kernels to extract the higher-order features of the traffic lights. The system extracts semantic features, such as the color change pattern of the traffic lights, to obtain 128 16×16 feature maps. Global average pooling is used to convert these 128 feature maps into 128-dimensional vectors. Two fully connected layers then map these 128-dimensional vectors to 64 and 16 dimensions respectively, resulting in a 16-dimensional image feature vector. A feature fusion layer sequentially concatenates the 16-dimensional electrical environment features output from the fully connected sub-network, the 16-dimensional temporal features output from the LSTM sub-network, and the 16-dimensional image features output from the CNN sub-network, forming a 48-dimensional fused feature vector. This 48-dimensional vector is then mapped to 32 dimensions through a fully connected layer, ultimately generating a multi-dimensional feature vector that comprehensively represents the operating status of the traffic lights.

[0046] Step 222: Input the multidimensional feature vector into the fault classification branch of the fault recognition model, perform fault pattern recognition analysis, and generate preliminary fault type judgment results. Specifically, the core architecture of the fault classification branch is 3 fully connected layers + 1 output layer. Based on historical fault data, sample data containing 16 common fault types are pre-trained. The first fully connected layer maps the 32-dimensional multidimensional feature vector to 64 dimensions and enhances the nonlinear expressive power of the features through the ReLU activation function. The second fully connected layer maps the 64-dimensional features to 32 dimensions and continues to filter effective features through the ReLU activation function. The third fully connected layer maps the 32-dimensional features to 16 dimensions, corresponding to 16 preset fault types. These 16 fault types are: traffic lights not lit (including red, green, and yellow lights not lit individually), traffic lights always on (including red, green, and yellow lights always on individually), traffic lights... The system handles various fault types, including: conflict (red and green lights on simultaneously, yellow and green lights on simultaneously, etc.), flashing traffic lights, dimming traffic lights, traffic signal malfunction, power failure of traffic signal, network outage of traffic signal, overvoltage power supply, undervoltage power supply, overcurrent power supply, leakage power supply, excessively high chassis temperature, excessively high chassis humidity, water immersion in chassis, and abnormal opening of chassis doors. The output layer uses a softmax function to convert the 16-dimensional feature vector into probability values ​​corresponding to 16 different fault types, with the sum of these probabilities equal to 1. For example, the probability of a red light not lighting for a certain set of features is 0.85, and the sum of the probabilities for other fault types is 0.15. The system takes the fault type with the highest probability value as the initial fault type judgment result, while setting a probability threshold of 0.6. If the highest probability value is lower than 0.6, it indicates that the current features are insufficient to clearly determine the fault type, and the system marks it as a suspected fault in the preliminary results. The system also records the three highest-probability fault types and their corresponding probability values.

[0047] Step 223: The preliminary fault type judgment result and the multi-dimensional feature vector are input into the location analysis branch of the fault identification model. The features and fault type are correlated to generate location information of the faulty component or traffic light corresponding to the fault type judgment result. Specifically, the location analysis branch consists of a feature adjustment layer and a location prediction layer. The feature selection is guided by the fault type to improve the positioning accuracy. The feature adjustment layer first selects features highly correlated with the fault type from the 32-dimensional multi-dimensional feature vector based on the preliminary fault type judgment result. For example, if the preliminary fault type is that the red traffic light is not lit, the focus is on extracting features related to the red light panel area from the features output by the video image sub-network, and current and voltage features corresponding to the red light power supply line from the electrical parameters. Then, the selected relevant features are concatenated with the one-hot encoded vector corresponding to the preliminary fault type. The one-hot encoded vector is 16-dimensional and corresponds to 16 fault types. If the preliminary fault type is that the traffic light is not lit, the feature adjustment layer first selects features highly correlated with the fault type from the 32-dimensional multi-dimensional feature vector. For example, if the preliminary fault type is that the red traffic light is not lit, the feature adjustment layer ... If the red light is off, the corresponding position is 1, and the other 15 positions are 0. After concatenation, a new fused feature vector is formed (the dimension is the filtered feature dimension + 16). The position prediction layer contains two fully connected layers. The first layer maps the fused feature vector to 32 dimensions and enhances the feature expression through the ReLU activation function. The second layer maps the 32-dimensional features to a 4-dimensional output vector, which corresponds to the longitude and latitude of the fault location, and the relative X and Y coordinates of the faulty component in the traffic light unit. The longitude and latitude are based on the GPS positioning data recorded when the traffic light was installed, and are corrected by combining the position association information in the feature vector, accurate to 6 decimal places. The relative X and Y coordinates are based on the center point of the traffic light unit as the origin, and the unit is meters. For example, if the red light panel is in the upper left corner of the traffic light unit, the relative coordinates may be (-0.3, 0.2). These two coordinates can be used to determine the specific location of the faulty component in the traffic light unit, and the positioning accuracy is controlled within 0.1 meters.

[0048] Step 224: Integrate the location information of the faulty component or signal light with the preliminary fault type judgment result to generate complete fault type and fault location information. Specifically, this includes: First, performing information matching verification. The system automatically checks whether the preliminary fault type and location information are logically consistent. For example, if the preliminary fault type is a signal malfunction, the longitude and latitude of the location information should point to the installation location of the signal malfunction housing. If the location information points to a specific signal light panel, it is determined to be an information conflict, the conflict type is marked, and both information is retained for subsequent verification. If the preliminary fault type is a continuously lit green signal light, the location information should point to a specific green light panel. If it points to other panels or signal lights, the information will be invalid. For the same faulty machine, a conflict is also marked. For information without conflict, the fault type, longitude and latitude of the fault location, relative coordinates of the faulty component, and unique identifier of the signal light unit are integrated. For preliminary results marked as suspected faults, the three fault types with the highest probability are retained during integration. Each fault type corresponds to a set of location information. At the same time, prompts that require further collaborative verification are marked, and the probability values ​​of the three fault types are recorded. Finally, a complete set of information is formed, including fault type, precise coordinates of the fault location, specific location of the faulty component, unique identifier of the signal light unit, whether there is information conflict, and whether it is a suspected fault. This set serves as the fault type and fault location information.

[0049] The feature extraction network designs dedicated subnetworks for different modal data, deeply mining features in various dimensions. Compared with traditional solutions that rely solely on electrical parameters, it can more comprehensively and three-dimensionally represent the operating status of traffic lights. The fault classification branch, based on pre-trained models and multi-dimensional features, achieves accurate identification of fault types. Through probability threshold screening and suspected fault marking, it reduces the risk of misjudgment. Through feature screening and accurate coordinate calculation, it makes fault location more targeted and accurate, avoiding low maintenance efficiency caused by fuzzy positioning, and determining the specific light panel and component where the fault occurred, providing clear guidance for maintenance personnel.

[0050] In a preferred embodiment of the present invention, step 3 above, which involves selecting multiple spatiotemporally correlated monitoring devices to form a diagnostic group based on fault location information, and performing fusion analysis and collaborative diagnosis on the multi-source data of the diagnostic group to generate a fault determination result, may include:

[0051] In this embodiment of the invention, step 330, based on the fault location information and combined with a preset geographical proximity threshold and time synchronization window, determines the spatiotemporal correlation range; specifically, it includes: firstly, extracting precise longitude and latitude data from the fault location information, using this as a reference point for spatial analysis; the preset geographical proximity threshold is divided according to the traffic flow level of the intersection where the fault is located; for core intersections with an average daily traffic flow exceeding 50,000 vehicles, the geographical proximity threshold is set to a circular area with a radius of 800 meters, which can cover the traffic lights of 3 to 5 surrounding related intersections; for major intersections with an average daily traffic flow of 20,000 to 50,000 vehicles, the threshold is set to a radius of 500 meters; for... For ordinary intersections with traffic flow of less than 20,000 vehicles, the threshold is set at a radius of 300 meters. The time synchronization window is determined by starting from the time of the first occurrence of the fault recorded in the preliminary fault identification results, tracing back 15 minutes and extending forward 15 minutes to form a 30-minute time interval. This interval includes both normal state data before the fault occurred and continuous state data after the fault occurred, ensuring that the complete change process before and after the fault can be captured. The area that meets the requirements of being within the corresponding radius in space and within the 30-minute interval in time is comprehensively defined as the spatiotemporal correlation range. This range accurately covers the monitoring equipment and data time periods that may be directly or indirectly related to the fault.

[0052] Step 331: Based on the spatiotemporal correlation range, select traffic light monitoring devices located within the spatiotemporal correlation range from the IoT monitoring network to form an initial device set. Specifically, this includes: the IoT monitoring network's device archive stores the fixed installation locations, unique device codes, and intersection information of all traffic light monitoring devices. The system sequentially traverses all devices in the archive, calculating the straight-line distance between the installation location and the fault reference point for each device. The calculation method involves using the device's longitude and latitude and the fault point's longitude and latitude, referring to the calculation method for the straight-line distance of geographical coordinates, to obtain the actual ground distance. If this distance is less than or equal to the geographical proximity threshold corresponding to the fault intersection, the device is determined to be within the spatial correlation range. Simultaneously, it checks whether the device has continuous data records within a 30-minute time synchronization window, i.e., at least one valid data record every 5 minutes. If this condition is met, the device is included in the initial device set. The initial device set includes information such as device code, intersection, installation location coordinates, and data record start and end times, ensuring that all devices that may have a spatiotemporal correlation with the fault are initially selected.

[0053] Step 332 involves verifying the communication status and data availability of the devices in the initial device set, and grouping the devices that pass the verification into a collaborative diagnostic group. Specifically, this includes: communication status verification is performed using the IoT communication management mode. The system sends three consecutive communication request signals to each device in the initial device set, with a two-second interval between each request. If the device returns a correct response signal within three seconds of receiving the signal, including the device code and the current time, the communication status is considered normal. If at least two of the three requests fail to receive a response or the response signal is incorrect, the communication is considered abnormal. Data availability verification targets records within the time synchronization window. First, calculate the data missing rate, which is the missing time of the device within 30 minutes divided by the total time of 30 minutes. If the missing rate is less than 8%, proceed to key parameter check. Key parameters include voltage, current, and light color status. If none of these parameters are missing for more than 3 consecutive minutes, and the cumulative missing time of a single parameter does not exceed 5 minutes, the data is determined to be usable. At the same time, the device that passes the communication status verification and data availability verification is retained, and the remaining devices are removed from the initial device set. The remaining devices form a collaborative diagnostic group. The group records the available data types of each device in detail, such as whether it contains video images and whether it has complete timing information.

[0054] Step 333: Collect multi-source monitoring data recorded by all devices in the collaborative diagnostic group within the same time period, and perform spatiotemporal alignment processing on the multi-source monitoring data to generate a standardized dataset with a unified timestamp and geographic coordinates. Specifically, the same time period is uniformly set to 30 minutes within the time synchronization window. The system collects all monitoring data within this time period from each device in the collaborative diagnostic group, including electrical parameters (three-phase voltage, three-phase current, active power) every 10 seconds, environmental data (internal temperature, humidity, external light intensity) every 30 seconds, millisecond-precision working timing information (on-hook, off-hook, and switching interval of red, yellow, and green lights), and video image data (including close-up of the light panel and surrounding environment) every 2 seconds. The time alignment processing combines the original timestamps of all devices. The time stamps are uniformly converted to Beijing time (UTC+8), accurate to milliseconds. If there is a deviation between the device's local time and the standard time, such as being 2 seconds ahead or 3 seconds behind, the deviation value is added or subtracted for calibration to ensure that the data from different devices at the same actual moment have the same time stamp. Spatial alignment processing converts the geographic coordinates of all devices to the national coordinate system 2000. If the original coordinates of the devices are in the WGS84 coordinate system, they are converted using preset coordinate transformation parameters, such as translation and rotation angles, to ensure that the location information of all devices is comparable in the same coordinate system. The aligned data is arranged in ascending order of timestamp. Data from different devices with the same timestamp form data units. All data units are combined into a standardized dataset, which includes timestamps, coordinates of each device, and multi-source monitoring data fields for each device.

[0055] Step 334: Perform a consistency comparison analysis on the electrical parameters from different monitoring devices in the standardized collaborative dataset to generate electrical parameter consistency analysis results; calculate the matching degree of display status and logical relationship on the video image data from different monitoring devices in the standardized collaborative dataset to generate visual status matching degree results; specifically, the electrical parameter consistency comparison analysis targets six parameters, namely three-phase voltage and three-phase current. One data point is taken every minute within a 30-minute period. The arithmetic mean of all devices in the collaborative diagnostic group is calculated for each parameter. That is, the sum of all device data for each parameter is divided by the number of devices. Then, the difference between each parameter value of each device and the average value of that parameter is calculated. The difference is divided by the average value to obtain the relative deviation percentage. If the absolute value of the relative deviation percentage of all six parameters of a device is less than 4%, the device is considered to have consistent electrical parameters with other devices. If the absolute value of the relative deviation percentage of one or two parameters is between 4% and 8%, and the remaining parameters are normal, it is considered to have slight inconsistency. If three or more parameters have a deviation exceeding 4%, or any parameter has a deviation exceeding 8%, it is considered to have serious inconsistency. The total number of consistent devices and slightly inconsistent devices is counted as a percentage of the total number of devices in the diagnostic group. The proportion of equipment is considered consistent. If the proportion exceeds 80%, the electrical parameter consistency analysis result is overall consistent; if the proportion is between 50% and 80%, it is locally consistent; if it is below 50%, it is overall abnormal. The equipment numbers and specific parameter deviations of severely inconsistent equipment are recorded. Visual state matching calculation first extracts the light color display status (red light on, green light on, yellow light on, all off, flashing) of each device at the same moment from the video image. Then, it is compared in two categories: one is the logical relationship of light colors between adjacent intersections, such as when the east-west main road displays a green light, the north-south secondary road should display a red light. If they match... If the logic is correct, it is recorded as a logical match; otherwise, it is a logical mismatch. Another category is the display status of different light groups in the same direction at the same intersection. For example, two traffic lights in the same direction should be green or red at the same time. If the status is the same, it is recorded as a display match; otherwise, it is a display mismatch. A comparison is performed every 10 seconds within 30 minutes. The proportion of logical matches to the total number of comparisons and the proportion of display matches to the total number of comparisons are calculated. If both proportions exceed 90%, the visual status matching result is a high match; if both are between 70% and 90%, it is a medium match; if either proportion is below 70%, it is a low match.

[0056] Step 335: Integrate the electrical parameter consistency analysis results with the visual state matching results, and perform a joint judgment based on preset fault discrimination rules to generate a comprehensive verification conclusion regarding the existence of a fault. Specifically, the preset fault discrimination rules are formulated based on the combination of the two types of results and include four core judgment logics: First, if the electrical parameter consistency analysis results are generally consistent and the visual state matching results are highly matched, and the initially identified faulty equipment shows no abnormalities in electrical parameters and visual state, then the fault is judged to be a false alarm, possibly caused by instantaneous sensor fluctuations in a single device. Second, if the electrical parameter consistency analysis results are generally abnormal or partially consistent, and the visual state matching results are low matched, and the abnormal parameters and mismatched states are concentrated in the initially identified faulty equipment, then the fault is judged to be present. The first type of fault is a real fault. The second type is a false alarm. The third type is a fault that is not real. If the electrical parameter consistency analysis result is consistent overall but the visual state matching result is low, then check the video image quality of the faulty equipment. If the image is unobstructed, clear, and the light color is clearly distinguishable, then the fault is real. If the image is severely obstructed or blurry, then it is determined to be unverified and the image needs to be manually reviewed. The fourth type is a fault that is not real. The electrical parameter consistency analysis result is abnormal overall but the visual state matching result is high. Then it is determined to be a suspected fault. It may be that the parameter abnormality is caused by power grid fluctuations but does not affect the light color display. According to the above rules, the two types of results are combined and analyzed, and finally one of the following is generated as a comprehensive verification conclusion: real, false alarm, unverified, or suspected fault. The specific data used in the judgment process is recorded in detail, such as the voltage deviation of a certain device or the logic mismatch of the light color at a certain moment.

[0057] Step 336: Based on the comprehensive verification conclusion, combined with the fault type and fault location information, generate the final fault judgment result; specifically, if the comprehensive verification conclusion is true, retain the initially identified fault type, such as the red light of the traffic light not turning on and the fault location information, precise coordinates and components, supplement the associated data of other devices in the collaborative diagnosis group, such as whether the surrounding devices have similar parameter anomalies, whether the light color logic is affected by this fault, and clarify the direct impact range of the fault, such as only this light group or affecting adjacent intersections; if the conclusion is a false alarm, mark the fault as an invalid alarm in the result, record the reason for the false alarm in detail, such as the instantaneous current fluctuation of the equipment but the light color is normal, and automatically terminate the subsequent fault handling process; if the conclusion is to be verified, in the fault type and location information, the fault type and location information are retained. The system adds a prompt message indicating that video images require manual review. Relevant video clips from the standardized dataset, 3 minutes before and after the fault, are pushed to the manual review terminal along with the fault information, awaiting the review results. If the conclusion is a suspected fault, the fault type and location information are retained, and it is marked that continuous monitoring is required for 30 minutes. The system initiates a short-term tracking mechanism, collecting data from the device and related devices every 5 minutes. If the abnormal state disappears within 30 minutes, it is determined that the temporary abnormality has been resolved. If the abnormality persists, it is upgraded to a real fault. The final fault determination result includes the authenticity of the fault, the confirmed fault type, the fault location accurate to the meter, the impact on related devices, the data verification basis, and the handling suggestions, forming a complete structured report.

[0058] The refined delineation of the spatiotemporal correlation range ensures that peripheral equipment that may be related to the fault and complete time series data are included in the analysis, avoiding misjudgments caused by insufficient data coverage. The generated fault judgment results integrate collaborative verification conclusions and original information, making the fault description more accurate and the basis more sufficient, thereby improving the efficiency and pertinence of fault response.

[0059] In a preferred embodiment of the present invention, step 4 above, based on the fault determination result and combined with the fault level rule base preset in the platform knowledge base, quantifies the impact range by calculating the area enclosed by the boundary coordinates of the traffic impact area, automatically determines the fault level and traffic impact range, generates structured fault alarm information, and pushes it to the visualization decision dashboard of the command center for dynamic display to obtain intelligent handling strategies, may include:

[0060] In this embodiment of the invention, step 440 involves querying the fault level rule base in the platform knowledge base based on the fault confirmation conclusion in the fault determination result, and obtaining the level determination rule and impact range calculation parameters corresponding to the fault type. Specifically, this includes: firstly, extracting a clear fault confirmation conclusion from the fault determination result, and only performing subsequent operations on faults that are marked as truly existing. For false alarms, pending verification, and suspected faults, the system automatically marks them as temporarily unprocessed and returns them to a state pending further confirmation. The fault level rule base in the platform knowledge base adopts a hierarchical storage structure. The top layer is divided according to major fault categories (traffic light faults, signal controller faults, electrical abnormality faults, environmental abnormality faults). Each major category is further subdivided into specific fault types. For example, traffic light faults include non-lighting, constantly lit, conflicting, flashing, dimming, etc. Each specific fault type corresponds to multiple rule entries. Each entry precisely matches the combination of fault type + intersection level + time period. The rule entry contains two core contents: one is the level determination rule, which clarifies the severity level range corresponding to different traffic impact range values; the other is the impact range calculation parameters, which include four specific values: basic diffusion radius, road network weight coefficient, intersection correlation coefficient, and time period correction coefficient.

[0061] To determine the basic diffusion radius, for each fault type, historical cases of that type of fault at different intersection levels were collected over the past 12 months. The actual road network radius affected by the fault in each case was statistically analyzed, and the average radius of all cases at the same intersection level was calculated. This was then adjusted based on the road density (proportion of main roads and secondary roads) at that intersection level. A 10% increase in the average radius was made for high road density, and a 10% decrease for low road density. This determined the basic diffusion radius for the fault type at different intersection levels. For example, for a red light malfunction, the average impact radius of historical cases at core intersections (daily traffic volume ≥ 50,000 vehicles) was 450 meters. Due to the high road density at core intersections, the adjusted basic diffusion radius was 500 meters. For ordinary intersections (daily traffic volume < 20,000 vehicles), the average impact radius of historical cases was 270 meters. Due to the low road density, the adjusted basic diffusion radius was 300 meters.

[0062] The road network weight coefficient is determined by collecting the average daily traffic flow of different road types (arterial roads, secondary arterial roads, and local roads) through a traffic flow monitoring system. The proportion of traffic flow of each road type in the total traffic flow of the area is calculated, and the higher the proportion, the greater the weight. The weight of arterial roads is set to 1.5, secondary arterial roads to 1.2, and local roads to 1.0. Then, based on the road network structure (the proportion of each type of road) in the area where the fault occurs, the average weight of the regional road network is calculated, which is the proportion of arterial roads multiplied by 1.5, plus the proportion of secondary arterial roads multiplied by 1.2, plus the proportion of local roads multiplied by 1.0. The result is the road network weight coefficient of the area. For example, in the area where the core intersection is located, the proportion of arterial roads is 40%, the proportion of secondary arterial roads is 30%, and the proportion of local roads is 30%. The road network weight coefficient is calculated as 40% multiplied by 1.5, 30% multiplied by 1.2, and 30% multiplied by 1.0, which gives a value of 1.3.

[0063] The determination of the intersection correlation coefficient involves analyzing the traffic flow correlation between the faulty intersection and surrounding intersections. This is achieved by statistically analyzing the rate of change in traffic flow at surrounding intersections during the fault (the percentage increase or decrease in traffic flow after the fault compared to normal times) using monitoring data. Higher correlation coefficients indicate a larger rate of change. The correlation is divided into three levels: high correlation (rate of change > 30%) with a coefficient of 1.3, medium correlation (rate of change 10%-30%) with a coefficient of 1.2, and low correlation (rate of change < 10%) with a coefficient of 1.0. For example, core intersections, due to their high traffic flow and wide radiation range, have a high correlation with more than three surrounding intersections, and their correlation coefficient is determined to be 1.2; ordinary intersections have only a medium to low correlation with one or two surrounding intersections, and their coefficient is determined to be 1.0.

[0064] The time-period correction factor is determined by collecting road network saturation (the ratio of actual traffic flow to road capacity) data for different time periods (morning peak 7:00-9:00, evening peak 17:00-19:00, off-peak 9:00-17:00, and nighttime 22:00-6:00 the next day). Higher saturation indicates a greater impact of the fault on traffic, resulting in a larger correction factor. For morning and evening peak saturation >80%, the correction factor is set at 1.1; for off-peak saturation 40%-80%, the correction factor is set at 1. 0; Nighttime saturation < 40%, correction factor is 0.9; For example, in the case of a traffic light red light malfunction, the parameters for the core intersection during the morning peak are: basic diffusion radius 500 meters, road network weight coefficient 1.3, intersection correlation coefficient 1.2, and time period correction coefficient 1.1; the parameters for the core intersection during off-peak hours are: basic diffusion radius 400 meters (historical case average radius 360 meters, adjusted to 400 meters for high road density), and road network weight coefficient 1.1 (regional main roads account for 30%, secondary roads 30%). For branch roads (40%), the calculated coefficients are: intersection correlation coefficient 1.2, time period correction coefficient 1.0; for ordinary intersections, the parameters for any time period are: basic diffusion radius 300 meters, road network weight coefficient 1.0, intersection correlation coefficient 1.0, and time period correction coefficient 1.0; for core intersections with signal malfunctions, the parameters for peak hours are: basic diffusion radius 800 meters (the historical average radius is 720 meters, adjusted to 800 meters due to high road density), road network weight coefficient 1.5 (calculated as 1.5 based on 50% main roads, 30% secondary roads, and 20% branch roads), intersection correlation coefficient 1.3 (high correlation with 5 surrounding intersections), and time period correction coefficient 1.2. When querying, the system first locates the sub-category under the corresponding major category in the rule base by the specific fault type in the fault judgment result, then matches the unique rule entry by the intersection level in the fault location information and the current time period (peak / off-peak) obtained by the system, and finally extracts the level judgment rule and four influence range calculation parameters under that entry.

[0065] Step 441: Based on the fault location information and the impact range calculation parameters, determine the affected traffic network area and generate a set of boundary coordinates for the traffic impact area. Specifically, this includes: first, extracting latitude and longitude coordinates accurate to six decimal places from the fault location information, using these as the center point for impact range calculation; combining the four parameters extracted in step 440, calculating the actual diffusion radius, which equals the basic diffusion radius multiplied by the road network weight coefficient, and then multiplied by the time period correction coefficient. For example, if the traffic light at a core intersection is red during the morning rush hour, the actual diffusion radius is 500 meters multiplied by 1.3, then multiplied by 1.1. The calculation process is: 500 meters multiplied by 1.3 to get 650 meters, then 650 meters multiplied by 1.1 to get 715 meters. Subsequently, the system calls the road network data interface of the electronic map to obtain the complete road network structure within a 715-meter radius of the center point, including the direction, width, and intersection relationships of main roads, secondary roads, and branch roads. The road network structure determines the boundary of the affected area. Based on the center point, it extends along the main roads in the four cardinal directions (east, south, west, and north) and the four diagonal directions (northeast, southeast, southwest, and northwest) to the boundary of the actual diffusion radius. Key inflection points on the boundary are selected. Inflection points must meet one of the following conditions: 1) the intersection of two or more roads (such as the intersection of a main road and a secondary road); 2) the intersection of the actual diffusion radius boundary and the road centerline; 3) the endpoint of the road network in the affected area (such as the end of a branch road). Each inflection point is obtained with latitude and longitude coordinates accurate to 6 decimal places through an electronic map and sorted clockwise. The sorting starts from the intersection of the eastward road and the diffusion radius boundary, and inflection points in the southeast, south, southwest, west, northwest, north, and northeast directions are selected in sequence. Finally, a set of boundary coordinates of the traffic impact area containing 8 to 12 vertices is formed to ensure that the boundary can completely surround all road networks and surrounding areas affected by the fault.

[0066] Step 442: Based on the boundary coordinate set of the traffic impact area, substitute the coordinates of each vertex in the boundary coordinate sequence into the shoelace formula to calculate the area of ​​the polygonal region enclosed by connecting the coordinate points sequentially, generating a quantified traffic impact range value. Specifically, this includes: first, confirming that the vertices in the boundary coordinate set are arranged in clockwise order, and that the first and last vertices are not repeated. The last vertex will automatically close with the first vertex during the calculation. Assume the coordinate set contains 5 vertices, labeled as vertex 1, vertex 2, vertex 3, vertex 4, and vertex 5. Each vertex has a defined longitude (denoted as longitude 1, longitude 2... longitude 5) and latitude (denoted as latitude 1, latitude 2... latitude 5). The calculation process consists of three steps. The first step is to calculate the first sum. This involves calculating the longitude of each vertex multiplied by the latitude of the next vertex, and then adding all the products together. Specifically, this involves multiplying the longitude of vertex 1 by the latitude of vertex 2, adding the longitude of vertex 2 by the latitude of vertex 3, adding the longitude of vertex 3 by the latitude of vertex 4, adding the longitude of vertex 4 by the latitude of vertex 5, and finally adding the longitude of vertex 5 by the latitude of vertex 1. First, sum the results of these five products to obtain the first sum. Second, calculate the second sum by multiplying the latitude of each vertex by the longitude of the next vertex, then summing all the products: (e.g., latitude of vertex 1 multiplied by longitude of vertex 2, latitude of vertex 2 multiplied by longitude of vertex 3, latitude of vertex 3 multiplied by longitude of vertex 4, latitude of vertex 4 multiplied by longitude of vertex 5, and finally latitude of vertex 5 multiplied by longitude of vertex 1). The third step is to calculate the area. Subtract the second sum from the first sum to obtain a difference; take the absolute value of this difference (if the difference is negative, take its opposite positive value); divide the absolute value by 2. The final result is the area of ​​the traffic impact zone, with the area unit uniformly set to square meters. For example, assuming the first sum is 156892.35 and the second sum is 145210.12, the difference is 11682.23, the absolute value is 11682.23, and dividing by 2 gives an area of ​​5841.115 square meters. This value is the quantified traffic impact range.

[0067] Step 443 involves inputting the traffic impact range value and fault type into the severity determination rules for matching analysis, automatically determining the fault severity level. Specifically, the severity determination rules, extracted in step 440, clearly define four severity levels based on a combination of fault type, intersection level, and time period: Level 1 is extremely severe, Level 2 is severe, Level 3 is relatively severe, and Level 4 is moderate. Each level corresponds to a specific traffic impact range range, which is based on historical fault handling data and traffic flow loss assessment. Boundary value handling principles are also specified: when the range value equals the upper limit of the range, the level is increased by one. For example, the severity determination rule for a red light malfunction is as follows: during the morning rush hour at a core intersection, an impact range value > 15,000 square meters is Level 1 (corresponding to historical cases causing large-scale congestion), 10,000-15,000 square meters is Level 2, 5,000-10,000 square meters is Level 3, and < 5,000 square meters is Level 4. During peak hours, an impact area greater than 12,000 square meters is classified as Level 1, 8,000-12,000 square meters as Level 2, 4,000-8,000 square meters as Level 3, and less than 4,000 square meters as Level 4. At ordinary intersections during peak hours, an impact area greater than 8,000 square meters is classified as Level 2, 5,000-8,000 square meters as Level 3, 2,000-5,000 square meters as Level 4, and less than 2,000 square meters as Level 4. The system first determines the current time of the fault (by matching peak / off-peak hours with system time) and the level of the intersection. Then, it compares the quantified traffic impact area with the corresponding intervals in the rules. If the range falls within an interval, the level corresponding to that interval is determined. If the range equals the upper limit of the interval, such as 15,000 square meters which is exactly the upper limit of Level 1 and Level 2 at a core intersection during morning peak hours, it is upgraded to Level 1. For example, if a red traffic light fails to illuminate at a core intersection during morning peak hours, with an impact area of ​​15,000 square meters, it is determined to be Level 1 (severe) after rule comparison.

[0068] Step 444 integrates the fault type, fault location information, fault severity level, and traffic impact range value to generate a standard-formatted structured fault alarm information. Specifically, the structured fault alarm information uses a standard format with 10 fixed fields to ensure the information is standardized and directly parsable by the system. The specific generation method for each field is as follows: the unique fault identifier consists of the intersection number + fault type code + timestamp. The intersection number is the city abbreviation + area number + intersection sequence number; the fault type code is a two-digit code preset in the rule base; the timestamp is the fault occurrence time; the fault type is the specific category in the fault determination result; the fault location includes precise latitude and longitude, the standard name of the intersection, and the specific location of the faulty component; the fault severity level is marked with the level number and name; the traffic impact range is marked with the area value and unit; the fault confirmation basis references the key results of collaborative diagnosis; the fault occurrence time is the time when the fault was first identified; the fault status is marked as pending handling; the associated equipment list is the code and status of all equipment in the collaborative diagnosis group. All fields are integrated in the above order to generate complete structured fault alarm information, ensuring that each field's information is complete and its source is clear.

[0069] Step 445 involves pushing structured fault alarm information to the command center's visual decision dashboard in real time for dynamic display. Combined with a pre-set handling strategy library, this generates intelligent handling strategies for the fault. Specifically, structured fault alarm information is pushed via a dedicated encrypted communication channel with a push delay strictly controlled within 2 seconds to ensure the command center receives information in real time. The visual decision dashboard is divided into four core display areas, each with a clearly defined function: First, a map visualization area using an electronic map as the base, marking fault locations with different style icons according to fault severity levels: Level 1 faults use flashing red icons, Level 2 uses solid orange icons, Level 3 uses solid yellow icons, and Level 4 uses solid blue icons. Clicking an icon pops up a card displaying the core fault information. Second, an information details area, displaying the complete content of the structured alarm information horizontally arranged by field, supporting filtering and sorting by fault level, intersection area, and time of occurrence. Fields can be clicked to view details. Third, a status tracking area, displaying the entire fault handling process progress in a timeline format. The initial status is "Pending Dispatch," and it automatically updates to "Dispatched," "Under Maintenance," "Repaired," and "Verified." Each status update is marked with a timestamp; fourth is the strategy suggestion area, which displays intelligent handling strategies for this fault. The system also queries the preset handling strategy library, which stores corresponding strategies according to fault level and fault type. The strategy includes four core contents: emergency measures, work order dispatch rules, notification targets, and status update frequency. All strategies are formulated based on historical fault handling experience and traffic diversion optimization plans. For example, the strategy for a level 1 fault, where the red light of a traffic light is not lit, is as follows: the emergency measures are to immediately activate the emergency diversion plan at the intersection and issue a warning information through the electronic screens of surrounding roads; the work order dispatch rules are to automatically select two maintenance personnel within 3 kilometers of the fault point, whose skill tag is traffic light repair, and who are currently not on duty, and dispatch an emergency work order (highest priority); the notification targets are the on-duty personnel of the traffic police squadron in the jurisdiction and the maintenance manager; the status update frequency is to automatically collect the status of the faulty equipment every 3 minutes and update the dashboard. The system accurately matches the corresponding entries in the strategy library according to the level 1 fault and the type of the red light not lit, generates an intelligent handling strategy, pushes it to the strategy suggestion area of ​​the dashboard, and automatically triggers work order dispatch and notification operations.

[0070] The quantitative calculation of traffic impact areas, through precise coordinate extraction, structured inflection point selection, and step-by-step calculation, transforms the ambiguous impact range into accurate area data, shifting the determination of fault severity from subjective experience to objective data support, thus improving the accuracy and fairness of the determination. The standardized generation of structured fault alarm information unifies the information format and source, avoiding the problem of messy information from multiple systems requiring manual comparison, and reducing the information organization burden on maintenance personnel.

[0071] In a preferred embodiment of the present invention, step 5 above, which generates a standardized electronic maintenance work order according to the intelligent handling strategy and assigns the standardized electronic maintenance work order to the mobile terminal of the corresponding maintenance personnel to initiate the fault handling process, may include:

[0072] In this embodiment of the invention, step 550 involves extracting handling instructions, priority identifiers, and required skill requirements based on the intelligent handling strategy to generate a set of core elements for the work order. Specifically, this includes: firstly, performing structured parsing of the intelligent handling strategy, separating content directly related to operation and maintenance from the strategy text. Handling instructions need to be detailed down to the operational steps and acceptance criteria. For example, for a level one fault where the red traffic light is not lit, the handling instruction is clearly stated as follows: upon arrival at the site, first use a multimeter to check the voltage of the light group's power supply line. If the voltage value is within 220V±10%, the line is considered normal, and further checks are needed to see if the light group's wiring terminals are loose. If the line is abnormal, check the status of the circuit breaker in the power distribution box, replace the damaged circuit breaker, and check the voltage again. After confirming the line is normal, remove the faulty red light unit, replace it with a new unit of the same model, connect the power supply, test the red light's lighting status, and observe for 5 minutes without any abnormalities. Then, take a video of the scene including the light group's status and the surrounding environment, with a video duration of no less than 10 seconds. The priority identifier is determined according to the severity level of the fault. The scope of impact is determined, with Level 1 faults corresponding to an emergency indicator, requiring no more than 15 minutes from work order receipt to on-site arrival; Level 2 faults correspond to a high indicator, with a time limit of 30 minutes; Level 3 faults correspond to a medium indicator, with a time limit of 60 minutes; and Level 4 faults correspond to a low indicator, with a time limit of 120 minutes. Each indicator is associated with a unique color code in the system. The required skills are determined based on the fault type and operational complexity. For example, a fault where a red traffic light is not lit requires three skills: first, holding a valid low-voltage electrician special operation certificate; second, having practical experience in replacing traffic light components (the system records no less than 50 practical operations); and third, being proficient in using a multimeter and wiring tools. Each skill requirement must specify the type of qualification certificate and the minimum standard. The extracted handling instructions, priority indicators (including time limits and color codes), and required skill requirements (including qualification standards) are combined in the order of operation instructions-priority-skill conditions to form a core element set for the work order, ensuring that each element can be directly used for work order generation.

[0073] Step 551: Integrate the core elements of the work order with the structured fault alarm information to generate a complete work order data package including fault location information, fault feature description, handling requirements, and priority level. Then, fill this data into a preset electronic work order template to generate an electronic maintenance work order with a standard format. Specifically, this includes: extracting handling instructions as handling requirements, priority identifiers as priority levels, and required skill requirements as skill qualification conditions from the core elements of the work order; extracting fault location information from the structured fault alarm information, including latitude and longitude accurate to six decimal places, standard intersection name, and specific location of the faulty component; extracting fault feature descriptions, including fault type, fault occurrence time, traffic impact range, and fault confirmation basis; and extracting fault... The unique fault identifier serves as the basis for the work order number. The integrated complete work order data package contains 12 fields: work order number, fault location information, fault characteristic description, handling requirements, priority level, skill qualification requirements, dispatching department, dispatching personnel, dispatching time, estimated completion time, associated fault number, and remarks. The preset electronic work order template adopts a table format. The table header contains the names of the above 12 fields. The content area is filled with data package information in the order of the fields. The system logo and traffic facility maintenance work order title are added to the top of the table, and an electronic signature area is reserved at the bottom. After completion, the system automatically generates a QR code and anti-counterfeiting code, and finally generates a standard format electronic maintenance work order in A4 paper size, ensuring uniform format and anti-counterfeiting traceability function.

[0074] Step 552: Based on the fault location information in the electronic maintenance work order and combined with the real-time acquired dynamic location data of maintenance personnel, calculate the Euclidean distance between each maintenance personnel and the fault point, and generate a distance matrix; according to the required skill requirements in the electronic maintenance work order and combined with the maintenance personnel skill qualification database, generate a candidate list of qualified maintenance personnel; specifically, this includes: extracting latitude and longitude coordinates from the fault location information in the electronic maintenance work order as a reference point, and acquiring the real-time dynamic location data of all on-duty maintenance personnel through the mobile terminal GPS positioning mode. The data must include... The system requires information such as personnel name, employee ID, current longitude, current latitude, and location update time. The interval between the location update time and the current time must not exceed 3 minutes. When calculating the Euclidean distance, the longitude difference is calculated by subtracting the longitude of the fault point from the current longitude of each maintenance worker. Similarly, the latitude difference is calculated by subtracting the latitude of the fault point from the current latitude of each maintenance worker. Both the longitude and latitude differences are then squared (multiplied by themselves), and the two squares are summed. The square root of the sum is then taken to find a specific value. Multiplying this number by itself equals the sum, and the result is the straight-line distance between the maintenance personnel and the fault point. For example, if a maintenance personnel's current longitude is 117.533456 and latitude is 34.858192, the longitude difference is 0.0012, which squares to 0.00000144; the latitude difference is -0.0011, which squares to 0.00000121; the sum is 0.00000265, and the square root is approximately 0.001628, which translates to an actual distance of about 180 meters. The names, employee numbers, and distance values ​​of all valid maintenance personnel are then used to calculate this distance. Distances are sorted from smallest to largest to form a distance matrix. The matrix needs to indicate the calculation time for each distance. At the same time, the required skill requirements are extracted from the electronic maintenance work order and the maintenance personnel skill qualification database is queried. This database stores each person's skill certificate number, issuance date, validity period, and practical operation records. The system compares them one by one. First, it checks whether the person holds a valid low-voltage electrician certificate. Then, it counts the number of practical operations for replacing traffic light components. If both conditions are met, the person is included in the candidate list. The candidate list includes the person's name, employee number, skill matching items, and current location, and is sorted in ascending order by employee number.

[0075] Step 553: Perform fusion analysis on the distance matrix and the candidate list of maintenance personnel, and determine the final maintenance personnel assignment scheme based on the current work order task load status. Specifically, this includes: filtering out personnel whose names and employee IDs both exist in the candidate list from the distance matrix to form a fusion list. The list retains the name, employee ID, distance value, and skill matching item. Query the current work order task load status of the personnel in the fusion list. The load status is calculated by the number of incomplete work orders and the estimated remaining time for each work order. Each incomplete work order is converted into a load value based on the estimated remaining time: less than 1 hour is calculated as 0.5, 1 to 2 hours as 1, and more than 2 hours as 1.5. Add up the load values ​​of all incomplete work orders of the same personnel to obtain the total load value. For example, if a personnel has one work order with an estimated remaining time of 40 minutes, the total load value is 0.5.

[0076] A comprehensive comparison is made among the personnel in the integration list. The comparison rule is that distance factor accounts for 60% of the weight and load factor accounts for 40%. The comparison method for distance factor is to take the minimum distance in the integration list as the benchmark, compare the distance of other personnel with this minimum distance to obtain the ratio, divide 1 by this ratio, and use the result to represent the proportion of distance factor. For example, if the minimum distance is 180 meters and a certain person's distance is 360 meters, the ratio is 2. 1 divided by 2 equals 0.5. This 0.5 represents the proportion of distance factor for that person.

[0077] The comparison method for load factors is as follows: a total load value of 0 corresponds to 0.1, a total load value of 0.5 corresponds to 0.08, a total load value of 1 corresponds to 0.06, a total load value of 1.5 corresponds to 0.04, and a total load value of 2 or above corresponds to 0.02. These values ​​represent the proportion of load factors. The comprehensive comparison result is equal to the proportion of distance factors multiplied by 60% plus the proportion of load factors multiplied by 40%. For example, if a person's proportion of distance factors is 0.8 and the proportion of load factors is 0.1, the comprehensive comparison result is 0.8 × 60% + 0.1 × 40% = 0.48 + 0.04 = 0.52.

[0078] The number of personnel to be assigned is determined based on the priority level of the electronic maintenance work order. Two personnel are assigned for urgent and high priority tasks, and one personnel is assigned for medium and low priority tasks. The personnel are sorted from highest to lowest according to the comprehensive comparison results, and the top N are selected, where N is the number of personnel to be assigned. If the comprehensive comparison results are the same, the closer the personnel are, the higher the priority is given. If the distances are also the same, the personnel with the smaller work ID is given priority. The final maintenance personnel assignment plan includes the name, work ID, contact number, current distance, and estimated arrival time of the selected personnel. The estimated arrival time is obtained by dividing the distance in meters by 60 meters per minute. For example, 180 meters divided by 60 equals 3 minutes. The plan must indicate the selection rules and calculation process.

[0079] Step 554: According to the final maintenance personnel assignment plan, push the electronic maintenance work order to the corresponding maintenance personnel's mobile terminal and trigger the work order receipt confirmation process; after receiving the work order confirmation signal from the maintenance personnel via the mobile terminal, update the work order status to "assigned" and simultaneously initiate the fault handling process; specifically, the system pushes the electronic maintenance work order through a dual-channel system of maintenance-dedicated APP and SMS based on the contact number in the assignment plan. The APP push content is the complete work order form and QR code, and the SMS content is "You have a new [urgent] maintenance work order (number XXX), please check the APP immediately for processing, deadline XX:XX"; immediately after the push, the receipt confirmation process is triggered, the mobile terminal APP automatically pops up a window to display the work order summary, and there are two buttons at the bottom of the pop-up window: "Confirm Receipt" and "Cannot Receipt". Clicking "Confirm Receipt" will provide a confirmation signal, and clicking "Cannot Receipt" requires filling in the reason and submitting. The system monitors the feedback signal in real time. If no confirmation is received within 1 minute, the system will automatically call the maintenance personnel to remind them; if no confirmation is received within 5 minutes and there is no "Cannot Receipt" feedback, the work order will be selected from the candidate list by comprehensive score. The system re-selects personnel and resubmits the work order. After receiving confirmation signals from all assigned personnel, the system updates the electronic maintenance work order status from "Pending Dispatch" to "Dispatched," with the update time accurate to the second. It also adds the confirmation time for each personnel to the work order record. Simultaneously, the fault handling process is initiated, which consists of three phases. The first phase is the departure phase (from dispatch to arrival at the site). The system acquires the location data of the maintenance personnel every 30 seconds, displays their movement trajectory in real time on the command center dashboard, automatically calculates the estimated arrival time, and compares it with the required time limit. An alert is issued if the time limit is exceeded. The second phase is the handling phase (from arrival at the site to submission of repair proof). Maintenance personnel must upload at least three photos of the site via the app, including the original fault appearance, the operation process, and the post-repair status. After completing each operation step, they click "Confirm" in the app, and the system synchronously updates the status to "Maintenance in Progress." The third phase is the verification phase. The system automatically sends a status verification command to the collaborative diagnostic equipment. After the equipment returns normal operating data and there are no abnormalities for 5 minutes, the work order status is updated to "Repaired," a handling report is generated, and the fault handling process ends.

[0080] The detailed extraction and standardized integration of core elements of work orders ensures the clarity of operation and maintenance instructions and the completeness of information, avoiding repetitive work caused by ambiguous operational requirements. Real-time push and multi-level confirmation processes ensure the timeliness and effectiveness of work order transmission, reduce the risk of information omission, and the phased tracking and automated verification of the fault handling process realizes transparent management of the entire process from dispatch to repair, making it easier for the command center to monitor the progress in real time and improving the efficiency and standardization of fault handling.

[0081] In a preferred embodiment of the present invention, step 6 above, during the fault handling process, involves recording diagnostic data and maintenance response information, and using the maintenance response information to adjust the diagnostic logic and handling strategy to achieve intelligent diagnosis and maintenance management of traffic light faults. This step may include:

[0082] In this embodiment of the invention, step 660 involves real-time collection and recording of diagnostic process data, maintenance personnel response time, on-site action sequence, and final result verification information related to the fault event throughout the entire process from fault handling to work order completion, generating a complete fault handling closed-loop record. Specifically, this includes: the system automatically initiates a full-process data collection mechanism when the fault is initially identified; the diagnostic process data covers four core types of information; electrical parameters are collected every 5 seconds by sensors in the signal light control box, including AC voltage (range 0-250V), DC current (range 0-5A), and power factor (range 0-1); and data is collected during abnormal fluctuations. The frequency is increased to once per second; environmental data is recorded every 30 seconds via the built-in temperature and humidity sensor in the chassis, including the internal temperature (range -20℃ to 60℃) and relative humidity (range 0-100%), while also recording outdoor weather conditions (sunny, rainy, foggy, etc.); video image data is recorded starting 10 minutes before the fault occurs, capturing a key frame every 30 seconds, focusing on preserving images of the lighting status of the light group, the appearance of the light panel, and the wiring connections, with a resolution of 1920×1080 pixels; collaborative diagnostic data records the synchronization parameters of three surrounding related signal lights, including the time difference of light color switching in the same direction (range 0-5 seconds) and the current fluctuation amplitude (percentage deviation from normal values).

[0083] The response time of maintenance personnel is recorded through a chain of timestamps automatically generated by the system. This chain includes the work order dispatch time, maintenance confirmation time, departure time, arrival time, start time of handling, and completion time of handling. Each time point is associated with a unique operation record ID. The time consumed in each stage is obtained by subtracting the previous time point from the subsequent time point. For example, the time consumed from confirmation to departure is the departure time minus the maintenance confirmation time. The time consumed in all stages is summarized to generate a detailed response time sheet. The sequence of handling actions performed on-site is selected by the maintenance personnel in the APP. The action options are subdivided according to the fault type. For example, for signal light faults, the options include checking the status of the power supply circuit breaker, measuring the continuity of the line, disassembling the light assembly housing, replacing the LED lamp module, re-crimping the wiring terminals, and testing the light color switching. Each action needs to record the start time, duration of the operation, and result. The system automatically arranges the actions in chronological order to form an action chain. Actions that are not performed according to the standard procedure will be marked as abnormal steps.

[0084] The verification information for the handling results includes three levels of verification data. Level 1 verification is the equipment self-test data, where electrical parameters (voltage stable at 220V±5%, current fluctuation ≤5%) and light color switching response time (≤0.5 seconds) are automatically uploaded for 5 consecutive minutes after the faulty equipment is restored. Level 2 verification is the feedback from the coordinating equipment, where the synchronization signal deviation of the surrounding associated traffic lights detecting the faulty equipment is ≤0.3 seconds. Level 3 verification is the manual uploading of materials, where maintenance personnel need to submit 3 on-site photos (one showing the front of the light group in its lit state, the wiring connection points, and the equipment nameplate) and 15-second video (recording the complete red, yellow, and green color switching process). All verification data must be uploaded within 30 minutes after the handling is completed. All the above data are linked by the unique fault identifier and timestamp, and organized into a structured document containing 32 fields. These fields cover basic fault information, detailed diagnostic data, response time chain, handling action chain, and level 3 verification results. Finally, a complete fault handling closed-loop record is generated and stored in a distributed database to ensure that the data is tamper-proof and traceable.

[0085] Step 661 involves structuring the closed-loop fault handling records, extracting key performance indicators, and generating a multi-dimensional performance evaluation dataset including fault diagnosis accuracy, response timeliness, and handling success rate. Specifically, this includes: firstly, preprocessing the closed-loop records using data cleaning tools to remove invalid data and correct outliers, ensuring data integrity of over 95%. When extracting key performance indicators, the calculation method for each indicator is as follows: For a single fault type, the preliminary diagnosis results and final confirmation results of the most recent 100 faults of that type are statistically analyzed. The number of correct preliminary diagnoses is divided by the total number of diagnoses. For example, in 100 cases of red traffic light failure, if 85 preliminary diagnoses and final confirmations are consistent, the diagnostic accuracy for that type is 85 divided by 100, which equals 0.85. If there are fewer than 100 records for a certain type of fault, the accuracy is calculated based on the actual number of records. For example, if 27 out of 30 records are correct, the accuracy is 27 divided by 30, which equals 0.9.

[0086] Response timeliness is calculated separately for each work order priority level. For urgent priority work orders, the ratio of the actual total response time to the standard time limit of 15 minutes is calculated. For example, if the actual time is 12 minutes, the ratio is 12 divided by 15, which equals 0.8, and the timeliness is considered met. If the actual time is 18 minutes, the ratio is 18 divided by 15, which equals 1.2, and the timeliness is considered unmet. High, medium, and low priorities are calculated using the same logic, corresponding to standard time limits of 30 minutes, 60 minutes, and 120 minutes, respectively. Finally, the compliance rate for each priority level is calculated. The success rate of handling a single fault type is calculated based on the success rate within 24 hours after the most recent 100 handling attempts. The equipment status is recorded as follows: if the equipment continues to operate normally, it is recorded as successful; if it recurs or the parameters are abnormal within 24 hours, it is recorded as a failure. The number of successes is calculated by dividing the total number of successes by the total number of failures. For example, if there are 92 successes, the success rate is 92 divided by 100, which equals 0.92. At the same time, the specific reasons for failure cases are recorded. These indicators are categorized and summarized by fault type, month, and intersection level. Each category includes fault type name, number of occurrences in the current month, diagnostic accuracy rate, response timeliness rate for each priority level, handling success rate, and distribution of failure reasons. A multi-dimensional performance evaluation dataset is generated and updated weekly to ensure data timeliness.

[0087] Step 662: Integrate the multidimensional performance evaluation dataset with the accumulated fault handling records to generate a training sample set for adjusting the fault identification model; and use the training sample set to adjust the parameters of the pre-trained fault identification model to generate the adjusted fault identification model. Specifically, this includes: when integrating the data, first associate the indicators in the multidimensional performance evaluation dataset with the accumulated fault handling records by fault type + occurrence date, and select two types of key samples. One type is fault cases with a diagnostic accuracy rate of less than 80%, which are used as samples to be optimized; the other type is cases with a diagnostic accuracy rate of more than 90% and successful handling, which are used as benchmark samples. The two types of samples are mixed in a 1:2 ratio, and complete multimodal feature data and labels are added to generate a training sample set with a total sample size of no less than 5,000, and each sample contains a 128-dimensional feature vector and a corresponding fault type label.

[0088] The specific construction steps of the pre-trained fault recognition model are as follows: First, construct a feature extraction layer. For electrical parameters (6 dimensions), design a 2-layer fully connected network. The first layer contains 128 neurons, and the second layer contains 64 neurons, outputting 32-dimensional electrical features. For video images, design a 3-layer CNN network. The first layer uses 64 3×3 convolutional kernels, the second layer uses 128 3×3 convolutional kernels, and the third layer uses 256 3×3 convolutional kernels. After global average pooling, outputting 64-dimensional visual features. For temporal data, design a 2-layer LSTM network. The first layer has 32 hidden units, and the second layer has 16 hidden units, outputting 16-dimensional temporal features. Second, construct a fusion layer to combine the electrical features (32 dimensions), visual features (64 dimensions), and temporal features (16 dimensions). The first step involves concatenating 6-dimensional data into a 112-dimensional vector, which is then mapped to a 32-dimensional fusion feature through a 1-layer fully connected network (64 neurons). The third step involves constructing a classification layer. A 3-layer fully connected network outputs the probability distributions of 16 fault types, trained using the cross-entropy loss function. This function is calculated as follows: for each sample, first obtain the probability distributions of the 16 fault types output by the fault identification model, find the probability value corresponding to the correct fault type for that sample, calculate the natural logarithm of that probability value, and then multiply this natural logarithm by -1 to obtain the loss value for a single sample. Finally, calculate the average of the loss values ​​for all training samples as the total loss. Initial parameters are generated through random initialization, and the first batch of 10,000 historical data are used for training until the loss value is below 0.15, completing the pre-training.

[0089] When adjusting the parameters of the fault identification model, the focus is on the samples to be optimized in the training sample set. The prediction bias of the fault identification model for these samples is calculated, and the weights of each layer are corrected through backpropagation. For example, for samples that misclassify poor line contact as faulty light sources, analysis revealed that the fault identification model did not pay enough attention to the instantaneous current fluctuation characteristics (a typical feature of poor line contact). The weights of the corresponding neurons in the electrical feature extraction layer were adjusted, increasing the weight value from 0.23 to 0.31 (a 35% increase). Simultaneously, the weight of the light group flicker frequency in the visual features was enhanced, and the CN... The weights of the convolutional kernels in the third layer of the N network are increased from 0.18 to 0.25 (a 39% improvement). In the fusion layer, the fusion ratio of electrical features and visual features is adjusted. The original ratio of electrical features to visual features is 40% and 40%, respectively. This is changed to 30% and 50% to enhance the influence of visual features. After each adjustment, the accuracy is tested using a validation set (20% of the total samples). If the recognition accuracy of the samples to be optimized increases to over 90% and the accuracy of the benchmark samples remains above 95%, the adjustment is stopped, and the adjusted fault recognition model is generated.

[0090] Step 663: Based on the fault feature analysis results obtained from the adjusted fault identification model, and combined with the handling efficiency and result data recorded in the training sample set, the fault level rule base and handling strategy base in the platform knowledge base are calibrated in terms of rules and parameters to generate a calibrated decision rule base. Specifically, this includes: firstly, extracting the fault feature analysis results of the adjusted fault identification model, that is, the core features and weights that the fault identification model focuses on when identifying various faults. For example, the core features of poor line contact are current fluctuation frequency > 3 times / minute and lamp group flickering times > 5 times / minute, with weights of 0.6 and 0.4 respectively; the core features of lamp source damage are current continuous < 0.1A and lamp group completely not lit, with weights of 0.7 and 0.3 respectively. These features are associated with the handling data in the training sample set, and the handling time and success rate corresponding to different feature combinations are statistically analyzed. For example, in the case of poor line contact with high current fluctuation frequency and many flickering times, the average time for handling by re-crimping the terminals is 15 minutes, with a success rate of 98%; while the average time for handling by replacing the lamp group is 30 minutes, with a success rate of 75%.

[0091] The calibration of the fault level rule base is reflected in three aspects. First, the level range of fault types is refined. In the original rules, the impact range of level one traffic light faults was uniformly set at >15,000 square meters. Now, based on feature analysis, the actual impact range of poor line contact is 30% smaller than that of light source damage. Therefore, the level one range of poor line contact is calibrated to >10,500 square meters (15,000 multiplied by 70%), and level two is 7,000-10,500 square meters. Second, the impact range calculation parameters are adjusted. The basic diffusion radius of poor line contact is calibrated from 500 meters to 350 meters (original radius multiplied by 70%), and the road network weight coefficient is calibrated from 1.3 to 1.0, because this type of fault has a smaller impact on the surrounding road network. Third, feature triggering conditions are supplemented. When the fault identification model identifies a current fluctuation frequency >5 times / minute, the level is automatically upgraded by one level, because high-frequency fluctuations are prone to triggering cascading faults.

[0092] The calibration of the handling strategy library includes: optimizing work order dispatch rules; previously, traffic light malfunctions were uniformly assigned to personnel based on priority, but now, for poor line contact (simple feature), one maintenance personnel is assigned as an emergency priority, while for light source damage + line aging (complex feature), two personnel are still assigned as an emergency priority; adjusting the handling action sequence, changing the recommended action for poor line contact from replacing the light group and testing to checking the terminals and re-crimping and testing, and clarifying the tool requirements for each action; updating the response time limit, based on sample data, the average handling time for poor line contact is 15 minutes, so the arrival time limit for its emergency priority is extended from 15 minutes to 20 minutes to avoid wasting resources; storing the calibrated level rules and handling strategies according to the hierarchical structure of fault type, feature combination, and level / strategy, with each rule associated with feature weight, sample size, and calibration date, generating a calibrated decision rule library to ensure that the rules are interpretable and traceable.

[0093] Step 664 involves updating the adjusted fault identification model and calibrated decision rule base to the central diagnostic platform, executing a new fault diagnosis and handling process, continuously recording operation and maintenance response information, and initiating a new adjustment cycle to achieve intelligent diagnosis and operation and maintenance management of traffic light faults. Specifically, this includes: the system performing the update operation during off-peak hours; first, verifying the file integrity of the adjusted fault identification model and the logical consistency of the rule base using a verification tool; after successful verification, creating backups of the old fault identification model and the old rule base; replacing the running files of the central diagnostic platform after backup; automatically loading the fault identification model file into memory after replacement; and rebuilding the index of the rule base after replacement to improve query speed; after the update, the central diagnostic platform runs according to the new process. When a fault occurs, the collected multimodal data is first input into the adjusted identification model, and the fault identification model outputs the fault type and feature weights; the platform calls the calibrated level rule base, combining feature weights to calculate the impact range and severity level; generating work orders based on the calibrated handling strategy base, specifying the assigned personnel, action sequence, and time limit; during work order execution, the system collects new response times, handling actions, and verification data in real time, with the storage format consistent with step 660.

[0094] The system sets a dual-trigger adjustment cycle: a quantitative condition, which automatically triggers a new round of adjustments every 100 new closed-loop fault handling records accumulated; and a timed condition, which automatically triggers on the last Sunday of each quarter, regardless of whether the record count has been met. After triggering, the system repeats steps 661-663 to generate a new performance evaluation dataset, adjust fault identification model parameters, and calibrate the rule base, forming a cyclical mechanism of data accumulation, fault identification model optimization, rule iteration, and practical verification. After each adjustment, an optimization report is output, including the accuracy improvement, efficiency comparison before and after strategy adjustment, and typical case analysis, providing a basis for continuous improvement in operation and maintenance management, and ultimately achieving intelligent and adaptive optimization of traffic light fault diagnosis and operation and maintenance.

[0095] The quantitative extraction of multidimensional performance evaluation indicators enables precise identification of weak links in diagnosis and treatment, improving the pertinence of improvements. The dynamic parameter adjustment of the fault identification model, combined with the weight optimization of multimodal features, makes the diagnostic results more in line with the actual fault characteristics. The calibration of the decision rule base based on real treatment efficiency data makes the fault level classification and treatment strategy more in line with the needs of the site, reducing resource misallocation and treatment delays.

[0096] like Figure 2 As shown, embodiments of the present invention also provide an intelligent diagnostic system for traffic light faults based on the Internet of Things, including:

[0097] The acquisition module is used to collect electrical parameters, environmental data, working timing information and video image data of the traffic light unit in real time, and to build a multimodal dataset of the traffic light's operating status.

[0098] The identification module is used to transmit multimodal datasets to the central diagnostic platform, use a pre-trained fault identification model to perform feature extraction and preliminary fault identification, and generate fault type and fault location information.

[0099] The diagnostic module is used to select multiple spatiotemporally correlated monitoring devices to form a diagnostic group based on fault location information, and to perform fusion analysis and collaborative diagnosis on the multi-source data of the diagnostic group to generate fault judgment results.

[0100] The judgment module is used to quantify the impact range by calculating the area enclosed by the boundary coordinates of the traffic impact area based on the fault judgment result and the fault level rule library preset in the platform knowledge base. It automatically determines the fault level and traffic impact range, generates structured fault alarm information, and pushes it to the visual decision-making dashboard of the command center for dynamic display in order to obtain intelligent handling strategies.

[0101] The assignment module is used to generate standardized electronic operation and maintenance work orders based on intelligent handling strategies, and assign the standardized electronic operation and maintenance work orders to the mobile terminals of the corresponding operation and maintenance personnel to start the fault handling process.

[0102] The management module is used to record diagnostic data and operation and maintenance response information during the fault handling process, and to adjust the diagnostic logic and handling strategy using the operation and maintenance response information to realize intelligent diagnosis and operation and maintenance management of traffic light faults.

[0103] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0104] Embodiments of the present invention also provide a computing device, including: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0105] Embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the method described above. All implementations in the above method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0106] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A smart fault diagnosis method for traffic lights based on the Internet of Things, characterized in that, The method includes: Step 1: Collect electrical parameters, environmental data, working timing information and video image data of the traffic light unit in real time to construct a multimodal dataset of traffic light operating status; Step 2: Transmit the multimodal dataset to the central diagnostic platform, use the pre-trained fault identification model to perform feature extraction and preliminary fault identification, and generate fault type and fault location information; Step 3: Based on the fault location information, and combined with a preset geographical proximity threshold and time synchronization window, determine the spatiotemporal correlation range, filter out the signal light monitoring devices within this range, and form a collaborative diagnosis group after verifying communication status and data availability; collect multi-source monitoring data from the collaborative diagnosis group during the same time period and perform spatiotemporal alignment processing to generate a standardized collaborative dataset with unified timestamps and geographical coordinates; perform consistency comparison analysis on the electrical parameters from different monitoring devices in the standardized collaborative dataset, and calculate the matching degree of display status and logical relationship of video image data from different monitoring devices; combine the above analysis and calculation results with preset rules for joint judgment to generate a comprehensive verification conclusion on whether the fault exists, and then combine the fault type and fault location information to generate the final fault judgment result; Step 4: Based on the final fault determination result, query the fault level rule library in the platform knowledge base to obtain the level determination rules and impact range calculation parameters corresponding to the fault type; determine the affected traffic network area and generate a boundary coordinate set according to the fault location information and calculation parameters; substitute the coordinate sequence of each vertex in the boundary coordinate set into the shoelace formula to calculate the area of ​​the polygonal area enclosed by the coordinate points connected in sequence, and generate a quantified traffic impact range value; input the traffic impact range value and fault type into the level determination rules to determine the fault severity level; integrate the fault type, fault location information, fault severity level and traffic impact range value to generate structured fault alarm information, push it to the command center's visual decision dashboard in real time, and generate intelligent handling strategies in combination with the preset handling strategy library; Step 5: Based on the intelligent handling strategy, extract handling instructions and skill requirements to generate a core element set of the work order, integrate structured fault alarm information, and generate a standardized electronic maintenance work order; calculate the Euclidean distance based on the fault location information in the electronic maintenance work order and the real-time acquired dynamic location data of maintenance personnel to generate a distance matrix, and combine the skill requirements with the maintenance personnel qualification database to generate a candidate list of qualified maintenance personnel; integrate the distance matrix, the candidate list, and the current work order task load status of maintenance personnel to determine the final maintenance personnel assignment scheme; push the electronic maintenance work order to the corresponding terminal according to the assignment scheme, and start the fault handling process after receiving the terminal confirmation signal; Step 6: During the fault handling process, record diagnostic data and operation and maintenance response information, and use the operation and maintenance response information to adjust the diagnostic logic and handling strategy to realize intelligent diagnosis and operation and maintenance management of traffic light faults.

2. The intelligent fault diagnosis method for traffic lights based on the Internet of Things according to claim 1, characterized in that, The multimodal dataset is transmitted to the central diagnostic platform, where a pre-trained fault identification model is used for feature extraction and preliminary fault identification, generating fault type and location information, including: The multimodal dataset is preprocessed to generate normalized multimodal data, and the normalized multimodal data is then input into a pre-trained fault identification model. By utilizing the feature extraction network in the fault identification model, deep feature learning is performed on the input normalized multimodal data to generate a multidimensional feature vector representing the state of the traffic lights. The multidimensional feature vector is input into the fault classification branch of the fault identification model to perform fault pattern recognition analysis and generate preliminary fault type judgment results. The preliminary fault type judgment result and the multi-dimensional feature vector are input into the location analysis branch of the fault identification model. The feature and fault type are correlated and analyzed to generate the location information of the faulty component or signal light corresponding to the fault type judgment result. The location information of the faulty component or signal light is integrated with the preliminary fault type judgment result to generate complete fault type and fault location information.

3. The intelligent fault diagnosis method for traffic lights based on the Internet of Things according to claim 2, characterized in that, The fault types include signal lights not lighting up, constantly lighting up, conflicting, flashing, dimming, signal lights freezing, power failure, network failure, and electrical abnormalities such as overvoltage, undervoltage, overcurrent, and leakage.

4. The intelligent fault diagnosis method for traffic lights based on the Internet of Things according to claim 3, characterized in that, Step 6 includes: Throughout the entire process from fault handling to work order completion, real-time data collection and recording are performed on diagnostic process data, maintenance personnel response time, on-site handling action sequence, and final handling result verification information related to the fault event, generating a complete fault handling closed-loop record. The closed-loop records of fault handling are structured and key performance indicators are extracted to generate a multi-dimensional performance evaluation dataset that includes fault diagnosis accuracy, response timeliness, and handling success rate. The multidimensional performance evaluation dataset is integrated with the accumulated fault handling records to generate a training sample set for adjusting the fault identification model; and the parameters of the pre-trained fault identification model are adjusted using the training sample set to generate the adjusted fault identification model. Based on the fault feature analysis results obtained from the adjusted fault identification model, and combined with the handling efficiency and result data recorded in the training sample set, the rules and parameters of the fault level rule base and handling strategy base in the platform knowledge base are calibrated to generate a calibrated decision rule base. The adjusted fault identification model and calibrated decision rule base are updated to the central diagnostic platform to execute the new fault diagnosis and handling process, and continuously record operation and maintenance response information, thus starting a new adjustment cycle and realizing intelligent diagnosis and operation and maintenance management of traffic light faults.

5. An intelligent diagnostic system for traffic light faults based on the Internet of Things, wherein the system implements the method as described in any one of claims 1 to 4, characterized in that, include: The acquisition module is used to collect electrical parameters, environmental data, working timing information and video image data of the traffic light unit in real time, and to build a multimodal dataset of the traffic light's operating status. The identification module is used to transmit multimodal datasets to the central diagnostic platform, use a pre-trained fault identification model to perform feature extraction and preliminary fault identification, and generate fault type and fault location information. The diagnostic module is used to select multiple spatiotemporally correlated monitoring devices to form a diagnostic group based on fault location information, and to perform fusion analysis and collaborative diagnosis on the multi-source data of the diagnostic group to generate fault judgment results. The judgment module is used to quantify the impact range by calculating the area enclosed by the boundary coordinates of the traffic impact area based on the fault judgment result and the fault level rule library preset in the platform knowledge base. It automatically determines the fault level and traffic impact range, generates structured fault alarm information, and pushes it to the visual decision-making dashboard of the command center for dynamic display in order to obtain intelligent handling strategies. The assignment module is used to generate standardized electronic operation and maintenance work orders based on intelligent handling strategies, and assign the standardized electronic operation and maintenance work orders to the mobile terminals of the corresponding operation and maintenance personnel to start the fault handling process. The management module is used to record diagnostic data and operation and maintenance response information during the fault handling process, and to adjust the diagnostic logic and handling strategy using the operation and maintenance response information to realize intelligent diagnosis and operation and maintenance management of traffic light faults.

6. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 4.

Images