A method, device, and system for determining faults in smart light poles.
By using high- and low-frequency differentiated data acquisition and constructing a fault propagation correlation topology map, the problem of low fault determination accuracy in smart light poles under multi-device integration conditions is solved, enabling accurate location and precise repair of the primary fault, and adapting to various application scenarios.
Patent Information
- Application Number
- CN202610930245.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-26
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-06-26
AI Technical Summary
Existing fault monitoring methods for smart light poles struggle to distinguish between primary and secondary faults in multi-device integrated environments, resulting in low fault determination accuracy and an inability to adapt to various application scenarios.
A high- and low-frequency differentiated acquisition mode is adopted to collect steady-state operation data and instantaneous electrical disturbance data of the equipment respectively. A fault propagation correlation topology map is constructed by using the transfer entropy formula and physical constraint function. By combining the health scores of the equipment and the topology propagation risk, the real fault source is screened out.
It improves the accuracy of fault diagnosis for smart light poles, enabling accurate location of the primary fault even when multiple devices issue simultaneous alarms, and adapts to various application scenarios.
Smart Images

Figure CN122471382B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of equipment testing technology, and in particular to a method, device and system for determining faults in smart light poles. Background Technology
[0002] Currently, existing fault monitoring solutions for smart light pole mounting equipment generally adopt an isolated monitoring mode with independent sensing for each device and fixed threshold alarms. The system monitors each mounting device (such as 5G base stations, lighting drivers, traffic signal controllers, etc.) individually, triggering an alarm once a limit is exceeded. However, in complex multi-device integrated operating conditions, this mode is prone to simultaneous alarms from multiple devices, making it impossible to distinguish between primary and secondary faults, and unable to determine which fault requires priority repair to minimize losses. Thus, the accuracy of existing smart light pole fault determination methods is low, making them difficult to adapt to various application scenarios. Summary of the Invention
[0003] This application provides a method, apparatus, and system for determining faults in smart light poles, which to some extent solves the problem that existing methods for determining faults in smart light poles have low accuracy and are difficult to adapt to various application scenarios.
[0004] According to one aspect of this application, a method for determining faults in smart light poles is provided. The method includes: for each mounted device of the smart light pole, acquiring first slowly varying steady-state data of the mounted device at a low frequency, and simultaneously acquiring first sudden transient data of the mounted device at a high frequency; based on the first derivative and / or effective value of the first slowly varying steady-state data, determining the triggering mounted device and a set of candidate faulty devices, and correspondingly obtaining second slowly varying steady-state data and second sudden transient data of all mounted devices in the candidate faulty device set from the first slowly varying steady-state data and the first sudden transient data, respectively; for each mounted device in the candidate faulty device set, based on the second sudden transient data and the second slowly varying steady-state data, using the transfer entropy formula and physical constraint functions, determining the directed edge of fault propagation between devices. Weights are assigned, and a directed correlation topology graph of equipment faults is constructed. Based on the second burst transient data and the second slowly changing steady-state data, a first health score is obtained. The directed correlation topology graph of equipment faults is traversed to determine the maximum fault propagation risk value in the candidate fault equipment set. Based on the first health score and the maximum fault propagation risk value, a second health score is determined. When there is a device in the candidate fault equipment set whose second health score is less than or equal to a preset first threshold, the device corresponding to the second health score is determined as the target fault equipment. When the number of target fault equipment is greater than or equal to two, the target fault equipment is updated and determined based on the weight of the directed edge of fault propagation and the triggering time of the triggering device.
[0005] Furthermore, according to one aspect of the method disclosed herein, the method further includes: predicting the health change trend of each mounted device over a future time period based on a directed correlation topology graph of device faults and a second health score; the prediction model includes at least one of the following: a temporal graph convolutional network model and a spatiotemporal prediction model; the spatiotemporal prediction model is a lightweight temporal prediction model combining graph convolution and dilated convolution.
[0006] Furthermore, according to one aspect of the method disclosed herein, the first slowly varying steady-state data includes: temperature, luminous flux, packet loss rate, and average operating current; the first burst transient data includes: power supply bus voltage and radio frequency current; for each mounted device of the smart light pole, the first slowly varying steady-state data of the mounted device is collected using a low frequency, and simultaneously, the first burst transient data of the mounted device is collected using a high frequency, including: based on a low-frequency sampling period, temperature, luminous flux, packet loss rate, and average operating current are collected sequentially using a temperature sensor, a photoelectric sensor, a mounted device communication module, and a current sensor to obtain the first slowly varying steady-state data; the low frequency includes: 0.1 Hz to 1 Hz; based on a high-frequency sampling period, power supply bus voltage and radio frequency current are collected sequentially using a mounted device voltage acquisition module and a mounted device radio frequency sampling circuit to obtain the first burst transient data, and the first burst transient data is cyclically cached and stored; the high frequency is greater than or equal to 10 Hz.
[0007] Furthermore, according to one aspect of the method of this disclosure, the candidate fault device set includes: a triggering mounted device, adjacent mounted devices, and mounted devices sharing a power supply bus with the triggering mounted device; based on the first derivative and / or effective value of the first slowly varying steady-state data, the triggering mounted device and the candidate fault device set are determined, and correspondingly, the second slowly varying steady-state data and the second sudden transient data of all mounted devices in the candidate fault device set are obtained from the first slowly varying steady-state data and the first sudden transient data, including: performing differentiation and / or absolute value processing on all the first slowly varying steady-state data to obtain the first derivative and / or effective value of the first slowly varying steady-state data, and when at any given time, at least one of the first slowly varying steady-state data... When the first derivative and / or effective value are greater than or equal to the corresponding preset second threshold and remain for a preset duration, the moment is determined as the trigger moment, and the mounted device is determined as the trigger mounted device. The trigger mounted device, the adjacent mounted devices of the adjacent trigger mounted devices, and the mounted devices sharing the power supply bus with the trigger mounted device are bound together and determined as a candidate fault device set. For each mounted device in the candidate fault device set, the second sudden transient data within a preset window length before and after the trigger moment is frozen from the first sudden transient data. At the same time, the second slowly changing steady-state data within a preset window length before and after the trigger moment is extracted from the first slowly changing steady-state data. The preset window length is less than or equal to the length of the circular cache storage.
[0008] Furthermore, according to one aspect of the method of this disclosure, the transfer entropy formula is used to calculate the fault causal association strength between devices in the candidate faulty device set; the physical constraint function includes a decay envelope function; for each mounted device in the candidate faulty device set, based on the second sudden transient data and the second slowly varying steady-state data, the transfer entropy formula and the physical constraint function are used to determine the weight of the directed edge of fault propagation between devices in the candidate faulty device set, and based on the weight of the directed edge of fault propagation, a directed association topology graph of device faults is constructed, including: for each mounted device in the candidate faulty device set, the second slowly varying steady-state data and the second sudden transient data are extracted and time series aligned to obtain the corresponding slowly varying feature time series and sudden feature time series, and the slowly varying feature time series and the sudden feature time series are concatenated to obtain the multidimensional feature time series corresponding to the mounted device; for the candidate faulty device For any two mounted devices within the set, the first multidimensional feature time series corresponding to the first mounted device and the second multidimensional feature time series corresponding to the second mounted device are input into the transfer entropy formula to calculate the fault causal association strength between the first mounted device and the second mounted device. Based on the installation spacing constraints, rated power constraints, and power supply bus subordination constraints of the first mounted device and the second mounted device, an attenuation envelope function is constructed. The fault causal association strength and the output value of the attenuation envelope function are multiplied to obtain the directed edge weight between the first mounted device and the second mounted device. The directed edge weights between any two mounted devices in the candidate fault device set are combined to obtain the fault propagation directed edge weights between each device in the candidate fault device set. All directed edge weights greater than or equal to a preset third threshold are selected from the fault propagation directed edge weights between each device and combined to obtain the device fault directed association topology graph.
[0009] Furthermore, according to one aspect of the method of this disclosure, a first health score is obtained based on second sudden transient data and second slowly varying steady-state data, and the maximum fault propagation risk value in the candidate fault device set is determined by traversing the directed association topology graph of equipment faults. A second health score is then determined based on the first health score and the maximum fault propagation risk value. This includes: for each mounted device in the candidate fault device set, mapping the second sudden transient data and the second slowly varying steady-state data using at least one normalization formula of a Sigmund function and a piecewise linear function to obtain the first health score corresponding to the mounted device; traversing all directed edges in the directed association topology graph of equipment faults, calculating the risk value propagated by a single edge of the corresponding mounted device based on the weight of each directed edge and the first health score of the corresponding mounted device, and taking the maximum value to obtain the maximum fault propagation risk value of the corresponding mounted device in the candidate fault device set; determining the normalized weight and penalty factor of the mounted device; multiplying the first health score of the mounted device by the normalized weight, and then subtracting the penalty factor multiplied by the maximum fault propagation risk value to obtain the second health score.
[0010] Furthermore, according to one aspect of the method of this disclosure, when the number of target faulty devices is greater than or equal to two, the target faulty devices are updated and determined based on the fault propagation directed edge weights and the triggering time of the triggering mounted devices, including: when the number of target faulty devices is greater than or equal to two, all target faulty devices are determined as an abnormal target fault set, and the minimum directed association subgraph corresponding to the abnormal target fault set is selected from the directed association topology graph of device faults; for any abnormal target faulty device in the minimum directed association subgraph and other abnormal target faulty devices, an exponent is calculated based on the fault propagation directed edge weights between the abnormal target faulty devices and other abnormal target faulty devices, and the time difference of the triggering time between the abnormal target faulty devices and other abnormal target faulty devices, and the source tracing score corresponding to the abnormal target faulty device is determined by summarizing; all source tracing scores in the abnormal target fault set are compared, and the abnormal target faulty device corresponding to the maximum value of the source tracing score is updated and determined as the target faulty device.
[0011] Furthermore, according to one aspect of the method disclosed herein, based on a directed correlation topology graph of equipment faults and a second health score, a prediction model is used to predict the health change trend of each mounted device over a future time period. This includes: acquiring the directed correlation topology graph of equipment faults and the second health score as training data in real time, and iteratively training at least one of a temporal graph convolutional network model and a spatiotemporal prediction model until at least one of a preset number of iterations and a preset convergence threshold is met, thus determining that the training is complete; inputting the directed correlation topology graph of equipment faults and the second health score of the smart light pole to be predicted into at least one of the trained temporal graph convolutional network model and spatiotemporal prediction model to obtain the health change trend; the health change trend includes at least one of the following: a health score change curve over a future time period, the time of fault, and the mounted device with the fault.
[0012] According to another aspect of this disclosure, a smart light pole fault determination device is provided. The device includes: a data acquisition unit, configured to acquire first slowly varying steady-state data of each mounted device on the smart light pole using a low-frequency method, and simultaneously acquire first sudden transient data of the mounted device using a high-frequency method; a first determination unit, configured to determine a triggering mounted device and a set of candidate faulty devices based on the first derivative and / or effective value of the first slowly varying steady-state data, and correspondingly obtain second slowly varying steady-state data and second sudden transient data of all mounted devices in the candidate faulty device set from the first slowly varying steady-state data and the first sudden transient data, respectively; and a construction unit, configured to, for each mounted device in the candidate faulty device set, based on the second sudden transient data and the second slowly varying steady-state data, use the transfer entropy formula and the physical... The system employs a constraint function to determine the weights of directed edges for fault propagation between devices and constructs a directed association topology graph of device faults. A second determining unit obtains a first health score based on second sudden transient data and second slowly varying steady-state data, traverses the directed association topology graph of device faults, determines the maximum fault propagation risk value in the candidate fault device set, and determines a second health score based on the first health score and the maximum fault propagation risk value. A third determining unit determines the device corresponding to the second health score as the target fault device when the second health score of any device in the candidate fault device set is less than or equal to a preset first threshold. Furthermore, when the number of target fault devices is greater than or equal to two, the unit updates and determines the target fault devices based on the weights of the directed edges for fault propagation and the triggering time of the triggered device.
[0013] According to another aspect of this disclosure, a smart light pole fault determination system is provided. The system includes: a mounting device for collecting first slowly varying steady-state data of the mounting device at a low frequency, and simultaneously collecting first sudden transient data of the mounting device at a high frequency; an edge gateway for determining the triggering mounting device and a set of candidate faulty devices based on the first derivative and / or effective value of the first slowly varying steady-state data, and correspondingly obtaining second slowly varying steady-state data and second sudden transient data of all mounting devices in the candidate faulty device set from the first slowly varying steady-state data and the first sudden transient data, respectively; and a regional management platform for determining the fault of each mounting device in the candidate faulty device set based on the second sudden transient data and the second slowly varying steady-state data using the transfer entropy formula. Based on physical constraint functions, the weights of directed edges for fault propagation between devices are determined, and a directed association topology graph of device faults is constructed. A first health score is obtained based on the second burst transient data and the second slowly varying steady-state data. The directed association topology graph of device faults is traversed to determine the maximum fault propagation risk value in the candidate fault device set. A second health score is determined based on the first health score and the maximum fault propagation risk value. When a device in the candidate fault device set has a second health score less than or equal to a preset first threshold, the device corresponding to the second health score is determined as the target fault device. When the number of target fault devices is greater than or equal to two, the target fault devices are updated and determined based on the weights of the directed edges for fault propagation and the triggering time of the triggered device.
[0014] This application provides a method, apparatus, and system for determining faults in smart light poles. This application, for each mounted device on a smart light pole, utilizes low-frequency acquisition of first slowly varying steady-state data of the mounted device, and simultaneously utilizes high-frequency acquisition of first sudden transient data of the mounted device. Based on the first derivative and / or effective value of the first slowly varying steady-state data, a set of triggering mounted devices and candidate faulty devices is determined. Correspondingly, second slowly varying steady-state data and second sudden transient data of all mounted devices in the candidate faulty device set are obtained from the first slowly varying steady-state data and the first sudden transient data, respectively. For each mounted device in the candidate faulty device set, based on the second sudden transient data and the second slowly varying steady-state data, the transfer entropy formula and physical constraint functions are used to determine the fault propagation between devices. The system assigns weights to directed edges and constructs a directed correlation topology graph of device faults. Based on the second burst transient data and the second slowly changing steady-state data, it obtains a first health score and traverses the directed correlation topology graph of device faults to determine the maximum fault propagation risk value in the candidate fault device set. Based on the first health score and the maximum fault propagation risk value, it determines a second health score. When there is a device in the candidate fault device set whose second health score is less than or equal to a preset first threshold, the device corresponding to the second health score is determined as the target fault device. When the number of target fault devices is greater than or equal to two, the target fault devices are updated and determined based on the weights of the directed edges for fault propagation and the triggering time of the device. In contrast to existing monitoring schemes that collect data from each mounted device individually, rely on single device parameter thresholds for independent fault determination, and completely sever the coupling and correlation between devices, this application adopts a high- and low-frequency differentiated acquisition mode. It collects steady-state operating data and instantaneous electrical disturbance data from the devices separately, and cyclically caches high-frequency data, completely preserving full-dimensional operational information before and after a fault occurs. Simultaneously, after a data anomaly is triggered, neighboring devices and devices sharing the same busbar form a candidate fault set, and data from the corresponding time period is uniformly extracted for joint analysis. Then, based on the collected data, the correlation between devices is calculated, and a fault propagation correlation topology is generated by combining physical constraints. A comprehensive health index is obtained by integrating the device's own data normalization score and the topology propagation risk. When multiple devices simultaneously detect a fault, the true fault source is further screened based on the fault propagation correlation between devices and the order of fault triggering time, effectively solving the pain point of not being able to distinguish between primary and secondary faults in simultaneous alarms from multiple devices. In summary, the technical solution provided in this application can improve the accuracy of fault determination for smart light poles and can adapt to various application scenarios.
[0015] It should be understood that both the foregoing general description and the following detailed description are exemplary and intended to provide further illustration of the claimed technology. Attached Figure Description
[0016] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments thereof in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain the application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.
[0017] Figure 1 A flowchart illustrating a method for determining faults in a smart light pole, as provided in an embodiment of this application; Figure 2 Another complete smart light pole fault determination system provided in the embodiments of this application; Figure 3 This is a structural block diagram of a smart light pole fault determination device provided in an embodiment of this application. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this application more apparent, exemplary embodiments according to this application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.
[0019] Currently, existing fault monitoring solutions for smart light pole mounting equipment generally adopt an isolated monitoring mode with independent sensing for each device and fixed threshold alarms. The system monitors each mounting device (such as 5G base stations, lighting drivers, traffic signal controllers, etc.) individually, triggering an alarm once a limit is exceeded. However, in complex multi-device integrated operating conditions, this mode is prone to simultaneous alarms from multiple devices, making it impossible to distinguish between primary and secondary faults, and unable to determine which fault requires priority repair to minimize losses. Thus, the accuracy of existing smart light pole fault determination methods is low, making them difficult to adapt to various application scenarios.
[0020] Therefore, to address the aforementioned problems, this application provides a method for determining faults in smart light poles. Compared to existing monitoring schemes that only collect data from each mounted device individually, rely on single device parameter thresholds to independently determine faults, and completely sever the coupling and correlation between devices, this application adopts a high- and low-frequency differentiated acquisition mode. It collects steady-state operating data and instantaneous electrical disturbance data of the devices separately and cyclically caches high-frequency data to completely retain full-dimensional operating information before and after a fault occurs. At the same time, after a data anomaly is triggered, the neighboring devices and devices sharing the busbar of the triggered device form a candidate fault set, and data from the corresponding time period is uniformly extracted for joint analysis. Then, the degree of correlation between devices is calculated based on the collected data, and a fault propagation correlation topology is generated by combining physical constraints. A comprehensive health index is obtained by combining the normalized score of the device's own data and the topology propagation risk. When multiple devices are simultaneously diagnosed as faulty, the true fault source is determined again based on the fault propagation correlation between devices and the order of fault triggering time, effectively solving the pain point that multiple devices simultaneously alarming cannot distinguish between primary and secondary faults.
[0021] This application provides a method for determining faults in smart light poles. Please refer to... Figure 1 , Figure 1 This is a flowchart illustrating a method for determining faults in a smart light pole, as provided in an embodiment of this application. Figure 1 As shown, the method includes: In step S101, for each mounted device of the smart light pole, the first slowly changing steady-state data of the mounted device is collected at a low frequency, and at the same time, the first sudden transient data of the mounted device is collected at a high frequency. In this application, a smart light pole can be understood as a multi-functional integrated urban pole infrastructure that integrates various types of electrical business equipment and has a unified power distribution layout. Mounted equipment can be understood as various electrical hardware such as 5G micro base stations, lighting drivers, information display screens, and video acquisition modules mounted on the light pole body and inside the chassis.
[0022] In this application, low frequency can be understood as a sampling frequency in the range of 0.1Hz to 1Hz. This allows for long-term tracking of the slow degradation trend of the equipment, reducing data redundancy and transmission overhead caused by routine sampling.
[0023] In this application, the first slowly varying steady-state data can be understood as steady-state operating parameters that reflect the slow aging and gradual changes in operating conditions of the equipment over a long period of time. The first slowly varying steady-state data in this application includes: temperature, luminous flux, packet loss rate, and average operating current. Among them, temperature reflects the degree of heating and aging of the components of the mounted equipment; luminous flux mainly applies to lighting mounted equipment, and shows a gradual decrease as the light source components age; packet loss rate is used to characterize the operational stability of communication modules, and a gradual upward trend will appear when the line or radio frequency components deteriorate; the average operating current reflects the overall load change of the equipment, and hardware aging will cause a slow shift in no-load or loaded current.
[0024] In this application, high frequency can be understood as a sampling frequency greater than 10Hz. This is because transient faults such as power supply disturbances and electromagnetic crosstalk have short durations, and only high-frequency sampling can capture short-term pulse anomalies.
[0025] In this application, the first burst transient data can be understood as dynamic parameters characterizing instantaneous electrical shocks and power supply coupling interference. The first burst transient data in this application includes: power supply bus voltage and radio frequency current. These two types of parameters are susceptible to short-term bursts caused by cross-device electromagnetic and power supply interference, and are also the focus of high-frequency data acquisition. It should be noted that the first burst transient data in this application is stored in a cyclic cache, which allows for rapid retrieval of data for a specified period before and after a fault after an anomaly is triggered, eliminating the need for continuous, massive uploading of high-frequency data.
[0026] In step S102, based on the first derivative and / or effective value of the first slowly changing steady-state data, the triggering mounted device and the candidate fault device set are determined, and the second slowly changing steady-state data and the second sudden transient data of all mounted devices in the candidate fault device set are obtained from the first slowly changing steady-state data and the first sudden transient data, respectively. In this application, the triggering mounted device can be understood as the mounted device whose steady-state parameter change exceeds the standard and triggers the anomaly judgment first.
[0027] In this application, the candidate faulty device set can be understood as a collection of related devices with a high probability of being affected by fault propagation. The candidate faulty device set includes: triggering mounted devices, adjacent mounted devices, and mounted devices that share a power supply bus with the triggering mounted devices. In this way, related devices are screened based on their spatial proximity and shared power supply, avoiding the omission of devices with propagated faults.
[0028] In this application, the second slowly varying steady-state data can be understood as the window segment data before and after the triggering time extracted from the full steady-state data.
[0029] In this application, the second burst transient data can be understood as high-frequency transient data of a specified period locked from the circular cache when an exception is triggered.
[0030] In step S103, for each mounted device in the candidate fault device set, based on the second burst transient data and the second slowly changing steady-state data, the weight of the directed edge for fault propagation between devices is determined using the transfer entropy formula and physical constraint function, and a directed association topology graph of device faults is constructed. In this application, the transfer entropy formula can be understood as a calculation formula that quantifies the strength of causal information flow between time-series data from different devices, as will be explained in detail later.
[0031] In this application, the physical constraint function can be understood as a calculation formula constructed by combining equipment installation spacing, rated power, and power supply subordination. The physical constraint function of this application includes an attenuation envelope function, which is used to correct spurious correlations introduced by pure data calculation and to match the actual physical conduction characteristics of the equipment. This will be elaborated upon later.
[0032] In this application, the directed edge weight for fault propagation between devices can be understood as a quantitative value that characterizes the ease with which a fault propagates from one device to another, after considering both data correlation and physical constraints.
[0033] In this application, the directed association topology graph of equipment faults can be understood as a topology graph with mounted equipment as nodes and fault propagation weights as directed connections.
[0034] In step S104, based on the second sudden transient data and the second slowly changing steady-state data, a first health score is obtained, and the directed association topology of equipment faults is traversed to determine the maximum fault propagation risk value in the candidate fault equipment set. Based on the first health score and the maximum fault propagation risk value, a second health score is determined. In this application, the first health score can be understood as a score that represents the health status of a single device based solely on the normalized conversion of its own various operating parameters.
[0035] In this application, the maximum fault propagation risk value can be understood as the highest fault cascading propagation risk coefficient of the entire group of devices obtained after traversing all propagation links of the topology.
[0036] In this application, the second health score can be understood as a comprehensive health score of the equipment that integrates the individual unit's own health level with the risk of external fault transmission.
[0037] In step S105, when there is a second health score of a mounted device in the candidate fault device set that is less than or equal to a preset first threshold, the mounted device corresponding to the second health score is determined as the target fault device. When the number of target fault devices is greater than or equal to two, the target fault devices are updated and determined based on the weight of the directed edge of the fault propagation and the triggering time of the mounted device.
[0038] In this application, the first threshold can be understood as a health score threshold that distinguishes between normal equipment operation and abnormal equipment failure. This application also provides that the first threshold can be determined by a weighted sum of health scores corresponding to factory failure threshold parameters and historical failure samples; no specific limitation is imposed.
[0039] In this application, the target faulty device can be understood as a mounted device whose comprehensive health score is lower than the critical threshold and which is preliminarily determined to have a fault or abnormality.
[0040] Specifically, the following steps can be included when determining faults in smart light poles: First, step S101 is executed, where low-frequency, slowly varying steady-state data and high-frequency, transient data are collected from all mounted equipment on the pole. The high-frequency transient data can be stored in a circular buffer. Next, step S102 is executed, where the first derivative and effective value of the steady-state data are calculated. If parameters exceed limits, the triggering device is locked, and adjacent and common busbar devices are associated to form a candidate fault set. Data for the corresponding time period is extracted and frozen from the original collected data to obtain the second slowly varying steady-state data and the second sudden transient data. Then, step S103 is executed, using the two types of extracted data, the causal correlation strength of the devices is calculated using the transfer entropy formula, and combined with… The physical constraint function in the form of attenuation envelope is used to solve for the fault propagation edge weights, and a directed association topology graph of the fault is built based on the effective weights. Then, step S104 is executed to normalize the two types of data of the device to generate the first health score, traverse the topology graph to extract the maximum fault propagation risk value, and calculate the corrected second health score by combining the aforementioned two parameters. Finally, step S105 is executed to compare the second health score with the preset first threshold, and select low-scoring devices as the initial target fault devices. If multiple devices fail at the same time, the source is traced again by combining the topology edge weights and the fault trigger time to accurately update the final fault device.
[0041] The method for determining faults in smart light poles in this application also includes: Based on the directed correlation topology graph of equipment faults and the second health score, a prediction model is used to predict the health change trend of each mounted equipment in the future time period. The prediction model includes at least one of the following: a temporal graph convolutional network model and a spatiotemporal prediction model. The spatiotemporal prediction model is a lightweight temporal prediction model that combines graph convolution with dilated convolution.
[0042] In this application, the prediction model can be understood as an intelligent computing model that infers the evolution of the subsequent health status of equipment based on the topological relationships between equipment and historical health data. The prediction model of this application includes at least one of the following: a temporal graph convolutional network model and a spatiotemporal prediction model. The temporal graph convolutional network model can be understood as a deep learning model that takes into account both the spatial topological relationships between equipment and the temporal variation patterns of health scores; the spatiotemporal prediction model can be understood as a lightweight network model that fuses graph convolution to extract spatial relationships between equipment and dilated convolution to mine long-term temporal variation features.
[0043] Specifically, the prediction process includes: using multiple sets of directed correlation topology graphs of equipment faults stored in history and the corresponding second health scores at the time to form training samples, training a time-series graph convolutional network or a lightweight spatiotemporal prediction model is completed; after training, the model parameters are saved; during the actual operation and maintenance phase, the real-time generated directed correlation topology graphs of equipment faults and the current second health scores of each device are input into the trained prediction model, and the model is used to calculate and output the future time period health score change trend of each mounted device, so as to predict the risk of equipment deterioration and failure in advance.
[0044] The following will explain in detail how to perform low-frequency sampling and high-frequency sampling, including: Based on a low-frequency sampling period, temperature, light flux, packet loss rate, and average operating current are collected sequentially using a temperature sensor, a photoelectric sensor, a mounted device communication module, and a current sensor to obtain the first slowly varying steady-state data; the low frequency includes 0.1 Hz to 1 Hz. Based on a high-frequency sampling period, the voltage acquisition module and the radio frequency sampling circuit of the mounted equipment are used to sequentially acquire the power supply bus voltage and radio frequency current to obtain the first burst transient data, and the first burst transient data is stored in a loop buffer; the high frequency is greater than or equal to 10 Hz.
[0045] In this application, during low-frequency data acquisition, various sensing and communication units can be periodically polled to continuously retain steady-state parameters that reflect the slow aging trend of the equipment. The data is archived long-term for subsequent trend analysis. Simultaneously, during high-frequency data acquisition, instantaneous power supply bus voltage and RF current can be periodically or in real-time acquired. The acquired data is stored in a circular buffer, and an in-first-out (FIFO) mechanism is used to cyclically overwrite old data, facilitating rapid retrieval of data for a specified time period after an anomaly is triggered.
[0046] In one embodiment of this application, a fixed sampling frequency of 0.5Hz can be selected for the low frequency, and all slowly changing parameters are collected every 2 seconds to continuously record the long-term operating conditions of the equipment. The high frequency uses a sampling rate of 10kHz to collect the bus voltage and radio frequency current in real time. The original high frequency data is stored in a ring buffer, and the earliest data is automatically overwritten when the buffer space is full.
[0047] In another embodiment of this application, the low frequency can be adaptively adjusted in the range of 0.1Hz to 1Hz. When the device is running smoothly, the sampling frequency is reduced, and when the device parameters fluctuate slightly, the sampling frequency is increased. The high frequency is uniformly configured to sample at 50Hz, and the stored content is refreshed cyclically according to the first-in-first-out rule only when the cache capacity is insufficient.
[0048] The following will elaborate on how to determine the candidate fault device set, as well as the second slowly varying steady-state data and the second sudden transient data, including: Differentiate and / or perform absolute value processing on all the first slowly changing steady-state data to obtain the first derivative and / or effective value of the first slowly changing steady-state data. When the first derivative and / or effective value of at least one of the first slowly changing steady-state data at any time is greater than or equal to the corresponding preset second threshold and is maintained for a preset duration, the time is determined as the trigger time and the mounted device is determined as the trigger mounted device. Bind the triggering device, the adjacent devices of the adjacent triggering device, and the devices that share the power supply bus with the triggering device to determine the set of candidate fault devices; For each mounted device in the candidate fault device set, freeze the second sudden transient data within a preset window length before and after the trigger time from the first sudden transient data. At the same time, extract the second slowly changing steady-state data within a preset window length before and after the trigger time from the first slowly changing steady-state data. The preset window length is less than or equal to the length of the circular cache storage.
[0049] In this application, differentiation and / or absolute value processing are mathematical operations performed on the time series of slowly varying parameters. The first derivative is used to calculate the rate of change of the parameter per unit time and identify the accelerated deterioration trend of the parameter. The absolute value operation is used to obtain the fluctuation amplitude of the parameter deviating from the reference value and to determine the degree of data deviation, thereby distinguishing between normal small fluctuations and abnormal changes caused by faults.
[0050] In this application, the second threshold is a critical value that distinguishes between normal fluctuations and abnormal degradation of equipment parameters, and independent thresholds are configured for different types of slowly varying steady-state data. This application also provides a specific method for determining the second threshold, including: summarizing historical monitoring data of the same model of mounted equipment operating without failure for a long time, combining it with the equipment's factory-specified performance limits, and determining the second threshold corresponding to each parameter after mathematically eliminating extreme disturbance data.
[0051] In this application, the triggering time and the triggering mounted device can be understood as the time node when the parameter change rate or offset amplitude exceeds the standard and the abnormal state is stable and continues for a preset duration, as well as the mounted device that first triggers the abnormal criterion and has the highest probability of being the source of the fault.
[0052] In this application, the candidate fault device set is a collection of candidate fault objects in the same batch, formed by aggregation based on physical coupling association rules. The devices in the set include light pole mounting hardware of different types and functions, covering cross-model mounting devices such as communication 5G micro base stations, lighting driver power supplies, and signal control equipment. Based on the electrical coupling relationship of spatially adjacent layout and shared power supply bus, the source triggering device, the nearest adjacent mounting device, and the mounting device powered by the same bus are bound and aggregated, and uniformly classified into the same candidate conducted fault event. This is used to extract fault time period data in batches from the association dimension, avoiding the problem of missed detection caused by fault cross-device propagation.
[0053] In this application, the preset window length is the time span of data to be extracted for fault tracing analysis. This application also provides a specific method for determining the window length, including: comprehensively calculating and calibrating by combining the average time taken for a fault to propagate from its inception to its spread in historical fault samples of the device, the hardware storage capacity of the circular buffer, and the high-frequency sampling rate; preferably, determining the base duration based on the duration of past fault waveforms of similar devices; and then combining this with the upper limit of the gateway's circular cache storage space for an upper limit constraint, without specifying a particular limit. It is important to note that the window length cannot exceed the upper limit of the storage duration of the high-frequency data circular cache to ensure that the high-frequency data before and after the trigger point to be retrieved is completely retained, preventing the loss of critical data due to cache overwriting.
[0054] Specifically, when determining the candidate fault device set, the second slowly varying steady-state data, and the second sudden transient data, the following may be included: First, derivatives and absolute values are calculated for all the first slowly varying steady-state time-series data collected from all devices to obtain the rate of change and effective value of fluctuation. The calculation results are compared with the second threshold of the corresponding parameters one by one. When any index exceeds the limit and remains continuously for a preset time, the current time is recorded as the trigger time, and the corresponding device is recorded as the trigger-attached device. Second, based on the two screening conditions of spatial proximity and shared power supply bus, the surrounding related devices of the trigger device are collected and integrated to form a complete set of candidate fault devices. Finally, according to the preset window duration, the window period data is extracted from the first slowly varying steady-state data in the original stored data as the second slowly varying steady-state data, and the data in the same time period is locked in the first burst transient data in the cyclic cache as the second burst transient data.
[0055] The following section will explain in detail how to construct a directed association topology graph of device faults, including: For each mounted device in the candidate fault device set, the second slowly changing steady-state data and the second sudden transient data are extracted and time series aligned to obtain the slowly changing feature time series and the sudden feature time series. The slowly changing feature time series and the sudden feature time series are then spliced together to obtain the multi-dimensional feature time series corresponding to the mounted device. For any two mounted devices in the candidate fault device set, input the first multidimensional feature time series corresponding to the first mounted device and the second multidimensional feature time series corresponding to the second mounted device into the transfer entropy formula to calculate the fault causal association strength between the first mounted device and the second mounted device. Based on the installation spacing constraints, rated power constraints, and power supply bus subordination constraints of the first and second mounted equipment, an attenuation envelope function is constructed. Multiply the fault causal correlation strength and the output value of the decay envelope function to obtain the directed edge weight between the first mounted device and the second mounted device. Combine the directed edge weights between any two mounted devices in the candidate fault device set to obtain the fault propagation directed edge weights between devices in the candidate fault device set. From the directed edge weights of fault propagation between devices, all directed edge weights greater than or equal to a preset third threshold are selected and combined to obtain a directed association topology graph of device faults.
[0056] In this application, extraction and time series alignment processing can be understood as removing invalid or missing data, and normalizing low-frequency steady-state data and high-frequency transient data with different sampling frequencies to the same time scale axis to achieve time series synchronization matching. The resulting slowly varying characteristic time series and burst characteristic time series can be understood as long-period time series data composed of slowly varying parameters such as temperature and current mean, and short-time pulse time series data composed of instantaneous sampling values of bus voltage and radio frequency current, respectively.
[0057] In this application, the multidimensional feature time series can be understood as an integrated fusion feature series of a single device after integrating steady-state degradation features and instantaneous electrical disturbance features and time-series alignment, which serves as the input data source for the calculation of the transfer entropy formula.
[0058] In this application, the transfer entropy formula can be understood as a calculation relationship that quantifies the strength of the unidirectional causal information flow between the time-series data of two devices, and is used to calculate the statistical correlation strength of fault propagation from one device to another at the pure data level.
[0059] In this application, the attenuation envelope function can be understood as a correction coefficient function constructed based on the actual physical installation conditions of the light pole, used to correct spurious correlations caused by pure data statistics; it is determined by installation spacing constraints, rated power constraints, and power supply bus subordination constraints. Specifically: the installation spacing constraint characterizes that the farther apart the equipment is, the weaker the electrical coupling effect of the fault, and the coefficient decays exponentially with distance; the rated power constraint relies on the ratio of the rated power of the equipment to the system reference power to limit the influence of power matching degree on fault coupling; the power supply bus subordination constraint is used to distinguish whether two devices are powered by the same bus, retaining a high coupling coefficient on the same bus and configuring a minimal correction coefficient on different buses.
[0060] In this application, the directed edge weight can be understood as a quantitative index that quantifies the difficulty of fault propagation from the starting device to the ending device after integrating data causal relationships and objective physical constraints. The larger the weight, the higher the probability of fault propagation across devices.
[0061] In this application, the third threshold can be understood as a significance threshold for screening effective transmission links, used to eliminate low-weight spurious association edges that have no actual physical transmission significance. This application also provides a specific method for determining the third threshold, including: calibrating the significance threshold value based on the weight distribution of historical fault samples and combining it with statistical hypothesis testing methods; no specific limitations are imposed.
[0062] Specifically, when determining the directed topology graph of equipment faults, it can include: First, perform time-series alignment and feature splicing on the two types of data extracted from all devices in the candidate set to generate a multi-dimensional time series; substitute each pair of devices into the transfer entropy formula to obtain the original causal association strength; then solve the attenuation envelope correction coefficient based on the three physical constraints of distance, power, and bus; multiply the two to obtain the original directed edge weights; finally, compare all weights with the third threshold, retain the qualified weights, and connect the devices as nodes and the effective weights to finally generate a fault directed association topology graph.
[0063] For example, this application provides specific formulas for the step-by-step calculation of directed weights and topology graphs: Step 1: Calculate the strength of the causal association between devices using the transfer entropy formula (candidate information flow metric):
[0064] in, For the equipment Pointing device The strength of the causal association of one-way failures, i.e., the candidate information flow metric; For equipment exist Multidimensional feature time-series sampled values at each moment; For equipment exist Multidimensional feature time-series sampled values at each moment; For equipment exist Multidimensional feature time-series sampled values at each moment; This refers to the joint probability and conditional probability of the occurrence of time-series combinations of corresponding multidimensional features.
[0065] Step 2: Construct a prior function for the attenuation envelope based on three physical constraints: installation spacing, rated power, and power supply bus.
[0066] in, This is an attenuation correction coefficient for merging three types of physical constraints: distance, power, and bus. The installation distance attenuation term characterizes the exponential decay of fault coupling strength with increasing equipment spacing, while This is the dimensionless distance attenuation adjustment coefficient; For equipment , The actual physical installation spacing between them; This is a rated power constraint term, characterizing the impact of equipment power matching on fault propagation, while This is the dimensionless power attenuation adjustment coefficient; For equipment Rated power; To establish a unified reference power for the system; For the power supply bus constraint function, The common busbar has a value of 1, and the non-common busbar has a minimum constant value. .
[0067] Step 3: Multiply and modulate the fault causality strength with the attenuation envelope coefficient to obtain the directed edge weights:
[0068] in, To integrate data causal relationships with actual physical constraints, the device Pointing device The final fault propagation is determined by the directed edge weights; the higher the weight value, the greater the probability of fault propagation across devices.
[0069] Step 4: Construct a topology graph and preset a third saliency threshold. Only when At that time, the directed edges from node to node are retained, and the candidate faulty devices are used as the topology node set V. The valid directed edges and their corresponding weights that are selected and retained constitute the edge set. Weight set This ultimately forms a directed correlation topology of faults. .
[0070] The following will explain in detail how to obtain the second health score, including: For each mounted device in the candidate faulty device set, the second sudden transient data and the second slowly changing steady-state data are mapped using at least one normalization formula of the Sigmund function and the piecewise linear function to obtain the first health score corresponding to the mounted device. Traverse all directed edges in the directed association topology graph of equipment faults. Based on the weight of each directed edge and the first health score of the corresponding mounted equipment, calculate the risk value of the single edge propagation of the corresponding mounted equipment item by item and take the maximum value to obtain the maximum fault propagation risk value of the corresponding mounted equipment in the candidate fault equipment set. Determine the normalized weights and penalty factors for the mounted equipment; The second health score is obtained by multiplying the first health score of the mounted equipment by the normalized weight, and then subtracting the penalty factor multiplied by the maximum failure propagation risk value.
[0071] In this application, the normalization formula can be understood as a numerical transformation formula that maps steady-state and transient monitoring parameters with different dimensions and numerical ranges to a unified health score range [0, 100]. The normalization formula in this application includes at least one of the Sigmund function and a piecewise linear function. The Sigmund function can be understood as a smooth nonlinear mapping function that can achieve continuous decay of the health score when the parameter deviates from the baseline value, adapting to the score change law of the gradual aging of the equipment; the piecewise linear function can be understood as a piecewise threshold linear mapping formula, in which different slopes can be configured in the normal parameter range, the slightly deteriorated range, and the severely faulty range, accurately matching the scoring logic of the phased deterioration of the equipment.
[0072] In this application, normalized weights can be understood as dimensionless weighted coefficients determined based on the power supply level and service importance of the equipment within the light pole system. The normalization converges after the weights of all equipment are weighted and summed. This application also provides a specific method for determining the calculation of normalized weights: basic weights can be assigned based on the service attributes (communication, lighting, control) and power supply topology level of the mounted equipment.
[0073] In this application, the penalty factor can be understood as a detrimental coefficient characterizing the impact of fault propagation risk on the overall health of equipment, used to quantify the health loss caused by adjacent faults propagating along the topology link. This application also provides a specific method for calculating the penalty factor: a benchmark value can be determined based on statistical data of health drops caused by fault propagation in a full set of historical fault samples, and then adjusted within intervals based on the light pole busbar power distribution architecture and the strength of coupling between equipment. A larger penalty factor is selected in the same power supply busbar network scenario, while a smaller penalty factor is selected in the cross-busbar weak coupling scenario.
[0074] Specifically, determining the second health score may include the following steps: First, a sigmoid function or piecewise linear normalization algorithm is selected to perform dimensionless mapping on the device's second slowly varying steady-state data and second sudden transient data, generating a first health score based solely on its own operating conditions. Second, all directed edges within the fault-directed topology pointing from external devices to the current device are traversed, and the propagation risk is calculated for each edge based on its weight and the source device's health value. The maximum value among all risk values is selected as the device's maximum fault propagation risk. Then, normalized weights and fixed propagation penalty factors are configured according to the device's service priority. Finally, based on the weighted penalty calculation logic, the device's own health score and propagation penalty term are combined to obtain the corrected second health score.
[0075] For example, this application also provides a specific formula for calculating the distribution of the second health score: Step 1: Normalize and map the monitoring parameters of device k to obtain the first health score, i.e., the node health level. .
[0076] Step 2: Determine the maximum transmission risk of the device when there is a single directed edge:
[0077] in, To maximize the risk of fault propagation; For equipment Pointing device The final fault propagation directed edge weight; Source device The first health score.
[0078] Step 3: Determine the second health score (overall health score for a single device):
[0079] in, For equipment The final second health score; For equipment Normalized key weights; For equipment The first health score obtained after normalization; This is the fault propagation penalty factor.
[0080] The following will explain in detail how to determine the target faulty device when multiple faulty devices occur, including: When the number of target faulty devices is greater than or equal to two, all target faulty devices are identified as an abnormal target fault set, and the minimum directed association subgraph corresponding to the abnormal target fault set is selected from the directed association topology graph of device faults. For any abnormal target fault device and other abnormal target fault devices within the minimum directed correlation subgraph, an exponent is calculated based on the weight of the directed edge of fault propagation between the abnormal target fault device and other abnormal target fault devices, and the time difference of the triggering time between the abnormal target fault device and other abnormal target fault devices. The source tracing score corresponding to the abnormal target fault device is then determined by summarizing the results. Compare all source tracing scores in the abnormal target fault set, and update and determine the abnormal target fault device corresponding to the maximum value in the source tracing score as the target fault device.
[0081] In this application, the minimum effective correlation subgraph can be understood as a topological subgraph consisting of all abnormal target faulty devices as nodes, retaining only the effective fault propagation directed edges that connect the abnormal nodes, eliminating idle devices and isolated topological branches that are unrelated to the abnormal set, and simplifying the calculation scope of fault tracing.
[0082] In this application, the source tracing score can be understood as the probability score of the initial fault obtained by comprehensively considering the fault propagation link weight and the device alarm timing difference. The higher the score, the higher the probability that the device is the source of the initial fault and induces secondary alarms in other devices.
[0083] Specifically, when multiple faulty devices are detected, identifying the target faulty device can include the following steps: The first step is to collect all the initially determined target fault devices ≥ 2 after comparing the second health score to form an abnormal target fault set. Based on the previously constructed global device fault directed association topology, extract only the fault nodes in the set and the effective fault propagation directed edges that connect the nodes to each other. Trim and generate the minimum directed association subgraph corresponding to the abnormal set, remove irrelevant normal devices and isolated topology branches, and narrow the scope of source tracing calculation.
[0084] The second step involves taking each abnormal device in the set as a suspected originating node, traversing all other abnormal faulty devices that can be directly reached from that node, retrieving the pre-calculated weights of the directed edges for fault propagation between each pair of devices, as well as the time difference between the fault triggering times of the two devices, and calculating the component coefficients of a single propagation path item by item using the exponential formula. Then, the component coefficients of all reachable paths are multiplied continuously, and the results are summed to obtain the source tracing score corresponding to the current candidate faulty device.
[0085] The third step is to perform a horizontal comparison and screening of the source tracing scores of all devices in the abnormal target fault set, select the device with the highest source tracing score as the original source of the fault, update and lock this device as the final accurate target fault device, and determine the other alarm devices in the same batch as secondary fault devices caused by the cross-link propagation of the fault.
[0086] For example, this application also provides a formula for calculating the traceability score:
[0087] in, For equipment The corresponding source tracing score; For abnormal target fault set; For equipment Pointing device The final fault propagation directed edge weight; For equipment With equipment The time difference between the time the fault is triggered.
[0088] The following will elaborate on how this application utilizes a predictive model for prediction, including: The directed association topology graph of equipment faults and the second health score are acquired in real time as training data, and at least one of the temporal graph convolutional network model and spatiotemporal prediction model is trained iteratively until at least one of the preset number of iterations and preset convergence threshold is met, and the training is determined to be complete. The directed association topology graph of the equipment faults of the smart light pole to be predicted and the second health score are input into at least one of the trained time-series graph convolutional network model and spatiotemporal prediction model to obtain the health change trend; the health change trend includes at least one of the following: the health score change curve in the future time period, the time of failure and the faulty mounted equipment.
[0089] In this application, the training data can be understood as multiple sets of samples archived in timestamp order. Each sample consists of a directed association topology of pole failure at the same time and a pairing of comprehensive second health scores corresponding to each device.
[0090] In this application, the preset number of iterations can be determined based on the total number of samples and the model convergence rate, preferably 50 to 200 rounds; the preset convergence threshold can be understood as the fluctuation critical value of the loss function in multiple consecutive iterations, preferably determined when the loss change amplitude is less than 0.001.
[0091] In this application, the health change trend can be understood as the result of the equipment's subsequent operating condition evolution derived by the model based on topological correlation and historical health evolution patterns. The health change trend of this application includes at least one of the following: a health score change curve for a future time period, the time of failure, and the equipment on which the failure occurred. Among them, the health score change curve visually displays the rise and fall trend of a single equipment's health score over time, predicting potential gradual aging risks; the time of failure and the equipment on which the failure occurred are used to locate potential failure points and the estimated time of failure in advance.
[0092] Specifically, when using a predictive model for prediction, the following steps may be included: The first step is to continuously collect historical time-series data, pair and organize the fault-directed correlation topology and the corresponding device second health score according to time, and divide the training sample set; train the time-series graph convolution or lightweight spatiotemporal prediction model in batch iterative way, calculate the model prediction loss after each round of training, iterate until the set number of iterations is reached or the loss is lower than the convergence threshold, end the model training and save the model parameters. The second step is to collect the current fault-directed association topology map of the target light pole and the real-time second health score of each device online in real time, and organize it into input data with the same format as the training samples. The third step is to import the processed real-time data into the converged prediction model. The model combines the spatial topology coupling relationship of the equipment with the health time series evolution law to output the subsequent health score change curve of the fully mounted equipment, the estimated time of failure, and the high-risk failure equipment, thus completing the early failure prediction.
[0093] For example, Figure 2 Another complete smart light pole fault determination system provided in this application embodiment includes: First, operational data collection is conducted on multiple mounted devices. Low-frequency inspections are used to obtain the first gradually changing steady-state data, and high-frequency cyclic storage is used to obtain the first sudden transient data. Then, the first derivative and effective value of the low-frequency data are calculated and compared with a preset trigger threshold. If the trigger condition is not met, continuous data collection resumes. If the trigger condition is met, a preset window of steady-state and transient data is frozen, generating the second gradually changing steady-state data and the second sudden transient data. Next, triggering devices, adjacent devices, and devices sharing the same busbar are bound to form a candidate fault device set. Finally, the fault causal correlation strength is calculated using the transfer entropy formula, and the fault causal correlation strength is determined by the device spacing / power / busbar ratio. The system constructs an attenuation envelope function, multiplies two values to obtain directed edge weights, and then generates a directed correlation topology graph of the fault. Based on the topology and the extracted data, it calculates the first health score and the maximum fault propagation risk of each device, and combines the weights and penalty factors to obtain the second health score. Then it determines whether there is a multi-device alarm: if there is no multi-device alarm, it directly enters the operation and maintenance processing and feedback; if there is a multi-device alarm, it extracts the minimum directed correlation subgraph corresponding to the abnormal device, completes the source tracing operation based on the subgraph, outputs the fault diagnosis report, and finally all the results are incorporated into the operation and maintenance processing and feedback process. The inspection data is used to reverse optimize the entire set of judgment parameters.
[0094] This application also provides a smart light pole fault determination system, the system comprising: The mounting device is used to acquire the first slowly varying steady-state data of the mounting device at a low frequency, and at the same time, to acquire the first sudden transient data of the mounting device at a high frequency. The edge gateway is used to determine the triggering mounted device and the candidate fault device set based on the first derivative and / or effective value of the first slowly varying steady-state data, and correspondingly obtain the second slowly varying steady-state data and the second sudden transient data of all mounted devices in the candidate fault device set from the first slowly varying steady-state data and the first sudden transient data, respectively. The regional management platform is used to determine the weights of directed edges for fault propagation between devices for each mounted device in the candidate fault device set, based on second burst transient data and second slowly varying steady-state data, using the transfer entropy formula and physical constraint functions, and constructing a directed association topology graph of device faults. Based on the second burst transient data and the second slowly varying steady-state data, a first health score is obtained, and the directed association topology graph of device faults is traversed to determine the maximum fault propagation risk value in the candidate fault device set. Based on the first health score and the maximum fault propagation risk value, a second health score is determined. When there is a mounted device in the candidate fault device set whose second health score is less than or equal to a preset first threshold, the mounted device corresponding to the second health score is determined as the target fault device. When the number of target fault devices is greater than or equal to two, the target fault devices are updated and determined based on the weights of the directed edges for fault propagation and the triggering time of the mounted device.
[0095] This application also provides a smart light pole fault determination device. Figure 3 This application provides a structural block diagram of a smart light pole fault determination device, as shown in the embodiments below. Figure 3 As shown, the smart light pole fault determination device 300 includes: The acquisition unit 301 is used to acquire the first slowly changing steady-state data of each mounted device of the smart light pole at a low frequency, and at the same time, acquire the first sudden transient data of the mounted device at a high frequency. The first determining unit 302 is used to determine the triggering mounted device and the candidate fault device set based on the first derivative and / or effective value of the first slowly changing steady-state data, and correspondingly obtain the second slowly changing steady-state data and the second sudden transient data of all mounted devices in the candidate fault device set from the first slowly changing steady-state data and the first sudden transient data, respectively. Construction unit 303 is used to determine the weight of the directed edge of fault propagation between each device based on the second burst transient data and the second slowly changing steady-state data, using the transfer entropy formula and physical constraint function, and to construct a directed association topology graph of device faults for each mounted device in the candidate fault device set. The second determining unit 304 is used to obtain a first health score based on the second sudden transient data and the second slowly changing steady-state data, and to traverse the directed association topology graph of equipment faults to determine the maximum fault propagation risk value in the candidate fault equipment set, and to determine a second health score based on the first health score and the maximum fault propagation risk value. The third determining unit 305 is used to determine the device corresponding to the second health score as the target fault device when the second health score of the mounted device in the candidate fault device set is less than or equal to the preset first threshold, and when the number of target fault devices is greater than or equal to two, update the determined target fault device based on the weight of the directed edge of fault propagation and the triggering time of the mounted device.
[0096] In one exemplary embodiment, the third determining unit 305 is further configured to: predict the health change trend of each mounted device over a future time period based on the directed correlation topology graph of device faults and the second health score; the prediction model includes at least one of the following: a temporal graph convolutional network model and a spatiotemporal prediction model; the spatiotemporal prediction model is a lightweight temporal prediction model that combines graph convolution with dilated convolution.
[0097] In one exemplary embodiment, the acquisition unit 301 is specifically used to: based on a low-frequency sampling period, use a temperature sensor, a photoelectric sensor, a mounted device communication module, and a current sensor to sequentially acquire temperature, luminous flux, packet loss rate, and average operating current to obtain first slowly varying steady-state data; the low frequency includes 0.1 Hz to 1 Hz; based on a high-frequency sampling period, use a mounted device voltage acquisition module and a mounted device RF sampling circuit to sequentially acquire power supply bus voltage and RF current to obtain first burst transient data, and cyclically cache and store the first burst transient data; the high frequency is greater than or equal to 10 Hz.
[0098] In one exemplary embodiment, the first determining unit 302 is specifically configured to: perform derivative and / or absolute value processing on all first slowly varying steady-state data to obtain the first derivative and / or effective value of the first slowly varying steady-state data; and when the first derivative and / or effective value of at least one first slowly varying steady-state data at any given time is greater than or equal to the corresponding preset second threshold and is maintained for a preset duration, determine that time as the trigger time, and determine the mounted device as the trigger mounted device; bind the trigger mounted device, adjacent mounted devices of adjacent trigger mounted devices, and mounted devices sharing the power supply bus with the trigger mounted device to determine a set of candidate fault devices; for each mounted device in the candidate fault device set, freeze the second sudden transient data within a preset window length before and after the trigger time from the first sudden transient data, and simultaneously extract the second slowly varying steady-state data within a preset window length before and after the trigger time from the first slowly varying steady-state data; the preset window length is less than or equal to the length of the circular cache storage.
[0099] In one exemplary embodiment, the construction unit 303 is specifically used for: extracting and aligning the second slowly varying steady-state data and the second sudden transient data for each mounted device in the candidate faulty device set, thereby obtaining a slowly varying feature time series and a sudden feature time series, and concatenating the slowly varying feature time series and the sudden feature time series to obtain a multidimensional feature time series corresponding to the mounted device; for any two mounted devices in the candidate faulty device set, inputting the first multidimensional feature time series corresponding to the first mounted device and the second multidimensional feature time series corresponding to the second mounted device into the transfer entropy formula to calculate the first mounted device and the second mounted device... The fault causal correlation strength between the devices is determined; based on the installation spacing constraints, rated power constraints, and power supply bus subordination constraints of the first and second mounted devices, an attenuation envelope function is constructed; the fault causal correlation strength and the output value of the attenuation envelope function are multiplied to obtain the directed edge weight between the first and second mounted devices; the directed edge weights between any two mounted devices in the candidate fault device set are combined to obtain the fault propagation directed edge weights between each device in the candidate fault device set; all directed edge weights greater than or equal to a preset third threshold are selected from the fault propagation directed edge weights between each device, and combined to obtain the device fault directed correlation topology graph.
[0100] In one exemplary embodiment, the second determining unit 304 is specifically configured to: for each mounted device in the candidate faulty device set, map the second sudden transient data and the second slowly changing steady-state data using at least one normalization formula of the Sigmund function and the piecewise linear function to obtain the first health score corresponding to the mounted device; traverse all directed edges in the directed association topology graph of device faults, calculate the risk value of single edge propagation of the corresponding mounted device item by item based on the weight of each directed edge and the first health score of the corresponding mounted device, and take the maximum value to obtain the maximum fault propagation risk value of the corresponding mounted device in the candidate faulty device set; determine the normalized weight and penalty factor of the mounted device; multiply the first health score of the mounted device by the normalized weight, and then subtract the penalty factor multiplied by the maximum fault propagation risk value to obtain the second health score.
[0101] In one exemplary embodiment, the third determining unit 305 is specifically configured to: when the number of target faulty devices is greater than or equal to two, determine all target faulty devices as an abnormal target fault set, and select the minimum directed association subgraph corresponding to the abnormal target fault set from the directed association topology graph of device faults; for any abnormal target faulty device in the minimum directed association subgraph and other abnormal target faulty devices, perform exponential calculation based on the fault propagation directed edge weight between the abnormal target faulty device and other abnormal target faulty devices and the time difference of the triggering time between the abnormal target faulty device and other abnormal target faulty devices, and summarize and determine the source tracing score corresponding to the abnormal target faulty device; compare all source tracing scores in the abnormal target fault set, and update and determine the abnormal target faulty device corresponding to the maximum value in the source tracing score as the target faulty device.
[0102] In one exemplary embodiment, the third determining unit 305 further includes: acquiring the directed correlation topology graph of device faults and the second health score as training data in real time, and iteratively training at least one of the time-series graph convolutional network model and the spatiotemporal prediction model until at least one of the preset number of iterations and the preset convergence threshold is met, and determining that the training is completed; inputting the directed correlation topology graph of device faults and the second health score of the smart light pole to be predicted into at least one of the trained time-series graph convolutional network model and the spatiotemporal prediction model to obtain the health change trend; the health change trend includes at least one of the following: the health score change curve for a future time period, the fault time, and the faulty mounted device.
[0103] In summary, this application provides a method, apparatus, and system for determining faults in smart light poles. This application, for each mounted device on a smart light pole, utilizes low-frequency acquisition of first slowly varying steady-state data of the mounted device, and simultaneously utilizes high-frequency acquisition of first sudden transient data of the mounted device. Based on the first derivative and / or effective value of the first slowly varying steady-state data, a set of triggering mounted devices and candidate faulty devices is determined. Correspondingly, second slowly varying steady-state data and second sudden transient data of all mounted devices in the candidate faulty device set are obtained from the first slowly varying steady-state data and the first sudden transient data, respectively. For each mounted device in the candidate faulty device set, based on the second sudden transient data and the second slowly varying steady-state data, the transfer entropy formula and physical constraint functions are used to determine the fault propagation between devices. The system assigns weights to directed edges and constructs a directed correlation topology graph of device faults. Based on the second burst transient data and the second slowly changing steady-state data, it obtains a first health score and traverses the directed correlation topology graph of device faults to determine the maximum fault propagation risk value in the candidate fault device set. Based on the first health score and the maximum fault propagation risk value, it determines a second health score. When there is a device in the candidate fault device set whose second health score is less than or equal to a preset first threshold, the device corresponding to the second health score is determined as the target fault device. When the number of target fault devices is greater than or equal to two, the target fault devices are updated and determined based on the weights of the directed edges for fault propagation and the triggering time of the device. In contrast to existing monitoring schemes that collect data from each mounted device individually, rely on single device parameter thresholds for independent fault determination, and completely sever the coupling and correlation between devices, this application adopts a high- and low-frequency differentiated acquisition mode. It collects steady-state operating data and instantaneous electrical disturbance data from the devices separately, and cyclically caches high-frequency data, completely preserving full-dimensional operational information before and after a fault occurs. Simultaneously, after a data anomaly is triggered, neighboring devices and devices sharing the same busbar form a candidate fault set, and data from the corresponding time period is uniformly extracted for joint analysis. Then, based on the collected data, the correlation between devices is calculated, and a fault propagation correlation topology is generated by combining physical constraints. A comprehensive health index is obtained by integrating the device's own data normalization score and the topology propagation risk. When multiple devices simultaneously detect a fault, the true fault source is further screened based on the fault propagation correlation between devices and the order of fault triggering time, effectively solving the pain point of not being able to distinguish between primary and secondary faults in simultaneous alarms from multiple devices. In summary, the technical solution provided in this application can improve the accuracy of fault determination for smart light poles and can adapt to various application scenarios.
[0104] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments claimed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0105] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details of the above application are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.
[0106] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.
[0107] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.
[0108] It should also be noted that in the system and method of this application, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions of this application.
[0109] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.
[0110] The above description of the claimed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be applied within the widest scope consistent with the principles and novel features of this application.
[0111] The above description has been given for illustrative and descriptive purposes. Furthermore, this description is not intended to limit the embodiments of this application to the forms described herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.
Claims
1. A method for determining faults in smart light poles, characterized in that, The method includes: For each mounted device of the smart light pole, the first slowly changing steady-state data of the mounted device is collected at a low frequency, while the first sudden transient data of the mounted device is collected at a high frequency. Based on the first derivative and / or effective value of the first slowly changing steady-state data, the triggering mounted device and the candidate fault device set are determined, and the second slowly changing steady-state data and the second sudden transient data of all mounted devices in the candidate fault device set are obtained from the first slowly changing steady-state data and the first sudden transient data, respectively. For each mounted device in the candidate fault device set, based on the second sudden transient data and the second slowly changing steady-state data, the weight of the directed edge for fault propagation between devices is determined using the transfer entropy formula and physical constraint function, and a directed association topology graph of device faults is constructed. Based on the second sudden transient data and the second slowly changing steady-state data, a first health score is obtained, and the directed association topology of the equipment fault is traversed to determine the maximum fault propagation risk value in the candidate fault equipment set. Based on the first health score and the maximum fault propagation risk value, a second health score is determined. When there is a device in the candidate fault device set whose second health score is less than or equal to a preset first threshold, the device corresponding to the second health score is determined as the target fault device. When the number of target fault devices is greater than or equal to two, the target fault device is updated and determined based on the fault propagation directed edge weight and the triggering time of the triggering device. The transfer entropy formula is used to calculate the fault causal correlation strength between devices in the candidate faulty device set; the physical constraint function includes a decay envelope function; for each mounted device in the candidate faulty device set, based on the second sudden transient data and the second slowly varying steady-state data, the transfer entropy formula and physical constraint function are used to determine the weights of the directed fault propagation edges between devices in the candidate faulty device set, and based on the weights of the directed fault propagation edges, a directed correlation topology graph of device faults is constructed, including: For each mounted device in the candidate fault device set, the second slowly changing steady-state data and the second sudden transient data are extracted and time series aligned to obtain the slowly changing feature time series and the sudden feature time series. The slowly changing feature time series and the sudden feature time series are then spliced together to obtain the multi-dimensional feature time series corresponding to the mounted device. For any two mounted devices in the candidate fault device set, the first multidimensional feature time series corresponding to the first mounted device and the second multidimensional feature time series corresponding to the second mounted device are input into the transfer entropy formula to calculate the fault causal association strength between the first mounted device and the second mounted device. Based on the installation spacing constraints, rated power constraints, and power supply bus subordination constraints of the first mounted equipment and the second mounted equipment, the attenuation envelope function is constructed. Multiply the fault causal correlation strength and the output value of the decay envelope function to obtain the directed edge weight between the first mounted device and the second mounted device, and combine the directed edge weights between any two mounted devices in the candidate fault device set to obtain the fault propagation directed edge weight between each device in the candidate fault device set. From the directed edge weights of the fault propagation between the devices, all directed edge weights greater than or equal to a preset third threshold are selected and combined to obtain the directed association topology graph of the device faults. The process involves obtaining a first health score based on the second sudden transient data and the second slowly changing steady-state data, traversing the directed association topology of the equipment faults to determine the maximum fault propagation risk value in the candidate faulty equipment set, and determining a second health score based on the first health score and the maximum fault propagation risk value, including: For each mounted device in the candidate fault device set, the second sudden transient data and the second slowly changing steady-state data are mapped using at least one normalization formula of the Sigmund function and the piecewise linear function to obtain the first health score corresponding to the mounted device. Traverse all directed edges in the directed association topology of the device faults, and calculate the risk value of single edge propagation for each device based on the weight of each directed edge and the first health score of the corresponding mounted device, and take the maximum value to obtain the maximum fault propagation risk value of the corresponding mounted device in the candidate fault device set. Determine the normalized weights and penalty factors for the mounted devices; The second health score is obtained by multiplying the first health score of the mounted device by the normalized weight and then subtracting the penalty factor multiplied by the maximum fault propagation risk value.
2. The method according to claim 1, characterized in that, The method further includes: Based on the directed correlation topology graph of the equipment faults and the second health score, a prediction model is used to predict the health change trend of each mounted equipment in the future time period; the prediction model includes at least one of the following: a temporal graph convolutional network model and a spatiotemporal prediction model; the spatiotemporal prediction model is a lightweight temporal prediction model that combines graph convolution with dilated convolution.
3. The method according to claim 1, characterized in that, The first gradually changing steady-state data includes: temperature, luminous flux, packet loss rate, and average operating current; the first sudden transient data includes: power supply bus voltage and radio frequency current; for each mounted device of the smart light pole, the first gradually changing steady-state data of the mounted device is collected at a low frequency, and simultaneously, the first sudden transient data of the mounted device is collected at a high frequency, including: Based on the low-frequency sampling period, the temperature, the luminous flux, the packet loss rate, and the average operating current are collected sequentially using a temperature sensor, a photoelectric sensor, a mounted device communication module, and a current sensor to obtain the first slowly varying steady-state data; the low frequency includes: 0.1 Hz to 1 Hz; Based on the high-frequency sampling period, the voltage acquisition module and the radio frequency sampling circuit of the mounted equipment are used to sequentially acquire the power supply bus voltage and the radio frequency current to obtain the first burst transient data, and the first burst transient data is stored in a loop buffer; the high frequency is greater than or equal to 10 Hz.
4. The method according to claim 1, characterized in that, The candidate fault device set includes: the triggering mounted device, adjacent mounted devices, and mounted devices sharing a power supply bus with the triggering mounted device; the step of determining the triggering mounted device and the candidate fault device set based on the first derivative and / or effective value of the first slowly varying steady-state data, and correspondingly obtaining the second slowly varying steady-state data and the second sudden transient data of all mounted devices in the candidate fault device set from the first slowly varying steady-state data and the first sudden transient data, includes: Differentiate and / or perform absolute value processing on all the first slowly changing steady-state data to obtain the first derivative and / or effective value of the first slowly changing steady-state data. When the first derivative and / or effective value of at least one of the first slowly changing steady-state data at any time is greater than or equal to the corresponding preset second threshold and is maintained for a preset duration, the time is determined as the trigger time and the mounted device is determined as the trigger mounted device. The triggering device, the adjacent devices of the triggering device, and the devices that share the power supply bus with the triggering device are bound together to determine the candidate fault device set. For each mounted device in the candidate fault device set, the second sudden transient data within a preset window length before and after the trigger time is frozen from the first sudden transient data. At the same time, the second slowly changing steady-state data within a preset window length before and after the trigger time is extracted from the first slowly changing steady-state data. The preset window length is less than or equal to the length of the circular cache storage.
5. The method according to claim 1, characterized in that, When the number of target faulty devices is greater than or equal to two, the target faulty devices are updated and determined based on the weight of the fault propagation directed edge and the triggering time of the triggering device, including: When the number of the target faulty devices is greater than or equal to two, all the target faulty devices are identified as an abnormal target fault set, and the minimum directed association subgraph corresponding to the abnormal target fault set is selected from the directed association topology graph of the device faults. For any abnormal target fault device and other abnormal target fault devices within the minimum directed correlation subgraph, an exponent is calculated based on the fault propagation directed edge weights between the abnormal target fault device and the other abnormal target fault devices, and the time difference between the triggering times between the abnormal target fault device and the other abnormal target fault devices. The source tracing score corresponding to the abnormal target fault device is then determined by summarizing the results. Compare all the source tracing scores in the abnormal target fault set, and update and determine the abnormal target fault device corresponding to the maximum value in the source tracing score as the target fault device.
6. The method according to claim 2, characterized in that, The step of predicting the health trend of each mounted device over a future time period using a prediction model based on the directed association topology graph of the device faults and the second health score includes: The device fault directed association topology graph and the second health score are acquired in real time as training data, and at least one of the temporal graph convolutional network model and spatiotemporal prediction model is trained iteratively until at least one of the preset number of iterations and preset convergence threshold is met, and the training is determined to be completed. The directed association topology graph of the device fault of the smart light pole to be predicted and the second health score are input into at least one of the trained temporal graph convolutional network model and spatiotemporal prediction model to obtain the health change trend; the health change trend includes at least one of the following: the health score change curve in the future time period, the fault time and the faulty mounted device.
7. A smart light pole fault determination device, characterized in that, The device includes: The data acquisition unit is used to acquire the first slowly changing steady-state data of each mounted device of the smart light pole at a low frequency, and at the same time, acquire the first burst transient data of the mounted device at a high frequency. The first determining unit is used to determine the triggering mounted device and the candidate fault device set based on the first derivative and / or effective value of the first slowly changing steady-state data, and correspondingly obtain the second slowly changing steady-state data and the second sudden transient data of all mounted devices in the candidate fault device set from the first slowly changing steady-state data and the first sudden transient data, respectively. The construction unit is used to determine the weight of the directed edge for fault propagation between devices based on the second burst transient data and the second slowly changing steady-state data, using the transfer entropy formula and physical constraint function, and to construct a directed association topology graph of device faults for each mounted device in the candidate fault device set. The second determining unit is used to obtain a first health score based on the second sudden transient data and the second slowly changing steady-state data, and to traverse the directed association topology graph of the equipment faults to determine the maximum fault propagation risk value in the candidate fault equipment set, and to determine a second health score based on the first health score and the maximum fault propagation risk value. The third determining unit is used to determine the device corresponding to the second health score as the target fault device when the second health score of the mounted device in the candidate fault device set is less than or equal to a preset first threshold, and when the number of the target fault devices is greater than or equal to two, update the determination of the target fault device based on the weight of the fault propagation directed edge and the triggering time of the triggering mounted device. The transfer entropy formula is used to calculate the fault causal correlation strength between devices in the candidate faulty device set; the physical constraint function includes a decay envelope function; for each mounted device in the candidate faulty device set, based on the second sudden transient data and the second slowly varying steady-state data, the transfer entropy formula and physical constraint function are used to determine the weights of the directed fault propagation edges between devices in the candidate faulty device set, and based on the weights of the directed fault propagation edges, a directed correlation topology graph of device faults is constructed, including: For each mounted device in the candidate fault device set, the second slowly changing steady-state data and the second sudden transient data are extracted and time series aligned to obtain the slowly changing feature time series and the sudden feature time series. The slowly changing feature time series and the sudden feature time series are then spliced together to obtain the multi-dimensional feature time series corresponding to the mounted device. For any two mounted devices in the candidate fault device set, the first multidimensional feature time series corresponding to the first mounted device and the second multidimensional feature time series corresponding to the second mounted device are input into the transfer entropy formula to calculate the fault causal association strength between the first mounted device and the second mounted device. Based on the installation spacing constraints, rated power constraints, and power supply bus subordination constraints of the first mounted equipment and the second mounted equipment, the attenuation envelope function is constructed. Multiply the fault causal correlation strength and the output value of the decay envelope function to obtain the directed edge weight between the first mounted device and the second mounted device, and combine the directed edge weights between any two mounted devices in the candidate fault device set to obtain the fault propagation directed edge weight between each device in the candidate fault device set. From the directed edge weights of the fault propagation between the devices, all directed edge weights greater than or equal to a preset third threshold are selected and combined to obtain the directed association topology graph of the device faults. The process involves obtaining a first health score based on the second sudden transient data and the second slowly changing steady-state data, traversing the directed association topology of the equipment faults to determine the maximum fault propagation risk value in the candidate faulty equipment set, and determining a second health score based on the first health score and the maximum fault propagation risk value, including: For each mounted device in the candidate fault device set, the second sudden transient data and the second slowly changing steady-state data are mapped using at least one normalization formula of the Sigmund function and the piecewise linear function to obtain the first health score corresponding to the mounted device. Traverse all directed edges in the directed association topology of the device faults, and calculate the risk value of single edge propagation for each device based on the weight of each directed edge and the first health score of the corresponding mounted device, and take the maximum value to obtain the maximum fault propagation risk value of the corresponding mounted device in the candidate fault device set. Determine the normalized weights and penalty factors for the mounted devices; The second health score is obtained by multiplying the first health score of the mounted device by the normalized weight and then subtracting the penalty factor multiplied by the maximum fault propagation risk value.
8. A smart light pole fault determination system, characterized in that, The system includes: The mounting device is used to collect the first slowly varying steady-state data of the mounting device at a low frequency, and at the same time, to collect the first burst transient data of the mounting device at a high frequency. The edge gateway is used to determine the triggering mounted device and the candidate fault device set based on the first derivative and / or effective value of the first slowly varying steady-state data, and correspondingly obtain the second slowly varying steady-state data and the second sudden transient data of all mounted devices in the candidate fault device set from the first slowly varying steady-state data and the first sudden transient data, respectively. The regional management platform is used to determine the weights of directed edges for fault propagation between devices for each mounted device in the candidate fault device set, based on the second sudden transient data and the second slowly varying steady-state data, using the transfer entropy formula and physical constraint functions, and constructing a directed association topology graph of device faults; based on the second sudden transient data and the second slowly varying steady-state data, a first health score is obtained, and the directed association topology graph of device faults is traversed to determine the maximum fault propagation risk value in the candidate fault device set, and a second health score is determined based on the first health score and the maximum fault propagation risk value; when there is a mounted device in the candidate fault device set whose second health score is less than or equal to a preset first threshold, the mounted device corresponding to the second health score is determined as a target fault device, and when the number of target fault devices is greater than or equal to two, the target fault devices are updated and determined based on the weights of the directed edges for fault propagation and the triggering time of the triggering mounted device; The transfer entropy formula is used to calculate the fault causal correlation strength between devices in the candidate faulty device set; the physical constraint function includes a decay envelope function; for each mounted device in the candidate faulty device set, based on the second sudden transient data and the second slowly varying steady-state data, the transfer entropy formula and physical constraint function are used to determine the weights of the directed fault propagation edges between devices in the candidate faulty device set, and based on the weights of the directed fault propagation edges, a directed correlation topology graph of device faults is constructed, including: For each mounted device in the candidate fault device set, the second slowly changing steady-state data and the second sudden transient data are extracted and time series aligned to obtain the slowly changing feature time series and the sudden feature time series. The slowly changing feature time series and the sudden feature time series are then spliced together to obtain the multi-dimensional feature time series corresponding to the mounted device. For any two mounted devices in the candidate fault device set, the first multidimensional feature time series corresponding to the first mounted device and the second multidimensional feature time series corresponding to the second mounted device are input into the transfer entropy formula to calculate the fault causal association strength between the first mounted device and the second mounted device. Based on the installation spacing constraints, rated power constraints, and power supply bus subordination constraints of the first mounted equipment and the second mounted equipment, the attenuation envelope function is constructed. Multiply the fault causal correlation strength and the output value of the decay envelope function to obtain the directed edge weight between the first mounted device and the second mounted device, and combine the directed edge weights between any two mounted devices in the candidate fault device set to obtain the fault propagation directed edge weight between each device in the candidate fault device set. From the directed edge weights of the fault propagation between the devices, all directed edge weights greater than or equal to a preset third threshold are selected and combined to obtain the directed association topology graph of the device faults. The process involves obtaining a first health score based on the second sudden transient data and the second slowly changing steady-state data, traversing the directed association topology of the equipment faults to determine the maximum fault propagation risk value in the candidate faulty equipment set, and determining a second health score based on the first health score and the maximum fault propagation risk value, including: For each mounted device in the candidate fault device set, the second sudden transient data and the second slowly changing steady-state data are mapped using at least one normalization formula of the Sigmund function and the piecewise linear function to obtain the first health score corresponding to the mounted device. Traverse all directed edges in the directed association topology of the device faults, and calculate the risk value of single edge propagation for each device based on the weight of each directed edge and the first health score of the corresponding mounted device, and take the maximum value to obtain the maximum fault propagation risk value of the corresponding mounted device in the candidate fault device set. Determine the normalized weights and penalty factors for the mounted devices; The second health score is obtained by multiplying the first health score of the mounted device by the normalized weight and then subtracting the penalty factor multiplied by the maximum fault propagation risk value.
Citation Information
Patent Citations
Systems and methods for improving identification of issues associated with detecting anomalous conditions
CA3143053A1
Fault detection method of smart lamp pole, terminal equipment and storage medium
CN113300918A
