Fault state intelligent identification method for urban street lamp network

Abstract

The invention relates to the technical field of artificial intelligence, discloses an intelligent fault state identification method and system for an urban street lamp network, and aims to solve the problems of high false alarm rate, resource waste and response delay caused by optical sensing performance attenuation and multi-source data fusion deficiency under complex meteorological conditions such as rain and fog in the prior art. The method comprises the following steps: monitoring environment humidity and visibility in real time; dynamically switching a standard sensing mode (based on visible light illuminance analysis) or a severe environment sensing mode (based on millimeter wave radar phase domain micro-vibration spectrum analysis); and the judgment threshold is dynamically adjusted in combination with a nonlinear calibration model of edge deployment. According to the technical scheme, all-weather high-precision fault recognition is achieved, the false alarm rate is remarkably reduced, and the robustness and energy efficiency of the system are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of artificial intelligence, and particularly relates to a fault state intelligent identification method for urban street lamp networks. BACKGROUND

[0002] With the in-depth promotion of smart city infrastructure construction, the intelligent operation and maintenance demand of urban street lamp networks as a key public lighting carrier is increasingly urgent. The current mainstream fault identification technology mainly relies on visible light cameras for brightness monitoring or judges the working state through single-point current signals. Such methods can still maintain basic functions under ideal weather conditions, but face serious challenges in complex environments such as rain and fog. Due to the strong scattering of light by suspended water droplets in the air and the high reflection interference caused by water vapor condensation on the lens surface, the optical system is prone to misjudging normally lit street lamps as extinguished. According to the 2023 "Urban Street Lamp Operation White Paper" of the China Municipal Engineering Association, the average fault false alarm rate of 32 key cities in China under rain and fog weather is as high as 37.8%, of which the proportion of false alarms caused by water vapor reflection is more than 82%, not only causing a large number of invalid inspection work orders, but also significantly increasing the operation and maintenance cost.

[0003] Among them, the existing technical solutions have obvious shortcomings in multi-source perception and dynamic response mechanism. Some improved systems try to introduce millimeter wave radars to make up for the shortcomings of optical identification, but generally use a "camera + radar" parallel processing architecture, lacking a modal switching logic triggered based on environmental parameters. When rain and fog occur, the system still forces the dual-channel to run synchronously, causing the edge computing chip to be severely overloaded, and the fault identification delay far exceeds the 500 millisecond threshold required by the industry standard. At the same time, the existing radar solution only uses point cloud data in the distance-velocity dimension, ignoring the sensitive characteristics of phase information to the micro-vibration of the lamp body (amplitude less than 0.5 mm), resulting in a fault detection rate of less than 65% for the radar channel under rain and fog noise interference. More importantly, although the concept of environment-adaptive threshold has been proposed in academia, there is still no engineering implementation path that can be quickly closed-loop executed on edge devices. The traditional linear calibration model cannot cope with the nonlinear coupling of humidity and visibility, and the cloud-dependent threshold adjustment mechanism violates the industry standard of localized real-time decision-making.

[0004] The existing technology is therefore trapped in the dilemma of "sensor stacking but coordination failure", failing to build a millisecond-level linkage mechanism between weather anomalies and perception modal switching. This makes the system unable to quickly switch to the radar phase analysis mode with high robustness when rain and fog appear, nor can it reset the decision threshold at a low delay on the physical layer, ultimately leading to high false alarm rates and serious resource waste. An intelligent identification method that deeply integrates environmental perception, hardware interrupt triggering, and phase dynamic calibration is urgently needed to fundamentally solve the reliability problem of street lamp fault identification under complex weather conditions. SUMMARY

[0005] The technical problem to be solved by the present application is to provide a fault state intelligent identification method for urban street lamp networks, aiming to overcome the problems of high false alarm rate of fault identification, serious waste of system resources and delayed decision response caused by performance degradation of optical sensors and lack of multi-source data fusion mechanism under complex weather conditions such as rain and fog.

[0006] To solve the above technical problems, the present application provides a fault state intelligent identification method for urban street lamp networks, which builds an asymmetric sensing and decision mechanism driven by environmental state, relies on high-resolution optical imaging for routine illumination analysis under normal weather conditions, and switches to millimeter wave radar phase domain micro-vibration spectrum analysis mode in real time through hardware interruption under adverse weather conditions, and combines with the edge-deployed nonlinear environmental parameter calibration model to dynamically compensate and adjust the decision threshold of the sensing data, thereby realizing all-weather and high-precision street lamp fault state identification.

[0007] The present application provides a fault state intelligent identification method for urban street lamp networks, which includes: Real-time monitoring of environmental and meteorological parameters of the area where the street lamp is located, the environmental and meteorological parameters including environmental humidity and atmospheric visibility; Based on the environmental and meteorological parameters, determine the current sensing mode, the sensing mode including standard sensing mode and adverse environment sensing mode; When it is determined that the current sensing mode is the standard sensing mode, execute the standard mode fault identification process, which includes: starting the visible light image sensor to collect ground image data of the street lamp lighting area; processing the ground image data to extract the average illumination value of the lighting area; comparing the average illumination value with the standard illumination threshold to determine whether the street lamp has failed; When it is determined that the current sensing mode is the adverse environment sensing mode, execute the adverse environment mode fault identification process, which includes: starting the millimeter wave radar sensor and working in high-precision detection mode, and switching the visible light image sensor to low-power or auxiliary monitoring state; the millimeter wave radar sensor transmits a frequency-modulated continuous wave signal and receives the echo signal from the street lamp body itself to generate raw complex intermediate frequency signal data; processing the raw complex intermediate frequency signal data to extract the phase domain vibration spectrum representing the micro-vibration characteristics of the street lamp body caused by abnormal operation of electrical components; comparing the characteristic frequency amplitude in the phase domain vibration spectrum with the adverse environment vibration threshold to determine whether the street lamp has failed; and dynamically calibrating the standard illumination threshold or the adverse environment vibration threshold based on the environmental and meteorological parameters through a pre-trained nonlinear model.

[0008] As an embodiment of the present application, the real-time monitoring of the environmental meteorological parameters of the area where the street lamp is located specifically includes: The relative humidity data of the air around the street lamp device is obtained by a capacitive humidity sensor, the measurement range of the capacitive humidity sensor is 0-100%, and the measurement accuracy is plus or minus 2%; The atmospheric visibility data around the street lamp device is obtained by a forward scattering visibility sensor, the forward scattering visibility sensor uses an infrared light source with a center wavelength of 850 nm, and the measurement range is 10 m-10000 m.

[0009] As an embodiment of the present application, based on the environmental meteorological parameters, the determination of the current belonging to the sensing mode specifically includes: The humidity threshold and the visibility threshold are preset in the edge computing unit, the humidity threshold is 85%, and the visibility threshold is 500 m; The real-time monitored environmental humidity is compared with the humidity threshold, and the real-time monitored atmospheric visibility is compared with the visibility threshold; When the environmental humidity is less than or equal to the humidity threshold, and the atmospheric visibility is greater than or equal to the visibility threshold, it is determined that the current belonging to the sensing mode is the standard sensing mode; When the environmental humidity is greater than the humidity threshold, or the atmospheric visibility is less than the visibility threshold, it is determined that the current belonging to the sensing mode is the harsh environment sensing mode, and a dedicated microcontroller sends a hardware interrupt signal to the main processor through a general input / output interface to trigger the switching of the sensing mode.

[0010] As an embodiment of the present application, in the standard mode fault identification process, the average illumination value of the lighting area is extracted, specifically including: In the ground image data, a fixed region of interest is calibrated according to the installation height and the irradiation angle of the street lamp; The image data in the region of interest is converted from the red-green-blue color space to the luminance, chrominance component color space; The pixel value of the luminance component channel is extracted, and the arithmetic mean value of all pixel luminance values in the region of interest is calculated as the quantitative representation of the average illumination value.

[0011] As an embodiment of the present application, in the harsh environment mode fault identification process, the millimeter wave radar sensor works in a high-precision detection mode, which is specifically configured as: Working in the frequency band of 77GHz-81GHz; Using a four-transmitting and four-receiving antenna array configuration to realize high-angle resolution detection of the street lamp body target; The sweep bandwidth of the transmitted frequency-modulated continuous wave signal is set to 4GHz, and the sweep period is 40μs.

[0012] As an embodiment of the present application, the phase domain vibration spectrum representing the micro-vibration characteristics of the street lamp body caused by abnormal operation of the electrical components specifically includes: A two-dimensional fast Fourier transform is performed on the original complex intermediate frequency signal data collected by the millimeter wave radar sensor to generate a range-Doppler matrix, and a target unit corresponding to the static position of the street lamp body is identified from the range-Doppler matrix; The complex signal sequence of the target unit in a complete coherent processing period is extracted; For each sampling point of the complex signal sequence, the phase value is calculated by the inverse tangent function to obtain an original phase time sequence; A high-pass digital filter is applied to the original phase time sequence to filter out low-frequency phase drift caused by environmental slow-changing factors to obtain a pure micro-vibration phase sequence; A short-time Fourier transform is performed on the pure micro-vibration phase sequence to generate a time-frequency distribution diagram, i.e., a phase domain vibration spectrum.

[0013] As an embodiment of the present application, the comparison of the characteristic frequency amplitude in the phase domain vibration spectrum with the harsh environment vibration threshold to determine whether the street lamp has failed specifically includes: In the phase domain vibration spectrum, a characteristic frequency monitoring frequency band introduced by alternating current power supply is preset, and the characteristic frequency monitoring frequency band is an interval of (100±2) Hz; The spectrum peak value in the characteristic frequency monitoring frequency band is extracted as the characteristic frequency amplitude; If the characteristic frequency amplitude continuously exceeds the harsh environment vibration threshold for a preset time length, it is determined that the street lamp has failed, and the preset time length is 300 ms.

[0014] As an embodiment of the present application, the dynamic threshold calibration step specifically includes: A pre-trained neural network model is deployed in the edge computing unit, and the input of the neural network model is the real-time monitored environmental humidity and atmospheric visibility; The output of the neural network model is an optical illumination calibration coefficient and a radar phase noise floor adjustment value; When performing the standard mode fault identification process, the standard illumination threshold is multiplied by the optical illumination calibration coefficient to generate a dynamic standard illumination threshold, and the subsequent comparison operations use this dynamic standard illumination threshold; When performing the harsh environment mode fault identification process, the radar phase noise floor adjustment value is added to the harsh environment vibration threshold to generate a dynamic harsh environment vibration threshold, and the subsequent comparison operations use this dynamic harsh environment vibration threshold.

[0015] As an embodiment of the present application, the pre-trained neural network model is a multi-layer perceptron, and the network structure thereof includes: An input layer containing two neurons, respectively receiving the normalized ambient humidity and atmospheric visibility values; Two hidden layers, each containing sixteen neurons, and using a linear rectifier function as the activation function; An output layer containing two neurons, respectively outputting the optical illumination calibration coefficient and the radar phase noise floor adjustment value, and using a sigmoid function as the activation function to ensure that the output value is within the interval of zero to one.

[0016] In summary, the present application includes at least one of the following beneficial technical effects: 1. The present application introduces environmental meteorological parameters as the core driving force for switching sensing modes, and constructs an asymmetric sensing decision architecture. Under adverse weather conditions such as rain and fog, the visible light sensing path, which is easily disturbed, is actively abandoned, and the millimeter wave radar phase domain analysis, which is not sensitive to water vapor, is switched to, thereby fundamentally solving the problem of failure and false reporting caused by poor environmental adaptability of optical recognition systems.

[0017] 2. The present application creatively proposes and implements a micro-vibration spectrum analysis technology based on millimeter wave radar phase information, which is used to directly detect the physical vibration characteristics of the lamp body caused by electrical faults in adverse environments. Compared with the traditional radar scheme relying on distance and speed information, this method has higher signal-to-noise ratio and detection sensitivity, and can accurately capture sub-millimeter level vibrations, significantly improving the fault detection rate in a strong scattering noise background.

[0018] 3. The present application deploys a lightweight nonlinear calibration model on the edge computing unit, realizes millisecond-level dynamic self-adaptive adjustment of the optical illumination threshold and radar vibration threshold, solves the problem of response delay and inaccurate decision caused by fixed threshold or dependence on cloud computing in the prior art, ensures the real-time and accuracy of fault recognition, and fully meets the industry specification requirements of localized decision-making.

[0019] 4. The sensing mode switching mechanism of the present application converts the computationally intensive image analysis task into a relatively efficient radar signal processing task under adverse weather conditions, and puts the image sensor into a low-power state, effectively avoiding the overload of edge computing resources, ensuring the stable operation and rapid response of the system under any weather conditions, and reducing the power consumption and hardware cost of the overall system. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a schematic diagram of the overall technical scheme architecture of a fault state intelligent identification method for urban street lamp networks provided by the present application. DETAILED DESCRIPTION

[0021] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined object of the present application, the specific embodiments of the present application are described in detail below in combination with the drawings and preferred embodiments.

[0022] Referring to Figure 1 The present application provides a city street lamp network-oriented intelligent fault state identification method, the core of which is to construct an asymmetric sensing and decision mechanism driven by environmental meteorological parameters. Through enabling a high-resolution visible light image sensor to perform illumination analysis under standard meteorological conditions, and triggering a millimeter wave radar sensor to enter a high-precision phase domain micro-vibration spectrum analysis mode through hardware interruption under severe meteorological conditions, and combining an edge-deployed nonlinear dynamic threshold calibration model, all-weather, low false alarm rate, and high response speed street lamp fault state identification is achieved.

[0023] A city street lamp network-oriented intelligent fault state identification method, comprising the following steps: Firstly, the S1 step of monitoring the environmental meteorological parameters of the area where the street lamp is located in real time is performed, and the environmental meteorological parameters include environmental humidity and atmospheric visibility. This step is completed by a special sensing unit integrated in the street lamp pole body or installed on the adjacent mounting bracket.

[0024] Among them, the environmental humidity data is obtained by a capacitive humidity sensor, which uses a high molecular film as a dielectric material, and its capacitance value changes linearly with the amount of water molecules adsorbed in the air, with a measurement range covering 0-100% relative humidity and a typical measurement accuracy of plus or minus 2%.

[0025] The atmospheric visibility data is obtained by a forward scattering type visibility sensor, which has an infrared light emitting diode with a center wavelength of 850nm as a light source, and a silicon photodiode receiver located at a scattering angle of 33°. By measuring the scattering light intensity, the atmospheric extinction coefficient is inverted, and then the visibility value in the range of 10m-10000m is calculated.

[0026] Both types of sensors output raw analog signals at a frequency of once per second. After being converted to sixteen-bit digital quantities by an analog-to-digital converter, they are transmitted to the input buffer area of the edge computing unit through the serial peripheral interface bus.

[0027] In order to ensure data reliability, the system performs sliding window median filtering on data of three consecutive sampling periods, eliminates abnormal jump points caused by instantaneous electromagnetic interference, and applies a time stamp to the filtered data to form a structured environmental meteorological parameter time sequence stream.

[0028] Subsequently, step S2 is performed: based on the environmental meteorological parameters, the current belonging to the sensing mode is determined, and the sensing mode includes a standard sensing mode and a harsh environment sensing mode. The determination logic is built into the firmware program of the edge computing unit, and the core is a set of preset threshold comparison rules.

[0029] Specifically, the system fixes the humidity threshold value as 85% and the visibility threshold value as 500m in the initialization phase, i.e. in the non-volatile memory. At the beginning of each sensing period, the edge computing unit reads the latest set of environmental humidity and atmospheric visibility values from the input buffer, and performs numerical comparison with the above-mentioned threshold values respectively.

[0030] When the environmental humidity is less than or equal to 85% and the atmospheric visibility is greater than or equal to 500m, the system determines that the current is in standard meteorological conditions, and activates the standard sensing mode; On the contrary, when the environmental humidity is greater than 85% or the atmospheric visibility is less than 500m, the system determines that the current is in harsh meteorological conditions, and activates the harsh environment sensing mode.

[0031] It is worth noting that the determination result is not only used for software logic branching, but also directly triggers the hardware level interrupt mechanism. Once the determination enters the harsh environment sensing mode, the edge computing unit immediately sends a high-level pulse signal to the dedicated microcontroller, which monitors this signal through the general input-output interface, and generates a hardware interrupt request within twenty microseconds after detecting the rising edge, forcing the main processor to pause the current task and enter the sensing mode switching subroutine.

[0032] This hardware interrupt mechanism ensures the determinacy and timeliness of mode switching, avoiding the response lag caused by operating system scheduling delay.

[0033] When the current belonging to the sensing mode is determined to be the standard sensing mode, the system performs step S3: the visible light image sensor is started to collect ground image data of the streetlight lighting area.

[0034] The visible light image sensor is a global shutter type complementary metal oxide semiconductor image sensor, with a resolution of 1920x1080 pixels, a frame rate of 30 frames / second, and a dynamic range of not less than 70dB. The sensor lens focal length is fixed at 4mm, and the aperture value is set to F2.0 to ensure that clear images can still be obtained in low-illumination nighttime environment.

[0035] The system pre-labels a rectangular region of interest in the image coordinate system according to the installation height of the streetlight (usually eight to twelve meters) and the light distribution curve of the lamp, which corresponds to a circular lighting surface with a diameter of about six meters directly below the lamp pole. Each time the sensor outputs a complete red-green-blue three-channel image data, which is transmitted to the image processing buffer of the edge computing unit via the mobile industry processor interface bus.

[0036] Then, step S4 is performed: the ground image data is processed to extract the average illuminance value of the illuminated area.

[0037] The processing flow first loads the complete frame data from the image processing buffer and then crops the target sub-image based on the pre-stored region of interest coordinates. Subsequently, the red-green-blue color space data of this target sub-image is converted to a luminance / chrominance component color space. The conversion formula adopts the BT.601 standard recommended by the International Telecommunication Union (ITU), where the expression for calculating the luminance component Y is:

[0038] After the conversion is complete, the system iterates through all pixels within the region of interest, extracts the value of its luminance component Y, and calculates the arithmetic mean of all Y values, denoted as . .Should The value is a quantitative representation of the average illuminance value, with a numerical range of 0 to 255, and it shows a monotonically increasing relationship with the actual physical illuminance.

[0039] To eliminate the long-term drift effects caused by lens contamination or slight misalignment, the system performs a self-calibration procedure once daily at 3:00 AM, recording the time periods when there are no vehicles or passengers. The baseline value is used to dynamically adjust the judgment threshold for subsequent days.

[0040] Then, step S5 is executed: the average illuminance value is compared with the standard illuminance threshold to determine whether the street light is malfunctioning.

[0041] The standard illuminance threshold was initially set at 70, a value derived from statistical analysis of extensive measured data on clear nights, corresponding to normal working conditions. The lower bound. The comparison operation is performed in the integer arithmetic unit of the edge computing unit, if If the value is less than this threshold, it is preliminarily determined that the street light is either out or has a serious light decay fault.

[0042] To prevent misjudgments caused by momentary occlusion (such as the passing of a large vehicle), the system introduces a continuous verification mechanism: only when five consecutive frames of images are occluded... A fault is confirmed and a fault event record is generated only when all values ​​are below the threshold and the time span exceeds 160ms.

[0043] Fault event logs include timestamps, Metadata such as sequence and environmental meteorological parameter snapshots are temporarily stored in local flash memory, awaiting uploading to the city lighting management platform.

[0044] When it is determined that the current belonging to the sensing mode is the harsh environment sensing mode, the system performs S6 step: start the millimeter wave radar sensor and make it work in high precision detection mode, and switch the visible light image sensor to low power consumption or auxiliary monitoring state at the same time.

[0045] The millimeter wave radar sensor works in the frequency band of 77GHz-81GHz, adopts a four-transmitting and four-receiving planar antenna array configuration, and the beam width in the horizontal and vertical directions is 5° and 10° respectively, so as to realize high angle resolution detection of the street lamp body target. The frequency-modulated continuous wave signal emitted by the radar adopts a triangular wave modulation mode, the sweep bandwidth is set to 4GHz, and the single up-sweep frequency and down-sweep frequency period is 40μs respectively, forming a complete modulation period.

[0046] After receiving the hardware interrupt signal, the radar front-end radio frequency module immediately wakes up from the sleep state, and completes the local oscillator locking, power amplifier preheating and reception link gain calibration within 50ms, and enters the stable working state.

[0047] At the same time, the visible light image sensor is instructed to enter standby mode, only basic power supply is reserved to maintain the register state, and its image acquisition function is completely closed, so that the overall power consumption of the system is reduced by about 40%.

[0048] Then, S7 step is performed: the millimeter wave radar sensor transmits a frequency-modulated continuous wave signal and receives a return signal from the street lamp body itself, to generate raw complex intermediate frequency signal data. Each receiving channel in the radar antenna array independently mixes the return signal with the local oscillator signal to down-convert it to a baseband intermediate frequency signal.

[0049] The intermediate frequency signal is digitized by a 12-bit high-speed analog-to-digital converter at a sampling rate of 2MSps to form a complex sequence composed of real and imaginary parts. The system continuously collects 256 modulation period data within one coherent processing period, each period contains 1024 sampling points, and finally constructs a three-dimensional complex tensor with dimensions of 4 (receiving channels) x 256 (slow time) x 1024 (fast time). The three-dimensional complex tensor, as the raw complex intermediate frequency signal data, is temporarily stored in the radar dedicated double data rate synchronous dynamic random access memory, waiting for subsequent signal processing.

[0050] Then, S8 step is performed: the raw complex intermediate frequency signal data is processed to extract the phase domain vibration spectrum representing the micro-vibration characteristics of the street lamp body caused by the abnormal operation of the electrical components.

[0051] The processing flow first performs two-dimensional fast Fourier transform on the data of each receiving channel independently. Among them, the Fourier transform along the fast time dimension generates the distance dimension spectrum, and the Fourier transform along the slow time dimension generates the Doppler frequency spectrum, and the combination of the two forms a range-Doppler matrix.

[0052] The system calculates the theoretical coordinate of the lamp body in the range-doppler matrix according to the known installation position (latitude, longitude and height) of the street lamp, combined with the installation attitude angle of the radar, and draws a 3x3 range-doppler cell search window around the coordinate. By searching for the energy peak in the window, the target cell corresponding to the static position of the lamp body is accurately located.

[0053] Subsequently, the system extracts the complex signal sequence of the target cell in 256 coherent processing periods, denoted as , n = 0, 1, …, 255. For each sampling point of the sequence, the system calculates its phase value by the inverse tangent function, that is , thereby obtaining the original phase time sequence.

[0054] Since slow-varying factors such as environmental temperature changes and wind loads will cause low-frequency drift of the phase, the system applies a high-pass digital filter with a cutoff frequency of 5 Hz to the original phase sequence to filter out the slow trend items below 0.5 Hz and retain the high-frequency phase fluctuations reflecting the micro-vibration of the lamp body, thereby obtaining the pure micro-vibration phase sequence.

[0055] Finally, the short-time Fourier transform is performed on the pure sequence, with the window length set to 64 points and the overlap rate set to 75%, to generate the time-frequency distribution diagram, i.e. the phase-domain vibration spectrum. The frequency axis of the spectrum covers 0-50 Hz, the time axis corresponds to 256 coherent periods, and the amplitude is expressed in decibels.

[0056] Then, step S9 is performed: comparing the amplitude of the characteristic frequency in the phase-domain vibration spectrum with the threshold of severe environmental vibration to determine whether the street lamp has failed. The core of this step lies in identifying the characteristic vibration frequency introduced by the alternating power supply system.

[0057] In China, the power grid frequency is 50 Hz, but the switching devices of the street lamp ballast or driving power supply will produce 100 Hz harmonic vibration when working, which will be conducted to the lamp body shell through the mechanical structure, forming a stable micro-amplitude vibration source.

[0058] Therefore, the preset characteristic frequency monitoring frequency band in the phase-domain vibration spectrum is the interval of (100±2) Hz. For all frequency points in this frequency band, the maximum value of the amplitude is calculated as the amplitude of the characteristic frequency.

[0059] The threshold of harsh environment vibration is initially set to -25dB, which is based on the statistical phase noise floor of a large number of normal street lamps in rain and fog environment. The comparison operation is completed in the floating point operation unit of the edge computing unit. If the amplitude of the characteristic frequency continuously exceeds the threshold for 300ms (corresponding to at least nine consecutive short-time Fourier transform frames), it is determined that the internal electrical components of the street lamp have failed, such as capacitor aging, poor contact or drive chip failure. At this time, the system generates a fault event record, which contains the phase domain vibration spectrum segment, the characteristic frequency amplitude sequence, the environmental meteorological parameters, etc., and is also temporarily stored in the local flash memory.

[0060] Further, the application also includes a dynamic threshold calibration step based on a nonlinear model, which runs through the judgment process of S5 and S9.

[0061] Specifically, a pre-trained multi-layer perception neural network model is deployed in the edge computing unit.

[0062] The input layer of the multi-layer perception neural network model includes two neurons, which respectively receive the normalized environmental humidity value (divided by one hundred) and the atmospheric visibility value (divided by ten thousand). The two hidden layers each include sixteen neurons, and the linear rectifier function is used as the activation function.

[0063] The output layer of the multi-layer perception neural network model includes two neurons, which respectively output the optical illumination calibration coefficient and the radar phase noise floor adjustment value , and the S-shaped function is used as the activation function to ensure , .

[0064] The multi-layer perception neural network model is trained in the offline stage using a dataset containing one hundred thousand samples, covering all working conditions with humidity from 40% to 98% and visibility from 20m to 8000m. The label is composed of the real fault state and the corresponding optimal threshold offset annotated by experts.

[0065] At runtime, the system calls the model for a forward inference every time the environmental meteorological parameters are updated, which takes no more than 8ms. In the standard sensing mode, the dynamic standard illumination threshold = standard illumination threshold x ; in the harsh environment sensing mode, the dynamic harsh environment vibration threshold = harsh environment vibration threshold + .

[0066] All subsequent comparison operations use this dynamic threshold, thereby achieving millisecond-level adaptive compensation for environmental interference.

[0067] In summary, the application constructs a closed-loop, environment-adaptive street light fault identification process through the above S1 to S9 steps and the dynamic threshold calibration mechanism. The process realizes millisecond-level switching of the sensing modalities under the driving of hardware interrupts, completes all data processing and decision-making on the edge side, does not need to rely on cloud computing, and fully meets the requirements of GB / T 38661-2020 standard for localized decision-making.

[0068] Corresponding to the above method, the application also provides a fault state intelligent identification system for a city street light network. The system is integrated in a single edge computing device and deployed in the control box of each street light, and specifically comprises: An environment state monitoring module, which is composed of a capacitive humidity sensor and a forward scattering visibility sensor, is responsible for real-time collection of environmental meteorological parameters.

[0069] A sensing mode decision and switching module, which is realized by a special microcontroller and a general input-output interface circuit, is responsible for executing threshold comparison and generating a hardware interrupt signal.

[0070] A standard mode data processing module, which contains a visible light image sensor and its image signal processor, is responsible for executing image acquisition, region of interest cropping, color space conversion, and average illumination calculation.

[0071] A harsh environment mode data processing module, which contains a millimeter wave radar sensor and its baseband signal processing unit, is responsible for executing intermediate frequency signal acquisition, two-dimensional fast Fourier transform, target unit positioning, phase demodulation, and short-time Fourier transform.

[0072] A dynamic threshold calibration module, which is an embedded neural network inference engine, is solidified in the read-only memory of the edge computing unit and is responsible for outputting calibration coefficients in real time according to environmental parameters.

[0073] A fault comprehensive decision module, which is a decision-making process running on a real-time operating system, is responsible for receiving preliminary judgment results from the standard mode or harsh environment mode data processing module, combining the currently activated sensing mode identifier, generating standardized fault alarm information, and uploading to the city lighting management platform through the narrowband Internet of Things communication module.

[0074] Through the deep cooperation of hardware and software, the system realizes efficient resource utilization, rapid response, and reliable precision all-weather street light fault monitoring capability.

[0075] It is obvious to those skilled in the art that the application is not limited to the details of the above exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or essential characteristics of the application. Therefore, from any point of view, the embodiments should be regarded as exemplary and non-limiting.

[0076] Furthermore, it should be understood that although the specification is described in terms of embodiments, not every embodiment includes only one independent technical solution, and the specification is described in this way only for the sake of clarity, and those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that those skilled in the art can understand.

Claims

1. A method for intelligent identification of fault state of urban street lamp network, characterized in that, The application comprises the following steps: Real-time monitoring of environmental meteorological parameters of the area where the street lamp is located, the environmental meteorological parameters including environmental humidity and atmospheric visibility; Based on the environmental meteorological parameters, determining the current sensing mode, the sensing mode including a standard sensing mode and a harsh environment sensing mode; When determining the standard sensing mode, executing a standard mode fault identification process: starting the visible light image sensor to collect ground image data of the street lamp lighting area, processing the ground image data, extracting the average illuminance value of the lighting area, comparing the average illuminance value with the standard illuminance threshold value to determine whether the street lamp has failed; When determining the harsh environment sensing mode, executing a harsh environment mode fault identification process: starting the millimeter wave radar sensor and working in a high-precision detection mode, at the same time, switching the visible light image sensor to a low-power or auxiliary monitoring state, the millimeter wave radar sensor emits a frequency-modulated continuous wave signal and receives the echo signal from the street lamp body itself to generate raw complex intermediate frequency signal data; processing the raw complex intermediate frequency signal data to extract the phase domain vibration spectrum representing the micro-vibration characteristics of the street lamp body caused by the abnormal operation of the electrical components; comparing the characteristic frequency amplitude in the phase domain vibration spectrum with the harsh environment vibration threshold value to determine whether the street lamp has failed; and based on the environmental meteorological parameters, dynamically calibrating the standard illuminance threshold value or the harsh environment vibration threshold value through a pre-trained nonlinear model. 2.The urban street light network oriented fault state intelligent identification method according to claim 1, characterized in that, The real-time monitoring of environmental meteorological parameters of the area where the street lamp is located specifically comprises: Obtaining the relative humidity data of the air around the street lamp equipment through a capacitive humidity sensor, the measurement range of the capacitive humidity sensor being 0-100%, and the measurement accuracy being plus or minus 2%; Obtaining the atmospheric visibility data around the street lamp equipment through a forward scattering type visibility sensor, the forward scattering type visibility sensor using an infrared light source with a center wavelength of 850 nm, and the measurement range being 10 m-10000 m. 3.The method of claim 1, wherein, Based on the environmental meteorological parameters, determining the current sensing mode specifically comprises: Presetting a humidity threshold value and a visibility threshold value in the edge computing unit, the humidity threshold value being 85%, and the visibility threshold value being 500 m; Comparing the real-time monitored environmental humidity with the humidity threshold value, and at the same time, comparing the real-time monitored atmospheric visibility with the visibility threshold value; When the environmental humidity is less than or equal to the humidity threshold value, and the atmospheric visibility is greater than or equal to the visibility threshold value, determining that the current sensing mode is the standard sensing mode; When the environmental humidity is greater than the humidity threshold value, or the atmospheric visibility is less than the visibility threshold value, determining that the current sensing mode is the harsh environment sensing mode, and sending a hardware interrupt signal to the main processor through a general input-output interface by a special microcontroller to trigger the switching of the sensing mode. 4.The method of claim 1, wherein, In the standard mode fault identification process, the extraction of the average illuminance value of the lighting area specifically comprises: In the ground image data, a fixed region of interest is pre-labeled according to the installation height and the irradiation angle of the street lamp; Converting the image data in the region of interest from the red-green-blue color space to the luminance-chrominance component color space; The pixel values of the luminance component channel are extracted, and the arithmetic mean of the luminance values of all pixels in the region of interest is calculated as a quantitative representation of the average illumination value. 5.The urban street light network oriented fault state intelligent identification method according to claim 1, characterized in that, In the harsh environment mode fault identification process, the millimeter wave radar sensor works in a high-precision detection mode, and the specific configuration is: Working in the frequency band of 77GHz-81GHz; Using a four-transmitting and four-receiving antenna array configuration to achieve high-angle resolution detection of the street lamp body target; The sweep bandwidth of the transmitted frequency-modulated continuous wave signal is set to 4GHz, and the sweep period is 40μs. 6.The urban street light network oriented fault state intelligent identification method according to claim 1, characterized in that, The phase domain vibration spectrum representing the micro-vibration characteristics of the street lamp body caused by abnormal operation of electrical components includes: Performing two-dimensional fast Fourier transform on the original complex intermediate frequency signal data collected by the millimeter wave radar sensor generates a range-Doppler matrix, and the target unit corresponding to the static position of the street lamp body is identified from the matrix; Extracting the complex signal sequence of the target unit within a complete coherent processing period; For each sampling point of the complex signal sequence, the phase value is calculated by the inverse tangent function to obtain an original phase time sequence; Applying a high-pass digital filter to the original phase time sequence to filter out low-frequency phase drift caused by environmental slow-changing factors to obtain a pure micro-vibration phase sequence; Performing short-time Fourier transform on the pure micro-vibration phase sequence generates a time-frequency distribution map, i.e., a phase domain vibration spectrum. 7.The urban street light network oriented fault status intelligent identification method according to claim 1, characterized in that, The comparison of the characteristic frequency amplitude in the phase domain vibration spectrum with the harsh environment vibration threshold to determine whether the street lamp has failed specifically includes: In the phase domain vibration spectrum, a characteristic frequency monitoring frequency band introduced by alternating current power supply is preset, and the characteristic frequency monitoring frequency band is an interval of (100±2)Hz; Extracting the spectrum peak value in the characteristic frequency monitoring frequency band as the characteristic frequency amplitude; If the characteristic frequency amplitude continuously exceeds the harsh environment vibration threshold for a preset time length, it is determined that the street lamp has failed, and the preset time length is 300ms. 8.The urban street light network oriented fault state intelligent identification method according to claim 1, characterized in that, The dynamic threshold calibration step specifically includes: Deploying a pre-trained neural network model in the edge computing unit, and the input of the neural network model is the real-time monitored environmental humidity and atmospheric visibility; The output of the neural network model is an optical illumination calibration coefficient and a radar phase noise floor adjustment value; When performing the standard mode fault identification process, multiply the standard illumination threshold by the optical illumination calibration coefficient to generate a dynamic standard illumination threshold, and the subsequent comparison operations use this dynamic standard illumination threshold; When performing the harsh environment mode fault identification process, add the radar phase noise floor adjustment value to the harsh environment vibration threshold to generate a dynamic harsh environment vibration threshold, and the subsequent comparison operations use this dynamic harsh environment vibration threshold.

Images