A street lamp network health state evaluation and fault prediction method and system

Abstract

This invention relates to the field of smart city lighting facility operation and maintenance, and discloses a method and system for assessing the health status and predicting faults in street light networks. The method includes the following steps: First, energy domain and information domain data of street light nodes are collected. A dynamic topology map with isomorphic information and energy domains is constructed based on electrical waveform correlation and channel characteristic similarity. Node features are reconstructed using a graph attention network, and passive monitoring residuals are calculated. Based on the dual-threshold judgment results of the residuals, when a node falls into a suspected grayscale range, an active micro-perturbation excitation mechanism is automatically triggered, breaking the steady state by issuing a fine-tuning light command. Subsequently, the degree of anomaly in the circuit physical layer is quantified by comparing the integral deviation between the actual physical response and the ideal response based on the transfer function. Finally, the passive residuals and active deviations are dynamically fused to output a full lifecycle health index. This invention solves the problem of difficulty in identifying early, weak faults under complex noise conditions, improving the accuracy of predictive maintenance of street light networks.

Description

Technical Field

[0001] This invention relates to the field of smart city lighting facility operation and maintenance technology, specifically to a method and system for assessing the health status and predicting faults in street light networks. Background Technology

[0002] With the rapid development of smart city infrastructure, urban street light networks are no longer limited to a single lighting function, but are gradually evolving into IoT terminal nodes that integrate lighting control, information dissemination, and environmental monitoring. To ensure the stable operation of large-scale street light networks, traditional maintenance methods are gradually shifting from reliance on periodic manual inspections to automated management based on remote monitoring systems. Existing monitoring technologies mostly adopt a Supervisory Control and Data Acquisition (SCADA) architecture, which collects steady-state electrical parameters such as voltage, current, and power of street lights and compares them with preset fixed thresholds to determine whether equipment malfunctions.

[0003] However, this passive monitoring method based on steady-state parameter thresholds has significant technical bottlenecks in practical applications. First, the insulation aging or electrolytic capacitor failure of key components such as street light driver power supplies is usually a slow, gradual process. Its early fault characteristics often manifest as extremely weak parameter drift or transient response anomalies. These minute changes are easily masked by grid voltage fluctuations, ambient temperature changes, or electromagnetic noise from the line itself, making it difficult for the system to identify potential hazards before complete functional failure, thus failing to achieve true fault prediction.

[0004] Secondly, as the physical medium for transmitting both electrical energy and information in street lighting networks, power lines have an extremely complex and time-varying channel environment. Existing technologies typically process energy domain data and information domain data separately, failing to fully utilize the high-frequency characteristics of power line carrier communication, such as signal-to-noise ratio and channel frequency response, to assist in diagnosing the health status of physical lines. This makes it difficult for the system to effectively distinguish between communication interruptions caused by external environmental noise and actual internal physical circuit failures when faced with complex field interference, easily leading to high false alarm or missed detection rates. Furthermore, existing monitoring methods lack proactive intervention mechanisms. For "grayscale" nodes on the verge of health or failure, passively received operational data cannot stimulate deep-seated physical characteristics within the equipment, making it difficult for maintenance personnel to confirm uncertain fault signals. This severely restricts the improvement of the refined and intelligent operation and maintenance level of urban lighting networks. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for assessing the health status of streetlight networks and predicting faults. This solves the technical problems of existing streetlight monitoring technologies, which rely on passively receiving steady-state parameters for threshold discrimination. In complex electromagnetic noise environments, these technologies struggle to accurately extract early, weak fault characteristics and cannot effectively distinguish between external environmental interference and internal physical hazards, resulting in high false alarm rates, high missed detection rates, and strong lag in fault prediction.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method and system for assessing the health status and predicting faults in a street light network. The present invention proposes a comprehensive assessment scheme that uses "information-energy isomorphism" to uncover the deep coupling relationship between electrical and communication systems and combines "active micro-perturbation" to achieve causal verification.

[0007] The first aspect of this invention provides a method for assessing the health status and predicting faults in a street light network.

[0008] This method mainly includes five core stages: data acquisition and model acquisition, dynamic topology construction, passive anomaly screening, active micro-perturbation verification, and multi-source evidence fusion.

[0009] I. The multidimensional data acquisition and ideal model construction system first acquires real-time multidimensional heterogeneous data from each node in the street light network. This multidimensional heterogeneous data is divided into two feature domains: energy domain data characterizing the electrical operating state. (Including input voltage) Current Active power Power factor and information domain data characterizing the physical properties of the communication channel. (Including signal-to-noise ratio) Signal attenuation Att, channel frequency response Simultaneously, a pre-defined ideal physical model for each node is obtained. This model is a linear time-invariant (LTI) system model established based on the healthy operating data of the nodes at the beginning of their entire life cycle, specifically represented by a transfer function describing the input-output relationship of the street light driver power supply. .

[0010] II. Dynamic Topology Generation Based on Energy Isomorphism: To capture the implicit electrical coupling and physical connection relationships between street light nodes, this invention constructs a dynamic graph. Adjacency matrix The construction no longer relies solely on geographic location, but is based on real-time signal similarity: Electrical isomorphism: Computational nodes , Voltage waveform sequence within the time window , The Pearson correlation coefficient is used to reflect the homogeneity of the power supply phases: Channel isomorphism: Calculating the channel frequency response vector , The cosine similarity is used to reflect the aging consistency of the physical circuit dielectric properties. Topology fusion: Weight the above two items and set a threshold. Filtering is used to form a sparse adjacency matrix, thereby accurately characterizing the abnormal propagation paths in the network: III. Passive Residual Evaluation Based on Graph Neural Networks: The dynamic topology and normalized heterogeneous features described above are input into a pre-trained graph neural network model. This model employs a Graph Attention Network (GAT) architecture, and its training objective is to minimize the feature reconstruction error of normal samples. During the inference phase, the attention mechanism is used to calculate the correlation strength between nodes. : Based on this coefficient, neighborhood information is aggregated and decoded to obtain the theoretical reconstructed features at the current time. Calculate actual characteristics With theoretical characteristics Euclidean distance as passive monitoring residual : The residual reflects the degree to which the current state of a node deviates from the normal pattern of its neighborhood.

[0011] IV. The determination of suspected grayscale regions and the active excitation trigger adopt a dual threshold determination logic: if Less than the first threshold or greater than the second threshold If the status is confirmed (healthy or diagnosed), a result is generated directly. The system determines that the node is in the "suspected grayscale range". At this point, the statistical characteristics of the passive data are insufficient to distinguish between faults and interference, and the system automatically triggers the active micro-perturbation incentive mechanism.

[0012] V. Active Micro-Perturbation Response Analysis: For grayscale nodes, the system issues a tiny dimming step command, generating a well-defined rectangular pulse excitation. For example, a 5% brightness adjustment. This is achieved using the Laplace transform principle, based on the aforementioned ideal physical model. Calculate the theoretical ideal current response : Simultaneously acquire the actual current transient waveform at a high sampling rate. The integral absolute error between the two in the time domain is calculated as the active response deviation. : This deviation can keenly identify early fault physical characteristics such as impedance mismatch or component parameter drift by observing changes in overshoot, oscillation frequency, or damping ratio in the transient response.

[0013] VI. Multi-source evidence fusion decision-making: Based on the triggering state of proactive incentives, the final health assessment result is dynamically generated. Dynamic confidence weights are introduced. If no active incentive is triggered, then If triggered, the signal-to-noise ratio will be dynamically adjusted based on the signal-to-noise ratio quality during the excitation period. Integrated Health Index The calculation is as follows: Ultimately By mapping to the fault probability space, the health status category of the node is determined. This fusion mechanism ensures low-power monitoring under normal conditions and high-precision diagnosis under complex problems, achieving an optimal balance between monitoring cost and accuracy.

[0014] A second aspect of the present invention provides a street light network health status assessment and fault prediction system. This system includes functional modules configured to perform the above-described methods: The data acquisition module is used to acquire dual-modal data in the energy and information domains; The graph construction module is used to calculate isomorphism relationships based on waveform and channel features and generate dynamic graphs; the passive evaluation module has a built-in graph attention network for calculating feature reconstruction residuals. The active verification module has closed-loop control capability and is used to implement micro-perturbation excitation and analyze transient deviations in grayscale state. The decision fusion module is used to perform direct or weighted fusion decisions based on logical branches.

[0015] A third aspect of the present invention provides an electronic device. It includes a processor and a memory, the memory storing a computer program, which, when executed by the processor, implements the steps of the method described in the first aspect.

[0016] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method described in the first aspect.

[0017] This invention provides a method and system for assessing the health status and predicting faults in a street light network. It offers the following advantages: 1. Utilizing the "information-energy isomorphism" mechanism, this invention enables the keen detection of early, subtle defects. Traditional monitoring often focuses only on high-voltage indicators such as voltage and current, and is insensitive to minute impedance changes in the early stages of line aging. This invention innovatively utilizes the physical characteristics of power lines—that they are both energy transmission conductors and communication carriers—incorporating the channel characteristics of high-frequency carrier communication into the monitoring dimension. Since high-frequency signals are far more sensitive to changes in dielectric loss and contact resistance than power frequency currents, this fusion mechanism allows the system to identify potential problems such as cable insulation aging or loose terminals in advance, before electrical functions fail and only manifest as slight degradation in communication quality, significantly advancing the fault warning time window.

[0018] 2. The construction of a dynamic topology graph effectively overcomes the misjudgment problem caused by spatial heterogeneity. Existing technologies mostly rely on static geographical location information to construct neighborhood relationships, which often leads to topology distortion due to phase switching or line modifications. This invention reconstructs a "signal-energy isomorphic dynamic topology" that reflects the true electrical connection relationships in real time by calculating electrical waveform correlation and channel feature similarity. This ensures that when the graph neural network performs feature aggregation, it only extracts information from neighboring nodes that are truly close in electrical distance, effectively shielding interference from nodes that are geographically close but have different electrical levels, thereby greatly improving the accuracy of the feature reconstruction benchmark in complex power grid environments.

[0019] 3. The hierarchical strategy of "passive initial screening + active verification" resolves the contradiction between false alarms and intrusiveness. Addressing the shortcomings of purely data-driven methods being susceptible to environmental noise interference leading to false alarms, and traditional active testing interfering with normal lighting, this invention designs a triggering logic based on residual intervals. Active micro-perturbations are only triggered when the passive monitoring result is in a "suspected grayscale interval" with low confidence. This strategy retains the advantages of passive monitoring being all-time and zero-perception, while introducing physical stimuli with causal verification capabilities at critical moments. This enables accurate "diagnosis" of difficult faults without affecting the normal lighting function of streetlights, significantly reducing the false alarm rate.

[0020] 4. A transient physical response-based analysis method enables in-depth device-level diagnostics. This invention utilizes fine-tuning light commands as step excitation signals. By comparing the actual transient current response with the output of an ideal model based on the transfer function, it can capture the dynamic features masked in the steady-state data through the integral difference. This method can deeply analyze the charging and discharging characteristics of energy storage components or the saturation trend of inductors within the streetlight driver power supply, thereby identifying component parameter drifts before complete equipment failure, providing a diagnostic basis with clear physical significance for condition-based maintenance (CBM).

[0021] 5. A dynamic confidence-weighted fusion mechanism ensures robustness of the assessment under all-weather conditions. Considering the complex and variable outdoor street light network environment, this invention does not simply superimpose active and passive results, but introduces a dynamic weight allocation mechanism based on signal-to-noise ratio. When severe weather or electromagnetic interference reduces the reliability of active stimulus data, the system automatically reduces its weight in decision-making, relying more on historical passive trends; conversely, it strengthens the decision-making power of active verification. This adaptive fusion logic effectively avoids the impact of single-dimensional data quality fluctuations on diagnostic results, ensuring the robustness and reliability of the final health status assessment results under various operating conditions. Attached Figure Description

[0022] Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention; Figure 2 This is the main flowchart of the method of the present invention; Figure 3 This is a schematic diagram illustrating the construction principle of the information-energy isomorphic dynamic topology graph of the present invention; Figure 4 This is a schematic diagram of the passive feature reconstruction principle based on graph attention network of the present invention; Figure 5 This is a schematic diagram of the state classification determination and active excitation triggering logic of the present invention; Figure 6 This is a schematic diagram of the transient response deviation analysis under active micro-perturbation according to the present invention; Figure 7 This is a hardware structure block diagram of the electronic device of the present invention. Detailed Implementation

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see the appendix Figures 1 to 7 This invention provides a method and system for assessing the health status and predicting faults in a street light network. At the physical layer, the system mainly consists of intelligent single-lamp controllers deployed on street light poles, concentrators or edge gateways responsible for area management, and a monitoring server deployed in the cloud. Each intelligent single-lamp controller communicates bidirectionally with the concentrator via power line carrier (PLC) technology, and the concentrator is connected to the monitoring server via a wide area network.

[0025] The street light network health status assessment and fault prediction method may include the following steps: First, step 1 is performed to collect real-time multidimensional heterogeneous data of each node in the street light network and obtain a preset ideal physical model for each node. In this step, the data acquisition is completed collaboratively by the metering module and PLC communication module integrated within the intelligent single-lamp controller. To comprehensively characterize the physical health status of the street light nodes, the collected data is strictly divided into energy domain data characterizing the electrical operating status and information domain data characterizing the physical characteristics of the communication channel.

[0026] The energy domain data This primarily reflects the steady-state electrical characteristics of the streetlight load (i.e., LED driver power supply and light source module) during power frequency energy transmission. The intelligent single-lamp controller samples the voltage and current waveforms of the power supply line at a preset high-frequency sampling rate and calculates statistical characteristics within a time window. Specifically, for any node... At any moment The energy domain data vector is represented as: in, This is the instantaneous effective value of the input voltage. This is the instantaneous effective value of the operating current. Active power This refers to the power factor. These parameters directly reflect whether the load has a short circuit, open circuit, or severe power attenuation.

[0027] The information domain data This involves using power line carrier communication signals as high-frequency probe signals to acquire channel state information. Since power lines serve as both energy transmission and communication media, aging lines, poor connections, or insulation damage can alter the high-frequency channel impedance characteristics, thus affecting the transmission quality of the carrier signal. Therefore, the communication module simultaneously extracts physical layer channel characteristics during heartbeat packet interaction or data reporting. For any node... Its information domain data vector representation is as follows: in, The signal-to-noise ratio (SNR) of the received signal characterizes the level of background noise in the channel. This represents signal attenuation, reflecting the medium loss along the transmission path. This represents the amplitude vector of the channel frequency response. In Orthogonal Frequency Division Multiplexing (OFDM) communication systems, It consists of the channel gain on each subcarrier, i.e. For the first The center frequency of each subcarrier This represents the total number of subcarriers. This vector can precisely characterize the microscopic impedance changes of the line in the frequency domain.

[0028] In step 1, obtaining the preset ideal physical model for each node serves as the benchmark for subsequent active micro-perturbation analysis. This ideal physical model is not a universal theoretical model, but rather a personalized benchmark model constructed based on data from the node during its initial installation or a period of assured health after maintenance. Considering that the street light driver power supply is essentially a dynamic system containing energy storage components such as inductors and capacitors, this embodiment models it as a linear time-invariant (LTI) system and employs a transfer function... This describes the dynamic relationship between the input control signal and the output response current.

[0029] The specific process of constructing the preset ideal physical model is as follows: During the system initialization phase, data is sent to nodes in a healthy state. A series of step dimming commands of different amplitudes are issued as input signals. Its current response was recorded at a microsecond sampling rate. The transfer function of the street light driver power supply is obtained by fitting using system identification methods (such as least squares method or subspace identification method). : in, Represents the Laplace transform. For complex frequency domain variables, and These are the model coefficients. and These are the orders of the polynomials in the denominator and numerator, respectively (usually...). The transfer function Stored in an edge gateway or cloud database, it serves as a dedicated "digital twin" benchmark for that node, used to calculate the expected response under ideal conditions in subsequent active micro-perturbation tests, thereby providing an accurate mathematical reference for the early detection of faults.

[0030] In this embodiment, after completing the acquisition of multidimensional heterogeneous data, the system executes step 2 to calculate the electrical waveform correlation between nodes in the energy domain data and the channel feature similarity in the information domain data, and constructs a dynamic topology graph of the street light network based on the calculation results. The core of this step is to break through the limitation of traditional methods that rely solely on geographical location to construct static topologies, and instead utilize the real-time coupling characteristics of physical signals to reconstruct the network structure, thereby ensuring that the feature aggregation of the subsequent graph neural network has clear physical homology.

[0031] Specifically, in order to quantize any two nodes and Regarding the correlation at the energy transmission level, this invention utilizes voltage waveform sequences within a time window to calculate the electrical waveform correlation. Let nodes... and The sequence of instantaneous voltage values ​​within the current time window are respectively vectors. and Its dimensions are The Pearson correlation coefficient between the two is calculated as an indicator of electrical waveform correlation. The calculation formula is as follows: in, and They represent the first in the sequence. Values ​​of each sampling point and These are the mean values ​​of the two voltage sequences. The closer this index is to 1, the more it indicates that the two nodes are in the same electrical phase and the power supply quality is affected by the same source disturbance, exhibiting a high degree of electrical isomorphism.

[0032] Meanwhile, to quantify the degree of correlation between nodes at the communication physical layer, this invention utilizes channel frequency response vectors to calculate channel feature similarity. Let nodes... and The channel frequency response amplitude vectors are respectively and These two vectors reflect the attenuation characteristics of the carrier signal at different frequency components. The cosine similarity between them is calculated as a channel feature similarity index. The calculation formula is as follows: Where · represents the vector dot product, The Euclidean norm of a vector. Represents a node In the Channel gain on each subcarrier. Since the characteristics of power line channels are affected by line impedance, branch structure, and aging, high similarity means that the two nodes have experienced similar physical media environments.

[0033] After obtaining the above two indicators, the system performs weighted fusion to generate connection weights, and constructs an adjacency matrix accordingly. For any pair of nodes in the network Its connection weight The calculation is as follows: in, To balance the contributions of the energy domain and the information domain, a weighting coefficient is assigned. Subsequently, a sparsity strategy is introduced, retaining only connection weights greater than a preset threshold. The edge, that is, if Then the adjacent matrix elements ,otherwise The resulting isomorphic dynamic topology graph can dynamically eliminate pseudo-neighbors that are geographically close but have significantly different electrical characteristics, providing an accurate graph structure basis for fault propagation analysis.

[0034] In this embodiment, step 3 is then executed, whereby the isomorphic dynamic topology graph and the multidimensional heterogeneous data are input into a pre-trained graph neural network model. The theoretical features of each node are reconstructed through neighborhood feature aggregation, and the passive monitoring residual is calculated. This model adopts a graph attention network architecture, which uses an attention mechanism to adaptively learn the contribution of neighboring nodes to the state reconstruction of the central node.

[0035] For the central node First, calculate its relationship with its neighboring nodes. (Right now Attention coefficient between nodes The process first performs a linear transformation on the node features, then concatenates the transformed features of the center node and its neighboring nodes, and passes the result through a shared attention vector. The value is mapped to a scalar and then processed using the LeakyReLU activation function: in, The weight matrix is ​​a learnable matrix. and They are nodes and The normalized input feature vector, This represents a vector concatenation operation. This is the transpose of the attention vector. Then, the attention coefficients are normalized using the Softmax function to obtain the final normalized attention weights. : in, Represents a node The set of neighboring nodes. This weight This intuitively reflects the neighboring nodes under the current network topology. The state of the inference node The importance of theoretical state.

[0036] Based on the calculated attention weights, the model performs weighted aggregation of neighborhood features and reconstructs nodes through a decoder (usually a fully connected layer). Theoretical characteristics : in, This is a non-linear activation function. Since this graph neural network model is pre-trained based on a large amount of historical health data, it possesses the ability to infer the expected state of the central node based on the normal patterns of its neighbors. Finally, the system calculates the actual collected features of the central node. Features of Reconstruction Theory The Euclidean distance between them is defined as the passive monitoring residual. : The passive monitoring residual directly quantifies the degree to which the current operating state of a node deviates from the normal mode of its local network, and serves as the basis for subsequent state classification judgment and triggering of active verification.

[0037] In this embodiment, after calculating the passive monitoring residuals of each node, the system executes step 4 to determine the node status based on the passive monitoring residuals. This step employs a dual-threshold hierarchical determination strategy, aiming to balance the real-time performance of the monitoring system with the accuracy of the diagnosis. The system presets a first threshold. Second threshold (in If the passive monitoring residual of a certain node Less than the first threshold This indicates that the actual operational characteristics of the node highly match the theoretical characteristics of neighborhood reconstruction, and the system directly determines that the node is in a healthy state; if Greater than the second threshold This indicates that the node's characteristics have deviated significantly from the statistical norm, and the system directly determines that the node is in a confirmed fault state.

[0038] However, when the passively monitored residual is in the range between the first threshold and the second threshold (i.e. When a node is in the "suspected grayscale range," the system determines that the node is in the "suspected grayscale range." At this time, it is difficult for passive monitoring data alone to distinguish whether this deviation is caused by environmental noise (such as power grid fluctuations or brief channel interference) or by early minor faults (such as slight capacitor drying or micro-oxidation of terminals). To eliminate this uncertainty, the system triggers step 5 for nodes in this grayscale range, that is, to execute the active micro-perturbation excitation mechanism.

[0039] In this embodiment, the active micro-perturbation excitation mechanism is not a traditional full-power test. Instead, it generates a rectangular pulse excitation signal with negligible impact on human vision but captureable electrical characteristics by sending a micro-adjustment command to the target node. Specifically, the system controls the street light driver power supply within an extremely short time window. A tiny brightness step is superimposed inside. (For example, if the adjustment range is 5%), the corresponding voltage control signal is represented by an excitation function in the time domain. The excitation signal is designed to disrupt the steady-state equilibrium of the system, forcing the circuit to exhibit a transient response with rich damping and frequency characteristics.

[0040] Simultaneously with the application of excitation, the system invokes the preset ideal physical model of the node obtained in step 1 to calculate the theoretical ideal current response. Since this preset ideal physical model uses a transfer function... The system first presents the time-domain excitation signal in the form of... Performing a Laplace transform yields its complex frequency domain representation. According to linear system theory, the complex frequency domain expression of the ideal output response is: Subsequently, the ideal current response curve in the time domain was calculated using the inverse Laplace transform algorithm. : in, This represents the inverse Laplace transform operator. This represents the standard transient trajectory that the node should exhibit when faced with this micro-perturbation excitation, assuming it is in a completely healthy state and its parameters have not drifted.

[0041] At the same time, the intelligent single-lamp controller synchronously collects the actual current response of the node at a high sampling rate (e.g., above 10kHz) during the excitation period. To quantify the difference between the actual physical state and the ideal model, the system calculates the integral difference between the actual current response and the ideal current response in the time domain, which is used as the active response deviation. The calculation formula is as follows: This integration operation covers the entire transient response process and can sensitively capture waveform distortions caused by changes in circuit parameters. For example, if the equivalent series resistance (ESR) of the electrolytic capacitors inside the drive power supply increases, it will cause a change in the damping ratio of the transient response, thereby affecting the actual waveform. In terms of overshoot or settling time Significant separation occurs. Therefore, As a physical indicator based on causal verification, it can effectively isolate the interference of environmental noise and reflect the true health level of the node hardware.

[0042] In this embodiment, after completing the active micro-perturbation excitation and calculating the active response deviation, the system executes step 6. For the node that triggered the active micro-perturbation excitation mechanism, the passive monitoring residual and the active response deviation are fused to update and generate the final health status assessment result. This step aims to solve the problem of inconsistent data reliability under different operating conditions through dynamic weight allocation. The system first defines a dynamic confidence weight mechanism, which adaptively adjusts the fusion ratio of passive and active features according to the signal quality during the active excitation process.

[0043] Specifically, for any node The system determines the weight allocation strategy based on the judgment result in step 4. If the node does not trigger the active micro-perturbation excitation mechanism (i.e., it is in a confident healthy or faulty range), the weight of the passive monitoring residual is set to 1, and the weight of the active response deviation is set to 0. If the node triggers the active micro-perturbation excitation mechanism, the system reads the signal-to-noise ratio reported by the communication module during the excitation period. And based on this, the confidence weight of active evidence is calculated. Since a higher signal-to-noise ratio means less noise contamination in the acquired transient response waveform, its reliability in reflecting physical reality is higher. and They show a positive correlation. This embodiment uses the Sigmoid function to map the signal-to-noise ratio to the interval (0,1): in, To adjust the coefficient for the steepness of the curve, This is the median threshold for the signal-to-noise ratio. At this point, the weight of the passive monitoring residual is adjusted accordingly. .

[0044] After determining the weights, the system monitors the passive monitoring residuals. and active response bias Normalization was performed separately to eliminate dimensional differences, resulting in... and Subsequently, a weighted sum is performed based on the assigned weights to calculate the fusion health index of the node. : The Integrated Health Index It integrates statistical trends from long-term passive monitoring with physical causal evidence from instantaneous active verification. Finally, the system uses a classifier (such as a Softmax regression layer) to map the fused health index to a failure probability distribution to determine the final health status category of the node. : in, Represents the status category (such as health, insulation aging, poor contact, drive failure, etc.). The total number of categories, These are the model parameters for the classifier. Through this probability mapping, the system can not only output discrete state results, but also provide the confidence probability of the judgment for operation and maintenance personnel to make decisions.

[0045] This invention also provides a street light network health status assessment and fault prediction system. The system logically includes a data acquisition module, a graph construction module, a passive evaluation module, an active verification module, and a decision fusion module. The data acquisition module is configured to periodically read dual-modal data in the energy and information domains from underlying sensors. The graph construction module generates a dynamic adjacency matrix reflecting electrical physical connections in real time based on the aforementioned waveform correlation and channel similarity algorithms. The passive evaluation module runs a graph neural network algorithm and outputs passive residuals. The active verification module includes closed-loop control logic, responsible for issuing dimming signals and simultaneously acquiring high-frequency current data upon receiving a grayscale determination command. The decision fusion module performs the aforementioned dynamic weight calculation and probability mapping functions and outputs a final report.

[0046] Furthermore, the present invention can also be implemented using an electronic device. This electronic device includes, at the hardware level, a processor, a memory, and a computer program stored in the memory and executable on the processor. The processor can be a central processing unit (CPU), a digital signal processor (DSP), or a field-programmable gate array (FPGA), etc., and interacts with the memory via an internal bus. The memory stores instruction code that implements steps 1 to 6 above. When the processor executes the program, the electronic device can process the accessed street light network data in real time, construct a homogeneous information and energy topology, perform active and passive collaborative diagnostics, and send the final health assessment results to a remote monitoring terminal via a network interface, thereby achieving intelligent management of the entire lifecycle of the street light network.

[0047] In this embodiment, the present invention also provides a computer-readable storage medium on which a computer program is stored. The storage medium can be any medium capable of storing program code, such as a read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk. When the computer program is executed by a processor, it can implement the various steps of the street light network health status assessment and fault prediction method described in the above embodiments. Specifically, the computer program is logically divided into multiple cooperating functional modules. These modules exchange data through a system bus or inter-process communication mechanism, collectively constituting the software core of intelligent diagnosis of the street light network.

[0048] The computer program mainly includes data acquisition module instructions, graph construction module instructions, passive evaluation module instructions, active verification module instructions, and decision fusion module instructions. When the processor runs the data acquisition module instructions, the system periodically reads dual-modal data of the energy domain and information domain from the underlying hardware registers and performs noise reduction and normalization processing on it. When running the graph construction module instructions, the system dynamically calculates the connection weights between nodes based on the electrical waveform correlation formula and the channel feature similarity formula, and updates the adjacency matrix of the energy-information isomorphic topology graph in memory. When the passive evaluation module instruction is executed, the system loads the preset graph neural network model parameters, performs feature aggregation operations based on the attention mechanism, and outputs the passive monitoring residual. When the active verification module instruction is executed, the system monitors the residual status. Once a grayscale range is detected, a fine-tuning optical interrupt request is immediately generated, the PWM (pulse width modulation) driver is controlled to output a rectangular pulse, and a high-speed ADC (analog-to-digital converter) is started to capture transient current. When the decision fusion module instruction is executed, the system calculates dynamic weights based on the signal-to-noise ratio, executes a weighted fusion algorithm and probability mapping, and finally generates a diagnostic report containing health status categories and confidence levels.

[0049] To ensure that the "pre-trained graph neural network model" in the aforementioned passive evaluation module can accurately reconstruct the theoretical features, the computer program also includes a sequence of instructions for offline training or online updating of the model. During the model training phase, the processor retrieves a large-scale, multi-dimensional, heterogeneous dataset accumulated by the streetlight network during historical healthy periods from the storage medium. A Graph Attention Network (GAT) is constructed as an autoencoder architecture, and a reconstruction loss function is defined. The mean squared error between the actual features and the reconstructed features of all nodes: in, for Time Node Input features, The reconstructed features of the network output, The set of learnable parameters for the network (including the attention layer weight matrix) and attention vector ), This is the regularization coefficient. The processor uses stochastic gradient descent (SGD) or Adam optimization algorithms to minimize the loss function through backpropagation. This continues until the model converges. The parameters after training are... This information is embedded and encapsulated into the passive evaluation module, serving as a benchmark for online inference. This training mechanism ensures that the model can deeply learn the high-dimensional coupling patterns between energy domain data and information domain data in the graph topology of the street light network under normal operating conditions, thus providing a high-precision reference background for anomaly detection.

[0050] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the device, system, and storage medium embodiments, since they are basically similar to the method embodiments, the descriptions are relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that the specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples.

[0051] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. For those skilled in the art, the present invention can have various modifications and variations. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention. For example, although this embodiment describes the acquisition of information domain data using power line carrier (PLC) communication as an example, in street light networks employing wireless communication technologies such as ZigBee, LoRa, or NB-IoT, the received signal strength indicator (RSSI) and link quality indicator (LQI) of the wireless channel are used as information domain features, and the principle of signal-energy homogeneity and active-passive fusion proposed in this invention is equally applicable.

[0052] This embodiment further describes in detail the hardware architecture of the electronic device for implementing the street light network health status assessment and fault prediction method of this invention. This electronic device can be implemented as an edge computing gateway deployed at the edge of the street light network, or as a data center server deployed in the cloud. This electronic device is designed to provide powerful computing capabilities to support the inference of graph neural networks and the real-time closed-loop control of the active incentive mechanism.

[0053] As shown in the figure, this electronic device is based on a bus architecture and includes at least one processor, memory, communication interface, bus, and input / output (I / O) interface. The bus serves as the system's data transmission backbone, used to transmit address, data, and control signals between the processor, memory, communication interface, and other components. This bus architecture can employ industry-standard bus protocols, such as PCIe (PCIe High-Speed ​​Peripheral Interconnect Standard) or AXI (Advanced Extensible Interface) bus, to ensure high-speed throughput of multidimensional heterogeneous data between various hardware units.

[0054] The processor is the computational core of the electronic device, configured to execute computer programs stored in memory to implement the aforementioned method steps. Considering that this invention relates to matrix operations (such as feature aggregation and attention coefficient calculation) and complex Laplace transforms in graph neural networks, the processor preferably employs a high-performance central processing unit (CPU) with a floating-point unit (FPU), or a heterogeneous computing architecture consisting of a CPU and a graphics processing unit (GPU), tensor processing unit (TPU), or field-programmable gate array (FPGA). In particular, when the processor includes a GPU or FPGA accelerator, it is configured to process the neighborhood feature aggregation operation in step 3 in parallel. Hardware parallel acceleration ensures that the system can maintain millisecond-level inference response speed even when the number of street light network nodes is large (e.g., tens of thousands of nodes).

[0055] The memory includes volatile memory (such as high-speed random access memory DRAM) and non-volatile memory (such as flash memory, hard disk, or solid-state drive SSD). Non-volatile memory is used for persistent storage of the operating system, application instructions, and critical data assets. Specifically, the "preset ideal physical model" (i.e., transfer function) of each node obtained in step 1... parameter coefficients and (and the parameters of the pre-trained graph neural network model from step 3) All data are stored in this non-volatile memory in the form of a structured database or serialized files. The volatile memory serves as main memory, used to load the aforementioned model parameters during processor runtime and to cache real-time multidimensional heterogeneous data streams, including voltage and current waveform sequences and channel frequency response vectors.

[0056] The communication interface serves as a physical bridge connecting the electronic device to the external street light network. In this embodiment, the communication interface includes a power line carrier (PLC) master module or a network adapter connected to it. This interface is responsible for receiving uplink data packets (containing energy domain data) from the underlying single-lamp controller. and information domain data It also undertakes the task of sending downlink control commands. Especially when the active micro-perturbation excitation mechanism is triggered in step 5, the processor sends a message containing fine-tuning optical parameters (such as...) to a node at a specific address through this communication interface. and The high-priority control frames are used. This communication interface integrates hardware-level timers and interrupt controllers to ensure microsecond-level synchronization accuracy between the transmission of excitation commands and subsequent transient response data acquisition, thereby preventing active response deviations. The accuracy of the calculation.

[0057] In addition, this electronic device is connected to a display unit and input devices via I / O interfaces to construct a human-machine interface. Maintenance personnel can intuitively view the generated dynamic topology diagram of the information and energy homogeneity, as well as the real-time health status of each node, through the display unit. And fault probability distribution. When the system detects a node with a "confirmed fault" or "suspected grayscale range", the I / O interface drives the alarm device to issue an audible and visual alert, and highlights the geographical location and electrical topology location of the abnormal node on the interface to assist maintenance personnel in performing rapid on-site maintenance.

[0058] In summary, the electronic device provided by this invention achieves an organic combination of "passive monitoring and initial screening" and "active stimulation for diagnosis" through the coordinated operation of a high-performance processor, a large-capacity hierarchical memory, and a high-real-time communication interface. This hardware architecture not only supports complex graph neural network algorithms but also precisely executes micro-perturbation control at the physical layer, thereby solving the problems of insufficient computing power leading to difficulty in feature extraction and control delays causing diagnostic errors in existing technologies. This provides a solid physical foundation for the intelligent health management of street light networks.

[0059] In this embodiment, further considering the challenges posed by large-scale street light networks to communication bandwidth and computing resources, the aforementioned street light network health status assessment and fault prediction method can be implemented not only with a centralized architecture in the cloud but also with a distributed architecture that combines cloud and edge computing. In the distributed implementation, the edge computing gateway is configured to undertake the inference task of the local subgraph. Specifically, the edge gateway only acquires street light node data within its jurisdiction, constructs a local isomorphic dynamic topology graph, and loads a lightweight graph neural network model to perform the passive residual calculation in step 3.

[0060] Under this architecture, only when the passive monitoring residual calculated by the edge gateway... Only when the preset upload threshold is exceeded, or when a node is determined to be in a "suspected grayscale range," are the relevant abnormal feature data and original waveforms uploaded to the cloud server, which then issues an active micro-perturbation command. This hierarchical processing mechanism can significantly reduce the amount of raw data sent back to the cloud, reducing network bandwidth utilization by more than an order of magnitude, while ensuring low-latency fault response. It is particularly suitable for city-level smart lighting projects containing tens of thousands of nodes.

[0061] Furthermore, the "preset ideal physical model" mentioned in this invention has the ability to adaptively update throughout its entire lifecycle. Although the model is initially built on the health data after node installation or major overhaul, the system is also equipped with a model calibration mechanism to cope with non-faulty natural aging in the street light driver power supply (such as the normal degradation of electrolytic capacitor capacity). When maintenance personnel complete the inspection or component replacement of a node on-site and enter the "maintenance completed" command through a handheld terminal, the system will automatically trigger the baseline reconstruction process for that node.

[0062] This reconstruction process will re-execute the system identification algorithm in step 1, collect the current step response data, and calculate the new transfer function parameters. and It also updates the ideal model library in memory. This mechanism ensures that active response bias is mitigated. The calculations are always based on the current optimal physical state of the equipment, avoiding false alarms caused by reference drift after long-term operation of the equipment, thus achieving precise health management throughout the entire process from "new installation" to "scrap".

[0063] It should be noted that although this invention uses brightness adjustment (dimming) as the specific form of active micro-perturbation excitation, in the application scenario of intelligent LED driver power supplies with multi-output functions (such as RGBW color-adjustable streetlights), the active micro-perturbation excitation mechanism can also be manifested as color temperature adjustment commands or spectral composition adjustment commands. As long as the control command can trigger a transient change in the energy state inside the driver circuit and produce an observable electrical response, it falls within the scope of "micro-perturbation excitation" as defined in this invention. For example, rapidly switching the load distribution of warm and cool color temperature channels will also excite the LC filter in the circuit to generate specific transient oscillations, and this oscillation characteristic is also applicable to the formula. The analytical framework.

[0064] Finally, it should be understood that the specific embodiments described in this specification are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of the present invention. The terms "comprising," "including," or any other variations thereof as used in this invention are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without conflict, the embodiments and features described in the present invention can be combined with each other.

[0065] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for assessing the health status and predicting faults in a street light network, characterized in that, Includes the following steps: Step 1: Collect real-time multidimensional heterogeneous data of each node in the street light network and obtain the preset ideal physical model of each node; The multidimensional heterogeneous data includes energy domain data characterizing the electrical operating state and information domain data characterizing the physical characteristics of the communication channel. Step 2: Calculate the electrical waveform correlation between nodes in the energy domain data and the channel characteristic similarity in the information domain data, and construct the information-energy isomorphic dynamic topology diagram of the street light network based on the calculation results; Step 3: Input the information-energy isomorphic dynamic topology graph and the multidimensional heterogeneous data into the pre-trained graph neural network model, reconstruct the theoretical features of each node through neighborhood feature aggregation, and calculate the passive monitoring residual between the theoretical features and the actual features. Step 4: Determine the node status based on the passive monitoring residual: If the passive monitoring residual is less than the first threshold or greater than the second threshold, a health or fault assessment result is directly generated based on the residual. If the passive monitoring residual is in the suspected grayscale range between the first threshold and the second threshold, the active micro-perturbation excitation mechanism is triggered by adjusting the street light dimming parameters. Step 5: For the node that triggers the active micro-perturbation excitation mechanism, collect its transient response data under excitation, and calculate the active response deviation between the transient response data and the output of the ideal physical model; Step 6: For nodes that trigger the active micro-perturbation incentive mechanism, fuse the passive monitoring residual and the active response deviation to update and generate the final health status assessment result.

2. The method for assessing the health status and predicting faults in a street light network according to claim 1, characterized in that, The multidimensional heterogeneous data specifically includes: The energy domain data includes the instantaneous RMS value of the input voltage, the instantaneous RMS value of the operating current, the active power, and the power factor; The information domain data includes the signal-to-noise ratio, signal attenuation, and amplitude vector of the channel frequency response for power line carrier communication.

3. The method for assessing the health status and predicting faults in a street light network according to claim 1, characterized in that, The construction of the information-energy isomorphic dynamic topology diagram of the street light network based on the calculation results specifically includes: Calculate the Pearson correlation coefficient of the voltage waveform sequences of any two nodes within a time window, and use it as the electrical waveform correlation. Calculate the cosine similarity between the channel frequency response vectors of any two nodes as the channel feature similarity; The electrical waveform correlation and the channel feature similarity are weighted and summed to obtain the connection weights. Edges with connection weights greater than a preset threshold are retained to generate the adjacency matrix of the signal-energy isomorphic dynamic topology graph.

4. The method for assessing the health status and predicting faults in a street light network according to claim 1, characterized in that, The process of reconstructing the theoretical features of each node through neighborhood feature aggregation and calculating the passive monitoring residual between the theoretical features and the actual features specifically includes: The attention coefficient between a node and its neighboring nodes is calculated using a graph attention mechanism. The attention coefficient represents the contribution of the neighboring nodes to the feature reconstruction of the central node. The features of neighboring nodes are weighted and aggregated based on the attention coefficients, and the Euclidean distance of the center node is calculated by the decoder to reconstruct the theoretical features. Calculate the Euclidean distance between the actual features of the central node and the reconstructed theoretical features, and use the Euclidean distance as the passive monitoring residual.

5. The method for assessing the health status and predicting faults in a street light network according to claim 1, characterized in that, The determination of node status based on the passively monitored residual specifically includes: If the passive monitoring residual is less than the first threshold, the node is determined to be in a healthy state. If the passive monitoring residual is greater than the second threshold, the node is determined to be in a confirmed fault state. If the passive monitoring residual is between the first threshold and the second threshold, the node is determined to be the suspected grayscale range, and the active micro-perturbation incentive mechanism is executed.

6. The method for assessing the health status and predicting faults in a street light network according to claim 1, characterized in that, The calculation of the active response deviation between the transient response data and the output of the ideal physical model specifically includes: A fine-tuning light command is sent to the node to generate a rectangular pulse excitation signal; Obtain the preset ideal physical model, which is specifically a street light driving power transfer function constructed based on the historical normal parameters of the nodes, and calculate the ideal current response under the action of the rectangular pulse excitation signal; The actual current response of the high-frequency sampling node during the excitation period is used to calculate the integral difference between the actual current response and the ideal current response in the time domain, which is taken as the active response deviation.

7. The method for assessing the health status and predicting faults in a street light network according to claim 1, characterized in that, The generation of health or failure assessment results and the updating and generation of final health status assessment results specifically include: Define dynamic confidence weights. If the active micro-perturbation excitation mechanism is not triggered, the weight of the passive monitoring residual is set to 1. If the active micro-perturbation excitation mechanism is triggered, the weights of the passive monitoring residual and the active response deviation are allocated according to the signal-to-noise ratio of the active excitation. Based on the assigned weights, the normalized passive monitoring residuals and active response biases are weighted and summed to obtain the fusion health index. The fused health index is mapped to the fault probability distribution to determine the health status category of the node.

8. A street light network health status assessment and fault prediction system, characterized in that, include: The data acquisition module is used to collect real-time multidimensional heterogeneous data from each node in the street light network. The multidimensional heterogeneous data includes energy domain data characterizing the electrical working state and information domain data characterizing the physical characteristics of the communication channel. The graph construction module is used to calculate the electrical waveform correlation between nodes in the energy domain data and the channel feature similarity in the information domain data, and to construct a dynamic topology graph of the street light network based on the calculation results. The passive evaluation module is used to input the information-energy isomorphic dynamic topology graph and the multidimensional heterogeneous data into a pre-trained graph neural network model to calculate the passive monitoring residual. The active verification module is used to trigger an active micro-perturbation excitation mechanism for the node when the passive monitoring residual is in a preset suspected grayscale range, and to calculate the active response deviation. The decision fusion module is used to directly generate evaluation results based on the passive monitoring residuals when the active micro-perturbation incentive mechanism is not triggered. And when the active micro-perturbation incentive mechanism is triggered, the passive monitoring residual and the active response deviation are fused together to output the final health status assessment result of each node.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images