A matrix layout lighting state detection method, device and equipment

Abstract

The application discloses a kind of matrix layout lighting state detection method, device and equipment, collect and analyze power bus current waveform and single lamp feedback signal, generate node feature cluster;Extract adjacent luminaire power consumption difference, identify abnormal observation point and construct fault diffusion chain;Mark time stamp to fault diffusion chain, determine fault source and through injection feature mark current demarcate influence domain;Monitoring anchor point is deployed at influence domain boundary, and the amount of change of light color parameter is collected and remaining life is predicted;Extract the oscillation envelope of starting current spike, assess the health of driver and generate maintenance density field;Fusion multidimensional detection data constructs three-dimensional detection field, finds singular point and accurately analyzes fault luminaire coordinates through fault positioning pulse.The application realizes stereoscopic monitoring from electrical characteristics, optical performance to driver state, can accurately locate single lamp fault, predict life trend, track fault propagation, and provides technical support for intelligent operation and maintenance of large-scale lighting system.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of lighting control and fault detection, in particular to a lighting state detection method, device and equipment of matrix layout. BACKGROUND

[0002] With the wide application of intelligent lighting systems in large buildings, industrial plants, urban roads and other scenes, the scale of lighting networks with matrix layout is increasingly large, and a single lighting matrix may contain hundreds or even thousands of lamp nodes. These lamps are connected to each other through complex power supply networks and control systems, forming a highly integrated lighting infrastructure.

[0003] However, the traditional lighting maintenance method mainly relies on manual inspection and post-fault maintenance, which has problems such as low detection efficiency, difficult fault location, high maintenance cost, etc. Especially for large-scale lighting systems with matrix layout, the failure of a single lamp may affect adjacent nodes through the power supply network, forming a fault diffusion effect, and the existing detection means cannot accurately track the fault propagation path and predict potential risks. At the same time, the light decay process of LED lamps has gradual characteristics, and traditional methods cannot realize life prediction and preventive maintenance, resulting in a decline in lighting quality and energy waste.

[0004] Therefore, there is an urgent need for a method to solve at least one of the above problems. SUMMARY

[0005] The present application provides a lighting state detection method, device and equipment of matrix layout, aiming to fully monitor and analyze the running state connotation of lighting matrix, and fuse electrical characteristics, optical parameters and spatio-temporal correlation knowledge, to comprehensively and accurately detect key state elements such as fault diffusion, life decay and drive health, and then reveal the fault propagation law and performance degradation mechanism between lighting nodes, and finally form a multi-dimensional, multi-level and high-precision lighting state detection system, providing comprehensive and real-time state diagnosis resources for intelligent lighting maintenance, predictive replacement, fault rapid positioning and other application scenarios.

[0006] The first aspect of the present application provides a lighting state detection method of matrix layout, comprising the following steps:

[0007] Collecting the power bus current waveform and single lamp feedback signal of the lighting matrix, analyzing the power bus current waveform to generate a power consumption profile, identifying an address code from the single lamp feedback signal, and dividing the power consumption profile into node feature clusters according to the address code;

[0008] Extracting the power consumption difference of adjacent lamps based on the node feature clusters, identifying abnormal observation points within the clustering radius according to the power consumption difference, connecting the abnormal observation points to form abnormal links, and generating a fault diffusion chain by backtracking along the abnormal links;

[0009] marking a timestamp sequence on the fault diffusion chain, judging a fault source according to the timestamp sequence, expanding anisotropically from the fault source to demarcate an influence domain, and generating a monitoring anchor point at the boundary of the influence domain;

[0010] acquiring a light color parameter variation through the monitoring anchor point, generating a decay trajectory based on the light color parameter variation, predicting a remaining life based on the decay trajectory, and generating a replacement priority list according to the remaining life ranking;

[0011] extracting a starting current spike in the node feature cluster, obtaining an oscillation envelope of the starting current spike, judging a driver health degree according to the oscillation envelope, and performing matrix coordinate mapping on the driver health degree to form a maintenance density field;

[0012] constructing a three-dimensional detection field based on the fault diffusion chain, the replacement priority list and the maintenance density field, finding a singular point in the three-dimensional detection field, generating a fault positioning pulse with the singular point as the center, and analyzing a fault lamp coordinate through the fault positioning pulse to complete the matrix layout lighting state detection.

[0013] The second aspect of the present application proposes a matrix layout lighting state detection device, comprising:

[0014] a signal acquisition module, configured to acquire a power bus current waveform of a lighting matrix and a single lamp feedback signal, perform waveform analysis on the power bus current waveform to generate a power consumption profile, and identify an address code from the single lamp feedback signal, and divide the power consumption profile into node feature clusters according to the address code;

[0015] an abnormality tracking module, configured to extract a power consumption difference value of adjacent lamps based on the node feature clusters, identify an abnormal observation point within a clustering radius according to the power consumption difference value, connect the abnormal observation points to form an abnormal link, and backtrack along the abnormal link to generate a fault diffusion chain;

[0016] an influence analysis module, configured to mark a timestamp sequence on the fault diffusion chain, judge a fault source according to the timestamp sequence, expand anisotropically from the fault source to demarcate an influence domain, and generate a monitoring anchor point at the boundary of the influence domain;

[0017] a life prediction module, configured to acquire a light color parameter variation through the monitoring anchor point, generate a decay trajectory based on the light color parameter variation, predict a remaining life based on the decay trajectory, and generate a replacement priority list according to the remaining life ranking;

[0018] a health assessment module configured to extract a starting current spike in the node feature cluster, obtain an oscillation envelope of the starting current spike, determine a driver health degree according to the oscillation envelope, and perform matrix coordinate mapping on the driver health degree to form a maintenance density field;

[0019] a fault positioning module configured to construct a three-dimensional detection field based on the fault diffusion chain, the replacement priority list and the maintenance density field, find a singular point in the three-dimensional detection field, generate a fault positioning pulse with the singular point as a center, and analyze a fault lamp coordinate through the fault positioning pulse to complete the matrix layout lighting state detection.

[0020] A third aspect of the present application provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the matrix layout lighting state detection method disclosed in the first aspect when executing the program.

[0021] The present application has the following beneficial effects: 1. The running data of each lighting node can be independently extracted by collecting single lamp feedback signals, combining address code recognition and power consumption profile segmentation; the fault diffusion chain and the propagation path can be tracked based on the analysis of the power consumption difference of adjacent nodes and the time stamp marking, so that the row and column coordinates of the fault lamp can be accurately positioned in a lighting matrix containing thousands of nodes, the positioning accuracy reaches the spacing of a single lamp, and the complete propagation process of the fault from the source to each node is restored. 2. By monitoring the time sequence changes of luminous flux, color temperature, color rendering index and other light color parameters, the initial drift and the late degradation characteristics are separated by using the stage decomposition method to construct the decay trajectory, so that the remaining use time of each lamp from the current time to the end of life can be predicted; by extracting the oscillation envelope of the starting current spike, analyzing the overshoot amplitude, decay rate and other characteristic parameters, the percentage quantitative score of the driver health degree is realized, so that the manager can make replacement plans in advance according to the prediction results. 3. By fusing the fault diffusion chain, the life prediction data and the health degree distribution to construct a three-dimensional detection field, the periodic absence of field strength is identified by using the spiral scanning path and the autocorrelation analysis method, so that the abnormal area in the lighting matrix can be automatically found; by using the row and column encoding technology and echo analysis of the fault positioning pulse, the fault lamp coordinate can be quickly solved by detecting the abnormal decay row and the echo missing position, which changes the traditional artificial one-by-one checking into automatic accurate positioning, and reduces the maintenance time.

[0022] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0023] The drawings here show the specific examples of the technical solutions of the present application, and constitute a part of the specification together with the specific embodiments, for explaining the technical solutions, principles and effects of the present application.

[0024] Unless specifically stated or otherwise defined, the same reference signs in different drawings represent the same or similar technical features, and different reference signs may also be used to represent the same or similar technical features.

[0025] Figure 1 is a flow diagram of a matrix layout lighting state detection method of the present application.

[0026] Figure 2 is a structural block diagram of a matrix layout lighting state detection device of the present application.

[0027] Figure 3 is a structural diagram of a computer device of the present application. DETAILED DESCRIPTION

[0028] In the following description, for the purpose of explanation and not limitation, specific details are set forth, such as specific system structures, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary details.

[0029] It should be understood that when used in the specification and the appended claims of the present application, the term "comprising" indicates the presence of the described features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0030] In addition, in the description of the present application and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0031] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0032] The technical solutions of the embodiments of this application will be described below.

[0033] like Figure 1 As shown, this embodiment of the invention provides a method for detecting the lighting state of a matrix layout, including the following steps S110-S160:

[0034] Step S110: Collect the power supply bus current waveform and single lamp feedback signal of the lighting matrix, perform waveform analysis on the power supply bus current waveform to generate power consumption profile, identify the address code from the single lamp feedback signal, and divide the power consumption profile according to the address code to form node feature clusters.

[0035] Specifically, the current waveform of the power supply bus and the feedback signal of each lamp in the lighting matrix are collected. A high-precision current transformer is installed at the power supply bus inlet of the lighting matrix, with a measurement range of 0-100A, an accuracy class of 0.2, and a sampling frequency of 10kHz, to collect the bus current waveform data in real time. The output signal of the current transformer is processed by a signal conditioning circuit, including amplification, filtering, and analog-to-digital conversion, to ensure the accuracy of the waveform data. The bus current data recording format is [timestamp, instantaneous current value, effective value, sampling number], and the data storage uses a circular buffer to retain the historical data of the most recent 30 minutes. The feedback signal of each lamp is collected via power line carrier communication, with a carrier frequency of 132kHz, FSK modulation, and a data transmission rate of 1200bps. Each lighting node is equipped with a carrier communication module to periodically send status feedback signals, including node address, working status, fault codes, etc. The feedback signal format is [start bit, address code, status byte, check code, end bit], where the address code is a 16-bit unique identifier that supports addressing 65,536 nodes. The signal receiver employs a carrier demodulation chip with a demodulation sensitivity of -95dBm, ensuring reliable communication even in noisy power grid environments. The data acquisition synchronization mechanism uses GPS clock calibration to ensure time alignment between the bus current and the feedback signal, achieving a synchronization accuracy of 1ms.

[0036] The power consumption profile is generated by waveform analysis of the bus current waveform. Using the collected bus current waveform data, combined with the bus voltage (usually 220V or 380V), the instantaneous power P(t) = U x I(t) is calculated, where P(t) is the instantaneous power (W), U is the bus voltage (V), and I(t) is the instantaneous current (A). The power consumption profile is generated by a sliding window method, with a window length of 1 second and a sliding step of 0.1 seconds. The average power in each window is calculated as the power consumption value at that time. Waveform feature extraction includes fundamental component, harmonic content, power factor, etc. The fundamental component is obtained by FFT transformation, reflecting the main power consumption component of the lighting load. Harmonic analysis focuses on 3rd, 5th, and 7th harmonics, which are usually generated by the switching power supply of LED lighting equipment. The power factor is calculated using the formula cosφ = P_avg / S, where P_avg is the average active power and S is the apparent power. The power factor reflects the power efficiency of the lighting system. The time resolution of the power consumption profile is 0.1 seconds, and the power resolution is 1 W, which can capture the dynamic changes of the lighting system. Profile smoothing uses a 5-point moving average filter to eliminate transient disturbances while preserving the power consumption trend. The anomaly detection mechanism identifies power mutation points, with a mutation threshold set to 20% of the normal power.

[0037] Address code recognition from single lamp feedback signal. First, frame synchronization detection is performed to determine the start position of the data frame by identifying a specific start bit sequence (0xAA55). Address code extraction starts from the 3rd byte after the start bit and reads 2 bytes (16 bits) as the node address code. Address code verification uses the CRC-16 check algorithm to ensure that the address code has not been corrupted during transmission. The address code mapping table maintains the correspondence between each address code and the physical location, in the format [address code, floor, area, luminaire number]. The repeated address detection mechanism prevents address conflicts, triggering an alarm and recording conflict information when the same address code is detected. The address code recognition success rate statistics show that the recognition success rate reaches more than 99.5% under normal power grid environment. Time stamping adds precise time stamps to each recognized address code, with a time stamp accuracy of 1 ms. Address code caching uses a hash table structure, with the address code as the key value, storing the latest state information and time stamp, supporting fast query and update.

[0038] The power consumption profile is divided into node feature clusters according to the address code. Based on the identified address code and the corresponding timestamp information, the continuous power consumption profile data is divided into segments corresponding to each lighting node. The segmentation algorithm uses an event-triggered mechanism. When the feedback signal of a certain address code is detected, the power consumption profile data of each 5 seconds before and after this time is extracted as the feature window of the node. Multi-node concurrent processing takes into account the possibility that multiple lighting nodes may work simultaneously, and uses a power consumption decomposition algorithm to estimate the independent contribution of each node. Power consumption decomposition is based on the rated power and historical power consumption pattern of each node, and solves the power distribution coefficient by least squares method. Node feature extraction includes startup power peak, steady-state power level, power fluctuation amplitude, shutdown delay time and other parameters. Feature vector construction organizes the feature parameters of each node into a fixed-dimensional vector, and the vector format is [peak power consumption, steady-state power consumption, fluctuation amplitude, power factor, harmonic distortion rate]. Feature cluster formation uses the K-means algorithm to classify nodes with similar features into a class, and the number of clusters is preset according to the lighting type. In-cluster consistency test ensures that nodes in the same feature cluster have similar electrical characteristics, and the consistency index uses the ratio of intra-cluster variance to inter-cluster variance. Feature cluster labels include normal operation cluster, efficiency reduction cluster, fault risk cluster and other categories, and each node is assigned to the corresponding cluster according to its feature vector.

[0039] Step S120, based on the node feature cluster, the power consumption difference of the adjacent lamps is extracted, the abnormal observation points within the clustering radius are identified according to the power consumption difference, the abnormal observation points are connected to form an abnormal link, and the fault diffusion chain is generated by backtracking along the abnormal link.

[0040] Specifically, the power consumption difference of adjacent luminaires is extracted based on node feature cluster. With node feature cluster data, the adjacent relationship is determined according to the physical topology of the lighting matrix, and the adjacent relationship is defined as the luminaires nodes that are physically adjacent or directly adjacent in electrical connection on the same power supply branch. The power consumption difference calculation uses the absolute difference of steady-state power consumption, ΔP = |P_i - P_j|, where ΔP is the power consumption difference (W), P_i and P_j are the steady-state power consumption of adjacent nodes i and j (W). The difference matrix is constructed to form an N x N symmetric matrix, N is the total number of lighting nodes, and the matrix element D[i,j] stores the power consumption difference between nodes i and j. The time window is set to 5 minutes, and the average power consumption difference is calculated within this time window to eliminate the influence of transient fluctuations. The difference normalization processing divides the power consumption difference by the average power consumption of the two nodes to obtain the relative difference percentage, and the normalization formula is ΔP_norm = 2 x ΔP / (P_i + P_j) x 100%. The difference threshold is determined according to the normal difference distribution within the same feature cluster, and the upper limit of the 95% confidence interval is taken as the abnormal judgment threshold. The adjacent expansion considers the secondary adjacent relationship, i.e. the adjacent nodes of the adjacent nodes, to form a local neighborhood network. The difference feature vector construction contains statistical features such as [maximum difference, average difference, difference standard deviation, and number of threshold values]. The difference time series analysis records the trend of power consumption difference over time, with a sampling interval of 1 minute, forming a difference time series.

[0041] Anomaly observation points are identified within the clustering radius according to the power consumption difference. The clustering radius is defined as the Euclidean distance threshold in the feature space, and the initial radius is set to 1.5 times the average distance within the cluster. The anomaly determination criteria include the power consumption difference exceeding the threshold and the duration exceeding 3 minutes, and the distance of the node deviating from the center of the feature cluster to which it belongs exceeds the clustering radius. The anomaly determination simultaneously refers to the difference value distribution of the node and multiple adjacent nodes in the difference matrix D[i,j] and the fluctuation characteristics in the difference feature vector, and the simultaneous occurrence of multi-dimensional anomalies improves the confirmation degree. The feature space mapping maps the multi-dimensional feature vector of the node to the principal component space, and the PCA dimension reduction to 3 dimensions facilitates visualization and distance calculation. The node anomaly evaluation increases the comparative analysis with the normal nodes in the same cluster, and the large deviation indicates the increase of the abnormal possibility. The anomaly degree quantification is calculated by the anomaly score S = w1 x AP norm + w2 x d cluster + w3 x t duration, wherein w1, w2, w3 are weight coefficients, d cluster is the distance to the cluster center, and t duration is the abnormal duration. The weight coefficients are dynamically adjusted according to the multi-dimensional anomaly detection results. The observation point marking marks the nodes with anomaly scores exceeding the set threshold as anomaly observation points, and the threshold is determined according to the anomaly distribution of historical data. The spatial aggregation analysis checks the distribution pattern of the anomaly observation points in the physical space, and the aggregation of adjacent anomaly points indicates that there may be associated faults. The time correlation analysis of the anomaly observation points analyzes the time sequence of the anomaly observation points, and the anomaly points with a time interval less than 10 minutes may have a causal relationship. The anomaly type classification classifies the observation points into three categories: sudden anomaly, gradual anomaly and periodic anomaly, and the classification basis is the time variation mode of the anomaly score, and a type label is added to each observation point.

[0042] The abnormal observation points form an abnormal link. The link construction adopts a minimum spanning tree algorithm, taking the abnormal observation points as the vertices of the graph, and the edge weight is defined as the weighted sum of the electrical distance and the time distance between nodes. The electrical distance calculation considers the power supply topology, and the electrical distance between nodes on the same branch is smaller, and the electrical distance between nodes across the branch is larger. The time distance is defined as the difference between the abnormal occurrence time, and a small time distance indicates a high possibility of abnormal propagation. The weight formula is W_edge=α×D_elec+β×D_time, where α and β are weight coefficients, and the values are 0.6 and 0.4 respectively, D_elec is the electrical distance, and D_time is the time distance. The connectivity constraint requires that the adjacent nodes in the link must be reachable on the power supply topology and cannot cross the electrically isolated area. Link optimization removes redundant connections through a pruning algorithm to retain the key links that best reflect the abnormal propagation path. The link direction is determined based on the time sequence of abnormal occurrence, from the node with early abnormal occurrence to the node with late abnormal occurrence. The link strength is evaluated by the average abnormal score and the abnormal duration of the nodes on the link, and the link with high strength represents the main path of abnormal propagation. When there are multiple independent abnormal links, the links are processed and labeled separately.

[0043] In some embodiments, the generating a fault diffusion chain along the abnormal link includes: identifying a node with the maximum abnormal strength from the abnormal link as a suspected source; tracing abnormal attenuation paths on both sides from the suspected source as a starting point; marking abnormal secondary enhancement points on the abnormal attenuation paths; and connecting the suspected source and the abnormal secondary enhancement points to form a fault diffusion chain.

[0044] A node with the maximum abnormal strength is identified from the abnormal link as a suspected source. The abnormal strength comprehensive evaluation includes three dimensions of abnormal score, power consumption deviation amplitude, and abnormal duration, and the strength calculation formula is I=S×ΔP_max×log(t+1), where I is the abnormal strength, S is the abnormal score, ΔP_max is the maximum power consumption deviation, and t is the duration (minutes). Global search traverses all nodes on the abnormal link to calculate the abnormal strength value of each node. Local extreme value processing identifies the local maximum value point of the abnormal strength, and when there are multiple local extreme values, the global maximum value is selected as the main suspected source. Source verification confirms the possibility of the node as a fault source by analyzing its historical fault records and maintenance logs. The time priority principle selects the node that first appears abnormal as the suspected source when the abnormal strengths are similar. Spatial rationality check checks the location of the suspected source in the power supply topology, and the fault source is usually located upstream or at a key node position. Source feature analysis extracts the electrical characteristics of the suspected source, including abnormal patterns of voltage fluctuation, current harmonic, power factor, and other parameters. When multiple independent suspected sources are identified, each diffusion path is labeled and analyzed independently.

[0045] The suspected source is taken as the starting point to trace the abnormal attenuation path along both sides. The breadth-first search algorithm is used to expand to the adjacent nodes layer by layer from the suspected source. The attenuation model is established based on the attenuation law of abnormal intensity with propagation distance, and the attenuation function is I(d) = I_0 x exp(-λ x d), where I_0 is the source abnormal intensity, d is the propagation distance, and λ is the attenuation coefficient. Bidirectional tracing traces in both upstream and downstream directions of power supply. Upstream tracing may find deeper fault causes, and downstream tracing reflects the scope of fault influence. Path node screening only retains nodes with abnormal intensity exceeding the background noise level, which is determined by the intensity fluctuation during normal operation. Propagation speed estimation calculates the fault propagation speed by the time difference of abnormality occurrence of adjacent nodes, and the speed v = Δd / Δt, where Δd is the distance between nodes, and Δt is the time difference. Path branch processing records the abnormal propagation of all branches at the power supply branch point to determine the main propagation path and secondary affected branches. Attenuation anomaly detection identifies nodes that do not conform to the normal attenuation law, which may have independent faults or amplification effects. Path integrity ensures that the tracing path is continuous on the power supply topology without jumps or breaks.

[0046] Abnormal secondary enhancement points are marked on the abnormal attenuation path. The secondary enhancement criterion is that the abnormal intensity of a node is higher than that of its upstream adjacent node, which is manifested as a local peak on the attenuation curve. The enhancement amplitude is calculated as ΔI = I_node - I_pred, where I_node is the abnormal intensity of the current node, and I_pred is the predicted intensity based on the attenuation model. The significance test requires that the enhancement amplitude exceeds 20% of the predicted intensity and is statistically significant, and the t-test is used to verify the significance. Enhancement point feature analysis identifies the possible causes of secondary enhancement, including local fault superposition, load characteristic amplification, and power supply quality deterioration. Spatial distribution statistics the distribution pattern of secondary enhancement points in the power supply network, and concentrated distribution may indicate systemic problems. Time delay analysis calculates the time delay of secondary enhancement relative to the upstream anomaly, and the delay pattern reflects the fault propagation mechanism. Enhancement type classification includes resonance enhancement, overload enhancement, and cascade enhancement, and the classification is based on the variation characteristics of electrical parameters. Correlation analysis studies the mutual relationship between secondary enhancement points to identify possible common triggering factors.

[0047] For example, the connection between the suspected source and the abnormal secondary enhancement point forms a fault diffusion chain, including: based on the suspected source and the abnormal secondary enhancement point, evaluating the connection complexity to determine the path strategy, the connection complexity includes node spacing, power supply topology and load distribution; according to the path strategy, planning the propagation path node; using the propagation path node to reconstruct the fault propagation process to generate the fault diffusion chain.

[0048] The connection complexity is evaluated based on the suspected source and abnormal secondary reinforcement points to determine the path strategy. The node distance calculation includes two dimensions of physical distance and electrical distance, the physical distance reflects the spatial layout, and the electrical distance reflects the power supply path length. The power supply topology analysis extracts all possible paths between the source and the reinforcement point, considering the main power supply path and the backup path. The load distribution evaluates the load type and capacity of each node on the path, and the high-power load node may change the fault propagation characteristics. The complexity quantification adopts the formula C=w1×D_avg+w2×N_branch+w3×L_var, where D_avg is the average node distance, N_branch is the branch number, L_var is the load distribution variance, and w1, w2, w3 are weight coefficients. The path strategy classification selects different strategies according to the complexity level: low complexity adopts direct connection strategy, medium complexity adopts key node connection strategy, and high complexity adopts segmented progressive strategy. The strategy parameter setting includes search depth, node screening threshold, path scoring standard, etc., which is dynamically adjusted according to the complexity level.

[0049] The propagation path nodes are planned according to the path strategy. According to the selected strategy, the shortest path algorithm is used for direct connection strategy, the key node strategy identifies the necessary nodes, and the segmented progressive strategy decomposes the path into multiple sub-sections. The node importance score considers factors such as the power supply location, load capacity, and historical fault frequency of the node, and the nodes with high scores are preferentially included in the path. The path constraint conditions include the need to follow the power supply topology structure, the inability to contain the already shut down nodes, and the total length not exceeding a reasonable range. When there are multiple feasible paths, the TOP-3 candidate paths are generated, and the optimal path is selected through path scoring. The path scoring standard includes abnormal propagation probability, path length, and node importance sum, and the path with the highest comprehensive score is selected as the main path. The path node sequence is arranged in the order from the source to the reinforcement point, recording the position and role of each node in the path. The path segmentation sets intermediate checkpoints on long paths to facilitate the analysis of the phased characteristics of fault propagation.

[0050] The path reconstruction of the fault propagation process using the propagation path nodes generates the fault diffusion chain. The path nodes are rearranged in the time sequence of the fault occurrence, and the propagation process of the fault from the source to each enhancement point is restored. The propagation model fitting fits the fault propagation model parameters based on the actual observation data, including the propagation speed, the attenuation coefficient, the enhancement factor, etc. The state transition analysis describes the transition process of each node from the normal state to the abnormal state, and constructs the state transition probability matrix. The fault diffusion chain structure includes the main chain and the branch chain, the main chain connects the source and the main enhancement point, and the branch chain represents the secondary influence path. The link attribute defines each link to contain [node address, abnormal intensity, occurrence time, propagation delay, state type] and other information. The propagation feature extraction includes the propagation speed variation, the intensity attenuation law, the enhancement point distribution, etc. The causal relationship verification verifies the causal relationship strength between the nodes on the chain through the Granger causality test and other methods. The diffusion chain visualization generates the spatiotemporal evolution diagram of the fault propagation, and intuitively displays the diffusion process of the fault from the source.

[0051] In step S130, the timestamp sequence is marked for the fault diffusion chain, the fault source is determined according to the timestamp sequence, the anisotropic expansion is performed from the fault source to delimit the influence domain, and the monitoring anchor points are generated on the boundary of the influence domain.

[0052] Specifically, the timestamp sequence is marked for the fault diffusion chain. The fault diffusion chain data is used to mark the accurate timestamp for each node abnormal event on the chain, and the timestamp accuracy reaches the millisecond level. The timestamp extraction starts from the power consumption mutation time of each node, and the first-order difference method is used for mutation detection. When the power consumption change rate exceeds the set threshold, the time is recorded. The multi-event processing records the first abnormal time as the main timestamp when multiple abnormal events occur in a single node, and the subsequent abnormalities as auxiliary timestamps. The time synchronization calibration ensures the clock synchronization of all nodes based on the GPS time system, and eliminates the timing errors caused by clock deviation. The timestamp sequence is constructed by arranging the timestamps of each node in the topological order of the diffusion chain to form an ordered time sequence. The time interval calculation calculates the timestamp difference of adjacent nodes, and the time interval reflects the speed characteristics of the fault propagation. The abnormal timing pattern recognition includes synchronous abnormality (multiple node timestamps are the same), cascading abnormality (timestamp increment), reverse abnormality (timestamp reverse order), etc. The time window analysis divides the continuous timestamp sequence into multiple time windows, and the window size is set according to the typical time of the fault propagation. The timing consistency test verifies the consistency of the timestamp sequence and the power supply topology, and the unreasonable timing relationship may indicate data errors or special fault patterns.

[0053] The time stamp sequence is used to determine the source of the fault. The primary principle of source determination is time priority, and the node with the earliest time stamp is most likely the source of the fault. The time sequence analysis algorithm compares the time stamps of each node to construct a time sequence priority matrix, and the matrix elements represent the time sequence relationship between nodes. The propagation path verification checks whether the fault propagation path starting from the candidate source is consistent with the observed time sequence, and if it is not, the candidate source is excluded. Multi-source identification determines whether it is a single-source or multi-source fault when there are multiple time-close early abnormal nodes. The source credibility evaluation considers factors such as time priority, propagation path rationality, and abnormal intensity to calculate the source credibility score. Reverse verification traces back from the end of the time sequence to verify whether the identified source can explain all downstream abnormalities. Source feature analysis extracts the electrical characteristic patterns of the fault source, including current waveform distortion, power mutation pattern, and harmonic characteristics. The source confirmation mechanism confirms the fault source when multiple criteria point to the same node, and initiates an artificial review process when there is a conflict.

[0054] In some embodiments, the anisotropic expansion from the fault source to delineate the impact domain includes: injecting a characteristic marker current at the fault source; detecting the occurrence delay of the characteristic marker current from the power consumption profile to form a delay distribution; connecting positions with the same delay to form an isochronous circle based on the delay distribution; and taking the outermost layer of the isochronous circle as the impact domain.

[0055] A characteristic marker current is injected at the fault source. An AC signal of a specific frequency is superimposed on the working current, and the marker frequency is selected as 317 Hz to avoid harmonic interference at the power frequency. The amplitude of the injected signal is controlled within 5% of the rated current to ensure that it does not affect normal lighting functions. The signal generator uses DDS technology to generate high-precision sine waves with a frequency stability better than 0.01%. The injection point is selected at the power input end of the fault source node, and the injection is coupled through a current transformer. The injection timing control uses pulse modulation, with an injection duration of 100 ms and an interval of 900 ms, forming a marker signal with a duty cycle of 10%. Injection power monitoring ensures that the injection process does not trigger overcurrent protection or affect power stability. The signal feature code is embedded in the marker current through FSK modulation. The injection synchronization mechanism is synchronized with the clock of the data acquisition system, and the injection starting time is recorded as the reference zero point for delay calculation. Multi-source injection processing uses different marker signals of different frequencies to distinguish different sources when multiple fault sources need to be analyzed simultaneously.

[0056] The time delay distribution is formed by detecting the occurrence of the signature current from the power consumption profile. The matched filter method is adopted, and the filter is designed to match the frequency characteristics of the 317 Hz signature signal. The signal extraction converts the time-domain power consumption data to the frequency domain through FFT transformation, and extracts the signal strength at the 317 Hz frequency point. The detection threshold is set to 3 times the noise floor, ensuring reliable detection while avoiding false alarms. The time delay is calculated as the time difference from the injection time to the first detection of the signature signal by each node, with a precision of 10 ms. The detection points cover all lighting nodes and key power supply nodes, forming a detection matrix for the entire network. The signal propagation path tracking records all intermediate nodes passed by the signature signal, and analyzes the signal attenuation and delay characteristics. Time delay compensation considers signal processing delay and device response time, and systematically compensates for the original time delay. Abnormal time delay identification detects time delay values that deviate significantly from the expected value, which may indicate transmission path abnormalities or device failures. Time delay distribution statistics form a time delay histogram, analyzing the concentration trend and dispersion degree of time delay.

[0057] The isochronal circle is formed by connecting the positions with the same time delay based on the time delay distribution. The isochronal definition allows nodes within a time delay deviation range of ±20 ms to be classified into the same time delay level. Time delay quantization discretizes continuous time delay values into multiple levels, with a quantization step of 50 ms. Considering the topology and load distribution characteristics of the power supply network, the time delay grows slowly along the main power supply direction, and grows quickly perpendicular to the power supply direction, forming an elliptical isochronal circle, which reflects the anisotropic characteristics of fault propagation. The anisotropy coefficient is determined according to the direction of the power supply backbone and the branch density, and the propagation speed in the main direction can reach 2-3 times that in the perpendicular direction. Spatial interpolation estimates the time delay values of positions that have not directly detected the signature signal through spatial interpolation of adjacent nodes. The isochronal line generation adopts the contour tracking algorithm, similar to the generation method of contour lines on a topographic map. The closed detection ensures that the isochronal line forms a closed curve, and the open endpoints are handled by the boundary conditions. The smoothing processing smoothes the generated isochronal line, eliminating the jagged boundaries caused by measurement noise. The multi-circle layer structure forms multiple concentric isochronal circles from the inside out, and the circle layer interval corresponds to a fixed time delay increment. The circle layer label labels the corresponding time delay value and the number of nodes contained in each isochronal circle. Topology consistency verification ensures that the distribution of isochronal circles conforms to the topology of the power supply network.

[0058] The outermost layer of the isochronous delay circle is taken as the influence domain. The farthest propagation boundary of the marker signal is determined by searching for the closed isochronous delay circle with the maximum time delay value. The boundary optimization performs morphological processing on the outermost isochronous delay circle to fill small gaps and smooth local concave-convex. The influence domain area is calculated by a polygon area algorithm to evaluate the scale of the fault influence. The boundary node identification extracts lighting nodes located on or adjacent to the boundary of the influence domain, which have special monitoring value. The buffer zone is set outside the boundary of the influence domain with a certain width to consider measurement error and safety margin. The influence domain partition divides the influence domain into multiple sub-regions according to the time delay level, and different regions adopt different monitoring strategies. The boundary feature analysis includes characteristic parameters such as regularity and symmetry of the boundary. The dynamic adjustment mechanism dynamically adjusts the boundary of the influence domain according to real-time monitoring data to adapt to the development and change of the fault. The boundary stability evaluation analyzes the trend of the influence domain boundary over time.

[0059] The monitoring anchor points are generated on the boundary of the influence domain. The anchor point arrangement strategy uniformly distributes monitoring anchor points on the boundary of the influence domain, and the spacing is adaptively adjusted according to the boundary curvature, with dense anchor points in areas with high curvature. The anchor point quantity optimization determines the minimum number of anchor points through coverage analysis to reduce deployment costs while ensuring monitoring effectiveness. The anchor point type classification includes main anchor points (deploying full-featured monitoring), auxiliary anchor points (deploying simplified monitoring), and virtual anchor points (through data calculation). The physical anchor point selection prioritizes node locations that are easy to install and maintain, such as power distribution boxes, lamp pole access points, etc. The anchor point function definition includes current monitoring, voltage monitoring, communication relay, data caching, etc. The monitoring parameter configuration configures different monitoring parameter sets according to the importance of the anchor point location, and the boundary key position is configured with a complete parameter set. The anchor point networking forms a monitoring network by connecting anchor points through wireless or power carrier methods, supporting data aggregation and collaborative analysis. The redundancy design deploys redundant anchor points in critical boundary areas to improve the reliability of the monitoring system. The anchor point addressing assigns a unique identification code to each anchor point, and the addressing rule contains regional information and function type.

[0060] In step S140, the light color parameter variation is collected through the monitoring anchor points, the decay trajectory is generated based on the light color parameter variation, the remaining life is predicted based on the decay trajectory, and the replacement priority list is generated according to the remaining life sorting.

[0061] Specifically, the light color parameter variation is collected by monitoring the anchor points. The light quality monitoring is performed by integrating the light color sensor module with the deployed monitoring anchor points. The sensor uses TCS3472 color sensor, which has RGB and clear channel detection capability. The luminous flux measurement uses a silicon photodiode with a cosine corrector, with a measurement range of 10-10000 lux, an accuracy of ±3%, and a response time of less than 1ms. The color temperature detection calculates the correlated color temperature (CCT) through RGB three-channel data. The calculation formula is based on the CIE1931 chromaticity coordinate system, and the color temperature range is 2700K-6500K. The color rendering index measurement uses a multi-spectral method to measure Ra value and R1-R15 components, with an accuracy of ±2 and a coverage of CRI70-95 range. The sampling strategy is set to combine timed sampling and event triggering. Under normal conditions, sampling is performed once an hour, and when rapid changes are detected, sampling is encrypted to every minute. The parameter variation calculation uses the relative change rate, ΔP=(P_current-P_initial) / P_initial×100%, where P represents luminous flux, color temperature or color rendering index. When multiple lamps are covered by a single anchor point, the time-sharing measurement strategy is used to realize single lamp parameter measurement by controlling the on-off state of each lamp. Environmental compensation considers the influence of environmental light, temperature and humidity on measurement, and establishes a compensation model to eliminate environmental factor interference. Data preprocessing includes steps such as outlier rejection, missing value interpolation and noise filtering to ensure the accuracy of the parameter variation.

[0062] In some embodiments, the generating the decay trajectory based on the light color parameter variation comprises: performing stage decomposition on the light color parameter variation to obtain an initial drift amount and a late degradation amount; performing trend correction on the late degradation amount using the initial drift amount to form a corrected variation amount; performing time series superposition on the corrected variation amount to obtain a cumulative decay amount; and performing trajectory fitting based on the cumulative decay amount to generate the decay trajectory.

[0063] The initial drift and late degradation are determined by the second derivative of the change rate. The initial drift is characterized by a rapid but gradually slowing parameter change, mainly caused by the initial aging of the LED chip and phosphor. The initial drift is extracted using an exponential fitting method, and the drift model is D_init=A×(1-exp(-t / τ)), where A is the drift amplitude and τ is the time constant. The late degradation is characterized by an approximately linear continuous decay, caused by irreversible material degradation. The late degradation is calculated from the cumulative change from the dividing point, and the degradation rate is obtained by linear regression. The decomposition algorithm uses least squares to fit both stages simultaneously, ensuring continuity at the dividing point. Time scale analysis shows that the initial drift typically occurs within the first 1000 hours, and the late degradation occurs throughout the remaining life. Parameter correlation analysis of the drift and degradation characteristics of different light color parameters establishes a correlation model between parameters. The decomposition accuracy is evaluated by fitting residuals and the coefficient of determination R², and R²>0.95 is considered effective decomposition. Abnormal decomposition handles data that do not conform to the two-stage model, considering multi-stage or other decomposition methods.

[0064] The corrected change is formed by trend correction of the late degradation using the initial drift. Based on the predictive effect of the initial drift on the late degradation, lamps with larger drifts typically degrade faster. The correction coefficient is established by analyzing the correlation between the initial drift and the late degradation rate based on a large amount of historical data, with a typical correlation coefficient of 0.6-0.8. The correction formula is C_corrected=C_late×(1+k×D_init / D_avg), where k is the correction intensity parameter and D_avg is the average drift. Nonlinear correction considers the nonlinear relationship between drift and degradation rate, using a polynomial or logarithmic function to improve the correction model. Parameter normalization normalizes the initial drift before correction, eliminating the effects of different parameter dimensions. Correction boundary limits avoid overcorrection, limiting the correction amplitude to within ±30% of the original value. Cross-validation verifies the generalization ability of the correction model using the leave-one-out method, optimizing the correction parameters. Time-varying correction considers the change of the correction coefficient over time, with a large correction intensity in the early stage and a gradual decrease in the late stage. Multi-parameter joint correction considers the drift of luminous flux, color temperature, and color rendering index for joint correction. Correction effect evaluation compares the prediction accuracy before and after correction, and effective correction should improve the life prediction accuracy by more than 10%.

[0065] The correction change amount is accumulated by time series superposition. The time series superposition gradually accumulates the correction change amount at each time point from the initial time, forming a monotonically increasing cumulative curve. The superposition algorithm uses the trapezoidal integration method, Cum(t) =∑[AC(i)+AC(i+1)] / 2×At, to ensure numerical stability. The time step is determined according to the sampling frequency, and the time step can be as small as 1 hour for dense sampling, and interpolation is used for sparse sampling. The cumulative reference is zero at full new state, and all decay amounts are negative, reflecting the continuous decline in performance. Multi-parameter accumulation accumulates luminous flux, color temperature, and color rendering index separately, maintaining the independence of each parameter. Normalization processing normalizes the cumulative amount of different parameters to the range of 0-100%, facilitating comprehensive evaluation. Abnormal accumulation detection identifies abnormal jumps or backtracking in the cumulative curve, which may be caused by measurement errors or maintenance operations. Smoothing processing uses moving average or spline interpolation to smooth the cumulative curve, eliminating local fluctuations. Cumulative rate analysis calculates the cumulative rate at different time periods to identify the acceleration period of decay. Threshold marking marks key threshold positions on the cumulative curve, such as 10%, 30%, and 50% decay points.

[0066] According to the cumulative decay amount, the trajectory fitting is implemented to generate the decay trajectory. The fitting model selects linear, exponential, power function, Weibull, etc. The model with the highest fitting degree is selected. The non-linear least squares method is used to estimate the model parameters, and the initial value is determined by grid search. Fitting constraints impose physical constraints, such as the decay trajectory must be monotonically increasing, the slope must be non-negative, etc. The segmented fitting strategy divides the cumulative curve into multiple intervals, and each interval uses the most suitable local model. The fitting quality evaluation calculates the mean square error MSE, the determination coefficient R², the Akaike information criterion AIC, etc. Trajectory extrapolation is based on the fitted model to extrapolate to future time, predicting the subsequent decay trend. The confidence band calculation is based on the fitting residual and parameter uncertainty to calculate the prediction confidence band, typically with 95% confidence level. Model updating uses the recursive least squares method to update the model parameters online as new data is added. Trajectory feature extraction includes decay rate, acceleration, inflection point position, etc. Multi-model integration uses model averaging or Bayesian model combination method to integrate the prediction results of multiple models, improving robustness.

[0067] The remaining life is predicted based on the decay trajectory. The end of life is defined as when the luminous flux decays to 70% of the initial value or the color temperature deviation exceeds 500K. The prediction model uses time series prediction methods, including ARIMA model, exponential smoothing method, machine learning regression, etc. Feature engineering extracts decay rate, acceleration, curvature, etc. as input variables of the prediction model. Historical data uses complete life data of the same type of lamps to train the prediction model to improve prediction accuracy. The prediction time window is set to the time span from the current time to the predicted end of life, with the unit being hours. Uncertainty quantification evaluates the probability distribution of the prediction results through Monte Carlo simulation, giving the expected value and standard deviation of the remaining life. Dynamic update updates the prediction model and remaining life estimate regularly as new data accumulates. Early warning triggers an alarm when the predicted remaining life is below a set threshold, which is set according to the maintenance cycle. Life influencing factor analysis analyzes the effects of factors such as operating temperature, switching frequency, and power quality on life, and establishes correction coefficients. Prediction accuracy evaluation compares with the actual failure time to calculate the prediction error and hit rate.

[0068] A replacement priority list is generated according to the remaining life ranking. The priority score takes into account factors such as remaining life, location importance, replacement cost, etc. The score S = w1 / L + w2 x I + w3 x C, where L is the remaining life, I is the importance coefficient, and C is the cost coefficient. The importance of location is defined according to the function of lighting area, and the importance coefficient of main channel and working area is high. Replacement cost considers factors such as lamp price, installation difficulty, downtime loss, etc. The cost coefficient of high-altitude operation and special environment is large. Batch replacement strategy When multiple lamps in adjacent areas are close to the end of life, batch replacement is recommended to reduce maintenance cost. Time window optimization arranges the replacement time in low-impact periods according to the maintenance plan and usage demand. Dynamic update of the list recalculates and sorts the list after each replacement is completed, maintaining the real-time nature of the list. Emergency level classification divides the list into three levels: emergency replacement (remaining life <100 hours), planned replacement (100-500 hours), and observation waiting (>500 hours). Spare parts management prepares spare parts in advance according to the replacement priority list to avoid temporary procurement delays. Replacement record tracking records the time, reason, new lamp information, etc. of each replacement, accumulating maintenance data. Economic analysis evaluates the cost-benefit of early replacement and replacement after failure to optimize the replacement timing.

[0069] In step S150, the starting current spike in the node feature cluster is extracted, the oscillation envelope of the starting current spike is obtained, the health degree of the driver is determined according to the oscillation envelope, and the health degree of the driver is mapped into a matrix coordinate to form a maintenance density field.

[0070] Specifically, the startup current spike in node feature cluster is extracted. With the node feature cluster data, the startup time of each lighting node is located, and the startup decision is based on the time when the power consumption jumps from zero to 10% of the rated value. The spike identification window is set to a 500ms time range after the startup time, which contains the complete startup transient process. The current sampling rate is raised to 20kHz to capture the rapidly changing spike details, and the sampling accuracy of 16 bits ensures the dynamic range. Spike feature extraction includes peak current, rise time, duration, peak factor, etc. The peak factor is defined as the ratio of peak current to steady-state current. Multiple startup statistics statistically analyze multiple startup events of the same node, calculate the mean and standard deviation of the spike features, and remove abnormal spikes caused by power grid disturbance or measurement error. The screening criteria are based on the 3σ principle. Spike waveform classification divides the startup current spike into single-peak, double-peak, and oscillation types according to waveform features. Time alignment uses the cross-correlation method to align the startup current waveforms of different nodes on the time axis, facilitating horizontal comparison. Spike energy calculation calculates the energy consumption during the spike period through the integral method, E=∫I²(t)Rdt, reflecting the startup impact strength. Temperature correlation analysis of the correlation between startup current spikes and environmental temperature, the spike is usually higher at low temperature.

[0071] In some embodiments, the startup current spike envelope includes: determining the envelope type based on the startup current spike identification oscillation feature, the oscillation feature including overshoot amplitude, decay rate and oscillation frequency; constructing an envelope extraction operator according to the envelope type; and using the envelope extraction operator to perform envelope tracking on the startup current spike to generate an oscillation envelope.

[0072] The envelope type is determined based on the oscillation characteristics identified from the inrush current spike. The overshoot amplitude is obtained by finding the difference between the maximum current value at the inrush moment and the steady-state current. The overshoot rate is defined as the percentage of the overshoot amplitude to the steady-state value. The decay rate analysis examines the current decrease process starting from the peak point by exponential fitting I(t) = I_peak x exp(-t / τ) + I_steady to extract the decay time constant τ. The oscillation frequency extraction uses the zero-crossing detection method to count the number of oscillation periods per unit time, supplemented by FFT spectrum analysis to confirm the dominant frequency. The oscillation mode classification divides the oscillation into three modes: under-damped (sustained oscillation), critically damped (fast convergence), and over-damped (slow convergence) based on three characteristic parameters. The characteristic parameter normalization normalizes the overshoot amplitude, decay rate, and oscillation frequency to the [0, 1] interval to eliminate the dimension effect. The feature space mapping plots a scatter plot in a three-dimensional feature space, and identifies typical oscillation modes through cluster analysis. The envelope type decision tree establishes decision rules based on characteristic parameters, such as frequency > 1 kHz and overshoot > 300% to determine high-frequency oscillation type. The mixed type processing uses fuzzy logic to determine the dominant type when the oscillation characteristics are between multiple types. The characteristic time-varying analysis examines the changes of oscillation characteristics during the startup process, which may present different types in the early and late stages.

[0073] The envelope extraction operator is constructed according to the envelope type. The operator library design predefines dedicated extraction operators for different envelope types, including Hilbert operator, morphological operator, spline operator, etc. The Hilbert operator is suitable for narrowband oscillation signals, and the envelope is obtained by analyzing the modulus of the signal, with a computational complexity of O(NlogN). The morphological operator uses the dilation operation of mathematical morphology, and the length of the structural element is adaptively adjusted according to the oscillation period. The spline operator constructs the envelope through spline interpolation of local extreme points, which is suitable for sparse sampling or irregular oscillation. The adaptive operator dynamically adjusts operator parameters such as window length and smoothing coefficient according to signal characteristics. The operator cascade uses multiple operators in cascade for complex oscillation, such as denoising first and then extracting the envelope. The parallel operator runs multiple operators in parallel for multi-modal oscillation and fuses the results. The operator optimization optimizes the operator parameters through genetic algorithm, with the objective function being the weighted sum of envelope smoothness and fitting degree. The calculation efficiency is realized by using fast algorithms, such as FFT to accelerate Hilbert transform and decomposition implementation of morphological operations.

[0074] The envelope extraction operator is used to track the envelope of the inrush current spike, generating an oscillatory envelope. Signal preprocessing is performed before envelope extraction, including denoising and baseline correction. Wavelet thresholding is used for denoising, and a polynomial fit is used to remove the baseline. The operator application calls the corresponding operator based on the determined envelope type, with the preprocessed current signal as input and the envelope curve as output. The tracking starting point is located at the maximum value of the spike, ensuring the capture of the complete oscillation decay process. Bidirectional tracking simultaneously tracks the envelope forward and backward, capturing the rising edge and decay process. Envelope connection connects the segmented envelopes, with weighted averaging used to ensure continuity. Boundary processing performs boundary extension at the start and end of the signal to avoid boundary effects affecting the envelope quality. Envelope verification checks whether the extracted envelope meets physical constraints, such as the envelope must surround the original signal and monotonic decay. Multi-resolution tracking uses a multi-scale method, with coarse scales capturing overall trends and fine scales capturing local features. The tracking termination condition is when the envelope amplitude decays to 105% of the steady-state value, avoiding overextension. The result output generates an envelope data sequence, including timestamps and corresponding upper and lower envelope values, with a sampling rate consistent with the original signal.

[0075] The driver health is determined based on the oscillatory envelope. The health score considers multiple dimensions such as envelope decay time, oscillation amplitude, and envelope regularity. The decay time score is full when the decay time τ < 50 ms, zero when τ > 200 ms, and linear interpolation in between. The oscillation amplitude score is based on the ratio of the initial amplitude to the steady-state value, with a ratio < 2 being excellent and > 5 being poor. The envelope regularity is evaluated by calculating the fractal dimension of the envelope, with regular envelopes having a fractal dimension close to 1. Frequency stability analyzes the time-varying characteristics of the oscillation frequency, with large frequency drift indicating unstable driver performance. Harmonic distortion evaluates the harmonic content of the oscillation waveform, with THD > 20% indicating severe nonlinearity of the driver. Health quantification uses a percentage system, with 100 points indicating complete health and scores below 60 requiring attention. Historical trend analysis tracks the change in health over time, with the rate of decline reflecting the degree of degradation. Fault mode identification identifies different envelope features corresponding to specific fault modes, such as capacitor aging and switch tube degradation. Health confidence interval calculates the confidence interval based on the dispersion of multiple measurements, evaluating the reliability of the diagnosis.

[0076] The driver health degree is mapped to a matrix coordinate to form a maintenance density field. The mapping strategy maps the physical layout of the lighting matrix to a two-dimensional coordinate system, with each node corresponding to a grid cell. The coordinate system is defined with the southwest corner of the lighting area as the origin, with the east-west direction as the X-axis and the north-south direction as the Y-axis. The grid resolution is determined according to the node density, with a typical value of 2m x 2m, ensuring that each grid contains at most one node. The health degree assignment assigns the driver health degree value of each node to the corresponding grid cell, and the grid with no node is assigned a null value. Spatial interpolation uses inverse distance weighting interpolation for null grids to generate a continuous health degree distribution field. The density calculation defines the maintenance demand density as (100-health degree) / 100, and the lower the health degree, the higher the maintenance demand. Hot spot identification identifies areas with concentrated maintenance demand through local density analysis, and areas with a density greater than 0.7 are considered maintenance hotspots. Density gradient calculation calculates the density difference between adjacent grids, and areas with large gradients indicate rapid changes in health degree. Field smoothing uses Gaussian filtering to smooth the density field, with a filter kernel size of 5x5, to eliminate local anomalies. Iso-density line generation generates iso-density lines similar to contour lines, which visually display the spatial distribution of maintenance demand.

[0077] Step S160, based on the fault diffusion chain, the replacement priority list and the maintenance density field, a three-dimensional detection field is constructed, and a singular point is found in the three-dimensional detection field, a fault positioning pulse is generated with the singular point as the center, and the fault lamp coordinates are analyzed through the fault positioning pulse, and the lighting state detection of the matrix layout is completed.

[0078] Specifically, based on the fault diffusion chain, the replacement priority list and the maintenance density field, a three-dimensional detection field is constructed. The three-dimensional coordinate system defines the physical layout of the lighting matrix as the XY plane, and the Z-axis represents the comprehensive detection intensity, with the coordinate origin set at the southwest corner of the matrix. Data normalization normalizes the value range of the three data sources to the [0, 1] interval, and the abnormal intensity of the fault diffusion chain, the urgency of the replacement priority, and the density value of the maintenance density field are normalized respectively. The field strength calculation formula is F(x, y, z) = α × S_fault(x, y) + β × P_replace(x, y) + γ × D_maint(x, y), where α, β, γ are weight coefficients, which are dynamically adjusted according to the detection focus, and the typical values are 0.4, 0.3, 0.3. The spatial resolution is set to a voxel grid of 1m x 1m x 0.1, ensuring that local anomalies can be captured. Data fusion uses a weighted superposition method, and the maximum value principle is used in the overlapping area to avoid information loss. The boundary condition is set to gradually change the field strength at the field boundary to zero, and a cosine window function is used to achieve smooth transition. Time stamp synchronization ensures the time consistency of the three data sources, and the latest data is used to construct a real-time detection field. Field strength gradient calculation calculates the partial derivatives in three directions ∇F=(∂F / ∂x,∂F / ∂y,∂F / ∂z), and the gradient size reflects the degree of change in the field. Abnormal enhancement enhances the field strength in the local anomaly area, and the enhancement coefficient is determined according to the severity of the anomaly.

[0079] In some embodiments, the finding singular points in the three-dimensional detection field comprises: constructing a spiral scanning path in the three-dimensional detection field; recording field strength along the spiral scanning path to form a field strength change sequence; performing autocorrelation analysis on the field strength change sequence to generate periodic missing points; and marking the periodic missing points as singular points.

[0080] A spiral scanning path is constructed in the three-dimensional detection field. Spiral parameter settings include a starting radius r0=1m, a pitch h=0.5m, and the number of spiral turns is determined according to the size of the detection field. The spiral equation adopts a parameterized form, x=r(θ)cosθ, y=r(θ)sinθ, z=hθ / 2π, where r(θ)=r0+aθ is a variable radius spiral. The scanning direction expands from the center outward, ensuring coverage of the entire detection field space and avoiding scanning blind areas. Path discretization discretizes the continuous spiral into a sequence of sampling points, with a sampling interval of 0.1m to ensure spatial resolution. Boundary processing uses reflection or truncation strategies to continue scanning when the spiral exceeds the detection field boundary. Path optimization adjusts the spiral parameters to achieve an optimal balance between path length and sampling point distribution. Multi-start strategy sets multiple spiral starting points in large detection fields and parallel scanning to improve efficiency. Path visualization generates a three-dimensional graph of the spiral path, visually displaying the scanning coverage. Adaptive adjustment dynamically adjusts the spiral density according to the complexity of the field strength distribution, with increased scanning in complex areas.

[0081] Field strength is recorded along the spiral scanning path to form a field strength change sequence. Field strength sampling extracts the field strength value F(x,y,z) of the three-dimensional detection field at each path point, forming a one-dimensional time sequence. Sampling synchronization ensures uniform scanning along the path at a speed of v=1m / s and a sampling frequency of f_s=10Hz. The data recording format is [index, position coordinates, field strength value, timestamp], which records the scanning process completely. Dynamic range compression performs logarithmic transformation on the field strength value to compress the dynamic range, facilitating detection of weak changes. Moving average uses a window length of 5 for moving average to smooth local noise and preserve the main trend of change. Outlier processing detects and marks outliers that exceed 3σ to prevent interference with subsequent analysis. Sequence segmentation divides the sequence into stable segments, gradual change segments, and abrupt change segments according to the field strength change characteristics. Difference sequence calculation calculates the first-order difference sequence ΔF[i]=F[i+1]-F[i] to highlight the change characteristics. Normalization processing normalizes the sequence to a standard form with a mean of 0 and a variance of 1. The sequence length is determined according to the total length of the spiral path, and a typical sequence contains thousands to tens of thousands of data points.

[0082] Periodicity detection finds local peaks in the autocorrelation function, and the interval between peaks corresponds to the period of the sequence. Periodicity strength evaluates the height of peaks to reflect the strength of periodicity, and a height greater than 0.5 indicates significant periodicity. Multi-period identification identifies multiple periodic components by spectral analysis, including the main period and harmonic periods. Periodicity missing detects missing peaks at the expected positions of periodic peaks, and marks them as missing. Missing pattern analysis analyzes the regularity of missing, and consecutive missing may indicate a large-scale anomaly. Phase analysis calculates the phase relationship of different periodic components, and phase mutations may exist at singular points. Window autocorrelation calculates local autocorrelation using a sliding window to capture the time-varying characteristics of periodicity.

[0083] Periodicity missing points are marked as singular points. Missing point positioning determines the position index of periodic missing in the original sequence based on the autocorrelation analysis results. Coordinate inversion finds the corresponding three-dimensional space coordinates (x, y, z) by path index. Missing degree quantification defines the missing degree as the difference between the expected peak and the actual value, and a large missing degree indicates strong singularity. Spatial clustering analysis analyzes the distribution of adjacent missing points in space, and clustered missing points form singular regions. Singular type inference infers the type of singular points based on the missing pattern, and single-point missing corresponds to point singularities, and consecutive missing corresponds to line singularities. Confidence evaluation calculates the confidence of singular points based on the strength of periodicity and the degree of missing. Multi-scale verification verifies the stability of singular points at different scanning densities, and stable appearance is a true singular point. Singular point attribute records complete information such as position coordinates, singularity strength, type, confidence, etc. Priority sorting sorts singular points according to singularity strength and confidence, and high-priority singular points are processed first. Result output generates a list of singular points, including all detected singular points and their attribute information.

[0084] The fault location pulse is generated with the singularity point as the center. The pulse is designed in the form of Gaussian pulse, and the pulse width is set according to the positioning accuracy requirement, and the typical value is 100 μs. The pulse amplitude is adaptively adjusted according to the singularity point strength, and the strong singularity point corresponds to the high amplitude pulse, which ensures the detection sensitivity. The center frequency is selected to avoid the 317 Hz of the power grid harmonic, which is consistent with the marker frequency of S130, so as to facilitate signal identification. The pulse code is embedded in the pulse. The spatial coordinate information of the singularity point is embedded in the pulse, and the Manchester code is used, and the code rate is 1 kbps. When there are multiple singularity points, a pulse sequence is generated, and the pulse interval is 500 ms to avoid signal aliasing. The transmission power control adjusts the transmission power according to the transmission distance and line attenuation, and ensures that the signal-to-noise ratio at the receiving end is greater than 20 dB. The pulse polarity is determined according to the singularity point type, and the source point uses a positive pulse, the sink point uses a negative pulse, and the saddle point uses a bipolar pulse. Each pulse carries an accurate time stamp. The pulse envelope adopts a raised cosine envelope to reduce spectrum leakage and improve spectrum utilization. The synchronization header is designed to add a synchronization header before the pulse sequence, which facilitates frame synchronization and decoding at the receiving end.

[0085] In some embodiments, the fault lamp coordinates are resolved by the fault location pulse, including: encoding the fault location pulse into a row detection pulse and a column detection pulse; sending the row detection pulse to the lighting matrix and identifying the abnormal attenuation row from the echo intensity; sending the column detection pulse to the abnormal attenuation row to detect the echo missing position; and determining the fault lamp coordinates according to the abnormal attenuation row and the echo missing position.

[0086] The fault location pulse is encoded into a row detection pulse and a column detection pulse. Orthogonal encoding is used, and the row pulse and the column pulse use different frequency or phase characteristics. The row pulse design uses a sweep frequency pulse, with a starting frequency of 300 Hz, an ending frequency of 400 Hz, and a sweep time of 10 ms. The column pulse design uses a fixed frequency pulse string, with a frequency of 350 Hz and a pulse number representing the column address. The address code encodes the row address through the sweep slope, and encodes the column address through the pulse number, supporting a 256x256 matrix. The pulse interval is 100 ms for the row pulse and 50 ms for the column pulse, which ensures correct decoding at the receiving end. The amplitude modulation adaptively adjusts the pulse amplitude according to the transmission distance, and the far-end node uses a high-amplitude pulse. The preamble design adds a fixed preamble before the address code, which is used for synchronization and identification of pulse type. The check code adds a CRC-8 check code to detect transmission errors and improve address decoding reliability. The pulse shaping uses a raised cosine pulse shaping to reduce inter-symbol interference and improve transmission quality. The coding efficiency is improved by compression coding, and adjacent addresses are differentially encoded. Multiplexing supports simultaneous transmission of multiple pulses, and frequency or code division multiplexing is used to improve detection speed.

[0087] The row detection pulse is sent to the lighting matrix, and the abnormal attenuation row is identified from the echo intensity. The pulse injection injects a row detection pulse at the row power line inlet of the lighting matrix, and impedance matching is achieved through a coupling transformer. Propagation monitoring synchronously collects pulse propagation signals at multiple monitoring points, which are distributed at different positions of the row. Echo acquisition collects reflected echoes at the injection point, with a sampling rate of 20 kHz and a collection time length covering the round-trip time of the farthest node. Echo separation separates the echoes of different nodes by matched filtering, and the filter is matched with the transmitted pulse. Intensity measurement calculates the peak intensity, energy and waveform distortion degree of each echo. Normal echo modeling establishes a normal echo model based on the echo characteristics of the fault-free row, including intensity distribution and time delay regularity. Abnormal detection compares the measured echo with the normal model, and the deviation exceeding the threshold is judged as abnormal. Attenuation mode analysis analyzes the spatial distribution of abnormal attenuation. Uniform attenuation may be a line problem, and local attenuation indicates node failure. Row positioning confirms the row number of the abnormal attenuation row through multiple features. The result record records the abnormal row list and its attenuation characteristics, providing targets for column scanning.

[0088] The column detection pulse is sent to the abnormal attenuation row to detect the missing position of the echo. The column scanning strategy only scans the column of the identified abnormal attenuation row, reducing the detection time. Pulse sequence generation generates a complete column address pulse sequence according to the number of matrix columns. Column-by-column transmission sends detection pulses in column address order, and waits for an echo time window after each transmission. Echo detection window calculates the expected echo arrival time according to the column position, and sets the detection window. Missing judgment determines that the echo is missing if no valid echo is detected in the detection window. Multiple confirmations detect suspected missing positions multiple times to reduce the false positive rate. Adjacent column reference refers to the echo characteristics of adjacent columns to exclude false missing caused by abnormal propagation path. Missing mode identification identifies single-point missing corresponding to lamp failure, and continuous missing may be a line break. Time delay compensation compensates for the time delay according to the propagation distance of different columns to ensure the accuracy of the detection window. Column coordinate determination converts the column number of the missing position to the actual column coordinates.

[0089] The abnormal attenuation row and the echo missing position are used to determine the fault lamp coordinates. Coordinate synthesis combines the abnormal row number and the missing column number to form a two-dimensional coordinate (row, col). Coordinate verification checks the validity of the synthesized coordinates to ensure that they are within the matrix range and correspond to the actual lamp positions. Physical mapping converts the matrix coordinates to physical space coordinates, taking into account the lamp spacing and arrangement. Fault type refinement subdivides the fault type according to the echo characteristics, with complete missing being an open circuit and severe attenuation being a short circuit. When multiple fault points are detected, a fault point list is established and priority is marked. Adjacent correlation analyzes the spatial correlation of fault points, and adjacent faults may have common causes. Coordinate accuracy evaluation evaluates the coordinate positioning accuracy through field verification, with a typical error of less than one lamp spacing. The result output generates a fault lamp coordinate table containing coordinates, fault type, detection confidence, etc. The positioning completion flag is set when all abnormal attenuation rows complete column scanning. The state update updates the detection results to the lighting management system, triggering the corresponding maintenance process. Through the aforementioned layer-by-layer analysis and accurate positioning, the lighting state detection of the matrix layout is completed.

[0090] In order to perform a lighting state detection method for a matrix layout corresponding to the above-mentioned method embodiment, to realize the corresponding functions and technical effects. Referring to Figure 2 , Figure 2 The structure block diagram of a lighting state detection device 200 for a matrix layout provided by an embodiment of the present application is shown. For ease of illustration, only the parts related to the present embodiment are shown. The lighting state detection device 200 for a matrix layout provided by an embodiment of the present application comprises:

[0091] The signal acquisition module 201 is configured to acquire the power bus current waveform and single lamp feedback signal of the lighting matrix, perform waveform analysis on the power bus current waveform to generate a power consumption profile, identify an address code from the single lamp feedback signal, and divide the power consumption profile according to the address code to form a node feature cluster.

[0092] The abnormal tracking module 202 is configured to extract the power consumption difference of adjacent lamps based on the node feature cluster, identify abnormal observation points within the clustering radius according to the power consumption difference, connect the abnormal observation points to form an abnormal link, and generate a fault diffusion chain by backtracking along the abnormal link.

[0093] The influence analysis module 203 is configured to mark a timestamp sequence on the fault diffusion chain, determine the fault source according to the timestamp sequence, perform anisotropic expansion from the fault source to delimit an influence domain, and generate a monitoring anchor point at the boundary of the influence domain.

[0094] The life prediction module 204 is configured to collect a light color parameter variation amount through the monitoring anchor point, generate a decay track based on the light color parameter variation amount, predict a remaining life based on the decay track, and generate a replacement priority list according to a ranking of the remaining life.

[0095] The health assessment module 205 is configured to extract a starting current spike in the node feature cluster, obtain an oscillation envelope of the starting current spike, determine a driver health degree according to the oscillation envelope, and perform matrix coordinate mapping on the driver health degree to form a maintenance density field.

[0096] The fault positioning module 206 is configured to construct a three-dimensional detection field based on the fault diffusion chain, the replacement priority list and the maintenance density field, find a singular point in the three-dimensional detection field, generate a fault positioning pulse with the singular point as a center, and analyze a fault lamp coordinate through the fault positioning pulse to complete the matrix layout lighting state detection.

[0097] The matrix layout lighting state detection device 200 described above can implement a matrix layout lighting state detection method according to the method embodiment described above. The optional items in the method embodiment described above are also applicable to the present embodiment, and will not be described in detail herein. The remaining content of the present embodiment can be referred to the content of the method embodiment described above, and will not be described in detail herein.

[0098] As shown in Figure 3 The third embodiment of the present application further provides a computer device, which comprises a memory 301, a processor 302, and a computer program stored in the memory 301 and capable of running on the processor 302, and characterized in that the processor 302 implements the steps of the matrix layout lighting state detection method according to the first embodiment of the present application when executing the program.

[0099] The purpose of the above embodiments is to exemplarily reproduce and deduce the technical solutions of the present application, and to completely describe the technical solutions, purposes and effects of the present application, so as to make the public more thoroughly and comprehensively understand the disclosed content of the present application, and not to limit the protection scope of the present application.

[0100] The above embodiments are not an exhaustive enumeration based on the present application, and there can be a plurality of other embodiments not listed in addition thereto. Any replacement and improvement made without violating the concept of the present application is within the protection scope of the present application.

Claims

1. A method of detecting lighting states of a matrix layout, characterized by, The method comprises the following steps: Collecting power supply bus current waveforms and single lamp feedback signals of a lighting matrix, performing waveform analysis on the power supply bus current waveforms to generate a power consumption profile, identifying an address code from the single lamp feedback signals, and dividing the power consumption profile according to the address code to form a node feature cluster; Extracting power consumption difference values of adjacent lamps based on the node feature cluster, identifying abnormal observation points within a clustering radius according to the power consumption difference values, connecting the abnormal observation points to form an abnormal link, and generating a fault diffusion chain by backtracking along the abnormal link; Labeling a time stamp sequence for the fault diffusion chain, judging a fault source according to the time stamp sequence, performing anisotropic expansion from the fault source to demarcate an influence domain, and generating a monitoring anchor point at the boundary of the influence domain; Collecting light color parameter change amounts through the monitoring anchor point, generating a decay trajectory based on the light color parameter change amounts, predicting a remaining life based on the decay trajectory, and generating a replacement priority list according to the remaining life sorting; Extracting a starting current spike in the node feature cluster, obtaining an oscillation envelope of the starting current spike, judging a driver health degree according to the oscillation envelope, and performing matrix coordinate mapping on the driver health degree to form a maintenance density field; Constructing a three-dimensional detection field based on the fault diffusion chain, the replacement priority list, and the maintenance density field, finding a singular point in the three-dimensional detection field, generating a fault positioning pulse with the singular point as the center, analyzing a fault lamp coordinate through the fault positioning pulse, and completing lighting state detection of a matrix layout.

2. The method of claim 1, wherein, The method of generating a fault diffusion chain by backtracking along the abnormal link comprises the following steps: Identifying a node with the maximum abnormal intensity in the abnormal link as a suspected source; Tracking an abnormal attenuation path from both sides with the suspected source as the starting point; Labeling an abnormal secondary enhancement point on the abnormal attenuation path; Connecting the suspected source and the abnormal secondary enhancement point to form a fault diffusion chain.

3. The method of claim 1, wherein, The method of performing anisotropic expansion from the fault source to demarcate an influence domain comprises the following steps: Injecting a feature marker current at the fault source; Detecting the appearance time delay of the feature marker current from the power consumption profile to form a time delay distribution; Connecting positions with the same time delay to form an isochronous circle based on the time delay distribution; Taking the outermost layer of the isochronous circle as the influence domain.

4. The method of claim 1, wherein, The method of generating a decay trajectory based on the light color parameter change amounts comprises the following steps: Performing stage decomposition on the light color parameter change amounts to obtain an initial drift amount and a late deterioration amount; Performing trend correction on the late deterioration amount using the initial drift amount to form a corrected change amount; Generating a cumulative decay variable by time series superposition on the corrected change amount; Implementing trajectory fitting to generate a decay trajectory according to the cumulative decay variable.

5. The method of claim 1, wherein, The method of obtaining an oscillation envelope of the starting current spike comprises the following steps: Identifying an oscillation feature to determine an envelope type based on the starting current spike, wherein the oscillation feature includes overshoot amplitude, decay rate, and oscillation frequency; Constructing an envelope extraction operator according to the envelope type; Performing envelope tracking on the starting current spike using the envelope extraction operator to obtain an oscillation envelope.

6. The method of claim 1, wherein, The method of finding a singular point in the three-dimensional detection field comprises the following steps: Constructing a spiral scanning path in the three-dimensional detection field; record field strength along the spiral scanning path to form a field strength variation sequence; perform autocorrelation analysis on the field strength variation sequence to generate a periodic missing point; mark the periodic missing point as a singular point.

7. The method of claim 1, wherein, the fault positioning pulse includes: encoding the fault positioning pulse into a row detection pulse and a column detection pulse; sending the row detection pulse to the lighting matrix and identifying an abnormally attenuated row from echo intensity; sending the column detection pulse to the abnormally attenuated row to detect a missing echo position; determining a fault lamp coordinate according to the abnormally attenuated row and the missing echo position.

8. The method of claim 2, wherein, the connection of the suspected source and the abnormal secondary enhancement point forms a fault diffusion chain, including: based on the suspected source and the abnormal secondary enhancement point, evaluating connection complexity to determine a path strategy, the connection complexity including node spacing, power supply topology, and load distribution; planning a propagation path node according to the path strategy; using the propagation path node to reconstruct a fault propagation process to generate a fault diffusion chain.

9. A matrix-arranged lighting state detection device, characterized by including: a signal acquisition module, configured to acquire a power bus current waveform of a lighting matrix and a single lamp feedback signal, perform waveform analysis on the power bus current waveform to generate a power consumption profile, identify an address code from the single lamp feedback signal, and divide the power consumption profile according to the address code to form a node feature cluster; an anomaly tracking module, configured to extract a power consumption difference value of adjacent lamps based on the node feature cluster, identify an abnormal observation point within a clustering radius according to the power consumption difference value, connect the abnormal observation points to form an abnormal link, and backtrack along the abnormal link to generate a fault diffusion chain; an influence analysis module, configured to mark a timestamp sequence on the fault diffusion chain, determine a fault source according to the timestamp sequence, perform anisotropic expansion from the fault source to delimit an influence domain, and generate a monitoring anchor point at the boundary of the influence domain; a life prediction module, configured to acquire a light color parameter change amount through the monitoring anchor point, generate a decay trajectory based on the light color parameter change amount, predict a remaining life based on the decay trajectory, and generate a replacement priority list according to the remaining life order; a health evaluation module, configured to extract a starting current spike in the node feature cluster, obtain an oscillation envelope of the starting current spike, determine a driver health degree according to the oscillation envelope, and perform matrix coordinate mapping on the driver health degree to form a maintenance density field; a fault positioning module, configured to construct a three-dimensional detection field based on the fault diffusion chain, the replacement priority list, and the maintenance density field, find a singular point in the three-dimensional detection field, generate a fault positioning pulse centered on the singular point, resolve a fault lamp coordinate through the fault positioning pulse, and complete lighting state detection of a matrix layout.

10. A computer device, comprising: comprising a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and realize the method of any one of claims 1 to 8 when executing the computer program.

Images