A smart lighting network state monitoring and fault self-diagnosis system

Abstract

The present application relates to the technical field of intelligent lighting network monitoring and fault diagnosis, and provides an intelligent lighting network state monitoring and fault self-diagnosis system. The system acquires electrical operation parameters, communication link quality parameters and gateway operation state data of lighting nodes through a data acquisition module, and an abnormal trigger module detects parameter changes based on a sliding time window and generates a diagnosis trigger signal. An active detection module performs micro-amplitude dimming detection and multi-path heartbeat detection on target nodes to obtain operation response data. A comparison and analysis module selects comparison nodes to generate a difference feature vector. A cross-layer root cause reasoning module fuses rule reasoning and Bayesian network calculation results to output fault root cause categories and confidence. An action arrangement module executes corresponding treatment strategies, and updates rule parameters and conditional probability tables through a feedback optimization module to form a closed-loop self-learning mechanism. The present application improves fault positioning accuracy and diagnosis reliability.

Description

Technical Field

[0001] This invention relates to the field of smart lighting network operation and maintenance and monitoring technology, specifically to a smart lighting network status monitoring and fault self-diagnosis system. Background Technology

[0002] Smart lighting networks, as an important application of IoT technology in urban infrastructure, connect a large number of lighting nodes to a unified management platform through wireless communication protocols, enabling centralized monitoring, remote dimming, and energy consumption management. With the expansion of network scale, the increase in the number of nodes, and the increasing complexity of the operating environment, network operation status monitoring and fault diagnosis capabilities have become key technical aspects for ensuring system stability and service continuity.

[0003] Existing smart lighting network status monitoring and fault diagnosis technologies generally suffer from the following technical deficiencies:

[0004] I. The monitoring method is mainly passive alarm, lacking an active detection mechanism. Some systems typically use passive monitoring, judging anomalies by periodically collecting electrical parameters such as voltage, current, and power reported by nodes, as well as communication indicators such as packet loss rate and signal strength. This method relies on nodes actively reporting data and can only trigger alarms when operating parameters significantly deviate from thresholds.

[0005] When voltage fluctuations are slight, power supply performance deteriorates, or communication link quality gradually degrades in its early stages, the changes in operating parameters are small and have not yet reached the alarm threshold. Passive monitoring methods are unable to detect potential faults in a timely manner, which can easily lead to the gradual accumulation of hidden problems and eventually evolve into sudden failures, affecting the continuity of lighting services.

[0006] Second, diagnostic analysis is limited to single-layer data and lacks cross-layer collaborative judgment capabilities. Traditional fault diagnosis methods often perform independent analysis on the electrical layer and the communication layer separately. The electrical layer focuses on threshold comparisons of parameters such as voltage, current, and power, while the communication layer focuses on statistical evaluation of network quality indicators such as packet loss rate, round-trip time, and signal strength.

[0007] This single-layer diagnostic approach ignores the coupling relationships between different layers. For example, unstable power supply may cause frequent node restarts, resulting in communication interruptions; communication link degradation may also cause dimming commands to fail to be executed in a timely manner, resulting in abnormal power. Without cross-layer data fusion and comprehensive reasoning mechanisms, the system is prone to misjudgments or inaccurate root cause localization.

[0008] Third, the ability to extract abnormal features is limited, and the noise resistance is insufficient. Smart lighting network operation data is characterized by strong time-varying properties and complex noise interference. Factors such as switching power supply ripple, electromagnetic interference, and transient fading of wireless channels can all disturb the sampled data. Existing technologies often use simple statistical indicators or fixed filtering algorithms for preprocessing, which makes it difficult to retain abnormal abrupt changes while suppressing noise.

[0009] While some systems introduce complex signal decomposition methods, these methods are highly dependent on parameter settings and have high computational complexity, making them unsuitable for real-time applications in large-scale node networks. Redundancy in feature dimensions and feature dispersion also reduce classification efficiency and accuracy.

[0010] Fourth, root cause analysis lacks integrated reasoning and interpretable mechanisms. Existing systems mostly use fixed rule bases or single machine learning models for fault classification. Rule-based methods struggle to cover complex and ever-changing anomaly patterns, while single models have limited generalization ability when faced with unseen anomaly types. Furthermore, most systems only output fault category labels, failing to provide diagnostic criteria, key feature intervals, or reasoning paths.

[0011] When diagnostic results differ from actual repair results, the system struggles to trace the source of the error and cannot optimize the model structure and inference weights based on historical feedback, affecting the maintenance personnel's trust in the system's results.

[0012] Fifth, the handling strategies lack differentiated decision-making and adaptive optimization capabilities. Existing systems typically execute fixed handling procedures for diagnosed faults, such as restarting equipment, switching channels, or dispatching repair orders, lacking a mechanism for adjusting strategies based on fault type, anomaly severity, and overall network status. Furthermore, most systems use static parameter configurations after deployment, with diagnostic thresholds, model weights, and inference rules remaining unchanged for extended periods, lacking a self-learning update mechanism based on operational feedback. When the network environment changes, equipment ages, or new fault modes emerge, the system's diagnostic accuracy gradually declines, requiring manual intervention for adjustments, resulting in high maintenance costs and delayed response times. Summary of the Invention

[0013] To address the common problems in existing smart lighting network status monitoring systems, such as passive data collection which makes it difficult to detect early anomalies in a timely manner, the disconnect between electrical and communication layer data leading to insufficient accuracy in root cause analysis, and the lack of verifiable evidence and self-learning optimization capabilities in diagnostic results, it is necessary to propose a smart lighting network status monitoring and fault self-diagnosis technology solution that can actively detect node operating status, integrate multi-layer data for cross-layer root cause reasoning, generate structured diagnostic evidence, and achieve closed-loop optimization.

[0014] To address this, a smart lighting network status monitoring and fault self-diagnosis system is proposed, comprising:

[0015] The data acquisition module is used to collect electrical operating parameters, control feedback data, communication link quality parameters, and gateway operating status data of the smart lighting nodes. The electrical operating parameters include voltage, current, and power data, and the communication link quality parameters include signal strength, packet loss rate, round-trip time, and number of retransmissions.

[0016] An anomaly triggering module, connected to the data acquisition module, is used to detect abnormal states based on the electrical operating parameters and communication link quality parameters. The anomaly triggering module uses a sliding time window statistical method to continuously analyze the operating parameters, calculates the parameter change rate and anomaly score within the window, and compares the anomaly score with a preset threshold. When the anomaly score exceeds the preset threshold, a diagnostic trigger signal is generated.

[0017] An active detection module, connected to the anomaly triggering module, sends a diagnostic probe command to the target illumination node after receiving the diagnostic trigger signal and collects probe response data; the active detection module includes a micro-amplitude dimming probe unit and a multi-path heartbeat detection unit.

[0018] The micro-amplitude dimming probe unit sends a step dimming command with an amplitude change less than a preset brightness threshold to the target lighting node, collects current, power and brightness feedback data within a preset time window, and determines whether there is an abnormality in the driving power supply based on the response deviation between the dimming command and the feedback data.

[0019] The multi-path heartbeat detection unit sends heartbeat data packets to the target node through different communication paths, records round-trip delay, packet loss rate and retransmission count, and determines whether there is communication link degradation by comparing the link quality parameters of different paths.

[0020] The comparison analysis module, connected to the active detection module, is used to select at least one comparison node based on the network topology adjacency matrix. The topology adjacency matrix is ​​constructed based on the signal strength and physical distance between nodes in the network. The comparison analysis module generates a difference feature vector by calculating the difference in operating parameters between the target node and the comparison node, which is used to reflect whether the anomaly of the target node has locality or loop-level characteristics.

[0021] A cross-layer root cause reasoning module, connected to the active detection module and the control analysis module, is used to fuse electrical operating parameters, communication link parameters, and control analysis results to output the fault root cause category and confidence level. The cross-layer root cause reasoning module uses a combination of rule-based reasoning and probabilistic reasoning models for reasoning. The rule-based reasoning model performs preliminary classification based on a preset fault rule base. The probabilistic reasoning model calculates the posterior probability of each fault category based on a Bayesian network. The final root cause category is determined by fusing the rule-based reasoning results and the posterior probability results.

[0022] The Bayesian network includes power supply state nodes, drive state nodes, link quality nodes, and gateway load nodes. By constructing a causal dependency matrix between nodes and training a conditional probability table based on historical operating data, cross-layer causal relationship modeling is achieved.

[0023] The evidence package generation module, connected to the cross-layer root cause reasoning module, is used to generate a structured diagnostic evidence package containing a summary of statistical features of operating parameters, active detection response results, comparative analysis results, and reasoning paths; the summary of statistical features of operating parameters includes mean, standard deviation, skewness, and kurtosis; the evidence package is used to support the verifiability and traceability of diagnostic results.

[0024] The action orchestration module is connected to the cross-layer root cause reasoning module and automatically executes the corresponding handling strategy according to the fault root cause category. When it is determined that the communication link is degraded, channel reselection, route reconstruction or adaptive adjustment of transmission power is executed. When it is determined that the drive power supply is abnormal, derating operation or periodic restart strategy is executed. When it is determined that the power supply circuit is abnormal, maintenance dispatch information carrying the evidence package is generated.

[0025] The feedback optimization module, connected to the action orchestration module, is used to collect operational data after the handling is completed, calculate the deviation between the diagnostic results and the actual maintenance results, and update the rule base parameters and the conditional probability table of the Bayesian network according to the deviation, so as to realize the system's self-learning closed-loop optimization. The feedback optimization module constructs a loss function by calculating the error between the predicted category and the actual operating state category, and updates the relevant parameters in the inference model, so that the system can adapt to changes in the operating environment and the evolution of fault modes.

[0026] Compared with the prior art, the beneficial effects of the present invention are:

[0027] First, this invention combines an anomaly triggering module with an active detection module to achieve a shift from passive monitoring to active verification. Traditional systems typically rely on periodic data reporting from nodes for anomaly detection, only identifying anomalies when their characteristics are obvious. This invention triggers active detection when an abnormal trend in operating parameters is detected, verifying the anomaly through a micro-dimming probe unit and a multi-path heartbeat detection unit. This obtains more discriminative operating data without affecting normal lighting services, thereby improving early anomaly identification capabilities.

[0028] This invention introduces a topological adjacency matrix and difference feature vector mechanism through a comparison analysis module, enabling the system to determine whether anomalies possess local or loop-level characteristics. By comparing the operating parameters of the target node with those of control nodes within the same loop or network group, it can distinguish between single-node faults and loop-level anomalies, reducing the probability of false positives and improving the accuracy of root cause localization.

[0029] This invention constructs a diagnostic mechanism combining rule-based reasoning and probabilistic reasoning by fusing electrical operating parameters, communication link parameters, and comparative analysis results through a cross-layer root cause reasoning module. The rule-based reasoning model can quickly match typical fault modes, while the probabilistic reasoning model calculates posterior probabilities using Bayesian networks to achieve quantitative judgment of complex coupled faults. By fusing the results of both types of reasoning, the stability and robustness of the diagnostic results are enhanced.

[0030] Furthermore, this invention uses an evidence package generation module to structurally encapsulate the diagnostic process, integrating and outputting statistical characteristics of operating parameters, active detection response results, comparative analysis data, and reasoning paths, thus ensuring the verifiability and traceability of the diagnostic results. Maintenance personnel can use the evidence package content to determine the diagnostic basis, thereby improving the credibility of the system's diagnostic results. Further, this invention uses an action orchestration module to execute differentiated handling strategies based on different fault categories, enabling targeted handling of different types of faults such as communication link degradation, drive power supply anomalies, and power supply circuit anomalies. This avoids resource waste or secondary fault risks that may result from a uniform handling strategy, improving handling efficiency.

[0031] In summary, this invention constructs a closed-loop self-learning mechanism through a feedback optimization module. After handling the issue, it collects operational data and updates the rule base parameters and Bayesian network conditional probability table based on actual maintenance results. This enables the system to continuously optimize its diagnostic capabilities as the operating environment changes and fault modes evolve, improving long-term operational stability and diagnostic accuracy. Through the synergistic cooperation of active detection, comparative analysis, cross-layer fusion reasoning, structured evidence generation, and the closed-loop feedback optimization mechanism, this invention improves the accuracy, interpretability, and adaptability of fault diagnosis in smart lighting networks while ensuring system operational stability. Attached Figure Description

[0032] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without creative effort. In the drawings:

[0033] Figure 1 This is an architecture diagram of a smart lighting network status monitoring and fault self-diagnosis system.

[0034] Figure 2 This is a flowchart of a smart lighting network status monitoring and fault self-diagnosis system. Detailed Implementation

[0035] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0036] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0037] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0038] Example 1

[0039] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0040] This embodiment provides a smart lighting network status monitoring and fault self-diagnosis system, which includes a data acquisition module, an anomaly triggering module, an active detection module, a comparative analysis module, a cross-layer root cause reasoning module, an evidence package generation module, and an action orchestration module.

[0041] The data acquisition module is used to collect electrical operating parameters, control feedback data, communication link quality parameters, and gateway operating status data of the smart lighting nodes. Electrical operating parameters include voltage, current, active power, power factor, and temperature. The acquisition frequency is set to 10 times per second. Control feedback data includes dimming command execution status, brightness feedback value, and command delay time. Communication link quality parameters include signal reception strength, packet loss rate, round-trip time, and retransmission count. Gateway operating status data includes gateway processor load rate, cache utilization rate, and gateway temperature.

[0042] The data acquisition module collects parameters through data acquisition units deployed at each lighting node. These units use an STM32F407 microcontroller as their processing core and are configured with a 12-bit analog-to-digital converter for voltage and current sampling. The voltage sampling range is set to 0 to 300 volts, and the current sampling range is set to 0 to 10 amperes. Temperature sampling uses a PT100 platinum resistance temperature sensor, with a measurement range of -40 degrees Celsius to +120 degrees Celsius and a measurement accuracy of ±0.5 degrees Celsius.

[0043] The acquisition of communication link quality parameters is accomplished through the wireless transceiver module. The wireless transceiver module uses an RF chip based on the IEEE 802.15.4 protocol, operating at a frequency of 2.4 GHz. Signal received strength is recorded in dBm, with a sampling period of once every 2 seconds. Packet loss rate is calculated by counting the number of unreceived packets out of every 100 packets. Round-trip time is obtained by timestamping data packets during transmission and calculating the time difference upon receiving response packets.

[0044] Gateway operational status data is collected by the gateway device's status monitoring program. The gateway device uses an ARM Cortex-A53 quad-core processor with a clock speed of 1.2 GHz and is equipped with 2 gigabytes of DDR4 memory. Processor load rate is obtained by reading the system CPU utilization rate, with a sampling period of once every 5 seconds. Cache utilization rate is obtained by calculating the ratio of used memory to total memory. Gateway temperature is measured by an onboard temperature sensor, with a sampling period of once every 10 seconds.

[0045] The anomaly triggering module is connected to the data acquisition module and is used to detect abnormal states based on electrical operating parameters and communication link quality parameters, and generate diagnostic trigger signals. The anomaly triggering module uses a sliding time window statistical method to detect abrupt changes in operating parameters. The sliding time window length is set to 60 seconds, and the window sliding step size is set to 10 seconds.

[0046] The exception triggering module calculates the rate of change of parameters within the window. The formula for calculating the rate of change of voltage is:

[0047] ;

[0048] This indicates the maximum voltage within the window. This indicates the minimum voltage value within the window. This represents the average voltage value within the window.

[0049] The formula for calculating the rate of change of current is:

[0050] ;

[0051] This indicates the maximum current within the window. Indicates the minimum current within the window. This represents the average current value within the window.

[0052] The formula for calculating the rate of change of power is:

[0053] ;

[0054] This indicates the maximum power within the window. This represents the minimum power value within the window. This represents the average power value within the window.

[0055] The formula for calculating the anomaly score is:

[0056]

[0057] Where S represents the anomaly score, to These represent the weighting coefficients of each parameter, ΔR represents the rate of change of signal received strength, and ΔL represents the rate of change of packet loss rate.

[0058] The initial value of the weighting coefficient is set to It equals 0.25. Equals 0.30, It equals 0.20. It equals 0.15. The value is equal to 0.10. The anomaly scoring threshold is set to 0.75. When the anomaly score value is greater than the anomaly scoring threshold, the anomaly triggering module generates a diagnostic trigger signal.

[0059] The formula for calculating the rate of change of received signal strength is:

[0060] ;

[0061] in, This indicates the current signal strength value. This represents the average signal reception strength within the window.

[0062] The formula for calculating the rate of change of packet loss rate is:

[0063] ;

[0064] in, Indicates the current packet loss rate This represents the average packet loss rate within the window.

[0065] The active detection module is connected to the anomaly triggering module. Upon receiving a diagnostic trigger signal, it sends diagnostic probe commands to the target illumination node and collects probe response data. The active detection module includes a micro-amplitude dimming probe unit and a multi-path heartbeat detection unit.

[0066] The micro-amplitude dimming probe unit sends a step dimming command with an amplitude change less than a preset brightness threshold to the target illumination node, and collects current, power, and brightness feedback data within a preset time window. The preset brightness threshold is set to 5%. The step dimming command adjusts the target brightness value of the illumination node from its current brightness value to a value increased by 3%. The preset time window length is set to 5 seconds.

[0067] After sending a step dimming command, the micro-amplitude dimming probe unit collects the current, power, and brightness feedback values ​​of the target lighting node at a sampling frequency of 20 times per second. The collected data forms a response curve. The theoretical response curve is calculated based on the driving power supply model of the lighting node. The driving power supply model is described using a first-order inertial element.

[0068] ;

[0069] in, Let represent the transfer function, K represent the gain coefficient, T represent the time constant, and s represent the complex frequency variable.

[0070] The gain factor K is set to 1.0, and the time constant T is set to 0.8 seconds. The theoretical response curve is calculated using the following formula:

[0071] ;

[0072] Where y(t) represents the response value at time t. Let represent the initial value, Δy represent the step amplitude, t represent the time variable, and e represent the base of the natural logarithm.

[0073] The formula for calculating response deviation is:

[0074] ;

[0075] Where E represents the response deviation Indicates the theoretical response curve at The value at time, Indicates the actual measured curve at The value at time N represents the number of sampling points.

[0076] The number of sampling points, N, is equal to 100. The response deviation threshold is set to 0.12. When the response deviation exceeds the response deviation threshold, the drive power supply is deemed to be malfunctioning.

[0077] The multipath heartbeat detection unit sends heartbeat data packets to the target node through different communication paths, recording round-trip time, packet loss rate, and retransmission count. These different communication paths include direct paths, one-hop relay paths, and two-hop relay paths. A direct path refers to the path where the gateway communicates directly with the target node. A one-hop relay path refers to the path where the gateway communicates with the target node through one relay node. A two-hop relay path refers to the path where the gateway communicates with the target node through two relay nodes.

[0078] The heartbeat data packet length is set to 64 bytes, the sending interval is set to 2 seconds, and the continuous sending duration is set to 30 seconds. The multipath heartbeat detection unit records the average round-trip delay, packet loss rate, and retransmission count for each path.

[0079] The formula for calculating packet loss rate is:

[0080] ;

[0081] Where PLR ​​represents packet loss rate, This indicates the number of heartbeat data packets sent. This indicates the number of heartbeat response packets received.

[0082] Link quality parameter comparison and analysis is completed by calculating the link quality differences between different paths. The formula for calculating the link quality difference is:

[0083] ;

[0084] Where D represents the link quality difference, and α and β represent the latency difference weight and packet loss rate difference weight, respectively. and These represent the average round-trip delays for the two paths, respectively. and These represent the packet loss rates of the two paths, respectively.

[0085] The latency difference weight α is set to 0.6, and the packet loss rate difference weight β is set to 0.4. The link quality difference threshold is set to 0.50. When the link quality difference exceeds the link quality difference threshold, communication link degradation is determined to exist.

[0086] The comparison analysis module is connected to the active detection module, and selects a comparison node within the same loop or network group to compare and analyze the operational responses of the target lighting node and the comparison node. The comparison analysis module selects at least one comparison node based on the topological adjacency matrix.

[0087] The topological adjacency matrix is ​​constructed based on the signal strength and physical distance between nodes in the network. Elements of the topological adjacency matrix A The calculation formula is:

[0088] ;

[0089] This represents the adjacency weight between node i and node j. Indicates signal strength weights. This represents the normalized strength value of the signal received by node i from node j. Indicates distance weight, This represents the physical distance between node i and node j. The maximum physical distance between points.

[0090] Signal strength weight Set to 0.7, distance weight Set to 0.3. Normalized intensity value. The calculation formula is:

[0091] ;

[0092] This represents the original strength value of the signal received by node i from node j. Minimum signal strength in the network The maximum signal strength in the network.

[0093] Minimum signal strength -100dBm, maximum signal strength -30dBm. Maximum physical distance. 500 meters.

[0094] The strategy for selecting control nodes is to choose the top three nodes with the highest adjacency weights as control nodes. The control analysis module collects the electrical operating parameters and communication link quality parameters of the control nodes within the same time window.

[0095] The formula for calculating the difference in operating parameters is:

[0096] ;

[0097] Where Diff represents the degree of difference in runtime parameters, This represents the normalized voltage value of the target node. This represents the normalized voltage value of the control node. This represents the normalized current value of the target node. This represents the normalized current value of the control node. This represents the normalized power value of the target node. This represents the normalized power value of the control node.

[0098] Voltage lower limit 180 volts, upper voltage limit 250 volts. Normalized current and normalized power values ​​are calculated using the same method. The lower limit of current is set to 0 amperes, and the upper limit of current is set to 10 amperes. The lower limit of power is set to 0 watts, and the upper limit of power is set to 2000 watts.

[0099] The differential feature vector consists of the differences in operating parameters between the target node and each control node.

[0100] ;

[0101] Where F represents the differential feature vector. , and These represent the differences in operating parameters between the target node and the first, second, and third control nodes, respectively.

[0102] The cross-layer root cause reasoning module is connected to the active detection module and the control analysis module. It is used to integrate electrical operating parameters, communication link parameters, and control analysis results to output the fault root cause category and confidence level. The cross-layer root cause reasoning module adopts a combination of rule-based reasoning model and probabilistic reasoning model.

[0103] The rule-based reasoning model performs initial classification based on a pre-defined fault rule base. This base includes rules for abnormal power supply circuits, abnormal drive power supplies, communication link degradation, and excessive gateway load.

[0104] The power supply circuit anomaly rule is defined as follows: if the voltage change rate is greater than 0.15 and the voltage change rate of all control nodes is greater than 0.15, it is initially classified as a power supply circuit anomaly. The drive power supply anomaly rule is defined as follows: if the response deviation is greater than the response deviation threshold and the voltage change rate is less than 0.10, it is initially classified as a drive power supply anomaly. The communication link degradation rule is defined as follows: if the link quality difference is greater than the link quality difference threshold and the response deviation is less than 0.08, it is initially classified as a communication link degradation. The gateway overload rule is defined as follows: if the gateway processor load rate is greater than 85% and the average round-trip latency is greater than 500 milliseconds, it is initially classified as a gateway overload.

[0105] The rule-based reasoning model outputs preliminary classification results and rule matching scores. The formula for calculating the rule matching score is:

[0106] ;

[0107] in, Indicates the rule matching degree. Indicates the number of conditions that are satisfied. This indicates the total number of conditions in the rule.

[0108] The probabilistic inference model calculates the posterior probability of each fault category based on a Bayesian network. The Bayesian network includes power supply state nodes, driver state nodes, link quality nodes, and gateway load nodes.

[0109] The power supply status node indicates the operating status of the power supply circuit, with a value of normal or abnormal. The drive status node indicates the operating status of the drive power supply, with a value of normal or abnormal. The link quality node indicates the operating status of the communication link, with a value of normal or degraded. The gateway load node indicates the load status of the gateway, with a value of normal or overloaded.

[0110] A causal dependency matrix describes the dependencies between nodes. Elements of the causal dependency matrix C. This represents the degree of influence of node i on node j. Matrix elements take values ​​of 0 or 1; a value of 1 indicates a causal dependency, and a value of 0 indicates no causal dependency.

[0111] There is a causal dependency between the power supply state node and the drive state node, corresponding to the matrix elements. Equals 1. The link quality node has a causal dependency on the gateway load node, corresponding to the matrix element. Equals 1. The driving state node and the link quality node are independent of each other, corresponding to matrix elements. and All are equal to 0.

[0112] The conditional probability table was trained using historical operational data. The training dataset includes 10,000 labeled samples. Each sample includes voltage, current, power, signal strength, packet loss rate, round-trip time, gateway load rate, and corresponding fault category labeling.

[0113] The prior probability of a power supply status node is obtained by statistically analyzing the proportion of abnormal power supply circuit samples in the training dataset. The prior probability P(abnormal power supply) is set to 0.12, and the prior probability P(normal power supply) is set to 0.88.

[0114] The conditional probability table P(Drive State|Power Supply State) for the drive state nodes is obtained by statistically analyzing the distribution of drive states under different power supply conditions. The conditional probability P(Drive Abnormal|Power Supply Abnormal) is set to 0.65, and the conditional probability P(Drive Abnormal|Power Supply Normal) is set to 0.08.

[0115] The prior probability P(link degradation) for link quality nodes is set to 0.18, and the prior probability P(link normal) is set to 0.82. The conditional probability P(gateway overload|link degradation) for gateway load nodes is set to 0.42, and the conditional probability P(gateway overload|link normal) is set to 0.05.

[0116] The posterior probability is calculated using Bayes' theorem:

[0117] ;

[0118] in, Indicates the fault category given evidence E. The posterior probability, Indicates the fault category The likelihood probability of observing evidence E when it occurs. Indicates the fault category The prior probability, P(E), represents the marginal probability of evidence E.

[0119] The formula for calculating the marginal probability P(E) is:

[0120] ;

[0121] The summation symbol indicates summation over all fault categories.

[0122] Likelihood probability Calculations are based on the correlation between observed evidence and each fault category. Observed evidence includes voltage change rate, response deviation, link quality variability, and gateway load rate.

[0123] The formula for calculating the likelihood probability of power supply circuit anomaly categories is as follows:

[0124] ;

[0125] in, Let f(ΔV) represent the likelihood probability of a power supply circuit malfunction, and let f(ΔV) represent the probability density function of the rate of voltage change. This represents the joint probability density function representing the difference among all control nodes.

[0126] The probability density function of the voltage change rate adopts a Gaussian distribution:

[0127] ;

[0128] in, This represents the mean rate of voltage change when the power supply circuit is abnormal. The standard deviation of the voltage change rate when the power supply circuit is abnormal is represented by π, where π represents the mathematical constant pi.

[0129] mean Set to 0.22, standard deviation Set to 0.05.

[0130] The formula for calculating the likelihood probability of the drive power supply anomaly category is as follows:

[0131] ;

[0132] in, Let h(E) represent the likelihood probability of a drive power supply malfunction, and let h(E) represent the probability density function of the response deviation.

[0133] The probability density function of the response bias adopts a Gaussian distribution:

[0134] ;

[0135] in, This represents the mean of the response deviation when the drive power supply is abnormal. This represents the standard deviation of the response deviation when the drive power supply is abnormal.

[0136] mean Set to 0.18, standard deviation Set to 0.04.

[0137] The formula for calculating the likelihood probability of communication link degradation category is:

[0138] ;

[0139] in, Let k(D) represent the likelihood probability of communication link degradation, and k(D) represent the probability density function of the link quality difference.

[0140] The probability density function for link quality variation adopts a Gaussian distribution:

[0141] ;

[0142] in, This represents the mean value of the link quality difference when the communication link degrades. The standard deviation represents the degree of difference in link quality when the communication link degrades.

[0143] mean Set to 0.68, standard deviation Set to 0.12.

[0144] The likelihood probability formula for the "gateway overload" category is as follows:

[0145] ;

[0146] in, Let m(L) represent the likelihood probability that the gateway is overloaded, and m(L) represent the probability density function of the gateway load rate. The probability density function represents the round-trip time delay.

[0147] The probability density function of the gateway load rate adopts a Gaussian distribution:

[0148] ;

[0149] in, This represents the average load rate when the gateway is under excessive load. This represents the standard deviation of the load rate when the gateway load is too high, and L represents the observed value of the gateway load rate.

[0150] mean Set to 0.90, standard deviation Set to 0.06.

[0151] The probability density function for round-trip time delay adopts a Gaussian distribution:

[0152] ;

[0153] in, This represents the average round-trip latency when the gateway is under high load. RTT represents the standard deviation of round-trip latency when the gateway is overloaded, and RTT represents the observed round-trip latency.

[0154] mean Set to 580 milliseconds, standard deviation Set to 80 milliseconds.

[0155] The formula for calculating the fusion weight between the fusion rule inference result and the posterior probability result is as follows:

[0156] ;

[0157] in Indicates the fault category The final confidence score, λ represents the rule inference weight. Indicates the fault category Rule matching degree Indicates the fault category The posterior probability.

[0158] The initial value of the rule inference weight λ is set to 0.4. The final root cause category is determined as the fault category with the highest final confidence. The confidence threshold is set to 0.60. When the maximum final confidence is less than the confidence threshold, the root cause category cannot be determined.

[0159] The evidence package generation module is connected to the cross-layer root cause reasoning module to generate a structured diagnostic evidence package containing a summary of operating parameters, probe response results, control analysis results, and reasoning path.

[0160] The operating parameter time series is compressed into a statistical feature summary. The statistical feature summary includes the mean, standard deviation, skewness, and kurtosis. The formula for calculating the mean of the voltage time series is:

[0161] ;

[0162] in, Indicates the average voltage. This represents the i-th voltage sample value, and N represents the total number of sampling points.

[0163] The total number of sampling points N equals 600, which corresponds to the number of data points collected at a frequency of 10 times per second within a 60-second time window.

[0164] The formula for calculating the standard deviation of a voltage time series is:

[0165] ;

[0166] in, This represents the standard deviation of voltage.

[0167] The formula for calculating the skewness of a voltage time series is:

[0168] ;

[0169] in, Indicates voltage deviation.

[0170] The formula for calculating the kurtosis of a voltage time series is:

[0171] ;

[0172] in, Indicates voltage kurtosis.

[0173] The statistical characteristics of the current time series and the power time series were summarized using the same method.

[0174] The probe response results record the residual curve between the active probe command and the response curve. The residual curve is composed of the residual values ​​at each sampling point. The formula for calculating the residual value is:

[0175] ;

[0176] in, express The residual value at time step, Indicates the actual measured curve at The value at time, Indicates the theoretical response curve at The value at any given moment.

[0177] The formula for calculating the root mean square value of the residual curve is:

[0178] ;

[0179] Here, RMSE represents the root mean square value of the residuals. The root mean square value of the residuals is used to quantify the degree of deviation between the response curve and the theoretical curve. The threshold for the root mean square value of the residuals is set to 0.15.

[0180] The comparative analysis results record the differential feature vectors and control node identifiers. The control node identifiers include the network address and physical location coordinates of the control node. The differential feature vectors are stored as numerical arrays.

[0181] The inference path records the rule number for rule inference matching, the posterior probability value for each fault category, the fusion weight value, and the final confidence value. The rule number is represented in string form, such as R001 for power supply circuit abnormality rule, R002 for drive power supply abnormality rule, R003 for communication link degradation rule, and R004 for gateway overload rule.

[0182] The structured diagnostic evidence package is organized in JSON format. The evidence package includes timestamps, target node identifiers, statistical feature summaries, root mean square residual values, differential feature vectors, a list of control nodes, rule matching results, a list of posterior probabilities, fusion weight values, final root cause categories, and final confidence values.

[0183] The action orchestration module connects to the cross-layer root cause reasoning module, automatically executing corresponding handling strategies based on the type of fault root cause. These handling strategies include network self-healing control strategies, equipment de-rating strategies, and maintenance dispatch strategies.

[0184] When the root cause is ultimately determined to be communication link degradation, the action orchestration module executes the network self-healing control strategy. The network self-healing control strategy includes channel reselection, route reconstruction, and adaptive transmit power adjustment.

[0185] The channel reselection strategy scans 16 channels in the 2.4 GHz band, measures the interference intensity of each channel, and selects the channel with the lowest interference intensity as the working channel. Channel interference intensity is obtained by measuring the noise floor power when the channel is idle. The noise floor power measurement duration is set to 1 second. After channel reselection is complete, the gateway broadcasts a channel switching command to all nodes in the network.

[0186] The route reconstruction strategy is implemented by recalculating the network topology and routing table. The route calculation uses the shortest path first algorithm. The path cost function is calculated as follows:

[0187] ;

[0188] Where Cost represents the path cost. and These represent the hop count weight, signal strength weight, and packet loss rate weight, respectively. Hop represents the path hop count, RSSI represents the path average signal strength, and PLR represents the path packet loss rate.

[0189] hop count weight Set to 1.0, signal strength weight Set to 0.5, packet loss rate weight Set to 2.0. After the route reconstruction is complete, the gateway sends the updated routing table to the relevant nodes.

[0190] The adaptive transmit power adjustment strategy dynamically adjusts the transmit power of the wireless transceiver module based on link quality. The transmit power adjustment step size is set to 2dBm. When the received signal strength is below -80dBm, the transmit power increases by 2dBm. When the received signal strength is above -50dBm, the transmit power decreases by 2dBm. The transmit power adjustment range is set from 0dBm to 20dBm.

[0191] When the root cause is ultimately determined to be a driver power supply malfunction, the action orchestration module executes the device derating strategy. The device derating strategy includes reducing the target brightness value and a periodic restart strategy.

[0192] The strategy of reducing the target brightness value reduces the operating brightness of the lighting nodes to 70% of the rated brightness. The duration of derating is set to 24 hours. During the derating period, a micro-dimming probe test is performed every 6 hours to assess whether the driver power supply status has returned to normal.

[0193] The periodic restart strategy performs a soft restart of the device at the start of derating operation. After restarting, it waits for 5 minutes, then collects operating parameters again and calculates the response deviation. If the response deviation recovers to less than 0.08, the drive power supply abnormality is considered to have been resolved, and derating operation is canceled. If the response deviation is still greater than 0.12, derating operation continues, and a maintenance dispatch order is generated.

[0194] When the root cause is ultimately determined to be a power supply circuit malfunction or cannot be handled by automatic strategies, the action orchestration module generates maintenance dispatch information. Maintenance dispatch information includes the fault node identifier, fault category, confidence level, structured diagnostic evidence package, and recommended maintenance measures.

[0195] Recommended maintenance measures are preset according to the fault category. For power supply circuit abnormalities, the recommended maintenance measures are to check the power supply line contacts and measure the circuit voltage stability. For driver power supply abnormalities, the recommended maintenance measures are to replace the driver power supply module. For communication link degradation, the recommended maintenance measures are to check the antenna connection status and surrounding radio frequency interference sources. For excessive gateway load, the recommended maintenance measures are to increase the number of gateway devices or upgrade the gateway hardware configuration.

[0196] Maintenance dispatch information is sent to the maintenance management platform via a message queue. The message queue uses the MQTT protocol. The MQTT broker server address is set to 192.168.1.100, and the port number is set to 1883. The dispatch message subject is set to maintenance / tickets. The dispatch message is encapsulated in JSON format.

[0197] The feedback optimization module is connected to the action orchestration module. It collects operational data after the treatment is completed, calculates the deviation between the diagnostic results and the actual maintenance results, and updates the rule base parameters and the conditional probability table of the Bayesian network according to the deviation to achieve self-learning closed-loop optimization.

[0198] After a maintenance work order is dispatched, the feedback optimization module collects the actual fault cause annotations submitted by the maintenance personnel. The actual fault cause annotations include five categories: power supply circuit abnormality, drive power supply abnormality, communication link degradation, gateway overload, and other faults.

[0199] The discrepancy between diagnostic results and actual repair results is statistically analyzed using a confusion matrix. The rows of the confusion matrix represent diagnostic results, and the columns represent the actual causes of the fault. Confusion matrix elements... This represents the number of samples that were diagnosed as category i but were actually in category j.

[0200] The formula for calculating diagnostic accuracy is:

[0201] ;

[0202] Accuracy represents the diagnostic accuracy rate. , represents the diagonal elements of the confusion matrix, , represents the number of correctly diagnosed samples, and the summation symbol represents summation over all categories. This represents the element in the i-th row and j-th column of the confusion matrix.

[0203] The initial diagnostic accuracy was 0.82. The target diagnostic accuracy was set at 0.92.

[0204] The rule base parameters are updated using backpropagation. The rule matching threshold is adjusted based on the number of misdiagnosed samples. If a rule causes the number of misdiagnosed samples to exceed the threshold, the weight coefficient of that rule is reduced. The weight coefficient adjustment formula is as follows:

[0205] ;

[0206] in, This represents the updated weight coefficients. This represents the weight coefficients before the update, and γ represents the learning rate. This represents the partial derivative of the loss function with respect to the weight coefficients.

[0207] The learning rate γ is set to 0.01. The loss function used is cross-entropy loss.

[0208] ;

[0209] Where L represents the loss value, Indicates the true label, The symbol represents the predicted probability, and the summation symbol represents summing over all samples.

[0210] The conditional probability table update of the Bayesian network uses the maximum a posteriori estimation method. The conditional probability update formula is:

[0211] ;

[0212] in, Let N(A, B) represent the updated conditional probability, N(A, B) represent the number of observations where events A and B occur simultaneously, and α represent the smoothing parameter. Let N(B) represent the conditional probability before the update, and let N(B) represent the number of observations that event B occurred.

[0213] The smoothing parameter α is set to 10. The conditional probability table is updated every 100 new samples.

[0214] The fusion weight λ is updated using the gradient descent method. The fusion weight update formula is:

[0215] ;

[0216] in, This indicates the updated fusion weights. This represents the fused weights before the update, and β represents the learning rate of the fused weights. This represents the partial derivative of the loss function with respect to the fusion weights.

[0217] The learning rate β for the fusion weights was set to 0.005. The partial derivatives of the loss function with respect to the fusion weights were calculated using numerical differentiation.

[0218] ;

[0219] Where δ represents the differential step size. The differential step size δ is set to 0.001.

[0220] In summary, this embodiment fully demonstrates the workflow of a smart lighting network status monitoring and fault self-diagnosis system. The system initiates active detection through an anomaly triggering mechanism, combining micro-dimming probe detection and multi-path heartbeat detection to acquire multi-dimensional operational characteristics, and generates differential feature vectors through comparative analysis. Based on this, the cross-layer root cause reasoning module integrates rule-based reasoning and Bayesian network probabilistic reasoning results to determine the final fault root cause category and confidence level. Subsequently, the evidence package generation module forms structured diagnostic evidence, and the action orchestration module executes differentiated handling strategies. After handling is completed, the feedback optimization module updates the rule parameters and conditional probability table, achieving closed-loop self-learning optimization.

[0221] This embodiment verifies the effectiveness of the data flow relationship and collaboration mechanism between the functional modules, demonstrating that the system can achieve accurate fault location and automatic handling while ensuring stable network operation, and has good engineering feasibility and promotional application value.

[0222] Example 2

[0223] This embodiment further illustrates the specific working process of the system based on Embodiment 1. In this embodiment, the smart lighting network comprises 200 lighting nodes, divided into 10 loops, with each loop containing 20 nodes. The network adopts a star topology, configured with 4 gateway devices, each managing 50 nodes. The lighting nodes use LED light sources, with a rated power of 80 watts and an operating voltage of 220 volts per node. Wireless communication uses the ZigBee protocol, operating at a frequency of 2.4 GHz, with channel number 15.

[0224] At the 128th second, the data acquisition module detected that the voltage of the lighting node with node number N045 decreased from 225 volts to 198 volts, the current decreased from 0.36 amperes to 0.31 amperes, and the power decreased from 78 watts to 58 watts. Simultaneously, the signal received strength decreased from -62 dBm to -68 dBm, and the packet loss rate increased from 2% to 6%.

[0225] The anomaly triggering module uses a 60-second sliding time window to statistically analyze operating parameters. Within the time window from the 128th to the 188th second, the calculated maximum voltage was 227 volts, the minimum voltage was 198 volts, and the average voltage was 218 volts. The voltage change rate is defined as:

[0226] ;

[0227] Substitute the data to obtain The maximum current is 0.37 amperes, the minimum current is 0.31 amperes, and the average current is 0.35 amperes. The rate of change of current is defined as:

[0228] ;

[0229] Substitute the data to obtain The maximum power is 79 watts, the minimum power is 58 watts, and the average power is 72 watts. The rate of change of power is defined as:

[0230] ;

[0231] Substitute the data to obtain .

[0232] The current received signal strength is -68 dBm, and the average value within the window is -63 dBm. The rate of change of received signal strength is defined as:

[0233] ;

[0234] Substitute the data to obtain The current packet loss rate is 6%, and the average rate within the window is 3%. The rate of change of packet loss rate is defined as:

[0235] ;

[0236] Substitute the data to obtain .

[0237] Anomaly score is defined as:

[0238] ;

[0239] in Substituting, we get .because:

[0240] ;

[0241] No active detection was triggered.

[0242] At time 256, the voltage at node N045 further decreased to 182 volts, the current decreased to 0.28 amperes, and the power decreased to 48 watts. The signal reception strength decreased to -75 dBm, and the packet loss rate increased to 12%. Within the time window from 256 to 316 seconds, the following data was obtained: Substituting into the anomaly scoring formula yields... .because No active detection was triggered.

[0243] At time 384, the voltage at node N045 drops to 168 volts, the current drops to 0.25 amperes, and the power drops to 38 watts. Within the time window from time 384 to time 444, the following values ​​are obtained: Substituting, we get .because No active detection was triggered.

[0244] At time 512, the voltage at node N045 suddenly recovered from 168 volts to 222 volts, but dropped again to 175 volts after 5 seconds and continued to fluctuate. Within the time window from time 512 to time 572, the following values ​​were obtained: Substituting, we get .

[0245] At time 640, the voltage at node N045 fluctuated between 160 volts and 190 volts, with a fluctuation frequency of one cycle every 8 seconds. Within the time window from 640 to 700 seconds, the voltage change rate... Voltage standard deviation Volts, rate of change of current Power change rate Rate of change of received signal strength Packet loss rate change rate .

[0246] The abnormal triggering module introduces voltage standard deviation as a supplementary criterion, with the voltage standard deviation threshold set at... express;

[0247] set up Fu, Dang At that time, the weighting coefficient for abnormal score values ​​was increased to At this point, the abnormal score is defined as:

[0248] ;

[0249] Substitute to get .because No active detection was triggered.

[0250] At time 768, the voltage fluctuation at node N045 increases, fluctuating between 150 volts and 200 volts. Within the time window from time 768 to time 828, the following is obtained: Fu, Calculated according to the above weighting rules .because No active detection was triggered.

[0251] At time 896, the voltage at node N045 suddenly drops to 135 volts and remains at that value. Within the time window from time 896 to time 956, the following values ​​were obtained: Voltage standard deviation Fu, satisfied:

[0252] ;

[0253] No weighted adjustment is needed. Calculated based on the baseline anomaly score. .because No active detection was triggered.

[0254] At time 1024, the voltage at node N045 returned to 210 volts, but the current abnormally increased to 0.52 amperes, and the power increased to 105 watts, exceeding the rated power. Within the time window from 1024 to 1084 seconds, the following was obtained: The basic anomaly score is calculated as follows: .because No active detection was triggered.

[0255] At 1152 seconds, the power of node N045 remained between 102 watts and 108 watts, the voltage between 208 volts and 215 volts, and the current between 0.49 amps and 0.52 amps. The power overload duration reached 128 seconds. The anomaly triggering module introduced a power overload duration criterion, setting a threshold of [value missing]. Seconds. When:

[0256] ;

[0257] Increased abnormal score, penalty item .

[0258] Within the time window from 1152 seconds to 1212 seconds, we obtain The basic abnormality score is The overall anomaly score is:

[0259] ;

[0260] get .because No active detection was triggered.

[0261] At time 1280, the voltage at node N045 suddenly dropped from 212 volts to 118 volts, the current decreased from 0.51 amperes to 0.18 amperes, and the power dropped from 106 watts to 21 watts. The signal reception strength decreased to -88 dBm, and the packet loss rate increased to 28%. Within the time window from 1280 to 1340 seconds, the following data was obtained: The basic anomaly score was calculated as follows. .because No active detection was triggered.

[0262] At time 1408, the voltage at node N045 fluctuated slightly between 110 volts and 125 volts, the current fluctuated between 0.16 amperes and 0.20 amperes, and the power fluctuated between 18 watts and 24 watts. The received signal strength fluctuated between -85 dBm and -92 dBm, and the packet loss rate fluctuated between 25% and 35%. Within the time window from 1408 seconds to 1468 seconds, the following data was obtained: Fu, The basic anomaly score was calculated as follows. .because No active detection was triggered.

[0263] At time 1536, the voltage values ​​of nodes N041, N042, N043, and N044 in the same circuit simultaneously decreased. The anomaly trigger module detected the simultaneous anomaly of multiple nodes and activated the circuit-level anomaly detection mechanism. The circuit-level anomaly detection mechanism calculates the proportion of anomaly nodes within the same circuit. The criteria for determining anomaly nodes are as follows:

[0264] ;

[0265] In loop 3, there are a total of 20 nodes. Five nodes were detected to meet the abnormal conditions. The proportion of abnormal nodes is:

[0266] ;

[0267] The loop-level anomaly ratio threshold is set to .

[0268] when ;

[0269] Multiply the abnormal score by the amplification factor. .

[0270] Node N045 obtains the following within the time window from second 1536 to second 1596: The basic abnormality score is The anomaly score after magnification is:

[0271] ;

[0272] get .because No active detection was triggered.

[0273] At time 1664, the number of anomalous nodes in loop 3 increased to 12, and the anomalous node ratio was:

[0274] ;

[0275] Node N045 obtains the following within the time window from 1664 seconds to 1724 seconds: The basic abnormality score is The anomaly score after magnification is:

[0276] ;

[0277] because This triggers active detection.

[0278] The anomaly trigger module generates a diagnostic trigger signal at 1724 seconds and sends it to the active detection module. Upon receiving the diagnostic trigger signal, the active detection module activates the micro-amplitude dimming probe unit and the multi-path heartbeat detection unit.

[0279] The micro-amplitude dimming probe unit sends a step dimming command to node N045. The current brightness of node N045 is 18%, and the step dimming command sets the target brightness value to 21%, with a brightness increase of 3%, satisfying the requirement that the increase is less than the preset brightness threshold of 5%. The dimming command is issued at 1726 seconds. The micro-amplitude dimming probe unit collects the current, power, and brightness feedback values ​​of node N045 at a sampling frequency of 20 times per second for 5 seconds, collecting a total of 100 data points.

[0280] The theoretical response curve is calculated based on the drive power supply model. The drive power supply model is as follows:

[0281] ;

[0282] in seconds. Initial current Ampere. The step amplitude is defined as:

[0283] ;

[0284] get Ampere. The theoretical response curve is:

[0285] ;

[0286] The theoretical and measured values ​​at each time point satisfy the definition of residual:

[0287] ;

[0288] And based on this, the response bias is obtained:

[0289] ;

[0290] in The calculations obtained in this embodiment are as follows: .because:

[0291] ;

[0292] The power supply was found to be faulty.

[0293] The multi-path heartbeat detection unit starts simultaneously, sending heartbeat data packets to node N045. The heartbeat data packet is 64 bytes long, sent at 2-second intervals, and continuously sent for 30 seconds. The multi-path heartbeat detection unit tests three communication paths.

[0294] Direct connection will send within 30 seconds One heartbeat data packet was successfully received. For each response data packet, the packet loss rate is defined as:

[0295] ;

[0296] get The average round-trip time delay is defined as:

[0297] ;

[0298] in ,get millisecond.

[0299] One-hop relay path obtained Milliseconds. The two-hop relay path is obtained. millisecond.

[0300] Link quality difference is determined by comparing direct paths with one-hop relay paths, and is defined as follows:

[0301] ;

[0302] in .consider The unit is milliseconds, normalized:

[0303] ;

[0304] get The link quality difference is taken in the normalized form:

[0305] ;

[0306] Calculated .because:

[0307] ;

[0308] The system determines that there is no communication link degradation, but marks the link quality assessment result as a decline in communication link quality.

[0309] The comparison analysis module selects comparison nodes based on the topological adjacency matrix. The adjacency weight is defined as follows:

[0310] ;

[0311] in Meters. The normalized signal strength is defined as:

[0312] ;

[0313] in .

[0314] The comparison analysis module selected the three nodes with the highest adjacency weights as comparison nodes, namely N044, N038, and N042. The module collected the electrical operating parameters of these three comparison nodes within a time window from 1664 seconds to 1724 seconds. Node N045 had an average voltage of 108 volts, an average current of 0.14 amperes, and an average power of 15 watts within the same time window.

[0315] The difference in operating parameters is defined as follows:

[0316] ;

[0317] The difference feature vector is defined as:

[0318] ;

[0319] The calculations obtained in this embodiment are as follows:

[0320] ;

[0321] The cross-layer root cause reasoning module receives data from the active detection module and the control analysis module, and initiates rule-based reasoning model and probabilistic reasoning model for fusion reasoning.

[0322] The rule-based reasoning model outputs the rule matching degree. The probabilistic inference model uses Bayes' theorem to calculate the posterior probability:

[0323] ;

[0324] The marginal probability is:

[0325] ;

[0326] Fusion confidence is defined as:

[0327] ;

[0328] This embodiment takes The root cause was ultimately determined to be a drive power supply malfunction, with a final confidence level of 0.799.

[0329] The evidence package generation module generates a structured diagnostic evidence package. The evidence package includes a timestamp of 1724 seconds, target node identifier N045, voltage statistical feature summary, current statistical feature summary, power statistical feature summary, root mean square residual value, differential feature vector, list of control nodes, rule matching results, posterior probability list, fusion weight value, final root cause category, and final confidence value.

[0330] The motion orchestration module receives the final root cause category as a driver power supply malfunction and initiates the device derating operation strategy, setting the target brightness value to 70%. The derating operation command is issued at 1732 seconds. At 1735 seconds, the voltage at node N045 recovers to 215 volts, the current rises to 0.48 amps, the power rises to 98 watts, and the brightness feedback value is 68%.

[0331] The motion orchestration module simultaneously executes a periodic restart strategy. Node N045 restarts at 1734 seconds and performs micro-amplitude dimming probe detection again at 2034 seconds. The response deviation is calculated as follows. ,satisfy The driver power supply malfunction has been resolved, the derating operation strategy has been canceled, and normal operation has resumed.

[0332] The feedback optimization module records that the diagnostic results are consistent with the actual cause of the fault, and the matrix elements are confused. Added 1. The system has completed a total of 56 diagnostic cases, with a diagnostic accuracy rate of:

[0333] ;

[0334] The statistics obtained in this embodiment .

[0335] The feedback optimization module adjusts the weight coefficients and introduces a communication delay compensation mechanism. During the Bayesian network parameter update, the mean response bias is updated as follows:

[0336] ;

[0337] in The standard deviation has been updated to:

[0338] ;

[0339] in .

[0340] The fusion weights are updated using gradient descent:

[0341] ;

[0342] The loss function is in the form of cross-entropy;

[0343] Authentic Labels The predicted probability is, .

[0344] The single-sample loss is:

[0345] ;

[0346] This embodiment obtains The numerical differential is:

[0347] ;

[0348] in Calculations yielded ,therefore .

[0349] This embodiment fully demonstrates the entire workflow of the system, from anomaly detection, active probing, comparative analysis, root cause reasoning, evidence package generation, action orchestration to feedback optimization, and verifies the effectiveness of the functional implementation and parameter settings of each module of the system.

Claims

1. A smart lighting network status monitoring and fault self-diagnosis system, characterized in that, include: The data acquisition module is used to collect electrical operating parameters, control feedback data, communication link quality parameters, and gateway operating status data of smart lighting nodes; An anomaly triggering module, connected to the data acquisition module, is used to detect abnormal states based on the electrical operating parameters and communication link quality parameters, and generate a diagnostic trigger signal. An active detection module, connected to the anomaly triggering module, sends a diagnostic probe command to the target lighting node after receiving the diagnostic triggering signal, and collects probe response data. The active detection module includes a micro-amplitude dimming probe unit and a multi-path heartbeat detection unit; The micro-amplitude dimming probe unit sends a step dimming command with an amplitude change less than a preset brightness threshold to the target lighting node, collects current, power and brightness feedback data within a preset time window, and determines whether there is an abnormality in the driving power supply based on the response deviation between the dimming command and the feedback data. The multi-path heartbeat detection unit sends heartbeat data packets to the target node through different communication paths, records round-trip delay, packet loss rate and retransmission count, and determines whether there is communication link degradation by comparing the link quality parameters of different paths. The comparison analysis module, connected to the active detection module, selects at least one comparison node within the same loop or network group based on the topological adjacency matrix. The topological adjacency matrix is ​​constructed based on the signal strength and physical distance between nodes in the network. By calculating the difference in operating parameters between the target node and the comparison node, a difference feature vector is generated to reflect whether the anomaly of the target node has locality or loop-level characteristics. The cross-layer root cause reasoning module is connected to the active detection module and the control analysis module. It is used to fuse electrical operating parameters, communication link parameters and control analysis results, and output the fault root cause category and confidence level. The evidence package generation module, connected to the cross-layer root cause reasoning module, is used to generate a structured diagnostic evidence package containing a summary of operating parameters, probe response results, control analysis results, and reasoning path. The action orchestration module is connected to the cross-layer root cause reasoning module and automatically executes the corresponding handling strategy according to the fault root cause category, including network self-healing control strategy, equipment de-rated operation strategy and operation and maintenance dispatch strategy.

2. The intelligent lighting network status monitoring and fault self-diagnosis system according to claim 1, characterized in that, The anomaly triggering module uses a sliding time window statistical method to detect abrupt changes in operating parameters and calculates the anomaly score using the following formula: ； Where ΔV is the rate of change of voltage, ΔI is the rate of change of current, ΔP is the rate of change of power, ΔR is the rate of change of received signal strength, and ΔL is the rate of change of packet loss rate. to These represent the weighting coefficients of each parameter; when the abnormal score value S is greater than the preset abnormal score threshold, a diagnostic trigger signal is generated.

3. The intelligent lighting network status monitoring and fault self-diagnosis system according to claim 1, characterized in that, After sending a step dimming command, the micro-amplitude dimming probe unit collects data that forms a response curve. The theoretical formula for calculating the response curve is as follows: ； in, Let t represent the initial value, Δy represent the step amplitude, and t represent the time variable. Represents the time constant; The response deviation E between the actual measured curve and the theoretical response curve is calculated: ； Where E represents the response deviation Indicates the theoretical response curve at The value at time, Indicates the actual measured curve at The value at time N represents the total number of sampling points collected at a preset sampling frequency within a preset time window; when the response deviation E is greater than the preset response deviation threshold, it is determined that there is an abnormality in the drive power supply.

4. The intelligent lighting network status monitoring and fault self-diagnosis system according to claim 1, characterized in that, The multipath heartbeat detection unit determines whether communication link degradation exists by calculating the link quality differences of different paths: ； Where D represents the link quality difference degree, and α and β represent the latency difference weight and packet loss rate difference weight, respectively. and These represent the average round-trip delays for the two paths, respectively. and These represent the packet loss rates of the two paths, respectively. When the link quality difference D is greater than the preset link quality difference threshold, it is determined that there is communication link degradation.

5. The intelligent lighting network status monitoring and fault self-diagnosis system according to claim 1, characterized in that, The comparison analysis module selects comparison nodes based on the topological adjacency matrix A, wherein the elements of the topological adjacency matrix are... Calculate using the following formula: ； in, Indicates signal strength weights. Indicates distance weight, This represents the normalized strength value of the signal received by node i from node j. This represents the physical distance between node i and node j. The maximum physical distance between points; Select the k nodes with the largest adjacency weights as control nodes; calculate the difference in operating parameters between the target node and each control node: ; in, This represents the normalized voltage value of the target node. This represents the normalized voltage value of the reference node; This represents the normalized current value of the target node. This represents the normalized current value of the reference node; This represents the normalized power value of the target node. This represents the normalized power value of the control node; all parameters are normalized to [value missing]. interval; Generate differential feature vectors ; in, This indicates the degree of difference in operating parameters between the target node and the first control node. This indicates the degree of difference in operating parameters between the target node and the second control node. The difference in operating parameters between the target node and the third control node is represented; the difference feature vector is used to reflect whether the anomaly of the target node has locality or loop-level characteristics.

6. The intelligent lighting network status monitoring and fault self-diagnosis system according to claim 1, characterized in that, The cross-layer root cause reasoning module employs a combination of rule-based reasoning and probabilistic reasoning models. The rule-based reasoning model performs preliminary classification based on a pre-set fault rule base and calculates the rule matching degree using the following formula: ； in, Indicates the number of conditions that are satisfied. This represents the total number of rule conditions; the probabilistic inference model, based on a Bayesian network, calculates the posterior probability of each fault category using the following formula: ； in, ; The final root cause category is determined using the following fusion formula: ; Where λ represents the rule reasoning weight.

7. The intelligent lighting network status monitoring and fault self-diagnosis system according to claim 6, characterized in that, The Bayesian network includes power supply state nodes, drive state nodes, link quality nodes, and gateway load nodes; a causal dependency matrix between nodes is constructed, and a conditional probability table is trained based on historical operating data.

8. The intelligent lighting network status monitoring and fault self-diagnosis system according to claim 1, characterized in that, The evidence package generation module compresses the time series of operating parameters into a statistical feature summary, which includes the mean, standard deviation, skewness, and kurtosis; and records the residual curve between the active detection command and the response curve, with the residual value calculated according to the following formula: ； in, Indicates that the active detection module is in The actual measurement curve values ​​collected at all times are used to reflect the true response of the target lighting node driving power supply; This indicates that the theoretical response curve calculated based on the first-order inertial element model of the driving power supply is in The value at a given moment reflects the expected response of the drive power supply under normal conditions; Calculate the root mean square value of the residuals using the following formula: ； in This represents the total number of sampling points; the evidence package is used to support the verifiability and traceability of the diagnostic results.

9. The intelligent lighting network status monitoring and fault self-diagnosis system according to claim 1, characterized in that, When the action orchestration module determines that a communication link has degraded, it performs channel reselection, route reconstruction, or adaptive adjustment of transmit power. Route reconstruction uses the shortest path first algorithm, and the path cost function is: ； in, and These represent the hop count weight, signal strength weight, and packet loss rate weight, respectively. Hop represents the path hop count, RSSI represents the average signal strength of the path, and PLR represents the path packet loss rate. When a driver power supply abnormality is detected, a derating operation or periodic restart strategy is executed. When a power supply circuit abnormality is detected, an operation and maintenance dispatch order carrying the aforementioned evidence package is generated.

10. The intelligent lighting network status monitoring and fault self-diagnosis system according to claim 1, characterized in that, It also includes a feedback optimization module; the feedback optimization module is connected to the action orchestration module; it collects the operation data after the treatment is completed, calculates the deviation between the diagnostic result and the actual maintenance result; and updates the rule weight parameters of the preset fault rule base and the conditional probability table of the Bayesian network in the cross-layer root cause reasoning module according to the deviation, so as to realize self-learning closed-loop optimization.

11. The intelligent lighting network status monitoring and fault self-diagnosis system according to claim 10, characterized in that, The feedback optimization module updates the rule base weights according to the following formula: ； in, This indicates the current value of the rule weight coefficient before the update; This indicates the new value of the rule weight coefficient after this update; The learning rate; The loss function used is cross-entropy loss: ; in, Indicates the first The sample belongs to the first The true label of the type of fault; Indicates the system predicts the first The sample belongs to the first The probability of a type of failure; The total number of samples, Total number of fault categories; γ represents the learning rate; the feedback optimization module updates the Bayesian network conditional probability table according to the following formula: ； Where α represents the smoothing parameter, N(A,B) represents the number of observations where events A and B occur simultaneously, and N(B) represents the number of observations where event B occurs; the fusion weight λ is updated according to the following formula: ; in, This represents the current value of the fusion weights before the update; This indicates the new value of the fusion weight after this update; The updated fusion weights; To integrate weights and learning rates.

12. The intelligent lighting network status monitoring and fault self-diagnosis system according to claim 2, characterized in that, The anomaly triggering module detects multiple nodes simultaneously exhibiting anomalies and activates a loop-level anomaly detection mechanism, multiplying the anomaly score by an amplification factor g to improve trigger sensitivity. ,in Based on the anomaly score.

13. The intelligent lighting network status monitoring and fault self-diagnosis system according to claim 2, characterized in that, The anomaly triggering module also introduces voltage standard deviation as a supplementary criterion; when the voltage standard deviation... Exceeding the preset voltage standard deviation threshold At that time, the abnormal score is multiplied by a weighting coefficient k to improve it: When the power exceeds the limit for a certain duration Exceeding the preset duration threshold At that time, add penalty items. Accumulate the abnormal scores.

14. The intelligent lighting network status monitoring and fault self-diagnosis system according to claim 1, characterized in that, The evidence package is organized in JSON format and includes timestamps, target node identifiers, statistical feature summaries, root mean square residual values, differential feature vectors, a list of control nodes, rule matching results, a list of posterior probabilities, fusion weight values, and final root cause categories and final confidence values. The inference path records the rule numbers of the rule inference matching pair, the posterior probability values, fusion weight values, and final confidence values ​​for each fault category.

15. A smart lighting network status monitoring and fault self-diagnosis system according to claim 9, characterized in that, During the derating operation, the action orchestration module performs a micro-dimming probe detection every preset time interval to assess whether the drive power supply status has returned to normal; the maintenance dispatch information is sent to the maintenance management platform through a message queue protocol.

Images