Method and system for monitoring navigation lights based on fail-safe

Abstract

The present invention relates to the field of traffic control technology, and in particular to a navigation light monitoring method and system based on fault insurance. The method comprises the following steps: collecting light operation data through a distributed sensor array, and extracting key parameters of the light operation data to obtain light operation characteristic data; performing real-time state modeling based on the light operation characteristic data to obtain an operation state model; performing anomaly detection based on the operation state model to obtain operation anomaly detection data; performing intelligent matching of insurance policies based on the operation anomaly detection data to obtain insurance policy execution data to perform navigation light monitoring auxiliary operations. The present invention can improve the fault detection capability and resource utilization efficiency of the monitoring system, and provide new possibilities for the intelligent development of the navigation light system.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic control, and in particular to a navigation light monitoring method and system based on fail-safe. Background Art

[0002] The navigation lighting system is a vital infrastructure in airport operations. Its core function is to provide clear and intuitive navigation support for aircraft at night or in low visibility conditions (such as rainy and foggy weather, sandstorms, etc.). The system helps pilots correctly identify the locations of taxiways, runways and parking spaces in complex and changing environments through the deployment of light arrays (such as runway edge lights, taxiway lights, entrance lights, etc.), ensuring that the take-off, landing and taxiing operations of aircraft are safe, accurate and efficient. However, with the continuous expansion of airport scale and the significant increase in operational complexity, traditional navigation lighting monitoring and management methods have gradually exposed many shortcomings and cannot fully meet the high requirements of modern airport operations. At present, the fault detection of most navigation lighting systems still relies on manual inspections, preset rules or subjective reports from pilots. This method usually finds problems only after the fault has occurred and the operation of the lights has been affected. It lacks foresight and early warning capabilities, which can easily lead to potential operational safety hazards. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a navigation light monitoring method and system based on fail-safe to solve at least one of the above technical problems.

[0004] The present application provides a method for monitoring navigation lights based on fail-safe, comprising the following steps:

[0005] Step S1: collecting lighting operation data through a distributed sensor array, and extracting key parameters of the lighting operation data to obtain lighting operation characteristic data;

[0006] Step S2: Perform real-time state modeling based on the lighting operation characteristic data to obtain an operation state model;

[0007] Step S3: performing anomaly detection according to the operation status model to obtain operation anomaly detection data;

[0008] Step S4: Intelligently match the insurance policy according to the abnormal operation detection data to obtain the insurance policy execution data to perform auxiliary operations for the navigation light monitoring.

[0009] The distributed sensor array in the present invention provides multi-dimensional real-time data, covering multiple parameters such as voltage, current, brightness, temperature, etc., to ensure the integrity and accuracy of the monitoring data. Dynamic data acquisition frequency and adaptive optimization technology ensure high-frequency monitoring during critical operating periods (such as nighttime or low-visibility environments) and capture abnormal changes in real time. Based on real-time state modeling technology, an operating state model is generated to comprehensively describe the current operating state and change trend of the navigation lights. Dynamic threshold adjustment reduces the false alarm rate and enhances the system's tolerance for normal fluctuations. Intelligent matching insurance strategies dynamically select the optimal strategy (such as switching backup lights, adjusting circuits, activating backup power supplies), respond quickly when a fault occurs, and ensure the continuous operation of the navigation lights.

[0010] Preferably, step S1 specifically includes:

[0011] Collect lighting operation data through distributed sensor arrays;

[0012] Extracting non-smoothed parameters of lighting operation data to obtain non-smoothed parameter data of lighting operation;

[0013] Perform non-smooth feature extraction according to the lighting operation non-smooth parameter data to obtain first lighting operation feature data;

[0014] Extract extreme skewness features from the lighting operation data to obtain lighting operation extreme skewness feature data;

[0015] Extract the kurtosis change feature of the lighting operation data to obtain the lighting operation kurtosis change feature data;

[0016] Get real-time environmental condition data;

[0017] Generate a feature vector for the lighting operation extreme skewness feature data and the lighting operation kurtosis change feature data according to the real-time environmental condition data to obtain second lighting operation feature data;

[0018] The first lighting operation characteristic data and the second lighting operation characteristic data are feature aligned to obtain the lighting operation characteristic data.

[0019] The distributed sensor array in the present invention covers the entire navigation lighting system, realizing comprehensive real-time monitoring of each lighting node, and there is no blind spot in data collection. The acquisition frequency is dynamically adjusted to adapt to real-time operating conditions (such as high traffic or bad weather) to ensure accurate acquisition of data at critical moments. Extracting non-smooth parameters (such as mutation points, short-term volatility) and non-smooth features can identify small but significant operating changes in the lighting system and improve the sensitivity to early faults. Extreme skewness feature extraction captures extreme values ​​in data distribution (such as sudden increase or decrease in brightness), which is suitable for analyzing lighting behavior under extreme operating conditions. Kurtosis change feature extraction describes abnormal peak and tail behaviors in the lighting system, revealing the risk of severe fluctuations faced by the system. Considering real-time environmental factors (such as temperature, humidity, wind speed, etc.) to enhance the practical significance of feature data, it can effectively separate anomalies caused by environmental changes from system-endogenous anomalies. According to environmental condition data, an enhanced feature vector is generated to construct the second lighting operation feature data related to the environment, which can more accurately describe the lighting operation status.

[0020] Preferably, the non-smoothing parameter extraction is specifically as follows:

[0021] Perform variational decomposition on the lighting operation data to obtain non-smooth variation interval data;

[0022] Perform non-smooth point detection based on non-smooth change interval data to obtain non-smooth point data for lighting operation;

[0023] Perform differential characteristic analysis on the non-smooth point data of lighting operation to obtain the characteristic data of non-smooth point change rate;

[0024] Perform time series gap analysis on the non-smooth point change rate characteristic data to obtain time interval distribution characteristic data;

[0025] Perform local discontinuity quantization according to time interval distribution characteristic data to obtain local non-smooth quantized data;

[0026] Perform spatial variation analysis based on the light source parameter data corresponding to the lighting operation data and the local non-smoothed quantized data to obtain the lighting operation spatial variation feature data;

[0027] Perform frequency domain mutation extraction based on the spatial variation feature data of lighting operation to obtain frequency domain mutation feature data of lighting operation;

[0028] The non-smooth point change rate characteristic data, local non-smooth quantization data and lighting operation frequency domain mutation characteristic data are integrated to obtain the lighting operation non-smooth parameter data.

[0029] In the present invention, variational mode decomposition is used to decompose the light data into smooth and non-smooth parts, ensuring accurate focus on non-smooth characteristics. Non-smooth point detection identifies mutation points and abnormal behaviors through adaptive thresholds, which is particularly suitable for capturing subtle anomalies that are difficult to detect in the system. Differential characteristic analysis captures the sudden dynamic changes in light operation by calculating the rate of change, revealing the characteristics of rapid fluctuations in the operating state. The time interval distribution characteristics reveal the frequency and interval regularity of non-smooth points, reflecting the stability of the system and the potential periodicity of operation. Local discontinuity quantification quantifies the severity of non-smooth points, providing an accurate description of anomalies for subsequent strategy matching and resource scheduling. Spatial variation analysis combines non-smooth behavior with the physical position and light source parameters of the light to accurately locate the abnormal distribution. Frequency domain mutation extraction captures high-frequency fluctuations and abnormal signals in light operation on the spectrum, revealing hidden problems. Multi-dimensional features are integrated to form a complete description of non-smooth parameters, covering abnormal information in the time domain, space domain and frequency domain.

[0030] Conventional means rely on preset fixed upper and lower limits based on threshold detection, and can only detect significant anomalies that exceed the set values. If the amplitude of the sudden change value in the lighting operation data is small, or occurs in a local range within a short period of time, traditional methods are often unable to detect it. And if the abnormality of the lighting system is gradually accumulated (such as the brightness gradually decreases over a long period of time, and the current fluctuation gradually increases), the traditional method cannot respond in the early stage of abnormal accumulation, resulting in delayed discovery of the problem. The present invention dynamically adapts to different types of lighting operation data, flexibly adjusts the decomposition ratio of smooth and non-smooth parts, organically combines time, space and frequency domain information, comprehensively characterizes the non-smooth characteristics of the system, enhances modeling capabilities, and provides strong technical support for the operation status analysis and anomaly detection of the lighting system to overcome problems that cannot be achieved by conventional means.

[0031] Preferably, the non-smooth feature extraction is specifically:

[0032] Density clustering calculation is performed according to the non-smooth parameter data of the light operation to obtain non-smooth density clustering data;

[0033] Generate characteristic spectrum for non-smooth density clustering data to obtain non-smooth characteristic spectrum data;

[0034] Perform non-smooth region segmentation according to local non-smooth quantized data to obtain non-smooth feature region data;

[0035] Cross-scale feature synthesis is performed according to the non-smoothed feature spectrum data and the non-smoothed feature region data to obtain first light operation feature data.

[0036] In the present invention, the distribution pattern of non-smooth anomalies is identified through a density clustering algorithm, and sparse anomalies and dense anomaly clusters are distinguished. There is no need to preset the number of clusters, and the algorithm can adaptively identify complex anomaly distributions based on data. Feature spectrum generation can extract high-frequency fluctuations and periodic anomaly characteristics in lighting operation from the perspective of frequency domain. Spectral analysis mines hidden behaviors that are difficult to capture with time domain data, providing more comprehensive input for anomaly detection, such as periodic flickering or voltage fluctuation characteristics in lighting operation, which can be clearly displayed through spectrum graphs. Through regional segmentation technology based on local non-smooth quantization, the spatial distribution area of ​​anomalies in the lighting system can be accurately located. Each segmented area corresponds to a unique abnormal behavior feature, which provides targeted support for subsequent system optimization and maintenance scheduling. Cross-scale features can reflect the relationship between anomalies at different scales, and improve the ability to identify and predict complex abnormal behaviors. The first lighting operation feature data provides high-quality input for state modeling and insurance policy matching.

[0037] Preferably, the extreme skewness feature extraction is specifically:

[0038] Perform non-uniform window segmentation on the lighting operation data to obtain lighting operation segmentation data;

[0039] Perform sliding least squares extreme value detection on the lighting operation segmented data to obtain the lighting operation extreme value point data;

[0040] Perform extreme value distribution fitting based on the extreme value point data of the lighting operation to obtain extreme value distribution fitting data;

[0041] The skewness is calculated based on the extreme value distribution fitting data to obtain the skewness characteristic data of the lighting operation;

[0042] According to the skewness characteristic data of the lighting operation, the extreme skewness characteristic data is extracted to obtain the extreme skewness characteristic data;

[0043] Multi-scale feature aggregation is performed based on the extreme skewness feature data to obtain the extreme skewness feature data of lighting operation.

[0044] In the present invention, the window size is automatically adjusted according to the rate of change of the light operation data to adapt to the dynamic changes of the system and avoid the loss of accuracy caused by the fixed window. The non-uniform segmentation is more suitable for capturing local drastic changes in the operation data and providing high-quality input for extreme value detection. The sliding least squares method smoothes the data while retaining important trends. The fluctuation characteristics of the light operation are accurately located by detecting local extreme points. Extreme point detection effectively identifies abnormal peaks or troughs of parameters such as brightness and power, which is suitable for capturing transient events. The extreme points are fitted by the generalized extreme value distribution to describe the distribution characteristics of extreme events in the lighting system, such as abnormal high brightness or complete extinguishing. Skewness describes the asymmetry of the data distribution and reflects the tendency of abnormal fluctuation direction during the operation of the light (such as too bright or too dark). Through in-depth analysis of the skewness feature, the extremely biased behavior features, such as extreme fluctuation amplitude and frequency, are extracted. The extreme skewness feature can reflect the extreme tendency of the light operation and is particularly suitable for the identification of serious faults such as light extinguishing and over-brightness. Multi-scale aggregation not only covers local details, but also describes the global pattern, providing comprehensive input for system modeling.

[0045] Preferably, the kurtosis change feature extraction is specifically as follows:

[0046] Perform weighted sliding average segmentation on the lighting operation data to obtain smooth segmentation data of lighting operation;

[0047] Perform preliminary kurtosis calculation on the smooth segmented data of lighting operation to obtain preliminary kurtosis data;

[0048] Perform local kurtosis anomaly detection based on preliminary kurtosis data to obtain local kurtosis anomaly data;

[0049] Perform time series kurtosis difference based on local kurtosis anomaly data to obtain kurtosis change rate data;

[0050] Perform multi-scale kurtosis change analysis based on kurtosis change rate data to obtain multi-scale kurtosis change data;

[0051] The multi-scale kurtosis change data is fitted with kurtosis Gaussian distribution to obtain kurtosis Gaussian distribution data;

[0052] Based on the multi-scale kurtosis change data and the kurtosis Gaussian distribution data, feature vectorization is performed to obtain the lighting operation kurtosis change characteristic data.

[0053] The sliding average method in the present invention smoothes the data, effectively removes random noise, and retains key trends. The weighted sliding average gives higher weights to the mutation area to ensure that the local abnormal characteristics will not be weakened by smoothing. The preliminary kurtosis calculation reflects the peak degree of the data distribution and captures the area with greater volatility in the operation of the light. Through the anomaly detection algorithm based on local kurtosis (such as Z-score analysis), the abnormal peak area in the operation of the light can be identified. The differential operation reveals the trend of kurtosis in the time series, reflecting the dynamic characteristics of the operation state of the light. The kurtosis change rate is more sensitive to anomalies and can identify trend anomalies and volatility faults. Multi-scale analysis combines short-term and long-term kurtosis change characteristics to provide a comprehensive description of anomalies. Through Gaussian distribution fitting, the distribution of kurtosis characteristics of the light operation data, especially the probability of occurrence of abnormal spikes, is described. The multi-scale analysis results and the distribution fitting results are vectorized to generate a set of unified high-dimensional feature vectors, which are suitable for modeling and strategy matching.

[0054] Preferably, step S2 specifically includes:

[0055] Get basic data of navigation lights;

[0056] The digital twin model is constructed based on the basic data of the navigation lights to obtain the digital twin model of the navigation lights;

[0057] Divide the data set according to the lighting operation feature data to obtain lighting operation training input data and lighting operation target output data;

[0058] Construct an input gate according to the lighting operation training input data to obtain lighting operation input gate data;

[0059] Construct a hidden layer according to the light operation input gate data to obtain the light operation hidden layer data;

[0060] The output gate is constructed according to the hidden layer data of the light operation to obtain the output gate data of the light operation;

[0061] Iteratively train the lighting operation output gate data using the lighting operation target output data to obtain a lighting operation deep model;

[0062] The lighting operation deep model and the navigation light digital twin model are integrated to obtain the operation status model.

[0063] In the present invention, the geometric parameters, physical characteristics, operation specifications and other basic data of the navigation lights are collected to ensure the authenticity and consistency of the model construction. The digital twin model realizes the interconnection of virtual and real data through real-time synchronization of the virtual model and the physical navigation lights, providing an accurate reference for the operation status modeling. The training data set (input and output) is automatically divided based on the light operation feature data to ensure the efficiency and effectiveness of the model training. The input gate can efficiently process the multi-dimensional feature data of the light operation (such as brightness, power consumption, voltage, etc.) and filter irrelevant information. The hidden layer learns the complex dynamic relationship of the light operation through a deep learning network (such as LSTM), including short-term fluctuations and long-term trends in the time series. The output gate maps the deep features extracted by the hidden layer to the output results (such as the predicted light status) through weight distribution and activation function optimization. Through multiple rounds of iterative training, the deep model gradually optimizes the parameters to improve the prediction accuracy of the light operation status. By integrating the real-time characteristics of the digital twin model and the learning ability of the deep model, the generated operation status model can dynamically respond to changes in the light operation status.

[0064] Preferably, step S3 specifically includes:

[0065] Outputting the operating state characteristics according to the operating state model to obtain operating state characteristic data;

[0066] Perform multimodal feature aggregation according to the running status feature data to obtain multimodal fusion feature data;

[0067] The environmental condition threshold is dynamically calculated based on the multimodal fusion feature data to obtain the operation anomaly detection data.

[0068] Based on the operating state model, the present invention extracts key features of lighting operation (such as brightness, current, voltage, environmental adaptability, etc.) to provide high-quality input for detection. Real-time monitoring of the operating state is achieved through dynamic feature extraction, capturing small changes in the lighting system and providing support for early fault warning. By integrating different modal features, the system can adapt to complex operating scenarios, including multiple light sources, multiple environments and dynamically changing conditions. By dynamically calculating the threshold in combination with real-time environmental conditions (such as temperature, humidity, wind speed, etc.), anomaly detection can adapt to the changing operating environment. The dynamic threshold avoids the high false positive and false negative problems caused by the static threshold, ensuring the accuracy of the anomaly detection results.

[0069] Preferably, step S4 is specifically:

[0070] Mapping is performed according to the operation anomaly detection data and a preset anomaly event library to obtain anomaly event mapping data;

[0071] Sort the processing priorities according to the abnormal event mapping data to obtain the abnormal event priority data;

[0072] Intelligent matching of insurance policies is performed based on the priority data of abnormal events to obtain insurance policy execution data for auxiliary operations of navigation lighting monitoring.

[0073] The present invention combines the operation anomaly detection data with the preset abnormal event library (including common fault types and their characteristics) to quickly identify the specific type of anomaly (such as light off, low brightness, unstable voltage, etc.). Priority data helps to dispatch limited maintenance resources (such as backup lights and maintenance personnel) to avoid resource waste. The most suitable insurance strategy (such as starting backup lights, switching redundant circuits, and notifying maintenance personnel) is matched according to the priority data to avoid invalid or excessive triggering.

[0074] Preferably, the present application further provides a failsafe-based navigation light monitoring system, which is used to execute the failsafe-based navigation light monitoring method as described above. The failsafe-based navigation light monitoring system includes:

[0075] A lighting operation feature extraction module is used to collect lighting operation data through a distributed sensor array, and extract key parameters of the lighting operation data to obtain lighting operation feature data;

[0076] A real-time state modeling module is used to perform real-time state modeling based on the light operation characteristic data to obtain an operation state model;

[0077] The lighting operation anomaly detection module is used to perform anomaly detection according to the operation status model and obtain operation anomaly detection data;

[0078] The lighting operation insurance policy module is used to intelligently match the insurance policy according to the operation anomaly detection data and obtain the insurance policy execution data to perform auxiliary operations for navigation lighting monitoring.

[0079] The beneficial effects of the present invention are as follows: the distributed sensor array realizes the real-time monitoring of the lighting system with full coverage, and can capture the multi-dimensional characteristics of the lighting operation (such as brightness, current, voltage, temperature, etc.), providing high-quality data for modeling and detection. By combining the non-smooth parameter extraction, extreme skewness feature extraction and kurtosis change feature extraction technology, the dynamic changes and extreme behaviors in the operation, especially the minor anomalies and potential faults in the system, are accurately captured. By combining deep learning technology (such as LSTM) with the digital twin model, the real-time modeling of the lighting operation state is realized, which can dynamically adapt to the operation changes of the system. The digital twin model virtualizes the physical navigation lighting system, provides real-time synchronization and state prediction capabilities, and enhances the robustness of the system. By combining multimodal feature aggregation and dynamic threshold calculation of environmental conditions, abnormal behavior can be accurately detected and the false alarm and missed alarm rates can be reduced. The dynamic threshold calculation adjusts the detection strategy according to the real-time environmental conditions (such as temperature, humidity, wind speed) to meet the needs of complex operating environments. The abnormal event mapping mechanism can quickly identify the fault type and match the most suitable insurance strategy (such as enabling backup lights, switching redundant circuits, and triggering maintenance tasks). Combined with the processing priority sorting mechanism, it ensures that resources are allocated first to the most critical abnormal events. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting implementations made with reference to the following drawings:

[0081] Figure 1 A flowchart of a method for monitoring navigation lights based on fail-safe according to an embodiment of the present invention is shown;

[0082] Figure 2 A flowchart showing the steps of a method for extracting light operation characteristics according to an embodiment is shown;

[0083] Figure 3 A flowchart showing a method for real-time state modeling according to an embodiment is shown;

[0084] Figure 4 A flowchart showing a method for detecting abnormal operation of lighting according to an embodiment of the present invention is provided;

[0085] Figure 5 A flowchart of the steps of a lighting operation insurance strategy method according to an embodiment is shown. DETAILED DESCRIPTION

[0086] The following is a clear and complete description of the technical method of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0087] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0088] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0089] The navigation lighting system of an airport consists of 50 lighting nodes. The system deploys a distributed sensor network to collect real-time operating data (brightness, current, voltage, temperature, etc.) of each lighting node. Light brightness data (unit: Lux): [100, 105, 102, 90, 85, 120, 200, 250, 245, 100]. Current data (unit: mA): [120, 122,119, 80, 78, 130, 210, 240, 235, 120]. Decompose the brightness data into smooth and non-smooth components. Smooth component: [100, 104, 103, 95, 92, 110, 190, 230, 220, 110] Non-smooth component: [0, 1, -1,-5, -7, 10, 10, 20, 25, -10]. Non-smooth point detection is performed, and non-smooth points are obtained: [6, 7, 8], corresponding to brightness data 120, 200, 250. The skewness calculation result is: 1.75 (right-skewed distribution, indicating that the main abnormality of lighting is increased brightness). The kurtosis calculation result is: 3.25, indicating that the data distribution has a significant peak characteristic. Lighting operation characteristic data: {Brightness: skewness 1.75, kurtosis 3.25; current: skewness 1.65, kurtosis 3.10}.

[0090] Generate a digital twin model using basic data of navigation lights (geometric parameters, power range, standard brightness range 100-200 Lux) and synchronize the physical system in real time. Train the extracted features with a deep learning model, using the light operation feature data as training input. Learn the time-varying patterns of brightness and current through LSTM. Predict the light brightness at the next time step. Input training data: [100, 105, 102, 90, 85, 120], predicted output:

[125] . Fusion of the deep learning model with the digital twin model to generate an operation status model for real-time prediction and anomaly detection.

[0091] Current brightness prediction value: 125 Lux, actual value: 250 Lux. Extract brightness features from the operating state model: {Current prediction deviation = 250-125 = +125}. Fusion of brightness and current features with environmental data (temperature: 30°C) generates multimodal fusion features: {Brightness deviation 125, Current deviation 115, Temperature anomaly 0}. Threshold calculation formula: Abnormality = Current deviation / Normal range = 125 / (200-100) = 1.25. Abnormal detection results are marked as "high risk abnormality".

[0092] Abnormal type: The light is too bright (brightness abnormality +125 Lux). Processing priority ranking: According to the impact of brightness on airport runway safety, the priority is 1. Match the fault type according to the abnormal event library: the light voltage is too high. Mapping result: It is necessary to switch the redundant circuit and adjust the power. Considering the impact of abnormal brightness, safety risks, and processing time, it is ranked as the highest priority. Trigger the strategy, switch the backup circuit -> reduce the light power -> notify the maintenance team. Insurance strategy execution data: {Strategy 1: switch circuit; Strategy 2: reduce the power to 180 Lux; Strategy 3: notify maintenance}.

[0093] See also Figures 1 to 5 The present application provides a method for monitoring navigation lights based on fail-safe, comprising the following steps:

[0094] Step S1: collecting lighting operation data through a distributed sensor array, and extracting key parameters of the lighting operation data to obtain lighting operation characteristic data;

[0095] Specifically, a distributed sensor array is used to monitor and collect parameters such as the brightness, flashing frequency, current, voltage, power factor, and temperature of the navigation lights in real time. Sensors are distributed at each lighting node, and the sampling interval of the sensors is set to 0.1 seconds. The data is transmitted to the data processing center through a wireless communication protocol.

[0096] The collected data is preprocessed, including denoising (using the sliding window method to remove high-frequency noise), outlier removal (filtering by setting upper and lower thresholds), and the following key parameters are calculated:

[0097] Average brightness ;in is the average brightness, which means the average brightness within a specified time range or within a specified area. is the total number of sampling points, indicating the number of data points used to calculate the average brightness. is the index of the data point, indicating the specific sampling point pointed to when the current brightness is calculated. For the The brightness value of the sampling point is expressed at time or spatial location The brightness measured at a point in time, expressed in lumens (lm) or similar units.

[0098] Flicker frequency standard deviation ;in is the standard deviation of flicker frequency, which is used to measure the flicker frequency of the lighting system The larger the value, the more significant the change in flicker frequency; the smaller the value, the more stable the flicker frequency. is the total number of sampling points, that is, the number of flicker frequency data points involved in the calculation of the standard deviation. The index (serial number) of the sampling point, used for traversal Flicker frequency data points, For the The flicker frequency value of the sampling point is expressed in The light flickering frequency measured at a certain time or location, is the average value of the flicker frequency, which means the arithmetic mean of the flicker frequencies of all sampling points.

[0099] The harmonic components of voltage and current are calculated by Fourier transform to obtain the proportion of fundamental wave and higher harmonics: ,in It is the proportion of harmonic components, indicating the ratio of the total amplitude of higher-order harmonics to the amplitude of the fundamental wave, and is used to measure the degree of harmonic distortion in the system voltage or current. The larger the value, the more significant the influence of higher-order harmonics and the more serious the harmonic distortion of the system. is the highest order number of the higher harmonics, that is, the highest order of the harmonics included in the Fourier transform analysis. is the harmonic number, used to traverse from the 2nd harmonic to Subharmonics, For the The amplitude of the subharmonic, indicating the The intensity of the subharmonic component in the signal, is the amplitude of the fundamental wave, indicating the intensity of the 1st harmonic (i.e. the main frequency component of the signal) in the signal, and is the most important component in Fourier transform.

[0100] Output lighting operation characteristic data, including brightness characteristics, frequency stability, and electrical characteristics (power factor, harmonic content).

[0101] Step S2: Perform real-time state modeling based on the lighting operation characteristic data to obtain an operation state model;

[0102] Specifically, the lighting operation state model is established through the state space expression: , ,in, is the light running state variable, is the input control variable (such as current regulation), , is the noise. The state matrix , , The initial value of is estimated by linear regression based on historical operation data. The operation status model is output to describe the dynamic behavior of the current lighting system.

[0103] Step S3: performing anomaly detection according to the operation status model to obtain operation anomaly detection data;

[0104] Specifically, based on historical data, set the normal range of lighting operating characteristics (such as brightness, flicker frequency, and electrical characteristics): ,in is the lower threshold, indicating the lowest value in the allowed range. For time The observed value at a certain moment represents the measurement data of the system at a certain moment or position (such as brightness, voltage, temperature, etc.). The upper threshold value indicates the highest value in the allowed range, and dynamically adjusts the threshold value to adapt to environmental changes. and running state model Compare and calculate the residual: ,if , it is judged as abnormal, where is the tolerance range. According to the abnormal characteristics, it is classified into brightness abnormality, electrical abnormality, frequency abnormality, etc., and the abnormal time, location and abnormality type are recorded. The operation abnormality detection data is output, including the abnormal classification results and specific numerical features.

[0105] Step S4: Intelligently match the insurance policy according to the abnormal operation detection data to obtain the insurance policy execution data to perform auxiliary operations for the navigation light monitoring.

[0106] Specifically, run anomaly detection data , including the following information, time : The time when the abnormality occurs; abnormal state : Whether it is abnormal (1 means abnormal, 0 means normal); abnormal characteristics : Numerical data of abnormal characteristics, such as brightness fluctuation, current instability, frequency drift, etc.

[0107] Detect features in data based on anomalies , the anomalies are classified into the following types: ,device anomalies, anomalies related to lighting hardware (such as ,bulb damage, voltage fluctuation exceeding the limit, etc.); ,the judgment condition is the characteristic The voltage characteristic or current stability parameter in the system exceeds the set threshold. Environmental abnormality, abnormality related to external environmental conditions (such as high humidity, strong wind, etc. causing unstable lighting); the judgment condition is that the environmental parameters (such as humidity, wind speed) are beyond the normal range. Control abnormality, abnormality related to lighting control logic (such as control signal delay, command loss); the judgment condition is that frequency drift is detected or the switch state is inconsistent with expectations. After each abnormality type is classified, the output , that is, it includes abnormal time, type and detailed characteristics.

[0108] Establish an insurance policy library, where each policy provides different response plans based on the type of anomaly and the severity of the anomaly. The device anomaly policy includes minor anomalies. If the anomaly is a slight fluctuation in the brightness of the bulb, a regular inspection reminder is generated; severe anomalies. If the bulb is damaged, a replacement work order is immediately triggered and the device is marked as unavailable. The environmental anomaly policy includes minor anomalies. If the humidity or wind speed exceeds the limit but does not cause serious impact on the operation, the real-time monitoring frequency is increased; severe anomalies. If the environmental conditions directly threaten the safety of operation (such as strong winds causing the lamp pole to shake), the lights are immediately turned off and an alarm is sounded. The control anomaly policy includes minor anomalies. If a short delay in the control signal is detected, the anomaly is recorded and an optimization suggestion is generated; severe anomalies. If the control signal fails continuously, an emergency shutdown policy is triggered and an alarm is sounded.

[0109] The anomaly detection results are associated with the policy library through matching rules, the severity of the anomaly is calculated according to the degree of deviation of the feature value, the feature value and the normal feature value are calculated, and the policy is matched according to the anomaly type and severity, where the priority is comprehensively evaluated based on the degree of anomaly impact and the policy execution cost. Generate specific policy execution steps, including the following information: execution content such as policy description (such as replacing equipment, strengthening monitoring, etc.); execution priority such as classification based on severity (such as high priority is 1, low priority is 2); execution schedule such as arranging specific time according to the anomaly type and current operating status; responsibility allocation such as designating responsible personnel or departments (such as maintenance engineers, system administrators). After each policy execution, record the execution results (such as whether the replacement is successful and whether the anomaly is eliminated).

[0110] Preferably, step S1 specifically includes:

[0111] Step S11: collecting lighting operation data through a distributed sensor array;

[0112] Specifically, the distributed sensor array consists of multiple sensor nodes, each of which is installed at a key location of the navigation light. The sensor collects various physical parameters of the light operation, such as current ( ),Voltage( ),brightness( ),temperature( ) etc. The sensor sampling interval is set to 0.1 seconds, and the time series data is collected , for The current data at the moment and the rest are the timing data of voltage, brightness and temperature, which are transmitted to the data center through the wireless communication module. The data center caches and verifies the transmitted data to ensure data integrity.

[0113] Step S12: extracting non-smoothed parameters from the lighting operation data to obtain non-smoothed lighting operation parameter data;

[0114] Specifically, the variational analysis method is used to decompose the lighting operation data into smooth and non-smooth parts to identify the non-stationary change interval. The points in the non-smooth interval are analyzed one by one, and the non-smooth points with significant changes are screened out by calculating the difference values ​​of adjacent points. The distribution and change trend of non-smooth points are quantified to generate non-smooth parameter data, including the location, amplitude and change rate of non-smooth points.

[0115] Step S13: performing non-smooth feature extraction according to the lighting operation non-smooth parameter data to obtain first lighting operation feature data;

[0116] Specifically, according to the non-smooth parameter data, the distribution density, change rules and dynamic characteristics of the non-smooth points are analyzed, and characteristic values ​​such as local discontinuity and change amplitude are extracted. Cluster analysis is performed on the non-smooth points by location or time to identify concentrated distribution areas and significant change areas. The first lighting operation characteristic data includes the statistical characteristics and clustering results of the non-smooth points.

[0117] Step S14: extracting extreme skewness features from the lighting operation data to obtain lighting operation extreme skewness feature data;

[0118] Specifically, the extreme points in the lighting operation data are identified through sliding windows and least squares fitting. The extreme point data are statistically analyzed and the skewness value is calculated to reflect the symmetry and tail characteristics of the distribution. The concentration and deviation of the extreme points are further quantified based on the skewness characteristics. The extreme skewness characteristic data of the lighting operation describes the extreme characteristics of the operation status.

[0119] Step S15: extracting the kurtosis change feature of the lighting operation data to obtain the lighting operation kurtosis change feature data;

[0120] Specifically, the weighted moving average method is used to perform piecewise smoothing on the operating data. The kurtosis value of each segment of data is calculated, the sharpness of the distribution is analyzed, and the significant change area is detected. The kurtosis change is differentially analyzed, the change rate is quantified, and trend analysis is performed in combination with multiple time scales. The characteristic data of the kurtosis change of the lighting operation is obtained to describe the peak characteristics and dynamic changes of the operating data.

[0121] Step S16: Acquire real-time environmental condition data;

[0122] Specifically, the external conditions such as temperature, humidity, wind speed, etc. are collected in real time through environmental sensors. The environmental data is cleaned and standardized to ensure that it is synchronized with the time series of the lighting operation data. The real-time environmental condition data is obtained and used to correct the lighting operation characteristics.

[0123] Step S17: generating a feature vector for the light operation extreme skewness feature data and the light operation kurtosis change feature data according to the real-time environmental condition data to obtain second light operation feature data;

[0124] Specifically, the extreme skewness and kurtosis change characteristics are corrected according to the real-time environmental conditions. For example, under high humidity conditions, the normal range of brightness fluctuation is relaxed. The corrected eigenvalues ​​are vectorized to form a unified feature representation to ensure that the environmental impact and operating characteristics are included. The second lighting operation feature data is obtained to describe the interactive characteristics of the lighting operation status and environmental conditions.

[0125] Step S18: aligning the first lighting operation characteristic data and the second lighting operation characteristic data to obtain the lighting operation characteristic data.

[0126] Specifically, the first feature data and the second feature data are aligned in time axis to ensure that the features from different sources correspond to each other at the same time point, and are spliced ​​according to the feature dimensions to generate a feature vector that fully describes the operating status of the light.

[0127] Preferably, the non-smoothing parameter extraction is specifically as follows:

[0128] Perform variational decomposition on the lighting operation data to obtain non-smooth variation interval data;

[0129] Specifically, by collecting real-time data of lighting operation, including changes in brightness, current and voltage, the data is divided into a stable change part and a sudden change part using an analysis method based on continuous changes. By observing the trend of data changes, when the continuity characteristics of the data in a certain period of time are broken, such as a sudden increase in the amplitude of brightness changes or a sharp fluctuation in current, the period is marked as a non-smooth change interval. The identification of non-smooth change intervals is based on the quantitative analysis of data continuity, aiming to find time periods that significantly deviate from the stable state.

[0130] Specifically, run data on the lights Perform variational decomposition to separate the data into smooth parts and the non-smooth part : . By optimizing the objective function, the following expression is minimized: ,in To optimize the value of the objective function, the lighting operation data is represented Middle smooth part The total cost of fitting error and smoothness of variation, is the time index, which indicates the sampling points, is the original observation value of the light operation data, indicating The actual data value at a time point, For the The smoothed data value fitted at each time point is obtained by optimizing the objective function, which means it is as close to the actual data as possible. While maintaining a smooth signal, For the The smoothed data value fitted at each time point, is the smoothing weight factor. The non-smooth change interval data of the non-smooth part is obtained.

[0131] Perform non-smooth point detection based on non-smooth change interval data to obtain non-smooth point data for lighting operation;

[0132] Specifically, within the non-smooth change interval, the data change characteristics are analyzed point by point to identify key change points, such as the point where the brightness drops rapidly or the peak point of the voltage. These points are called non-smooth points. By comparing the degree of change between adjacent data points one by one, when the degree of change of a point far exceeds the average change of the previous and next data points, it is determined to be a non-smooth point. The detection of non-smooth points is completed by a point-by-point comparison method to ensure that the specific location of the data mutation can be accurately captured.

[0133] Perform differential characteristic analysis on the non-smooth point data of lighting operation to obtain the characteristic data of non-smooth point change rate;

[0134] Specifically, the change trend of non-smooth points is further analyzed to observe the change rate characteristics of these points. Specifically, by comparing the change amplitudes of non-smooth points one by one, the change rate of these points can be quantified. For example, if a certain brightness mutation point switches from a high brightness state to a low brightness state in a very short time, its change rate will be significantly higher than other change points. Through this rate analysis, the dynamic characteristics of non-smooth points can be captured more accurately.

[0135] Perform time series gap analysis on the non-smooth point change rate characteristic data to obtain time interval distribution characteristic data;

[0136] Specifically, based on the change rate analysis of non-smooth points, the time interval distribution characteristics between these points are further studied. For example, the time interval between every two non-smooth points is calculated, and the distribution law of these intervals is statistically analyzed. Through this analysis, the distribution density and regularity of non-smooth points can be understood. For example, if the time interval between non-smooth points gradually shortens, it indicates that the lighting system is experiencing a gradual deterioration.

[0137] Perform local discontinuity quantization according to time interval distribution characteristic data to obtain local non-smooth quantized data;

[0138] Specifically, based on the time interval distribution, the continuity characteristics of the lighting system within the local time range are analyzed. If the non-smooth points in a certain area are densely distributed, the area is marked as a local discontinuity area, and its discontinuity is quantitatively analyzed. For example, the number of non-smooth points in the area and the degree of fluctuation of the time interval are counted to obtain a quantitative result reflecting the local discontinuity.

[0139] Specifically, more importantly, gamma distribution fitting is performed on the time interval distribution characteristic data to obtain abnormal time interval distribution data; variance feature extraction and range feature extraction are performed on the abnormal time interval distribution data to obtain time interval variance feature data and time interval range feature data respectively; cluster calculation is performed on the time interval variance feature data and the time interval range feature data to obtain variance feature clustering data and range feature clustering data respectively; feature fusion is performed on the variance feature clustering data and the range feature clustering data to obtain interval feature clustering data; local discontinuity identification is performed on the interval feature clustering data to obtain local discontinuity data; complexity quantification is performed on the local discontinuity data to obtain local non-smooth quantized data.

[0140] Collect the time interval distribution characteristics of the light operation data, including the time intervals of changes in light brightness, frequency, etc. Next, model these time interval data through the gamma distribution fitting method, and the fitting results reflect the distribution pattern of the data. The gamma distribution model can accurately capture the abnormal fluctuations and skewness characteristics in the time interval data, especially the long-tail distribution part of the data, and obtain abnormal time interval distribution data. Extract the variance and range characteristics of the abnormal time interval distribution data. The variance feature calculates the volatility of the data in the time interval and reflects the degree of dispersion between data points; while the range feature reveals the extreme fluctuation of the data by calculating the difference between the maximum and minimum values ​​in the data. These two features can comprehensively describe the change pattern of the time interval data. The variance feature data and range feature data extracted separately provide two different angles of time interval fluctuation analysis. The variance feature data and range feature data are analyzed by clustering calculation. Clustering algorithms (such as K-means or DBSCAN) group the variance data and range data of the time interval, and the data in each group have similar characteristics. Through this operation, abnormal patterns or change trends in the data can be identified. Variance feature clustering data reflects the fluctuation pattern of data in different time periods, while range feature clustering data reveals specific intervals with large changes in time intervals. Interval feature clustering data is obtained by fusing variance feature clustering data and range feature clustering data. Feature fusion uses weighted average, splicing or other methods to integrate the clustering results of these two features. Based on interval feature clustering data, local discontinuity identification is performed, and anomaly detection techniques, such as distance measurement or threshold setting methods, are used to identify time periods that deviate significantly from the normal pattern. In these time periods, the time interval changes more drastically or shows sudden changes. The identified local discontinuity data will serve as a signal to indicate the existence of irregular fluctuations or mutations in the system. The complexity of "local discontinuity data" is quantified. The complexity quantification methods include calculating the Hurst index or fractal dimension, which can quantify the complexity of data changes. The Hurst index can reveal whether the data presents a long-term dependency, while the fractal dimension can measure the nonlinear and multi-scale characteristics of data changes. By quantifying the complexity of local discontinuity data, the local non-smooth quantified data further reflects the non-smooth and discontinuous change characteristics existing in the lighting operation data.

[0141] Perform spatial variation analysis based on the light source parameter data corresponding to the lighting operation data and the local non-smoothed quantized data to obtain the lighting operation spatial variation feature data;

[0142] Specifically, the distribution characteristics of non-smooth points in space are analyzed in combination with the light source parameters of the lighting system, such as the light position, brightness distribution, etc. By focusing on the analysis of specific areas of the light position, such as whether the distribution of non-smooth points in certain areas is more dense or more significant, the abnormal change characteristics of the lighting system in space can be identified. For example, if the brightness of lights in a certain area is frequently abnormal, it reflects that the environmental impact of this area is greater.

[0143] Perform frequency domain mutation extraction based on the spatial variation feature data of lighting operation to obtain frequency domain mutation feature data of lighting operation;

[0144] Specifically, the results of spatial variation analysis are further processed to study the mutation characteristics of lighting operation data in the frequency domain. For example, the frequency characteristics of brightness fluctuations are observed to identify whether there are sudden high-frequency or low-frequency components. Through this analysis, the spatial variation information can be converted into feature expressions in the frequency domain, revealing the periodic or sudden problems of system operation.

[0145] The non-smooth point change rate characteristic data, local non-smooth quantization data and lighting operation frequency domain mutation characteristic data are integrated to obtain the lighting operation non-smooth parameter data.

[0146] Specifically, all the characteristic data obtained from the analysis are integrated, including the change rate of non-smooth points, the quantitative results of local discontinuities, and the frequency domain mutation characteristics. These data are combined into a non-smooth parameter data set to fully describe the non-smooth characteristics of the lighting operation state.

[0147] Preferably, the non-smooth feature extraction is specifically:

[0148] Density clustering calculation is performed according to the non-smooth parameter data of the light operation to obtain non-smooth density clustering data;

[0149] Specifically, the non-smooth parameter data of the lighting operation, including time, position and change amplitude, are input as the characteristic values ​​of the data points. For example, the characteristics of the data points include brightness fluctuation, voltage deviation and current time. According to the spatial and temporal distribution of the non-smooth points, the data points are clustered according to their density. The high-density areas are marked as clusters, and the sparse areas are marked as backgrounds. During the clustering process, it is determined whether the number of points in a certain area reaches the specified threshold. If it does, a cluster is formed. Each cluster represents a group of adjacent non-smooth points, which correspond to local abnormal characteristics in the same lighting system. By analyzing the center position, number of points and range of each cluster, the non-smooth area with dense clustering is identified. Non-smooth density clustering data is obtained, including the position, size and characteristic distribution of each cluster.

[0150] Generate characteristic spectrum for non-smooth density clustering data to obtain non-smooth characteristic spectrum data;

[0151] Specifically, for each density cluster, its time series data is converted into a frequency domain signal to extract the frequency characteristics of its changing pattern. For example, the periodicity and suddenness of light brightness changes are analyzed. The time series data in the cluster is frequency decomposed to identify the main frequency component and the secondary frequency component, and their energy distribution is calculated. The more frequency components there are, the more complex the changing pattern of the light operation; the more obvious the main frequency, the stronger the periodicity of the abnormal light operation. The following features are extracted from the spectrum: main frequency value, frequency concentration and energy distribution trend. These features are used to describe the volatility and stability of light operation. Non-smoothed feature spectrum data are obtained, including frequency distribution diagram and main eigenvalues.

[0152] Perform non-smooth region segmentation according to local non-smooth quantized data to obtain non-smooth feature region data;

[0153] Specifically, based on the local non-smooth quantization data, the non-smooth characteristics of the lighting operation are segmented according to the continuity of space or time. When the number of non-smooth points in a certain area is higher than the specified threshold, the area is marked as a non-smooth feature area. For each non-smooth feature area, its size, shape, distribution density, etc. are calculated. The larger the area and the higher the density, the more significant the non-smooth characteristics of the area. The spatial or temporal associations between different non-smooth feature areas are analyzed, for example, whether there are multiple areas concentrated in a specific lighting system location. The non-smooth feature area data are obtained, including the range, density and correlation information of each area.

[0154] Cross-scale feature synthesis is performed according to the non-smoothed feature spectrum data and the non-smoothed feature region data to obtain first light operation feature data.

[0155] Specifically, the non-smooth feature spectrum data and the non-smooth feature regional data are aligned in time or space. For example, the main frequency in the spectrum is associated with the time range of the regional data to form a cross-modal feature correspondence. The aligned features are multi-scale fused to analyze the changing trends and correlations of the non-smooth characteristics in different time periods and spatial regions. The fusion results can reveal the characteristic scale and diffusion pattern of the anomaly. The spectrum features (such as main frequency, energy distribution) and the regional features (such as area, density) are weighted and combined to generate feature data that comprehensively describes the non-smooth characteristics of the lighting operation. The first lighting operation feature data is obtained, including comprehensive features across scales, which are used to further analyze the lighting operation status.

[0156] Preferably, the extreme skewness feature extraction is specifically:

[0157] Perform non-uniform window segmentation on the lighting operation data to obtain lighting operation segmentation data;

[0158] Specifically, the data is divided into non-uniform windows according to the changing trend of the lighting operation data. When the change amplitude of the brightness, current or voltage in a certain area exceeds the preset threshold, the area is marked as a segment starting point until the change amplitude returns to a stable state to form a complete segment. The length of each window is related to the intensity of the change of the lighting operation data. The window in the area of ​​​​drastic changes is shorter, and the window in the stable area is longer. Each segmented data includes the time range and specific values ​​of the operating parameters, such as the brightness change range and current fluctuations. The segmented data set records the time range and operating parameters of each segment.

[0159] Perform sliding least squares extreme value detection on the lighting operation segmented data to obtain the lighting operation extreme value point data;

[0160] Specifically, in each segment, a sliding window of fixed length is set, and the data points in the window are analyzed point by point. For the data in each sliding window, a quadratic curve is fitted to capture the local change trend, and the local extreme value is identified by the position of the extreme point of the fitting curve. By comparing the extreme points of adjacent windows, the extreme points with small fluctuations or repetitions are removed, and significant extreme values ​​are retained. The extreme point data is obtained, including the time, position and characteristic value of the extreme point.

[0161] Perform extreme value distribution fitting based on the extreme value point data of the lighting operation to obtain extreme value distribution fitting data;

[0162] Specifically, statistical analysis is performed on the data of extreme points to observe their numerical distribution characteristics, including whether there is an obvious trend of deviation from the central value. Using historical data and real-time data, a probability distribution model of extreme points is established to fit the mean, variance and tail characteristics of the distribution to ensure that the fitting results can reflect the operating rules of the lighting system. The key features of the fitted distribution are extracted, such as the degree of deviation of the distribution, the tail length and the offset of the central value. The fitting data of the extreme value distribution is obtained, including the distribution model and its key parameters.

[0163] The skewness is calculated based on the extreme value distribution fitting data to obtain the skewness characteristic data of the lighting operation;

[0164] Specifically, skewness reflects the symmetry of the distribution. By analyzing the deviation of the distribution of extreme points, the skewness characteristics of the lighting operation data are calculated. The distribution of extreme points is calculated one by one to determine the direction and magnitude of their deviation from the center value. If the data points are more biased to one side, the skewness is high. When the skewness is positive, it means that the extreme points are more concentrated on the right side of the distribution; when the skewness is negative, it means that the extreme points are more concentrated on the left side of the distribution. The skewness characteristic data is obtained to describe the symmetry and central tendency of the lighting operation data.

[0165] According to the skewness characteristic data of the lighting operation, the extreme skewness characteristic data is extracted to obtain the extreme skewness characteristic data;

[0166] Specifically, extreme skewness describes the degree of deviation of data within the extreme value range, which is a deeper analysis of the skewness feature. Combine the skewness feature and the location data of the extreme value points to analyze the degree of deviation and concentration of the distribution tail. For example, if the extreme value points are concentrated on one side of the tail, the extreme skewness is significant. The extreme skewness is quantified into a numerical value, reflecting the concentrated deviation of the extreme value points in the lighting operation data. The extreme skewness feature data is obtained, including the quantitative results of the degree of deviation and the concentration characteristics.

[0167] Multi-scale feature aggregation is performed based on the extreme skewness feature data to obtain the extreme skewness feature data of lighting operation.

[0168] Specifically, at different time scales, the changing trend of the extreme skewness characteristics is observed. For example, extreme skewness in a short period of time reflects instantaneous anomalies, while extreme skewness in a long period of time reveals systematic trends. The extreme skewness characteristics of different time scales are weighted averaged, and the short-term characteristics and long-term characteristics are combined to generate comprehensive extreme skewness characteristics. In the aggregation process, significantly deviated extreme points are given priority, while abnormal or irrelevant data points are filtered out. Comprehensive lighting operation extreme skewness characteristic data is obtained to describe the extreme deviation of the system at different time scales.

[0169] Preferably, the kurtosis change feature extraction is specifically as follows:

[0170] Perform weighted sliding average segmentation on the lighting operation data to obtain smooth segmentation data of lighting operation;

[0171] Specifically, the lighting operation data is segmented according to the time series, and the data points of each segment are smoothed by weighted sliding average. The weight distribution method gradually decreases according to the distance between the time point and the current time, so that the recent data contributes more to the smoothing result. In each time period, the original data is smoothed to eliminate noise and random fluctuations, and only the trend change characteristics of the data are retained. The lighting operation smooth segmented data is obtained, and each segment contains the characteristics such as smoothed brightness, current or voltage, which is used for kurtosis calculation.

[0172] Perform preliminary kurtosis calculation on the smooth segmented data of lighting operation to obtain preliminary kurtosis data;

[0173] Specifically, kurtosis is used to measure the sharpness of data distribution. In each smooth segment, the kurtosis of the distribution is calculated to reflect whether the segment has a peak characteristic. Statistical analysis is performed on the smooth segment data to extract its local mean, standard deviation and other characteristics, and calculate the kurtosis value of the distribution. The kurtosis value of a sharp distribution is higher, while the kurtosis value of a flat distribution is lower. Preliminary kurtosis data, including the kurtosis value of each segment of data, is used for anomaly detection.

[0174] Perform local kurtosis anomaly detection based on preliminary kurtosis data to obtain local kurtosis anomaly data;

[0175] Specifically, if the kurtosis value of a certain segment of data deviates significantly from the normal range, it is marked as a local anomaly. A threshold is set according to the distribution range of historical kurtosis data, and the preliminary kurtosis data is analyzed segment by segment. If the kurtosis value of a certain segment of data is higher or lower than the set range, it is marked as an anomaly. The abnormal data is classified according to the magnitude of the anomaly, such as slight anomaly, significant anomaly, etc. The local kurtosis anomaly data is obtained, including the time period when the anomaly occurs and the corresponding kurtosis value.

[0176] Perform time series kurtosis difference based on local kurtosis anomaly data to obtain kurtosis change rate data;

[0177] Specifically, the kurtosis data of continuous time periods are differentially processed to quantify the rate of change of kurtosis. Time periods with high change rates reflect sudden abnormalities in the operation of lights. By calculating the mean and range of the differential values, the trend and amplitude of data changes are analyzed. High-value areas in the differential results are marked to capture the dynamics of abnormal occurrence and recovery process. The kurtosis change rate data is obtained to describe the speed and trend of the change of the kurtosis value in the time dimension.

[0178] Perform multi-scale kurtosis change analysis based on kurtosis change rate data to obtain multi-scale kurtosis change data;

[0179] Specifically, the kurtosis change rate data is analyzed at different time scales, such as short time windows to capture instantaneous changes, and long time windows to capture systematic trends. The short-term, medium-term and long-term kurtosis change rate data are statistically analyzed respectively, and the mean, extreme value and fluctuation range of the change at each time scale are extracted. The analysis results of each time scale are compared to observe the regularity of the change rate and the abnormal concentration area at different scales. Multi-scale kurtosis change data are obtained, including the kurtosis change characteristics of each time scale.

[0180] The multi-scale kurtosis change data is fitted with kurtosis Gaussian distribution to obtain kurtosis Gaussian distribution data;

[0181] Specifically, in general, due to the stability of power amplifier efficiency, the multi-scale kurtosis change data conforms to the Gaussian distribution, and the mean and standard deviation of the distribution are determined by statistical fitting methods. If the tail characteristics of the distribution deviate significantly from the standard Gaussian distribution, it is marked as an abnormal distribution. The fitting parameters are used to quantify the central tendency and dispersion of the distribution. The fitted Gaussian distribution is compared with the original data to verify the accuracy of the fit. The kurtosis Gaussian distribution data is obtained, including the mean, standard deviation and outliers of the distribution.

[0182] Based on the multi-scale kurtosis change data and the kurtosis Gaussian distribution data, feature vectorization is performed to obtain the lighting operation kurtosis change characteristic data.

[0183] Specifically, the multi-scale kurtosis change data and the kurtosis Gaussian distribution data are fused to generate a feature vector describing the kurtosis change of the light operation. The kurtosis change characteristics in each time period are quantified, and key feature values ​​are extracted, such as the mean of the change rate, the degree of deviation from the Gaussian distribution, and the distribution of abnormal areas. All features are combined into a vector form to provide a systematic description. The characteristic data of the kurtosis change of the light operation is obtained, which contains a comprehensive description of the light operation status, which is used for state modeling and anomaly detection.

[0184] Preferably, step S2 specifically includes:

[0185] Step S21: Obtain basic data of navigation lights;

[0186] Specifically, the specifications, positions and corresponding electrical parameter data of the navigation lights are obtained.

[0187] Step S22: constructing a digital twin model according to the basic data of the navigation light to obtain a digital twin model of the navigation light;

[0188] Specifically, the physical design parameters of the navigation lights (such as light structure, installation location, etc.) are used to build a geometric model to simulate the spatial distribution and connection relationship of the lighting system. The dynamic model of the navigation light system is established based on the collected operation data. For example, the dynamic behavior of the light operation is defined using the relationship between power, current and voltage. The collected external environmental data is introduced into the model to build an interactive relationship between the impact of the environment on the lighting system. The model simulates the impact of changes in environmental conditions on the operating status. The digital twin model of the navigation light is obtained, which can reflect the operating status of the navigation light system in real time and predict the future status.

[0189] Step S23: dividing the data set according to the lighting operation characteristic data to obtain lighting operation training input data and lighting operation target output data;

[0190] Specifically, the model input data set is constructed using the real-time characteristic data of the lighting operation, such as brightness, current, and voltage. The target output data is set, such as the operating status label of the lighting (normal or abnormal) or the specific prediction value (brightness attenuation value). The label is collected through a preset database or manually set during the processing. The data is divided into a training set and a test set in chronological order, usually 80% as training data and 20% as test data. Ensure that the training data covers all types of operating states. The lighting operation training input data and target output data are used for model construction and evaluation, respectively.

[0191] Step S24: constructing an input gate according to the lighting operation training input data to obtain lighting operation input gate data;

[0192] Specifically, the input gate filters and transforms the input features. By weighted calculation of the importance of each feature, the original input is mapped to the implicit space of the model. Weights are assigned according to the importance of the input features, and features that contribute less to the target output are filtered out. The selected features are nonlinearly transformed to generate input gate data. The light operation input gate data is obtained to provide high-quality input for the construction of the model hidden layer.

[0193] Step S25: constructing a hidden layer according to the lighting operation input gate data to obtain lighting operation hidden layer data;

[0194] Specifically, the hidden layer converts the input gate data into a deeper feature representation through nonlinear calculations, thus enhancing the model's expressiveness. Using a multi-layer neural network structure, the hidden layer output is calculated layer by layer, and each layer output depends on the input data and weight parameters of the previous layer. The hidden layer can capture the complex relationship in the input data. The hidden layer data of the light operation is obtained, which contains a deep feature representation.

[0195] Step S26: constructing an output gate according to the lighting operation hidden layer data to obtain lighting operation output gate data;

[0196] Specifically, the output gate maps the features of the hidden layer to the target output space, and calculates the final output through weights and biases. The hidden layer output data is linearly transformed and mapped to the target range (such as probability values ​​in classification tasks or numerical predictions in regression tasks) through an activation function. The light operation output gate data is obtained to explain the model output.

[0197] Step S27: iteratively train the lighting operation output gate data using the lighting operation target output data to obtain a lighting operation deep model;

[0198] Specifically, the model error is calculated based on the output gate data and the target output data, and the loss function is such as the cross entropy loss of the classification task or the mean square error of the regression task. The model parameters are adjusted using the gradient descent method, and the loss function is minimized through multiple rounds of iterations. The training set data is traversed multiple times (epochs), and the model weight parameters are updated in each round until the loss function converges. The deep model of lighting operation is obtained, which can accurately predict the operating status of the lighting system.

[0199] Step S28: Fuse the lighting operation deep model and the navigation light digital twin model to obtain an operation status model.

[0200] Specifically, the simulation capability of the digital twin model is combined with the feature learning capability of the deep model. For example, the digital twin model provides real-time environment and physical information, and the deep model provides data-driven state prediction. The outputs of the two models are combined in a weighted average or cascade manner to generate an operating state model. The operating state model is obtained, which integrates the digital twin and deep learning capabilities and can comprehensively monitor and predict the operating state of the lighting.

[0201] Preferably, step S3 specifically includes:

[0202] Step S31: outputting the operating state characteristics according to the operating state model to obtain operating state characteristic data;

[0203] Specifically, the real-time data of the lighting operation (such as brightness, current, voltage, temperature and humidity, wind speed, etc.) is input through the operation status model. The model calculates the output feature value based on the input data to describe the operation status of the system. The operation status characteristics are divided into static characteristics and dynamic characteristics. Static characteristics describe the state value of the system at the current moment, such as the average brightness and current stability; dynamic characteristics describe the trend of the system operation status over time, such as the brightness change rate and voltage fluctuation amplitude.

[0204] Step S32: performing multimodal feature aggregation according to the running state feature data to obtain multimodal fusion feature data;

[0205] Specifically, multimodal features include data from different sources or types, for example, physical modes: brightness, current, voltage. Environmental modes: temperature, humidity, wind speed. Time modes: short-term features and long-term features. Standardize the data of each modality to ensure that different features have the same scale. For example, normalize brightness, current, temperature and humidity to the same range. Assign weights according to the importance of the features and perform weighted fusion of different modal features. For example, calculate the weight coefficient of each modality and set it according to the volatility or abnormal contribution of the modality in historical data. Generate fused features in the form of weighted summation or concatenation vector. The aggregated multimodal fusion feature data should be representative and able to fully describe the operating status of the system. Obtain multimodal fusion feature data to provide input for environmental condition threshold calculation.

[0206] Step S33: Dynamically calculate the environmental condition threshold based on the multimodal fusion feature data to obtain operation anomaly detection data.

[0207] Specifically, the anomaly detection threshold is dynamically adjusted according to historical data and real-time environmental parameters to ensure that the threshold can adapt to different operating conditions. For example, under high humidity or strong wind conditions, the brightness fluctuation increases, and the brightness abnormality threshold needs to be relaxed. Under low temperature conditions, the current stability decreases, and the current-related threshold needs to be adjusted. The basic threshold is set to set the basic threshold according to the normal range of the multimodal fusion feature data. The environmental correction combines the current environmental data to dynamically adjust the basic threshold. For example: dynamic threshold = basic threshold + f (environmental parameter), where f (environmental parameter) is the correction amount of the environment to the threshold, and the function expression is a weighted function or a preset database for parameter matching. The multimodal fusion feature data is compared with the dynamic threshold item by item. If a feature exceeds the threshold range, it is marked as abnormal. The detected anomalies are classified according to the severity, such as minor anomalies and severe anomalies, and the time, feature value and environmental conditions of the anomaly are recorded. The operation anomaly detection data is obtained, including the label, timestamp and detection result of the abnormal feature.

[0208] Preferably, step S4 is specifically:

[0209] Step S41: mapping is performed according to the operation abnormality detection data and the preset abnormal event library to obtain abnormal event mapping data;

[0210] Specifically, the input operation anomaly detection data includes abnormal characteristic values, occurrence time, location and environmental conditions. Each record describes the current specific abnormal condition of the system. The abnormal event library contains detailed descriptions of known anomalies, trigger conditions and corresponding processing measures. Each abnormal event record includes the abnormality type (such as damaged bulb, unstable current, decreased brightness, etc.); abnormal threshold (used to match operation anomaly detection data); event level (such as minor, medium, severe). According to the characteristic value in the operation anomaly detection data, it is matched with the trigger conditions in the abnormal event library: if the characteristic value exceeds the threshold range, it is judged as the corresponding abnormal type. If multiple abnormal types are matched, the matching degree of each type is recorded. Abnormal event mapping data includes the mapping relationship between the operation anomaly detection data and the abnormal event library, including the abnormality type, event level and matching degree.

[0211] Step S42: sorting the processing priorities according to the abnormal event mapping data to obtain abnormal event priority data;

[0212] Specifically, the priority of each abnormal event is determined according to the severity, impact scope and processing complexity of the abnormal event. The sorting rules include severity, the higher the abnormality level, the higher the priority; impact scope, abnormalities involving multiple lighting points or key areas have higher priority; environmental conditions, abnormalities occurring in harsh environments have higher priority. Score each abnormal event in the mapped data, and calculate the priority score based on various factors. Obtain the priority data of abnormal events, sort them from high to low according to priority, including detailed information and priority score of each event.

[0213] Step S43: Intelligently match the insurance policy according to the abnormal event priority data to obtain the insurance policy execution data to perform auxiliary operations for navigation light monitoring.

[0214] Specifically, the policy library contains corresponding processing solutions for different types of exceptions. Each policy includes applicable exception types; processing measures (such as notifying maintenance, replacing equipment, turning off lights, etc.); and response time requirements. According to the priority data of abnormal events, select the best policy from the insurance policy library, with matching priority. According to the type and level of abnormality, give priority to the adapted policy; time requirements, consider the response time of policy execution, and match it with the urgency of the abnormality; resource conditions, ensure that the selected policy is executable under the current resource conditions. Generate a specific processing strategy for each abnormal event, including processing content (such as replacing light bulbs, adjusting voltage, etc.); processing steps (specific execution process); time schedule (such as immediate execution, regular inspection, etc.). Obtain insurance policy execution data, including the processing strategy and execution plan for each abnormal event, to provide guidance for navigation lighting monitoring auxiliary operations.

[0215] Preferably, the present application further provides a failsafe-based navigation light monitoring system, which is used to execute the failsafe-based navigation light monitoring method as described above. The failsafe-based navigation light monitoring system includes:

[0216] A lighting operation feature extraction module is used to collect lighting operation data through a distributed sensor array, and extract key parameters of the lighting operation data to obtain lighting operation feature data;

[0217] A real-time state modeling module is used to perform real-time state modeling based on the light operation characteristic data to obtain an operation state model;

[0218] The lighting operation anomaly detection module is used to perform anomaly detection according to the operation status model and obtain operation anomaly detection data;

[0219] The lighting operation insurance policy module is used to intelligently match the insurance policy according to the operation anomaly detection data and obtain the insurance policy execution data to perform auxiliary operations for navigation lighting monitoring.

[0220] Therefore, from any point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the attached application documents rather than the above description, and it is intended that all changes falling within the meaning and scope of equivalent elements of the application documents are included in the present invention.

[0221] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A navigation light monitoring method based on fail-safe, characterized in that: The following steps are involved: Step S1: collecting lighting operation data through a distributed sensor array, and extracting key parameters of the lighting operation data to obtain lighting operation characteristic data; Step S2: Perform real-time state modeling according to the lighting operation characteristic data to obtain an operation state model; Step S3: performing anomaly detection according to the operation status model to obtain operation anomaly detection data; Step S4: intelligently matching the insurance policy according to the abnormal operation detection data to obtain the insurance policy execution data for performing auxiliary operations of the navigation light monitoring; Step S1 is specifically as follows: Collect lighting operation data through distributed sensor arrays; Extracting non-smoothed parameters of lighting operation data to obtain non-smoothed parameter data of lighting operation; Perform non-smooth feature extraction according to the lighting operation non-smooth parameter data to obtain first lighting operation feature data; Extract extreme skewness features from the lighting operation data to obtain lighting operation extreme skewness feature data; Extract the kurtosis change feature of the lighting operation data to obtain the lighting operation kurtosis change feature data; Get real-time environmental condition data; Correcting the lighting operation extreme skewness characteristic data and the lighting operation kurtosis change characteristic data according to the real-time environmental condition data to obtain corrected data, and generating a characteristic vector for the corrected data to obtain second lighting operation characteristic data; Performing feature alignment on the first lighting operation feature data and the second lighting operation feature data to obtain the lighting operation feature data; The extreme skewness feature extraction is specifically as follows: Perform non-uniform window segmentation on the lighting operation data to obtain lighting operation segmentation data; Perform sliding least squares extreme value detection on the lighting operation segmented data to obtain the lighting operation extreme value point data; Perform extreme value distribution fitting based on the extreme value point data of the lighting operation to obtain extreme value distribution fitting data; The skewness is calculated based on the extreme value distribution fitting data to obtain the skewness characteristic data of the lighting operation; According to the skewness characteristic data of the lighting operation, the extreme skewness characteristic data is extracted to obtain the extreme skewness characteristic data; Multi-scale feature aggregation is performed based on the extreme skewness feature data to obtain the extreme skewness feature data of lighting operation.

2. The method according to claim 1, characterized in that The non-smoothing parameter extraction is as follows: Perform variational decomposition on the lighting operation data to obtain non-smooth variation interval data; Perform non-smooth point detection based on non-smooth change interval data to obtain non-smooth point data for lighting operation; Perform differential characteristic analysis on the non-smooth point data of lighting operation to obtain the characteristic data of non-smooth point change rate; Perform time series gap analysis on the non-smooth point change rate characteristic data to obtain time interval distribution characteristic data; Perform local discontinuity quantization according to time interval distribution characteristic data to obtain local non-smooth quantized data; Perform spatial variation analysis based on the light source parameter data corresponding to the lighting operation data and the local non-smoothed quantized data to obtain the lighting operation spatial variation feature data; Perform frequency domain mutation extraction based on the spatial variation feature data of lighting operation to obtain frequency domain mutation feature data of lighting operation; The non-smooth point change rate characteristic data, local non-smooth quantization data and lighting operation frequency domain mutation characteristic data are integrated to obtain the lighting operation non-smooth parameter data.

3. The method according to claim 2, characterized in that The non-smooth feature extraction is specifically as follows: Density clustering calculation is performed according to the non-smooth parameter data of the light operation to obtain non-smooth density clustering data; Generate characteristic spectrum for non-smooth density clustering data to obtain non-smooth characteristic spectrum data; Perform non-smooth region segmentation according to local non-smooth quantized data to obtain non-smooth feature region data; Cross-scale feature synthesis is performed according to the non-smoothed feature spectrum data and the non-smoothed feature region data to obtain first light operation feature data.

4. The method according to claim 1, characterized in that: The kurtosis change feature extraction is specifically as follows: Perform weighted sliding average segmentation on the lighting operation data to obtain smooth segmentation data of lighting operation; Perform preliminary kurtosis calculation on the smooth segmented data of lighting operation to obtain preliminary kurtosis data; Perform local kurtosis anomaly detection based on preliminary kurtosis data to obtain local kurtosis anomaly data; Perform time series kurtosis difference based on local kurtosis anomaly data to obtain kurtosis change rate data; Perform multi-scale kurtosis change analysis based on kurtosis change rate data to obtain multi-scale kurtosis change data; The multi-scale kurtosis change data is fitted with kurtosis Gaussian distribution to obtain kurtosis Gaussian distribution data; Based on the multi-scale kurtosis change data and the kurtosis Gaussian distribution data, feature vectorization is performed to obtain the lighting operation kurtosis change characteristic data.

5. The method according to claim 1, characterized in that Step S2 is specifically as follows: Get basic data of navigation lights; The digital twin model is constructed based on the basic data of the navigation lights to obtain the digital twin model of the navigation lights; Divide the data set according to the lighting operation feature data to obtain lighting operation training input data and lighting operation target output data; Construct an input gate according to the lighting operation training input data to obtain lighting operation input gate data; Construct a hidden layer according to the light operation input gate data to obtain the light operation hidden layer data; The output gate is constructed according to the hidden layer data of the light operation to obtain the output gate data of the light operation; Iteratively train the lighting operation output gate data using the lighting operation target output data to obtain a lighting operation deep model; The lighting operation deep model and the navigation light digital twin model are integrated to obtain the operation status model.

6. The method according to claim 1, characterized in that Step S3 is specifically as follows: Outputting the operating state characteristics according to the operating state model to obtain operating state characteristic data; Perform multimodal feature aggregation according to the running status feature data to obtain multimodal fusion feature data; The environmental condition threshold is dynamically calculated based on the multimodal fusion feature data to obtain the operation anomaly detection data.

7. The method according to claim 3, characterized in that Step S4 is specifically as follows: Mapping is performed according to the operation anomaly detection data and a preset anomaly event library to obtain anomaly event mapping data; Sort the processing priorities according to the abnormal event mapping data to obtain the abnormal event priority data; Intelligent matching of insurance policies is performed based on the priority data of abnormal events to obtain insurance policy execution data for auxiliary operations of navigation lighting monitoring.

8. A navigation light monitoring system based on fail-safe, characterized in that: Used to execute the navigation light monitoring method based on fail-safe as claimed in claim 1, the navigation light monitoring system based on fail-safe comprises: The light operation feature extraction module is used to collect light operation data through a distributed sensor array; perform non-smooth parameter extraction on the light operation data to obtain light operation non-smooth parameter data; perform non-smooth feature extraction based on the light operation non-smooth parameter data to obtain first light operation feature data; perform extreme skewness feature extraction on the light operation data to obtain light operation extreme skewness feature data; perform kurtosis change feature extraction on the light operation data to obtain light operation kurtosis change feature data; obtain real-time environmental condition data; perform feature vector generation on the light operation extreme skewness feature data and the light operation kurtosis change feature data based on the real-time environmental condition data to obtain second light operation feature data; the first light The operation characteristic data and the second light operation characteristic data are feature aligned to obtain the light operation characteristic data; wherein the extreme skewness feature extraction is specifically as follows: performing non-uniform window segmentation on the light operation data to obtain light operation segmented data; performing sliding least squares extreme value detection on the light operation segmented data to obtain light operation extreme point data; performing extreme value distribution fitting according to the light operation extreme point data to obtain extreme value distribution fitting data; performing skewness calculation according to the extreme value distribution fitting data to obtain light operation skewness characteristic data; performing extreme skewness feature extraction according to the light operation skewness characteristic data to obtain extreme skewness characteristic data; performing multi-scale feature aggregation according to the extreme skewness characteristic data to obtain light operation extreme skewness characteristic data; A real-time state modeling module is used to perform real-time state modeling based on the light operation characteristic data to obtain an operation state model; The lighting operation anomaly detection module is used to perform anomaly detection according to the operation status model and obtain operation anomaly detection data; The lighting operation insurance policy module is used to intelligently match the insurance policy according to the operation anomaly detection data and obtain the insurance policy execution data to perform auxiliary operations for navigation lighting monitoring.

Images