Rail transit variable frequency power supply intelligent operation and maintenance fault early warning method

Abstract

The application discloses a kind of track traffic variable frequency power supply intelligent operation and maintenance fault early warning method, it is related to power electronics and intelligent operation and maintenance technical field, comprising the following steps: setting up multipoint synchronous acquisition device to the voltage, current signal of variable frequency power supply system is high-precision real-time acquisition, and data is digitally encoded processing;The voltage current signal collected is analyzed in frequency domain using a preset spectral analysis algorithm, and the harmonic amplitude and phase characteristic parameters of different orders are extracted.The application realizes the rapid and accurate identification of harmonic anomaly through high-precision synchronous acquisition and spectral analysis;Fusion machine learning model and dynamic threshold adjustment, enhance the adaptability and intelligent identification ability of system to fault mode under multiple working conditions;Introduce trend prediction and graphical early warning platform, realize the early prediction and visual response of fault risk, significantly improve the operation efficiency and the operation safety of track traffic power supply system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power electronics and intelligent operation and maintenance, and particularly relates to a rail transit variable frequency power supply intelligent operation and maintenance fault early warning method. BACKGROUND

[0002] Rail transit variable frequency power supply intelligent operation and maintenance fault early warning refers to, in a rail transit system, using intelligent means (such as sensor collection, edge computing, artificial intelligence algorithms, big data analysis, etc.) to perform real-time monitoring of the running state and continuous analysis of historical data of variable frequency power supply equipment that undertakes key functions such as traction power supply and power conversion, to timely identify potential fault signs, realize early warning and fault positioning, thereby guiding maintenance personnel to carry out targeted maintenance, reducing the rate of sudden failures, improving the reliability and safety of system operation, and ultimately realizing the intelligent, precise and efficient operation and maintenance.

[0003] The prior art has the following disadvantages: In the existing rail transit variable frequency power supply intelligent operation and maintenance fault early warning process, the timely identification of harmonic abnormalities is a far-reaching problem. Harmonics caused by nonlinear loads can cause voltage and current distortion, although the initial impact is not significant, but long-term existence will cause equipment overheating, insulation aging, protection device misoperation and other hidden dangers, and in severe cases may cause power supply system instability, communication interference and even train operation interruption. Due to the weak early characteristics of harmonic faults, it is easy to be confused with normal fluctuations, and is often ignored by traditional monitoring methods, thereby forming a hidden risk in the operation and maintenance system and affecting the safety and reliability of the rail transit system.

[0004] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present application is to provide a rail transit variable frequency power supply intelligent operation and maintenance fault early warning method, which realizes rapid and accurate identification of harmonic abnormalities through high-precision synchronous collection and spectrum analysis, enhances the adaptability and intelligent identification ability of the system to fault modes under multiple working conditions by fusing machine learning models and dynamic threshold adjustment, and introduces trend prediction and graphical early warning platform to realize early judgment and visual response to fault risks, significantly improving operation efficiency and the operation safety of the rail transit power supply system, to solve the problems in the background technology.

[0006] In order to achieve the above-mentioned purpose, the present application provides the following technical scheme: a rail transit variable frequency power supply intelligent operation and maintenance fault early warning method, comprising the following steps:

[0007] A multi-point synchronous acquisition device is arranged to collect voltage and current signals of the variable frequency power supply system in high precision and real time, and to perform digital encoding processing on the data.

[0008] The collected voltage and current signals are analyzed in frequency domain by using a preset spectrum analysis algorithm to extract harmonic amplitude and phase characteristic parameters of different orders.

[0009] An abnormality identification module based on a machine learning model is constructed, and current characteristic parameters are inputted and compared with historical operation data to mine potential harmonic abnormality patterns.

[0010] A dynamic threshold mechanism is introduced, and adaptive identification thresholds are set for each order of harmonic in combination with equipment operation environment, load type and time period factors to improve the accuracy of abnormality identification.

[0011] A time series trend prediction algorithm is used to predict the development trend and potential risk level of abnormal harmonic parameters.

[0012] The judgment result is displayed in real time to an operation and maintenance platform through a graphical interface, and an early warning mechanism is triggered to push diagnosis results and recommended measures to a maintenance personnel terminal.

[0013] Preferably, in the high-precision real-time acquisition step, the acquisition device is a synchronous multi-channel analog-to-digital conversion unit, which is respectively arranged at the input end, the output end and the key load node of the variable frequency power supply system, and the sampling precision is not less than 16 bits and the sampling frequency is not less than 10 kHz.

[0014] The data encoding processing adopts a discrete time window segmentation encoding method to ensure the time alignment of the data of each channel.

[0015] Each data segment is automatically labeled with a time stamp after being collected, sent to a local cache and uploaded to a central processing unit through a CAN bus or an optical fiber channel to ensure high real-time performance and transmission stability of data acquisition.

[0016] Preferably, the spectrum analysis algorithm adopts a joint processing mechanism of windowed Fourier transform and short-time Fourier transform to perform multi-scale frequency domain analysis on the sampled signals, and the specific steps are as follows:

[0017] The original sampled signal is windowed and divided by using a Hamming window to reduce the spectrum leakage phenomenon.

[0018] The amplitude and phase of the fundamental wave and 2-25th order harmonics are extracted from the window function output data based on fast Fourier transform (FFT) operation.

[0019] Each order of harmonic is subjected to mean value normalization in multiple time windows, and a harmonic feature matrix is constructed.

[0020] Preferably, the machine learning model is based on a supervised learning structure, and gradient boosting decision trees and support vector machines are used for fusion training. The training data comes from labeled harmonic feature samples in historical normal operation periods and fault periods.

[0021] The model training process includes feature selection, sample balancing, and cross-validation steps to ensure that the training process is not affected by data bias. The GBDT model is used to identify strong nonlinear patterns, and the SVM model is used to accurately divide the abnormal boundary in the feature space.

[0022] In actual deployment, the model uses a sliding window to input real-time feature sequences, and based on the model output probability value, it realizes the abnormal scoring of each order harmonic.

[0023] Preferably, based on the extracted spectral feature parameters, the following processes are sequentially executed to accurately identify weak harmonic abnormal trends in the operation of rail transit variable frequency power supply, as follows:

[0024] For each order harmonic signal, the amplitude data in the period is continuously sampled, and a time series is constructed. The mean square error of the time series with respect to its average amplitude is calculated to evaluate the fluctuation stability of the harmonic in the current time period. The calculation formula is as follows:

[0025] , where is the amplitude of the order harmonic in the period, is the average amplitude of the order harmonic in the period, is the time window length for calculation, in units of cycles, is the mean square error of the order harmonic;

[0026] To capture whether the harmonic is in an upward trend, the difference between the current cycle and the data before the previous 5 cycles is taken for mean processing, and the formula is as follows:

[0027] , where is the amplitude of the order harmonic in the current cycle, representing the latest sampling value, is the amplitude of the order harmonic 5 cycles ago, is the average growth rate of the order harmonic in the last 5 cycles, and the denominator represents the cycle interval between two samplings;

[0028] The mean square error obtained is weighted and fused with the average growth rate to generate an anomaly score value, and the generation formula is as follows:

[0029] , wherein is the weight coefficient of the mean square error index, is the weight coefficient of the growth rate index, is the comprehensive anomaly score of the m-th order harmonic.

[0030] Preferably, the dynamic threshold mechanism adopts a multi-factor fusion judgment strategy, and the identification threshold of the harmonic anomaly is adjusted in real time according to the current running state. The specific steps are as follows:

[0031] Extract the average value and fluctuation amplitude of the target order harmonic in the past multiple working cycles as the initial reference benchmark;

[0032] Combine the current running load level of the device, the ambient temperature, and the change of the interference factor in the historical data of the same period to respectively evaluate the influence degree of the harmonic behavior, and set the adjustment factor accordingly;

[0033] Comprehensively reflect the state changes of each adjustment factor through the built-in weight strategy to generate a new identification reference interval and dynamically correct the original threshold range.

[0034] Preferably, the trend prediction algorithm adopts an improved long short-term memory network, and combines a gating mechanism to perform attention weighting on key features in the input sequence;

[0035] The network includes three layers of LSTM units and one layer of fully connected output layer, the input is the amplitude of each order harmonic and its derivative sequence in the last M cycles, and the output is the trend classification label in the future K cycles;

[0036] The training data introduces three trends of "risk rising", "fluctuation stable", and "recovery falling" through the extended label method, and the model outputs the risk level of each order when reasoning.

[0037] Preferably, after completing the classification output of the machine learning model, the precision and interpretability of anomaly identification are improved, high-order statistics and spectral dynamics indexes are introduced for clustering analysis and fine-grained classification judgment of the harmonic signal, and the specific steps are as follows:

[0038] The signal in each cycle is converted into a frequency distribution form through fast Fourier transform to obtain a power spectrum, and the information entropy of the power spectrum is calculated to obtain the uniformity feature of the current harmonic signal energy distribution, and the calculation formula is as follows:

[0039] ​, wherein, is the power ratio of the harmonic signal at the th frequency point, is the number of frequency components analyzed in the spectrum, is the th order harmonic signal,

[0040] The maximum, minimum and median of the harmonic in the selected time window are extracted to construct a standardized nonlinear index, and the expression is as follows:

[0041] , wherein, is the nonlinear fluctuation index of the th order harmonic signal in a given period, is the maximum value of the th order harmonic amplitude sequence in the current analysis period window, reflecting the peak energy, is the minimum value in the same period, used to measure the signal base change, is the median of the sequence, as a normalized reference point for the fluctuation intensity;

[0042] The spectrum power entropy , the nonlinear fluctuation index and the comprehensive abnormal score are weighted and fused to form the final discriminant score, and the weighted fusion formula is as follows:

[0043] , wherein, is the fusion feature score of the th order harmonic signal, is the weight of the spectrum power entropy in the fusion feature score, is the weight of the nonlinear fluctuation index in the fusion score, is the weight of the comprehensive abnormal score .

[0044] Preferably, the early warning mechanism includes local audio and video prompts, cloud message push and automatically generated maintenance suggestions;

[0045] The audio and video prompts include voice broadcast to report the alarm level and fault type, red highlight flashing border prompt; the cloud push is based on the MQTT protocol, and the fault information is pushed to the maintenance personnel's handheld terminal APP in real time, and is accompanied by the occurrence position, harmonic index change graph;

[0046] The maintenance suggestion system calls similar cases through the pre-set knowledge graph engine to realize automatic recommendation of fault level, cause and preferred troubleshooting path.

[0047] In the above technical solutions, the present application provides technical effects and advantages:

[0048] The present application realizes high-precision and low-delay extraction of harmonic signals in the rail transit variable frequency power supply system through the technical path of multi-point synchronous acquisition and spectrum joint analysis. Traditional methods are difficult to stably identify harmonic changes in a complex electromagnetic environment, are easily affected by sampling drift and phase deviation, and cause early warning failure. The present application uses a high-precision analog-to-digital converter and a time synchronization mechanism, and combines short-time Fourier transform to extract multi-scale features in the frequency domain, ensuring that weak harmonic anomalies are accurately captured within milliseconds. This method significantly improves the accuracy and response time of fault identification, providing solid foundation data support for the early warning system.

[0049] The present application effectively improves the intelligent level and adaptability of abnormal identification by introducing a multi-model fusion and dynamic threshold adjustment mechanism. The traditional fixed threshold method cannot cover the fluctuation range under different operating conditions, and is prone to false positives or false negatives. The present application classifies harmonic anomaly patterns by combining GBDT and SVM models, and dynamically generates identification thresholds based on device state parameters, realizing a dual enhancement method of "model self-learning + threshold self-adjustment". This mechanism not only improves the system's ability to handle sudden complex harmonic problems, but also ensures robustness and reliability in multiple operating conditions and scenarios.

[0050] The present application realizes the system transformation from traditional "fault response" to "active prediction" by constructing a trend prediction model and a graphical early warning platform. The improved LSTM neural network is used to analyze the time evolution trend of harmonic characteristics, and a multi-dimensional risk scoring system is formed by combining high-order statistical indicators, so that the system has the ability to judge the future risk trend. Once potential abnormal trends occur, the platform can immediately issue a graphical early warning and provide corresponding maintenance recommendations, significantly shortening the maintenance response time. This method not only reduces the occurrence rate of sudden failures, but also significantly improves the forward-looking and precision of operation and maintenance work, enhancing the safety protection capability of the rail transit power supply system. BRIEF DESCRIPTION OF DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments or prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0052] Figure 1 The method flowchart of the present application is a rail transit variable frequency power supply intelligent operation and maintenance fault early warning method. DETAILED DESCRIPTION

[0053] Example implementations are now described with reference to the drawings. Example implementations can, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the example implementations to those skilled in the art.

[0054] The present application provides a rail transit variable frequency power supply intelligent operation and maintenance fault early warning method as shown in Figure 1 The present application provides a rail transit variable frequency power supply intelligent operation and maintenance fault early warning method as shown in

[0055] A multi-point synchronous acquisition device is arranged to acquire voltage and current signals of the variable frequency power supply system in high precision and real time, and to process the data by digital encoding;

[0056] In the high-precision real-time acquisition step, the acquisition device is a synchronous multi-channel analog-to-digital conversion unit, which is respectively arranged at the input end, the output end and the key load node of the variable frequency power supply system, and the sampling precision is not less than 16 bits and the sampling frequency is not less than 10 kHz;

[0057] The data encoding processing adopts a discrete time window segmentation encoding mode to ensure the time alignment of the data of each channel and avoid the phase drift problem in the acquisition process;

[0058] Each data segment is automatically labeled with a time stamp after acquisition, sent to a local cache and uploaded to a central processing unit through a CAN bus or an optical fiber channel to ensure high real-time performance and transmission stability of data acquisition.

[0059] The synchronous acquisition mechanism enables the subtle changes in the early stage of the fault to be accurately captured within a millisecond time range, enhancing the response sensitivity of the early warning system to sudden harmonic events.

[0060] The collected voltage and current signals are analyzed in the frequency domain by using a preset spectrum analysis algorithm to extract harmonic amplitude and phase characteristic parameters of different orders;

[0061] The spectrum analysis algorithm adopts a joint processing mechanism of windowed Fourier transform and short-time Fourier transform to perform multi-scale frequency domain analysis on the sampled signals, and the specific steps are as follows:

[0062] First, the Hamming window is used to divide the original sampled signal into windows to reduce the spectrum leakage phenomenon;

[0063] Second, the amplitude and phase of the fundamental wave and the 2-25th harmonic are extracted based on the FFT operation of the window function output data;

[0064] Third, the mean value of each order of harmonic is normalized in multiple time windows, and a harmonic feature matrix is constructed.

[0065] The processing method not only improves the anti-interference ability of spectrum recognition, but also maintains the accurate modeling effect of harmonic structure under non-stationary load conditions, thereby enhancing the robustness of the subsequent anomaly recognition step.

[0066] An anomaly recognition module based on a machine learning model is constructed, and current feature parameters are input and compared with historical operation data to mine potential harmonic anomaly patterns.

[0067] The machine learning model is based on a supervised learning structure and uses gradient boosting decision trees (GBDT) and support vector machines (SVM) for fusion training. The training data comes from labeled harmonic feature samples of historical normal operation periods and fault periods.

[0068] The model training process includes feature selection, sample balancing, and cross-validation steps to ensure that the training process is not affected by data bias. The GBDT model is used to identify strong non-linear patterns, and the SVM model is used to accurately divide the anomaly boundary in the feature space.

[0069] In actual deployment, the model uses a sliding window to input real-time feature sequences, and based on the model output probability value, an anomaly score is calculated for each order harmonic. The score value plays a key role in subsequent dynamic threshold adjustment.

[0070] The model output is not only used for classification and judgment, but also used to predict whether the input signal is in a risk rising trend interval.

[0071] Based on the extracted frequency spectrum feature parameters, the following processes are sequentially executed to accurately identify weak harmonic anomaly trends in the operation of rail transit variable frequency power supplies. The processes are as follows:

[0072] For each order harmonic signal, the amplitude data in periods is continuously sampled, and a time series is constructed. The fluctuation stability of the harmonic in the current time period is evaluated by calculating the mean square error of the time series with respect to its average amplitude. The calculation formula is as follows:

[0073] , where is the amplitude of the order harmonic in the period, and is the harmonic feature quantity extracted from the spectrum analysis, with units of voltage (V) or current (A), depending on the monitoring object, is the average amplitude of the order harmonic in periods, is the time window length for calculation, with units of cycles, usually selected as multiple monitoring periods (such as 10 or 20 cycles) of continuous sampling, is the amplitude of the The root mean square error of a harmonic order represents the intensity of that harmonic fluctuation within the monitoring period. The larger the value, the more drastic the harmonic changes, which may indicate an anomaly.

[0074] Larger The value usually indicates high harmonic activity or the presence of structurally abnormal fluctuations, which helps to screen for potential risk signals.

[0075] To detect whether harmonics are on the rise, the current period is taken. The average difference between the data and the data from the previous 5 periods is calculated. If the average growth rate is significantly positive, it indicates that the amplitude of that harmonic order is rapidly increasing. This trend may represent an early sign of a fault. The formula is as follows:

[0076] In the formula, It is the current number The first cycle The order harmonic amplitude represents the latest sampled value. It is the 5th cycle ago The amplitude of the first harmonic is used as a past reference value; the difference between the two represents the change in amplitude. It is the first The average growth rate of the order harmonics over the past 5 cycles reflects whether the harmonics are in a continuous upward trend. (The denominator is...) This represents the interval between two sampling periods, and calculates the average rate of change per unit period, with the unit being the change in amplitude per period.

[0077] The obtained mean square error Compared with average growth rate Weighted fusion is performed to generate an anomaly score, which reflects whether a certain harmonic is currently deviating from its typical fluctuation range and showing an increasing risk. It is an important basic indicator for triggering subsequent threshold judgments and alarm actions. The generation formula is as follows:

[0078] In the formula, It is the weighting coefficient of the mean squared error index, used to adjust... The degree of influence on the final score. The value is selected based on historical experience with actual harmonic anomalies (e.g., 0.6). It is the weighting coefficient of the growth rate indicator, used for adjustment. The influence can be adjusted according to the need for sensitivity to trend changes (e.g., 0.4). It is the first The comprehensive anomaly score of order harmonics is the key indicator used to determine whether a harmonic has entered the warning range. The higher the score, the greater the harmonic fluctuation and the upward trend.

[0079] By constructing a triple computer mechanism of harmonic fluctuation intensity (mean square deviation), trend growth rate and its weighted fusion anomaly score, early, weak but potentially risky harmonic abnormal behavior in variable frequency power supply system is accurately identified. This mechanism not only considers the severity of the current harmonic fluctuation, but also comprehensively judges whether it is in a continuous deterioration trend, thus having more dynamic perception and prediction ability than the traditional fixed threshold method. It significantly enhances the fault warning system's ability to capture "precursor faults", effectively avoids false positives or false negatives due to early weak signals, and improves the forward-looking and proactive nature of rail transit power supply system operation and maintenance.

[0080] A dynamic threshold mechanism is introduced, which sets adaptive identification thresholds for each order harmonic in combination with the device operating environment, load type and time period factors, improving the accuracy of anomaly identification;

[0081] The dynamic threshold mechanism adopts a multi-factor fusion decision strategy to adjust the harmonic anomaly identification threshold in real time according to the current operating state. The specific steps are as follows:

[0082] First, the average value and fluctuation amplitude of the target order harmonic in the past multiple working cycles are extracted as the initial reference;

[0083] Second, combined with the current operating load level of the device, the ambient temperature and the change of interference factors in the same period historical data, the influence of each factor on the harmonic behavior is evaluated respectively, and the adjustment factor is set;

[0084] Third, by using the built-in weight strategy, the state changes reflected by each adjustment factor are integrated to generate a new identification reference interval, and the original threshold range is dynamically corrected.

[0085] This method avoids the problem of false positives caused by ignoring the changes in the operating state of the device, enhances the adaptability of the system in dealing with high-frequency disturbances, seasonal load fluctuations and special operating modes (such as morning and evening peak, night maintenance), and effectively improves the accuracy and robustness of fault identification.

[0086] Through a time series trend prediction algorithm, the abnormal harmonic parameters are predicted for multiple cycles to judge their development trend and potential risk level;

[0087] The trend prediction algorithm uses an improved long short-term memory network (LSTM) combined with a gating mechanism to weight the key features in the input sequence;

[0088] The network contains three layers of LSTM units and one layer of fully connected output layer, the input is the amplitude of each order harmonic and its derivative sequence in the last M cycles, and the output is the trend direction classification label in the future K cycles;

[0089] The training data introduces three trends of "risk increase", "fluctuation stability", and "recovery decrease" through extended labels, and the model outputs the risk level of each order when reasoning.

[0090] The prediction results are dynamically marked with different colors when graphically displayed, making it easy for maintenance personnel to identify the severity and development trend of the problem, and achieving a full closed-loop response system of "running, monitoring, and early warning".

[0091] After completing the classification output of the machine learning model, the accuracy and interpretability of anomaly identification are improved, high-order statistics and spectral dynamics indicators are introduced, and harmonic signals are clustered and classified in detail. The specific steps are as follows:

[0092] The signal in each cycle is converted into a frequency distribution form by Fast Fourier Transform (FFT), and the power spectrum is obtained. The information entropy of the power spectrum is calculated to obtain the uniformity characteristics of the current harmonic signal energy distribution, reflecting the complexity of the system state. High entropy value often corresponds to the state of abnormal high-frequency vibration or harmonic disorder of the equipment. The formula is as follows:

[0093] , where is the power ratio of the harmonic signal at the th frequency point, which is calculated by normalizing the square amplitude obtained by Fast Fourier Transform (FFT), and satisfies , forming a probability distribution, is the number of frequency components analyzed in the spectrum, which usually depends on the FFT resolution and sampling rate. For example, when sampling a 10kHz signal with 512 points, generally has 256 positive frequency components, is the spectral power entropy of the th order harmonic signal, which is used to measure the complexity and uncertainty of the energy distribution of the harmonic in the frequency domain. The larger the value, the more dispersed the frequency components of the signal, which may indicate that the system has multiple frequency disturbances or harmonic energy splitting phenomena.

[0094] Extract the maximum, minimum, and median of the harmonic in the selected time window to construct a standardized nonlinear index. This index can effectively identify strong fluctuations or areas of intense change, and is particularly suitable for judging occasional pulse-type harmonic interference. The formula is as follows:

[0095] , where is the nonlinear fluctuation index of the th order harmonic signal in the given cycle, which is used to evaluate the intensity of harmonic change. The larger the value, the more intense the fluctuation, and the worse the stability, is the nonlinear fluctuation index of the The maximum value of the sequence of the amplitude of the harmonic of the order, reflecting the peak energy, is the minimum value in the same cycle, used to measure the signal base change, is the median of the sequence, as a normalized reference point of fluctuation intensity, which helps to eliminate the influence of abnormal extreme values on the fluctuation index;

[0096] The spectral power entropy , the nonlinear fluctuation index and the comprehensive abnormal score are weighted and fused to form the final discriminant score, which is not only used to judge the severity of the harmonic abnormality, but also serves as a support vector machine (SVM) input feature to participate in the multi-class classification task, so as to realize accurate positioning of the abnormal category and the occurrence trend, and the formula is as follows:

[0097] , wherein, is the fusion feature score of the harmonic signal of the order , which is used to comprehensively reflect whether it is in a high-risk abnormal state, and the higher the score, the more the signal deviates from the normal range, which is the key input for final classification and judgment, is the weight of the spectral power entropy in the fusion feature score, which controls the influence of the spectral entropy on the final score, and the spectral entropy reflects the complexity of the harmonic energy distribution, The greater the weight, the more the system values the abnormal situation of the frequency distribution structure, and it is suitable for identifying problems with obvious changes in frequency domain characteristics, such as harmonic diffusion or multi-frequency disturbance, is the weight of the nonlinear fluctuation index in the fusion score, which controls the influence of the signal intensity fluctuation characteristics on the abnormal judgment, and the stronger the nonlinear fluctuation, the more it usually indicates that the system enters an unstable state, The greater the weight, the more emphasis on the importance of fluctuation intensity in fault warning, and it is more suitable for dealing with sudden spikes or impact load interference, is the weight of the comprehensive abnormal score , which reflects the degree of attention of the system to the time series behavior (such as growth trend and historical deviation), and if is greater, it emphasizes that the trend evolution and fluctuation risk are the core of the abnormal judgment, and it is suitable for identifying gradual degradation or chronic failure in advance.

[0098] By introducing three types of advanced feature parameters, including spectral power entropy, nonlinear fluctuation index and previous abnormal score, a multi-dimensional fusion feature score is constructed to realize fine identification and high-precision classification of harmonic abnormal signals. This method breaks through the limitations of traditional abnormal detection relying on single threshold or single-point index, and can comprehensively reflect the change characteristics of signals in frequency domain distribution, fluctuation amplitude and time trend, thereby more comprehensively describing the internal mechanism of harmonic instability process. The scoring results not only serve as the key input of the final classification model, but also provide the basis for abnormal level judgment and risk assessment for the subsequent operation and maintenance system, significantly improving the intelligent judgment and adaptive decision-making ability of the early warning system under complex operating conditions.

[0099] The judgment results are displayed in real time to the operation and maintenance platform through a graphical interface, and the early warning mechanism is triggered, pushing the diagnosis results and recommended measures to the maintenance personnel terminal;

[0100] The early warning mechanism includes local audio and video prompts, cloud message pushing and automatically generated maintenance recommendations;

[0101] The audio and video prompts include voice broadcast of warning level and fault type, red highlight flashing border prompt; the cloud pushing is based on MQTT protocol, which pushes the fault information to the maintenance personnel's handheld terminal APP in real time, and attaches the occurrence location and harmonic index change graph;

[0102] The maintenance recommendation system calls similar cases through the pre-set knowledge graph engine to realize automatic recommendation of fault level, cause and preferred troubleshooting path.

[0103] This mechanism ensures timely information coverage and clear instruction transmission, which helps to shorten the decision response time and assists operation and maintenance personnel to quickly locate and handle harmonic hidden dangers.

[0104] Embodiment one: This embodiment mainly deploys high-precision synchronous acquisition devices at key nodes of rail transit variable frequency power supply system, and realizes real-time acquisition of voltage and current signals and spectral analysis to accurately capture harmonic abnormal characteristics. In the rail transit power supply network, the frequency converter usually plays an important role in adjusting power frequency and improving energy efficiency. However, its operation process is disturbed by complex power grid environment and nonlinear load, which easily introduces power quality problems such as harmonics and waveform distortion. In order to improve the sensitivity and processing efficiency of the operation and maintenance system to such faults, a multi-node, multi-dimensional and real-time synchronous data acquisition system must be established.

[0105] Specifically, the system sets three types of data acquisition terminals at the input, output and typical load branch of the frequency converter. These terminals are equipped with high-precision analog-to-digital conversion modules, which can achieve millisecond-level data sampling and time synchronization. The collected voltage and current raw signals are processed in batches through window segmentation technology, labeled with accurate time, and cached to the local edge computing unit. These terminals communicate data to the central controller through industrial Ethernet or high-speed optical fiber communication for unified data processing and spectrum conversion.

[0106] At the spectrum analysis level, the collected time-domain signals are first divided into several fixed time windows, and the data in each window is processed with windowing to reduce spectral leakage. Then, using short-time spectral analysis technology, the frequency domain amplitude and phase characteristics of multiple harmonics are extracted, and harmonic spectrum and feature matrix are constructed. These feature matrices not only have good anti-interference ability, but also can be used for subsequent intelligent identification and trend prediction analysis. The whole process realizes the closed-loop path of "data acquisition-real-time processing-harmonic feature generation", which can discover abnormal signal fluctuations and issue preliminary alarms in the acquisition stage, greatly improving the response speed to sudden power quality problems.

[0107] In addition, the system also has a data fault-tolerant mechanism and a backup acquisition path to ensure that the acquisition task does not be interrupted in the event of network failure or terminal failure. In this way, the rail transit frequency power supply system realizes the global power monitoring with high time resolution and high frequency coverage, providing a solid data foundation for subsequent fault warning models and forming the first technical defense line of active defense.

[0108] Embodiment two: This embodiment focuses on improving the intelligent level of harmonic anomaly identification, using multiple machine learning models to identify potential abnormal patterns in the operation data, and building a dynamically adjusted warning threshold system to enhance the system's ability to cope with the complexity and suddenness of actual working conditions.

[0109] In the model training phase, a historical sample library is constructed from long-term acquisition of operation data from the rail transit field system. The samples cover normal state, slight disturbance state and typical fault state. Each sample record contains statistical features of multiple harmonic orders, such as amplitude mean, fluctuation trend, periodic change, etc. These features are extracted and formatted as standard vectors as input for machine learning algorithms. In terms of model structure, two types of models are combined: one is the gradient boosting tree model with strong feature fitting capability, which is used to capture the complex relationships between non-linear features; the other is the support vector machine model, which focuses on high-dimensional feature boundary identification to accurately judge marginal abnormal samples.

[0110] In the actual operation stage, the terminal collects the latest feature vector and inputs it into the model to output the abnormal score of each harmonic. The score can be understood as the confidence of the model in judging whether the data point is abnormal. The higher the score, the more serious the deviation from the normal state. In order to adapt to the influence of external factors such as device load changes, seasonal fluctuations and power grid interference, the system introduces a dynamic threshold adjustment strategy. This strategy not only refers to historical statistical indicators, but also introduces the current operating environment parameters of the device, such as load level, temperature, voltage fluctuation frequency and other factors, to dynamically calculate the warning threshold.

[0111] When the model score exceeds the current threshold, the system triggers an alarm and uploads the alarm result to the central server for synchronization and archiving. The dynamic threshold mechanism effectively prevents false positives or false negatives caused by traditional fixed value judgments under dynamic operating conditions, making the entire fault identification process more accurate and reliable. At the same time, by continuously optimizing the model training samples and strategy parameters, the system can continuously improve the identification accuracy and stability, providing a reliable basis for daily operation and management.

[0112] Embodiment three: This embodiment builds an intelligent operation and maintenance platform around the concept of "early warning and proactive prevention", uses deep learning technology to predict harmonic trends, and realizes real-time display and operation guidance of fault information through a graphical interface, greatly improving the efficiency and accuracy of operation and maintenance.

[0113] The platform consists of four main functional modules: data access module, intelligent prediction module, risk assessment module and user interaction interface. The data access module connects the front-end collection system to receive and process all real-time harmonic parameters. Once the data is transmitted, it is parsed into a standard format and sent to the intelligent prediction module for trend judgment.

[0114] The trend prediction module uses a time series learning model based on a neural network architecture, which can mine non-linear trend change patterns in real-time data streams and judge the harmonic change trend in the near future. This module can identify the subtle fluctuations of harmonic levels in different working states of the system and distinguish between stable fluctuations and potential risk rising states. The prediction model introduces a key feature selection mechanism to give higher weights to representative harmonic changes, focusing on the root cause of the problem and improving prediction accuracy.

[0115] The system also has a risk assessment module that classifies the trend results. Combining the platform's historical experience model and expert knowledge base, it judges the potential harm level of each type of abnormal trend and pushes different levels of alarm information based on the severity. Alarm information is not only displayed on the platform's main interface, but also sent to the handheld terminal of maintenance personnel through a cloud push mechanism to ensure that critical decision-making information is conveyed in the shortest possible time.

[0116] The user interaction interface adopts a modular visual design, and operation and maintenance personnel can view the trend line, historical fluctuation graph and future risk prediction of each order harmonic in the interface. When the system identifies that a node has a risk trend, it automatically generates maintenance suggestions including fault type, possible impact, and priority troubleshooting path, significantly reducing the judgment burden of operation and maintenance personnel and improving the disposal efficiency.

[0117] Through this embodiment, the system not only has the ability of passive response, but also has the comprehensive ability of active perception, prospective judgment and intelligent recommendation, realizing the qualitative leap from "problem post-processing" to "problem early warning" in the maintenance of the rail transit power supply system.

[0118] The present application realizes high-precision and low-delay extraction of harmonic signals in the rail transit variable frequency power supply system through the technical path of multi-point synchronous acquisition and spectrum joint analysis. Traditional methods are difficult to stably identify harmonic changes in a complex electromagnetic environment, are easily affected by sampling drift and phase deviation, and cause early warning failure. The present application uses a high-precision analog-to-digital converter and a time synchronization mechanism, and combines short-time Fourier transform to extract multi-scale features in the frequency domain, ensuring that weak harmonic anomalies are accurately captured within milliseconds. This method significantly improves the accuracy and response time of fault identification, providing solid foundation data support for the early warning system.

[0119] The present application effectively improves the intelligent level and adaptability of abnormal identification by introducing a multi-model fusion and dynamic threshold adjustment mechanism. Traditional fixed threshold methods cannot cover the fluctuation range under different operating conditions, and are prone to false positives or false negatives. The present application classifies harmonic anomaly patterns by combining GBDT and SVM models, and dynamically generates identification thresholds based on device state parameters, realizing a dual enhancement method of "model self-learning + threshold self-adjustment". This mechanism not only improves the system's ability to handle sudden complex harmonic problems, but also ensures robustness and reliability in multiple operating conditions and scenarios.

[0120] The present application realizes the system transformation from traditional "fault response" to "active prediction" by constructing a trend prediction model and a graphical early warning platform. The improved LSTM neural network is used to analyze the trend of harmonic characteristics over time, and a multi-dimensional risk scoring system is formed by combining high-order statistical indicators, so that the system has the ability to judge the future risk trend. Once a potential abnormal trend occurs, the platform can immediately issue a graphical early warning and provide corresponding maintenance suggestions, significantly shortening the maintenance response time. This method not only reduces the occurrence rate of sudden failures, but also significantly improves the forward-looking and precision of operation and maintenance work, enhancing the safety protection capability of the rail transit power supply system.

[0121] The above formulas are all de-dimensioned to calculate the numerical values, the formulas are obtained by collecting a large amount of data to simulate a formula of the most recent real situation, and preset parameters in the formulas are set by a person skilled in the art according to actual conditions.

[0122] The above merely describes certain exemplary embodiments of the present application by way of illustration, and it is needless to say that those skilled in the art can modify the described embodiments in various manners without departing from the spirit and scope of the present application. Therefore, the above drawings and descriptions are merely illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present application.

[0123] It should be noted that, in this document, relational terms such as first and second and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0124] It should be understood that, in various embodiments of the present application, the size of the sequence number of each process described above does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0125] Those skilled in the art can appreciate that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0126] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be described here.

[0127] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Part or all of the units can be selected to achieve the purpose of the embodiment scheme according to actual needs.

[0128] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0129] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any skilled person in the art can easily think of changes or replacements within the technical range disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0130] The above only describes some exemplary embodiments of the present application by way of illustration, and it is needless to say that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present application. Therefore, the above figures and description are illustrative in nature and should not be understood as limiting the scope of protection of the claims of the present application.

Claims

1. A rail transit variable frequency power supply intelligent operation and maintenance fault early warning method, characterized in that, The method comprises the following steps: A multi-point synchronous acquisition device is arranged to acquire voltage and current signals of the variable frequency power supply system in high precision and real time, and to perform digital encoding processing on the data; A preset spectrum analysis algorithm is used to perform frequency domain analysis on the acquired voltage and current signals, and to extract harmonic amplitude and phase characteristic parameters of different orders; An abnormality recognition module based on a machine learning model is constructed, current characteristic parameters are input, and historical operation data are compared, harmonic abnormality scores of each order are calculated and output; A dynamic threshold mechanism is introduced, and adaptive recognition thresholds are set for each order of harmonic in combination with equipment operation environment, load type and time period factors; Through a time series trend prediction algorithm, the abnormal harmonic parameters are predicted for multiple cycles to judge their development trend and potential risk level; The judgment result is displayed in real time to an operation and maintenance platform through a graphical interface, and an early warning mechanism is triggered to push diagnosis results and recommended measures to a maintenance personnel terminal; Based on the extracted spectrum characteristic parameters, the following processes are sequentially executed for accurately identifying weak harmonic abnormality trends in the operation process of the rail transit variable frequency power supply, and the processes are as follows: For each order harmonic signal, continuously sampling amplitude data within one period, constructing a time series, evaluating the fluctuation stability of the harmonic in the current time period by calculating the mean square error of the time series for its average amplitude, the calculation formula is as follows: wherein is the amplitude of the th harmonic at the th cycle, is the average amplitude of the th harmonic over the th cycle, is the length of the time window used for the calculation, is the mean square error of the th harmonic; Take the current cycle The data difference value before 5 cycles is processed by mean value, and the formula is as follows: In the formula, It is the current number The first cycle The order harmonic amplitude represents the latest sampled value. It is the 5th cycle ago First harmonic amplitude, It is the first The average growth rate of the order harmonics over the past 5 periods, denominator This represents the interval between two sampling periods; The mean square error obtained The average growth rate Weighted fusion is performed on the average growth rate to generate an anomaly score value, and the generation formula is as follows: wherein, is a weight coefficient of the mean square deviation indicator, is a weight coefficient of the growth rate indicator, is the synthetic anomaly score of the harmonic of order n. 2.The rail transit variable frequency power supply intelligent operation and maintenance fault early warning method according to claim 1, characterized in that, In the high-precision real-time acquisition step, the acquisition device is a synchronous multi-channel analog-to-digital conversion unit, which is respectively arranged at the input end, the output end and the key load node of the variable frequency power supply system, the sampling precision is not less than 16 bits, and the sampling frequency is not less than 10 kHz; the data encoding processing adopts a discrete time window segmentation encoding mode, each data segment is automatically labeled with a time stamp after acquisition, and is sent to a local buffer and uploaded to a central processing unit through a CAN bus or an optical fiber channel. 3.The rail transit variable frequency power supply intelligent operation and maintenance fault early warning method of claim 1, wherein, The spectrum analysis algorithm adopts a joint processing mechanism of windowed Fourier transform and short-time Fourier transform to perform multi-scale frequency domain analysis on the sampled signals, and the specific steps are as follows: The original sampled signal is windowed and divided by using a Hamming window; The amplitude and phase of the fundamental wave and 2-25th harmonic are extracted based on fast Fourier transform (FFT) operation on the window function output data; The average value of each order of harmonic in multiple time windows is normalized, and a harmonic feature matrix is constructed.

4. The intelligent operation and maintenance fault early warning method for a rail transit variable-frequency power supply according to claim 1, characterized in that, The machine learning model is based on a supervised learning structure, and gradient boosting decision trees and support vector machines are used for fusion training, and the training data come from labeled harmonic feature samples in historical normal operation periods and fault periods; The model training process includes feature selection, sample balancing and cross-validation steps to ensure that the training process is not affected by data bias, wherein the GBDT model is used to identify strong nonlinear patterns, and the SVM model is used to accurately divide the abnormal boundary in the feature space; in actual deployment, the model uses a sliding window to input real-time feature sequences, and the model output probability value is used to realize abnormal scoring of each order of harmonic. 5.The intelligent operation and maintenance fault early warning method of a rail transit variable-frequency power supply according to claim 1, characterized in that, The dynamic threshold mechanism adopts a multi-factor fusion judgment strategy, and the harmonic abnormality recognition threshold is adjusted in real time according to the current operation state, and the specific steps are as follows: The average value and fluctuation amplitude of the target order harmonic in the past multiple working periods are extracted as the initial reference; The influence degree of the current operating load level, ambient temperature and interference factors in the same period historical data on the harmonic behavior is evaluated respectively, and an adjustment factor is set based on the evaluation results; A new identification reference interval is generated by combining the state changes reflected by each adjustment factor through the built-in weight strategy, and the original threshold range is dynamically corrected. 6.The intelligent operation and maintenance fault early warning method of a rail transit variable-frequency power supply according to claim 1, characterized in that, The trend prediction algorithm uses an improved long short-term memory network combined with a gating mechanism to give attention to key features in the input sequence; The network contains three layers of LSTM units and one layer of fully connected output layer. The input is the amplitude of each order harmonic and its derivative sequence in the last M periods, and the output is the trend classification label in the future K periods. The training data introduces three trends of "risk rising", "volatility stable" and "recovery falling" through the extended label method. The model outputs the risk level of each order when reasoning. 7.The intelligent operation and maintenance fault early warning method of a rail transit variable-frequency power supply according to claim 1, characterized in that, After the classification output of the machine learning model is completed, the precision and interpretability of anomaly detection are improved by introducing high-order statistical quantities and spectral dynamics indicators for clustering analysis and fine-grained classification judgment of harmonic signals. The specific steps are as follows: The signal in each period is converted into frequency distribution form by fast Fourier transform to obtain the power spectrum. The power spectrum is calculated for information entropy to obtain the uniformity feature of the current harmonic signal energy distribution. The calculation formula is as follows: , wherein, is the power ratio of the harmonic signal at the th frequency point, is the number of frequency components analyzed in the spectrum, is the spectral power entropy of the th order harmonic signal; The maximum, minimum and median of the harmonic in the selected time window are extracted to construct a standardized nonlinear index. The expression is as follows: In the formula, It is the first The nonlinear fluctuation index of an order harmonic signal within a given period Within the current analysis period window, the first The maximum value of the order harmonic amplitude sequence reflects its peak energy. It is the minimum value within the same period, used to measure the base variation of the signal. It is the median of the sequence, serving as a normalized reference point for fluctuation intensity; The spectral power entropy , the nonlinear fluctuation index , and the comprehensive anomaly score are weighted and fused to form a final discrimination score, and the weighted fusion formula is as follows: wherein is a fusion feature score of the mth order harmonic signal, is a spectral power entropy weight in the fusion feature score, is a nonlinear fluctuation index weight in the fusion score, is a weight of the composite anomaly score . 8.The rail transit variable frequency power supply intelligent operation and maintenance fault early warning method of claim 1, characterized in that, The early warning mechanism includes local audio and video prompts, cloud message pushing and automatic maintenance suggestions. The audio and video prompts include voice broadcast of warning level and fault type, red highlight flashing border prompt; The cloud pushing is based on the MQTT protocol, which pushes the fault information to the maintenance personnel's handheld terminal APP in real time, and attaches the occurrence location and harmonic index change graph. The maintenance suggestion system calls similar cases through the pre-set knowledge graph engine to realize the automatic recommendation of fault level, cause and preferred troubleshooting path.

Images