Fault prediction method and device for track power supply panel

Abstract

The present invention provides a method and device for predicting the fault of a track power supply panel, the method comprising: performing anomaly detection on circuit parameter timing samples based on an anomaly detection algorithm to obtain abnormal samples, and marking the abnormal samples with fault types; obtaining target circuit parameter samples according to the abnormal samples and fault type marking data; obtaining a fault prediction model according to the target circuit parameter samples; inputting the circuit parameters and prediction period of the target power supply panel in the test set into the fault prediction model to obtain the probability of failure of the target power supply panel. The device is used to execute the above method. The method and device for predicting the fault of a track power supply panel provided by the present invention, by performing anomaly detection on the circuit parameters of the track power supply panel, marking the fault type of the abnormal data, and combining the timing feature method, construct an accurate power supply panel fault prediction model, realize the automatic prediction of power supply panel faults, and improve the accuracy of fault prediction of the track power supply panel.

Description

Technical Field

[0001] The present invention relates to the field of rail transit technology, and in particular to a fault prediction method and device for a rail power supply panel. Background Art

[0002] Railway power panels are important equipment for supplying power to railway signal equipment. Whether the power panel operates safely and stably can directly affect driving safety. The power panel is affected by the aging of its own components, the environment and other factors during operation, which can cause frequent failures and seriously affect driving safety. Therefore, monitoring the operating status of the power panel, predicting failures, and discovering the fault points in advance can play an extremely important role in the maintenance of the power panel.

[0003] At present, the monitoring of power panels is mostly completed by the centralized signal monitoring system. The various parameters monitored by it are far from satisfying the prediction of the time and type of power panel failure. In addition, the current fault prediction methods mostly stay at setting relevant thresholds and performing comparisons. There are great limitations and limited data utilization.

[0004] For the operation and maintenance of urban rail transit power supply panel equipment, the operation and maintenance data relied on by the operation and maintenance personnel are mainly concentrated on the input and output monitoring of each power distribution module of the power supply panel. The data is relatively simple, and most of the existing power supply panel fault prediction technologies use the threshold interval to compare the collected data with the threshold to predict the time of fault occurrence. This method is heavily dependent on the threshold range set by the prior experience of the operation and maintenance personnel, and needs to be set separately for power supply panels of different manufacturers and power supply panels in different states. It is inefficient, has low versatility, and is not accurate enough in predicting power supply panel faults. Summary of the invention

[0005] The fault prediction method for a track power panel provided by the present invention is used to overcome the defects of low efficiency, low versatility and insufficient accuracy in the prior art in predicting power panel faults. It can realize the automated prediction of power panel faults, improve the prediction efficiency, and at the same time build an accurate fault prediction model based on the circuit parameters of the power panel, thereby improving the applicability and accuracy of power panel fault prediction.

[0006] The present invention provides a fault prediction method for a track power supply panel, comprising:

[0007] Performing anomaly detection on circuit parameter timing samples based on an anomaly detection algorithm to obtain abnormal samples, and marking the fault types of the abnormal samples;

[0008] Acquire a target circuit parameter sample according to the abnormal sample and data after the abnormal sample is labeled with a fault type;

[0009] Acquiring a fault prediction model according to the target circuit parameter sample;

[0010] Input the circuit parameters and prediction period of the target power panel in the test set into the fault prediction model to obtain the probability of failure of the target power panel;

[0011] Wherein, the circuit parameter timing samples are determined according to the circuit parameters of the target power panel in the training set;

[0012] The collection of circuit parameters of the target power panel in the training set precedes the collection of circuit parameters of the target power panel in the test set.

[0013] According to a fault prediction method for a track power supply panel provided by the present invention, the circuit parameter timing samples are obtained in the following manner:

[0014] According to the types of the circuit parameters of the target power panel in the training set, the circuit parameters of the target power panel in the training set are divided into time series at preset intervals to obtain the circuit parameter time series samples.

[0015] According to a fault prediction method for a track power supply panel provided by the present invention, the abnormality detection of circuit parameter timing samples based on an abnormality detection algorithm to obtain abnormal samples includes:

[0016] Acquire statistical characteristics of the circuit parameter timing samples, and determine a characteristic matrix of the circuit parameter timing samples according to the statistical characteristics;

[0017] Performing anomaly detection on the feature matrix based on an isolation forest anomaly detection algorithm or a local outlier coefficient anomaly detection algorithm to obtain an anomaly score of the circuit parameter timing sample;

[0018] Determining the abnormal sample according to the abnormal score;

[0019] The statistical characteristics include at least mean, standard deviation and entropy.

[0020] According to a fault prediction method for a track power supply panel provided by the present invention, the fault prediction model is obtained according to the target circuit parameter sample, comprising:

[0021] Analyzing the target circuit parameter samples to determine the abnormal time of each fault type;

[0022] Obtaining a maximum value of the abnormal time, and using the maximum value as the prediction period;

[0023] Scaling the target circuit parameter sample geometrically according to the prediction period to obtain a prediction vector;

[0024] Determining parameters of the fault prediction model according to the prediction vector;

[0025] Determining the fault prediction model according to the parameters of the fault model;

[0026] The prediction vector is used to characterize the probability of failure of the target power supply panel.

[0027] According to a method for predicting a fault of a track power supply panel provided by the present invention, the method of obtaining the probability of the target power supply panel failing comprises:

[0028] According to the circuit parameters of the target power panel in the test set and the prediction period, obtaining a slope vector within a prediction period from the start time to the current time;

[0029] Matching the slope vector with the prediction vector output by the fault prediction model, and determining the probability of failure of the target power panel according to the matching result;

[0030] Wherein, the slope vector is determined according to adjacent data points in the circuit parameters;

[0031] The prediction vector is an average value of the slope vectors of the target circuit parameter samples after geometric scaling.

[0032] According to a fault prediction method for a track power supply panel provided by the present invention, the fault prediction model is obtained according to the target circuit parameter sample, and further includes:

[0033] According to the target circuit parameter samples, dividing the target circuit parameter samples at intervals of the preset time to obtain target circuit parameter timing samples;

[0034] Performing frequency domain decomposition on the target circuit parameter time series samples to obtain seasonal time series items, trend time series items, and error items corresponding to the target circuit parameter time series samples;

[0035] Determining parameters of the fault prediction model according to the seasonal time series term, the trend time series term and the error term;

[0036] The fault prediction model is determined according to the parameters of the fault prediction model.

[0037] According to a fault prediction method for a track power supply panel provided by the present invention, the circuit parameters include:

[0038] Voltage, current, frequency, phase and harmonics of track power supply;

[0039] Voltage, current, frequency, phase and harmonics of local power supply;

[0040] The phase angle of the local power supply and the phase angle of the track power supply;

[0041] The pickup and release time and return coefficient of the relay in the track power supply panel;

[0042] Track relay occupancy and idle status data;

[0043] Voltage, current, phase, harmonics and frequency at the supply end;

[0044] Power on / off data between the transformer box and the signal end of the choke transformer;

[0045] The voltage, current, phase, harmonics and frequency provided by the power transformer to the target power panel; and the temperature and humidity inside the transformer box.

[0046] The present invention also provides a fault prediction device for a track power supply panel, comprising: an abnormal sample marking module, a target sample acquisition module, a fault prediction model determination module and a fault prediction module;

[0047] The abnormal sample labeling module is used to perform abnormal detection on the circuit parameter timing samples based on the abnormal detection algorithm to obtain abnormal samples, and label the abnormal samples with fault types;

[0048] The target sample acquisition module is used to acquire a target circuit parameter sample according to the abnormal sample and the data after the abnormal sample is labeled with the fault type;

[0049] The fault prediction model determination module is used to obtain a fault prediction model according to the target circuit parameter sample;

[0050] The fault prediction module is used to input the circuit parameters and prediction period of the target power panel in the test set into the fault prediction model to obtain the probability of failure of the target power panel;

[0051] Wherein, the circuit parameter timing samples are determined according to the circuit parameters of the target power panel in the training set;

[0052] The collection of circuit parameters of the target power panel in the training set precedes the collection of circuit parameters of the target power panel in the test set.

[0053] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the fault prediction method for the track power supply panel as described in any one of the above are implemented.

[0054] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the fault prediction method for the track power supply panel as described in any one of the above are implemented.

[0055] The fault prediction method for a track power supply panel provided by the present invention performs abnormal detection on the circuit parameters of the track power supply panel to obtain abnormal samples, and constructs an accurate power supply panel fault prediction model by marking the fault types of the abnormal samples in combination with the timing feature method. The method can realize automatic prediction of power supply panel faults based on the construction of an accurate fault prediction model, and at the same time improve the accuracy of power supply panel fault prediction. Compared with the prior art in which operation and maintenance personnel predict power supply panel faults based on prior experience, the efficiency of power supply panel fault prediction is improved, and operation and maintenance personnel need to perform fault prediction for different power supply panels separately when performing fault prediction for different power supply panels, which has low versatility. Fault prediction for different power supply panels can be realized by constructing a fault prediction model, which improves the applicability of power supply panel fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 It is a schematic flow chart of a fault prediction method for a track power panel provided by the present invention;

[0058] Figure 2 It is a schematic diagram of the flow of circuit parameter data collection of the track power supply panel provided by the present invention;

[0059] Figure 3 It is a structural schematic diagram of a fault prediction device for a track power supply panel provided by the present invention;

[0060] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] Figure 1 It is a flow chart of the fault prediction method of the track power panel provided by the present invention, such as Figure 1 As shown, the method includes:

[0063] S1. Perform anomaly detection on circuit parameter timing samples based on anomaly detection algorithm to obtain abnormal samples, and mark the fault types of the abnormal samples;

[0064] S2. Obtain target circuit parameter samples based on abnormal samples and data after fault type annotation of the abnormal samples;

[0065] S3. Obtain a fault prediction model based on the target circuit parameter sample;

[0066] S4. Input the circuit parameters and prediction period of the target power panel in the test set into the fault prediction model to obtain the probability of failure of the target power panel;

[0067] Among them, the circuit parameter timing samples are determined according to the circuit parameters of the target power panel in the training set;

[0068] The collection of circuit parameters of the target power panel in the training set precedes the collection of circuit parameters of the target power panel in the test set.

[0069] It should be noted that the execution subject of the above method may be a computer device.

[0070] Optionally, the circuit parameter related data of the track power supply panel is collected through the data collection module, and the collected circuit parameter related data are divided into a training set and a test set according to the collection time.

[0071] Optionally, the circuit parameter-related data collected first is used as a training set, and the circuit parameter-related data collected after the training set is used as a test set. The circuit parameters of the power panel in the training set are used to obtain the target circuit parameter samples required to build the fault prediction model, and the circuit parameters of the power panel in the test set are used as input data to the fault prediction model. According to the trained fault prediction model, the probability of the power panel failure is obtained.

[0072] Optionally, the circuit parameters of the power panel in the collected training set are divided into time series to obtain circuit parameter time series samples of the power panel, and anomaly detection is performed on the circuit parameter time series samples based on an anomaly detection algorithm to obtain abnormal class data in the circuit parameter time series samples, and abnormal samples are constructed based on the abnormal class data.

[0073] The operation and maintenance personnel shall mark the fault type of the power panel based on the circuit parameter timing samples corresponding to the fault type in the expert database and the screened abnormal samples. The expert database may be pre-established based on the historical power panel fault information and its corresponding fault data. For example, the expert database may be established by further refining the fault data type corresponding to the power panel fault information.

[0074] Optionally, the fault information refers to a description of the power supply panel fault type. For example, the fault type is "starting circuit failure" of the power supply panel, and the fault information corresponding to the fault type is "indoor equipment failure" or "outdoor equipment failure". It can be seen that the fault information actually includes the cause of the fault, but the cause is not specified to a more specific device. The fault information is refined step by step through an expert database, such as a pre-established relational database, that is, the cause of the power supply panel failure is determined step by step until the power supply panel fault data corresponding to a certain fault information that cannot be refined is traced back.

[0075] Optionally, when the fault information is "indoor equipment failure", it can be further refined into: "track power supply failure", "local power supply failure" and "track circuit relay failure"; when the fault information is "outdoor equipment failure", it can be further refined into: "power transmission end equipment failure", "transformer box and choke transformer equipment failure" and "power supply voltage failure".

[0076] It should be noted that the "track power failure", "local power failure", "track circuit relay failure", "power transmission end equipment failure", "transformer box and choke transformer equipment failure" and "power supply voltage failure" in the present invention can still be further refined according to operation and maintenance experience until the corresponding source of the faulty equipment is found. The present invention does not make specific limitations on this.

[0077] According to the relationship between fault type, fault information and fault data, an expert database describing the fault types of power supply panels is finally established.

[0078] Optionally, the abnormal samples and the fault type annotation data corresponding to the abnormal samples are integrated to form a data set, and the obtained data set is used as a target circuit parameter sample for subsequent construction of a fault prediction model based on the target circuit parameter sample. Optionally, the target circuit parameter sample is input into a pre-established fault prediction model for training, the parameters of the pre-established fault prediction model are determined, and then a fault prediction model for power panel fault prediction is constructed according to the parameters of the pre-established fault prediction model.

[0079] Among them, the pre-established fault prediction model can be a corresponding decision tree machine learning model or a corresponding support vector machine model, and the present invention does not make specific limitations on this.

[0080] According to the constructed fault prediction model, the circuit parameters and prediction cycle of the power supply panel in the collected test set are input into the trained fault prediction model, and the probability of failure of the power supply panel is determined according to the output result of the fault prediction model, such as the prediction vector.

[0081] The fault prediction method for a track power supply panel provided by the present invention performs abnormal detection on the circuit parameters of the track power supply panel to obtain abnormal samples, and constructs an accurate power supply panel fault prediction model by marking the fault types of the abnormal samples in combination with the timing feature method. The method can realize automatic prediction of power supply panel faults based on the construction of an accurate fault prediction model, and at the same time improve the accuracy of power supply panel fault prediction. Compared with the prior art in which operation and maintenance personnel predict power supply panel faults based on prior experience, the efficiency of power supply panel fault prediction is improved, and operation and maintenance personnel need to perform fault prediction for different power supply panels separately when performing fault prediction for different power supply panels, which has low versatility. Fault prediction for different power supply panels can be realized by constructing a fault prediction model, which improves the applicability of power supply panel fault prediction.

[0082] Furthermore, in one embodiment, the circuit parameter timing samples are obtained in the following manner:

[0083] A1. According to the type of the circuit parameters of the target power panel in the training set, the circuit parameters of the target power panel in the training set are divided into time series at preset intervals to obtain circuit parameter time series samples;

[0084] Optionally, data related to circuit parameters of the power supply panel are obtained according to a data acquisition module, wherein the data acquisition module may be an industrial computer, and the circuit parameters of the power supply panel in a continuous period of time may be acquired based on the industrial computer.

[0085] In the actual acquisition process, as many different types of circuit parameters of the power supply panel as possible are collected, and the collected data related to various types of continuous circuit parameters are divided into paths according to the types of circuit parameters of the power supply panel.

[0086] Optionally, the same type of circuit parameters of the power supply panel are taken as the same path sequence, such as circuit parameters such as voltage, current, and phase; different types of circuit parameters of the power supply panel are divided into different path sequences, for example, the relevant circuit parameters of the temperature and humidity of the corresponding outdoor transformer box of the power supply panel are taken as one path sequence, and the power on and off between the corresponding transformer box of the power supply panel and the signal end of the choke transformer is taken as another path sequence.

[0087] Optionally, the circuit parameters of the power supply panel corresponding to each of the above-mentioned sequences are divided into time series at preset time intervals. For example, the sequences corresponding to various types of continuous circuit parameter related data are divided every T seconds to obtain timing samples corresponding to various types of circuit parameter related data. The obtained timing samples corresponding to different types of circuit parameter related data are integrated to obtain the circuit parameter timing samples of the power supply panel.

[0088] The fault prediction method for a track power supply panel provided by the present invention divides the collected circuit parameters of different types of power supply panels into time series, refines the circuit parameters of the power supply panel, and obtains circuit parameter time series samples required for building a fault prediction model based on the refined circuit parameters. Since the refined circuit parameters can characterize the fault type of the power supply panel in more detail and accuracy, the accuracy of predicting the power supply panel fault based on the constructed fault prediction model is higher, and an accurate description of the fault type can be achieved.

[0089] Furthermore, in one embodiment, step S1 may specifically include:

[0090] S11, obtaining statistical characteristics of the circuit parameter timing samples, and determining a characteristic matrix of the circuit parameter timing samples according to the statistical characteristics;

[0091] S12, performing anomaly detection on the feature matrix based on an isolation forest anomaly detection algorithm or a local outlier coefficient anomaly detection algorithm to obtain an anomaly score of the circuit parameter timing sample;

[0092] S13, determining abnormal samples according to the abnormality score;

[0093] The statistical characteristics include at least mean, standard deviation and entropy.

[0094] Optionally, according to the circuit parameter timing samples obtained above, Circuit parameter timing samples in t i For example, an unsupervised timing anomaly detection algorithm based on statistical features is used to perform anomaly detection, and the statistical features of the circuit parameter timing samples corresponding to each path i, such as the mean, standard deviation, and entropy features, are extracted to form a feature matrix F of the circuit parameter timing samples.

[0095] A statistical learning anomaly detection algorithm based on isolation forest and / or local outlier coefficient is used to calculate the anomaly score of each circuit parameter timing sample, and based on the anomaly score of the circuit parameter timing sample, the abnormal samples in the circuit parameter timing sample are determined based on statistical significance. This process is repeated to filter out abnormal sample data from massive circuit parameter data, so that the fault type of the power supply panel can be labeled based on the abnormal samples.

[0096] Among them, the process of calculating the anomaly score of the isolation forest anomaly detection algorithm is as follows:

[0097] B1. Take 80% of the circuit parameter timing samples as training data, randomly select Ψ points from the training data as sub-samples, and put them into the root node of an isolated tree;

[0098] B2. Randomly specify a dimension and randomly generate a cutting point p within the current node data range, where the cutting point is generated between the maximum and minimum values ​​of the specified dimension in the current node data;

[0099] B3. The selection of this cutting point generates a hyperplane, which divides the current node data space into two subspaces: points less than p under the currently selected dimension are placed on the left branch of the current node, and points greater than or equal to p are placed on the right branch of the current node;

[0100] B4, recursively repeat steps B2 and B3 on the left and right branches of the node, and continuously construct new leaf nodes until there is only one data on the leaf node (no further cutting is possible) or the tree has grown to the set height;

[0101] B5. Repeat steps B1-B4 until a preset number of isolated trees are obtained, integrate all isolated trees, use the generated isolated trees to evaluate the remaining 20% ​​of the circuit parameter timing samples, and calculate the abnormal score of the circuit parameter timing samples using formula (1):

[0102]

[0103] In the formula, x represents the circuit parameter timing sample, h(x) represents the height of x in each tree, and c(ψ) represents the average path length when the number of samples is ψ, which is used to standardize the path height of sample x.

[0104] If the anomaly score is close to 1, then it must be an outlier;

[0105] If the anomaly score is much less than 0.5, then it is definitely not an anomaly;

[0106] If the anomaly score for all points is around 0.5, then there is a high probability that there are no anomalies in the sample.

[0107] Similarly, the anomaly score of the circuit parameter timing samples can be calculated by using a local outlier coefficient anomaly detection algorithm such as a LOF algorithm.

[0108] The fault prediction method for the track power supply panel provided by the present invention obtains the statistical characteristics of the circuit parameter timing samples of the power supply panel, such as the mean, standard deviation or entropy, so that the characterization of the characteristics of the circuit parameter timing samples is more accurate, and performs anomaly detection on the circuit parameter timing samples based on the isolated forest anomaly detection algorithm or the local outlier coefficient anomaly detection algorithm. Compared with the traditional Kmeans and DBScan algorithms, since it does not need to calculate distance-related indicators, it can quickly complete the anomaly detection of the circuit parameter timing samples, greatly improving the anomaly detection speed, and further improving the speed of obtaining abnormal samples.

[0109] Furthermore, in one embodiment, step S3 may specifically include:

[0110] S31, analyzing the target circuit parameter samples to determine the abnormal time of each fault type;

[0111] S32, obtaining the maximum value of the abnormal time, and taking the maximum value as the prediction period;

[0112] S33, scaling the target circuit parameter samples geometrically according to the prediction period to obtain a prediction vector;

[0113] S34, determining the parameters of the fault prediction model according to the prediction vector;

[0114] S35, determining a fault prediction model according to the parameters of the fault model;

[0115] The prediction vector is used to characterize the probability of failure of the target power panel.

[0116] Optionally, based on business logic, the fault prediction of the power supply panel is mainly divided into two parts: short-term prediction and long-term prediction. Among them, short-term faults mainly predict the probability of a certain type of fault within a unit time in the future, and long-term fault prediction mainly predicts the probability of a certain type of fault in the next week or even longer.

[0117] Optionally, the specific implementation process of short-term prediction is as follows:

[0118] The obtained target circuit parameter samples are analyzed to obtain the time from normal state to fault state of each type of circuit parameter that determines each fault type of the power panel, and the time from normal state to fault state of the circuit parameter is defined as the abnormal time, which is recorded as T1~T num , num represents the total number of each type of fault samples in the target circuit parameter samples.

[0119] By taking the maximum value of the abnormal time zone as the prediction period T for short-term prediction s , and T s =max[T1,T2,...,Tnum ], and the num fault sample data are calculated according to the prediction period T s Scaling to the same dimension, calculating the slope k between adjacent data points, we can get the slope vector of num fault data, which can be expressed as k 1～num =(k1,k2,...,k num ), and determine the prediction vector of the fault prediction model for short-term prediction based on the obtained slope vector.

[0120] The prediction vector of the short-term prediction fault prediction model is used as the parameter of the fault prediction model. According to the parameter of the fault prediction model, a fault prediction model is constructed for short-term fault prediction of the power supply panel.

[0121] The fault prediction method for a track power supply panel provided by the present invention analyzes the target circuit parameter samples to obtain the parameters of a fault prediction model for short-term prediction, and determines the fault prediction model according to the obtained fault model parameters, thereby realizing accurate prediction of short-term faults of the power supply panel. Since the abnormal time of the fault type of the power supply panel is used when determining the fault prediction model, when the fault prediction model is used to predict short-term faults of the power supply panel, the time of occurrence of each fault type of the power supply panel can be accurately predicted.

[0122] Furthermore, in one embodiment, step S4 may specifically include:

[0123] S41, according to the circuit parameters and prediction period of the target power panel in the test set, obtaining a slope vector within a prediction period from the start time to the current time;

[0124] S42, matching the slope vector with the prediction vector output by the fault prediction model, and determining the probability of failure of the target power panel according to the matching result;

[0125] Among them, the slope vector is determined based on adjacent data points in the circuit parameters;

[0126] The prediction vector is the average value of the slope vector of the target circuit parameter samples after geometric scaling. Optionally, the circuit parameters of the target power panel in the test set and the prediction period T s Input to the fault prediction model to obtain the prediction vector. The prediction vector is obtained by calculating according to the prediction period T s The average value of the slope vector of the geometrically scaled target circuit parameter samples.

[0127] Optionally, according to the circuit parameters of the target power panel in the test set and the prediction period T s , calculate a prediction period T from the start time to the current time sThe slope vector in the calculation will obtain a prediction period T from the start time to the current time. s The slope vector in is matched with the above-obtained prediction vector. If the calculated prediction period T from the start time to the current time s If the difference between the slope vector in and the above-obtained prediction vector is within the preset threshold range, it is determined that the probability of the power panel failing at the next time point is high, where the difference between the next time point and the current time point is a prediction period T s .

[0128] Optionally, different threshold ranges can be set according to different prediction accuracies. For example, if a prediction period T from the start time to the current time is s If the difference between the slope vector in and the above-obtained prediction vector is within 0-0.05, the probability of the power panel failing at the next time point is determined to be more than 95%; if a prediction period T from the start time to the current time is s If the difference between the slope vector in and the above-obtained prediction vector is within 0-0.5, it is determined that the probability of the power panel failing at the next time point is more than 50%. A prediction period T from the start time to the current time s The relationship between the difference between the slope vector within and the above-mentioned prediction vector and the probability of power supply panel failure can be set according to the prior experience of professional operation and maintenance personnel, and the present invention does not make specific limitations on this.

[0129] The fault prediction method for a track power supply panel provided by the present invention matches the prediction vector output by the fault prediction model with the slope vector of the circuit parameters of the power supply panel in the test set obtained by calculation within a prediction period from the start time to the current time, so as to determine the probability of the power supply panel failing within a prediction period from the current time to the next time, and at the same time, quantifies the probability of the power supply panel failing according to the result of matching the prediction vector with the slope vector, thereby improving the accuracy of the short-term fault prediction of the power supply panel.

[0130] Furthermore, in one embodiment, step S3 may also include:

[0131] S36, dividing the target circuit parameter samples at preset time intervals according to the target circuit parameter samples to obtain target circuit parameter timing samples;

[0132] S37, performing frequency domain decomposition on the target circuit parameter time series samples to obtain seasonal time series items, trend time series items, and error items corresponding to the target circuit parameter time series samples;

[0133] S38, determining the parameters of the fault prediction model according to the seasonal time series term, the trend time series term and the error term;

[0134] S39. Determine the fault prediction model according to the parameters of the fault prediction model.

[0135] Optionally, when performing long-term fault prediction on the power supply panel, since the long-term prediction of various types of faults in the power supply panel can be converted into a prediction of various basic data trends of the power supply panel, the time of occurrence of the fault can be reversed based on the predicted trend to achieve the purpose of long-term fault prediction.

[0136] Therefore, when predicting long-term faults of power panels, the target circuit parameter samples are firstly obtained and divided into target circuit parameter samples at preset time intervals to obtain target circuit parameter timing samples y t ;

[0137] Secondly, since a time series of basic data can be composed of several time series superimposed or coupled in the frequency domain, the decomposition model of the time series evolved based on this is used to convert the target parameter time series sample y into t Perform frequency domain decomposition to obtain the seasonal time series terms, trend time series terms, and error terms corresponding to the target circuit parameter time series samples:

[0138] y t =S t +T t +R t (2)

[0139] y t =S t *T t *R t (3)

[0140] In the formula, S t represents the seasonal time series term, T t represents the trend time series term, R t represents the error term;

[0141] The seasonal time series term S t The trend time series term T changes with seasonal changes such as day, week, month and year. t Indicates the changing trend of the time series in the non-periodic state, R t Representation refers to other factors that affect time series data, such as the precision of the device used to collect the data.

[0142] In addition, the target circuit parameter timing samples can be decomposed into an addition / multiplication mixed model composed of formula (2) and formula (3). When designing features, it is possible to define the workload of a single week (month, quarter) as a periodic variable according to different degrees for calculation, that is, different long-term fault prediction cycles can be set.

[0143] According to the seasonal time series item St , trend time series term T t and the error term R t , constitute the parameters of the fault prediction model, and determine the fault prediction model for long-term fault prediction based on the parameters of the fault model.

[0144] Optionally, when performing long-term fault prediction on a power panel, the circuit parameters of the target power panel in the collected test set and the prediction period of the long-term fault prediction are input into the fault prediction model, and the seasonal time series item S is obtained. t , trend time series term T t and the error term R t .

[0145] Professional operation and maintenance personnel obtain seasonal time series items S t Seasonal timing items when the power panel is operating normally, and trend timing items T t The trend timing items when the power panel is operating normally and the error item R obtained according to t The error term is compared with that when the power panel is normal to determine whether the power panel will fail within the prediction period of the long-term fault prediction.

[0146] The fault prediction method for a track power supply panel provided by the present invention obtains the seasonal timing items, trend timing items and error items required for long-term fault prediction of the power supply panel by performing frequency domain decomposition on the target circuit parameter timing samples, and compares the acquired seasonal timing items, trend timing items and error items with the seasonal timing items, trend timing items and error items during normal operation of the power supply panel, thereby realizing long-term fault prediction of the power supply panel.

[0147] Furthermore, in one embodiment, the circuit parameters of the power panel include the following data:

[0148] Voltage, current, frequency, phase and harmonics of track power supply;

[0149] Voltage, current, frequency, phase and harmonics of local power supply;

[0150] The phase angle of the local power supply and the phase angle of the track power supply;

[0151] The pickup and release time and return coefficient of the relay in the track power supply panel;

[0152] Track relay occupancy and idle status data;

[0153] Voltage, current, phase, harmonics and frequency at the supply end;

[0154] Power on / off data between the transformer box and the signal end of the choke transformer;

[0155] The voltage, current, phase, harmonics and frequency provided by the power transformer to the target power panel; and the temperature and humidity inside the transformer box.

[0156] Alternatively, if Figure 2 As shown, when predicting faults of a power supply panel, since the power supply panel relies on different circuit parameters corresponding to different fault types, in actual scenarios, a data acquisition module is often used to collect data on the circuit parameters of the power supply panel from both outdoor equipment and indoor equipment.

[0157] Optionally, the indoor equipment collection mainly collects the following circuit parameters of the power panel:

[0158] Track power supply, the collected data include: voltage, current, frequency, phase and harmonics;

[0159] Local power supply, the collected data include: voltage, current, frequency, phase and harmonics;

[0160] The phase angle of the local power supply and the phase angle of the track power supply;

[0161] Performance of key relays in track circuits, pickup and release time, return coefficient, etc.;

[0162] Track relay occupancy and idle status data;

[0163] Optionally, the outdoor equipment collects the following circuit parameters of the main power panel:

[0164] Power quality at the transmission end, including voltage, current, phase, harmonics and frequency;

[0165] Power on / off data between transformer XB box and choke transformer BE signal terminal;

[0166] The quality of the power supplied to the equipment by the power transformer BG output, including voltage, current, phase, harmonics and frequency;

[0167] Temperature and humidity inside transformer XB box.

[0168] The fault prediction method for the track power supply panel provided by the present invention collects the circuit parameters of the power supply panel through the data acquisition module, which can provide abundant circuit parameter samples of the power supply panel for constructing a fault prediction model, thereby enabling the fault prediction model established based on this to more comprehensively and accurately reflect the operating status of the power supply panel.

[0169] The fault prediction device of the track power supply panel provided by the present invention is described below. The fault prediction device of the track power supply panel described below and the fault prediction method of the track power supply panel described above can be referred to each other.

[0170] Figure 3Schematic diagram of the structure of the fault prediction device of the track power panel provided by the present invention. Figure 3 As shown, it includes: an abnormal sample marking module 310, a target sample acquisition module 311, a fault prediction model determination module 312 and a fault prediction module 313;

[0171] The abnormal sample marking module 310 is used to perform abnormal detection on the circuit parameter timing samples based on the abnormal detection algorithm to obtain abnormal samples, and mark the fault types of the abnormal samples;

[0172] The target sample acquisition module 311 is used to acquire target circuit parameter samples according to abnormal samples and fault type annotation data;

[0173] A fault prediction model determination module 312 is used to obtain a fault prediction model according to a target circuit parameter sample;

[0174] A fault prediction module 313 is used to input the circuit parameters and prediction period of the target power panel in the test set into the fault prediction model to obtain the probability of the target power panel failing;

[0175] Among them, the circuit parameter timing samples are determined according to the circuit parameters of the target power panel in the training set;

[0176] The circuit parameters of the target power panel in the training set are collected before the circuit parameters of the target power panel in the test set.

[0177] The fault prediction device for the track power supply panel provided by the present invention performs abnormal detection on the circuit parameters of the track power supply panel based on the abnormal sample labeling module 310, obtains abnormal samples, and labels the fault types of the abnormal samples. In combination with the timing feature method, the target sample acquisition module 311 is used to integrate the abnormal samples and the fault type labeling data to obtain the target circuit parameter samples, and the fault prediction model determination module 312 is used to build an accurate power supply panel fault prediction model. The fault prediction module 313 can realize automatic prediction of power supply panel faults based on the construction of an accurate fault prediction model, and at the same time improves the accuracy of power supply panel fault prediction. Compared with the prior art in which operation and maintenance personnel predict power supply panel faults based on prior experience, the efficiency of power supply panel fault prediction is improved, and the operation and maintenance personnel need to perform fault prediction on different power supply panels separately when predicting faults for different power supply panels, which has low versatility. By constructing a fault prediction model, fault prediction of different power supply panels can be realized, and the applicability of power supply panel fault prediction is improved.

[0178] Figure 4 is a schematic diagram of the physical structure of an electronic device provided by the present invention, such as Figure 4As shown, the electronic device may include: a processor 410, a communication interface 411, a memory 412 and a bus 413, wherein the processor 410, the communication interface 411 and the memory 412 communicate with each other through the bus 413. The processor 410 may call the logic instructions in the memory 412 to execute the following method:

[0179] Based on the anomaly detection algorithm, the circuit parameter timing samples are detected for anomalies to obtain abnormal samples, and the fault types of the abnormal samples are marked;

[0180] Obtain target circuit parameter samples based on abnormal samples and data after fault type annotation of the abnormal samples;

[0181] Obtaining a fault prediction model according to a target circuit parameter sample;

[0182] Input the circuit parameters and prediction period of the target power panel in the test set into the fault prediction model to obtain the probability of failure of the target power panel;

[0183] Among them, the circuit parameter timing samples are determined according to the circuit parameters of the target power panel in the training set;

[0184] The collection of circuit parameters of the target power panel in the training set precedes the collection of circuit parameters of the target power panel in the test set.

[0185] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0186] Furthermore, the present invention discloses a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, when the program instructions are executed by a computer, the computer can execute the fault prediction method of the track power supply panel provided by the above-mentioned method embodiments, for example, comprising:

[0187] Based on the anomaly detection algorithm, the circuit parameter timing samples are detected for anomalies to obtain abnormal samples, and the fault types of the abnormal samples are marked;

[0188] Obtain target circuit parameter samples based on abnormal samples and data after fault type annotation of the abnormal samples;

[0189] Obtaining a fault prediction model according to a target circuit parameter sample;

[0190] Input the circuit parameters and prediction period of the target power panel in the test set into the fault prediction model to obtain the probability of failure of the target power panel;

[0191] Among them, the circuit parameter timing samples are determined according to the circuit parameters of the target power panel in the training set;

[0192] The collection of circuit parameters of the target power panel in the training set precedes the collection of circuit parameters of the target power panel in the test set.

[0193] On the other hand, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the computer program is executed by a processor to execute the fault prediction method of the track power supply panel provided in the above embodiments, for example, including:

[0194] Based on the anomaly detection algorithm, the circuit parameter timing samples are detected for anomalies to obtain abnormal samples, and the fault types of the abnormal samples are marked;

[0195] Obtain target circuit parameter samples based on abnormal samples and data after fault type annotation of the abnormal samples;

[0196] Obtaining a fault prediction model according to a target circuit parameter sample;

[0197] Input the circuit parameters and prediction period of the target power panel in the test set into the fault prediction model to obtain the probability of failure of the target power panel;

[0198] Among them, the circuit parameter timing samples are determined according to the circuit parameters of the target power panel in the training set;

[0199] The collection of circuit parameters of the target power panel in the training set precedes the collection of circuit parameters of the target power panel in the test set.

[0200] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Those of ordinary skill in the art may understand and implement it without creative effort.

[0201] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0202] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fault prediction method for a track power supply panel, characterized in that: include: Performing anomaly detection on circuit parameter timing samples based on an anomaly detection algorithm to obtain abnormal samples, and marking the fault types of the abnormal samples according to a preset expert database; The expert database is established by refining the fault type corresponding to the fault information of the power supply panel; the fault information includes the cause of the fault, and the cause of the fault is determined by gradually refining the fault information according to the expert database; Acquire a target circuit parameter sample according to the abnormal sample and data after the abnormal sample is labeled with a fault type; Acquiring a fault prediction model according to the target circuit parameter sample; Input the circuit parameters and prediction period of the target power panel in the test set into the fault prediction model to obtain the probability of failure of the target power panel; Wherein, the circuit parameter timing samples are determined according to the circuit parameters of the target power panel in the training set; The collection of circuit parameters of the target power panel in the training set precedes the collection of circuit parameters of the target power panel in the test set; The method of performing anomaly detection on the circuit parameter timing samples based on an anomaly detection algorithm to obtain an abnormal sample includes: Acquire statistical characteristics of the circuit parameter timing samples, and determine a characteristic matrix of the circuit parameter timing samples according to the statistical characteristics; Performing anomaly detection on the feature matrix based on an isolation forest anomaly detection algorithm or a local outlier coefficient anomaly detection algorithm to obtain an anomaly score of the circuit parameter timing sample; Determining the abnormal sample according to the abnormal score; Wherein, the statistical characteristics include at least: mean, standard deviation and entropy; The acquiring the fault prediction model according to the target circuit parameter sample comprises: Analyze the target circuit parameter samples to determine the change time of each fault type; the change time is used to characterize the time for each different type of circuit parameter on which the fault type depends to change from a normal state to a fault state; Obtaining a maximum value of the abnormal time, and using the maximum value as the prediction period; Scaling the target circuit parameter sample geometrically according to the prediction period to obtain a prediction vector; Determining parameters of the fault prediction model according to the prediction vector; Determining the fault prediction model according to the parameters of the fault prediction model; Wherein, the prediction vector is used to characterize the probability of failure of the target power panel; The circuit parameter timing samples are obtained in the following manner: According to the types of the circuit parameters of the target power panel in the training set, the circuit parameters of the target power panel in the training set are divided into time series at preset intervals to obtain time series samples of the circuit parameters; The obtaining the probability of the target power supply panel failing includes: According to the circuit parameters of the target power panel in the test set and the prediction period, obtaining a slope vector within a prediction period from the start time to the current time; Matching the slope vector with the prediction vector output by the fault prediction model, and determining the probability of failure of the target power panel according to the matching result; Wherein, the slope vector is determined according to adjacent data points in the circuit parameters; The prediction vector is the average value of the slope vectors of the target circuit parameter samples after geometric scaling; The acquiring of a fault prediction model according to the target circuit parameter sample comprises: According to the target circuit parameter samples, dividing the target circuit parameter samples at intervals of the preset time to obtain target circuit parameter timing samples; Performing frequency domain decomposition on the target circuit parameter time series samples to obtain seasonal time series items, trend time series items, and error items corresponding to the target circuit parameter time series samples; Determining parameters of the fault prediction model according to the seasonal time series term, the trend time series term and the error term; The fault prediction model is determined according to the parameters of the fault prediction model.

2. The fault prediction method for the track power supply panel according to claim 1 is characterized in that: The circuit parameters include: Voltage, current, frequency, phase and harmonics of track power supply; Voltage, current, frequency, phase and harmonics of local power supply; The phase angle of the local power supply and the phase angle of the track power supply; The pickup and release time and return coefficient of the relay in the track power supply panel; Track relay occupancy and idle status data; Voltage, current, phase, harmonics and frequency at the supply end; Power on / off data between the transformer box and the signal end of the choke transformer; The voltage, current, phase, harmonics and frequency provided by the power transformer to the target power panel; and the temperature and humidity inside the transformer box.

3. A fault prediction device for a track power panel, characterized in that: include: Abnormal sample labeling module, target sample acquisition module, fault prediction model determination module and fault prediction module; The abnormal sample labeling module is used to perform abnormal detection on the circuit parameter timing samples based on the abnormal detection algorithm to obtain abnormal samples, and label the abnormal samples with fault types according to a preset expert database; The expert database is established by refining the fault type corresponding to the fault information of the power supply panel; the fault information includes the cause of the fault, and the cause of the fault is determined by gradually refining the fault information according to the expert database; The target sample acquisition module is used to acquire a target circuit parameter sample according to the abnormal sample and the data after the abnormal sample is labeled with the fault type; The fault prediction model determination module is used to obtain a fault prediction model according to the target circuit parameter sample; The fault prediction module is used to input the circuit parameters and prediction period of the target power panel in the test set into the fault prediction model to obtain the probability of failure of the target power panel; Wherein, the circuit parameter timing samples are determined according to the circuit parameters of the target power panel in the training set; The collection of circuit parameters of the target power panel in the training set precedes the collection of circuit parameters of the target power panel in the test set; The abnormal sample labeling module is also used for: Acquire statistical characteristics of the circuit parameter timing samples, and determine a characteristic matrix of the circuit parameter timing samples according to the statistical characteristics; Performing anomaly detection on the feature matrix based on an isolation forest anomaly detection algorithm or a local outlier coefficient anomaly detection algorithm to obtain an anomaly score of the circuit parameter timing sample; Determining the abnormal sample according to the abnormal score; Wherein, the statistical characteristics include at least: mean, standard deviation and entropy; The fault prediction model determination module is also used for: Analyze the target circuit parameter samples to determine the change time of each fault type; the change time is used to characterize the time for each different type of circuit parameter on which the fault type depends to change from a normal state to a fault state; Obtaining a maximum value of the abnormal time, and using the maximum value as the prediction period; Scaling the target circuit parameter sample geometrically according to the prediction period to obtain a prediction vector; Determining parameters of the fault prediction model according to the prediction vector; Determining the fault prediction model according to the parameters of the fault prediction model; Wherein, the prediction vector is used to characterize the probability of failure of the target power panel; The circuit parameter timing samples are obtained in the following manner: According to the types of the circuit parameters of the target power panel in the training set, the circuit parameters of the target power panel in the training set are divided into time series at preset intervals to obtain time series samples of the circuit parameters; The fault prediction module is further used for: According to the circuit parameters of the target power panel in the test set and the prediction period, obtaining a slope vector within a prediction period from the start time to the current time; Matching the slope vector with the prediction vector output by the fault prediction model, and determining the probability of failure of the target power panel according to the matching result; Wherein, the slope vector is determined according to adjacent data points in the circuit parameters; The prediction vector is the average value of the slope vectors of the target circuit parameter samples after geometric scaling; The fault prediction model determination module is also used for: According to the target circuit parameter samples, dividing the target circuit parameter samples at intervals of the preset time to obtain target circuit parameter timing samples; Performing frequency domain decomposition on the target circuit parameter time series samples to obtain seasonal time series items, trend time series items, and error items corresponding to the target circuit parameter time series samples; Determining parameters of the fault prediction model according to the seasonal time series term, the trend time series term and the error term; The fault prediction model is determined according to the parameters of the fault prediction model.

4. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the fault prediction method for the track power supply panel as described in any one of claims 1 to 2 are implemented.

5. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the fault prediction method for the track power supply panel as claimed in any one of claims 1 to 2 are implemented.

Images