Anomaly monitoring data diagnosis method and system combining prior physical information

Abstract

The application provides an abnormal monitoring data diagnosis method and system combining prior physical information, belongs to the technical field of railway operation and maintenance monitoring, and proposes a correlation prior hypothesis for a sensor combination with high correlation, calculates the correlation between multiple sensors in a fixed time window in real time based on the hypothesis, and judges the abnormality; for sensors without correlation, an observation regression model based on prior physical information is established; in combination with the Bayesian theory and the prior regression model, the number and position of abnormal points in the observation area are assumed, the most likely value combination is obtained through Markov chain sampling, and the number and position of abnormal points are determined. The application proposes a prior correlation assumption and a prior physical regression model based on the prior physical information of monitoring data, avoids the strong dependence of traditional abnormal detection algorithms on data features and distribution, thereby realizing the rapid and accurate capture of abnormal monitoring data, and providing a strong guarantee for the data reliability of the railway monitoring system.

Description

Technical Field

[0001] This invention relates to the field of railway operation and maintenance monitoring technology, specifically to a method and system for diagnosing abnormal monitoring data that combines prior physical information. Background Technology

[0002] Objective and effective monitoring data is the foundation for condition assessment and safety early warning of railway infrastructure. Although monitoring systems are highly reliable, data anomalies caused by factors such as temperature changes, rain and snow, strong winds, train vibrations, sensor aging, structural expansion and contraction, or human factors cannot be ruled out under long-term monitoring conditions. Research by the U.S. Center for Intelligent Maintenance Systems indicates that over 40% of fault alarms in automated systems are false alarms caused by sensor system malfunctions. Abnormal monitoring data often affects the results of numerical analysis. Therefore, before analysis, preprocessing steps such as anomaly diagnosis and repair of massive amounts of data are necessary to obtain effective information that accurately reflects the service status of seamless railway lines and facilitates analysis and evaluation, ensuring the accuracy and effectiveness of the monitoring data.

[0003] A sensor malfunction is considered to occur when there is an unacceptable deviation between the measured value and the true value. Sensor malfunctions are generally defined from two perspectives: a system-centric perspective and a data-centric perspective. From a system-centric perspective, sensor malfunctions can be categorized into calibration failures, connectivity issues, and insufficient battery power, among others. From a data-centric perspective, sensor malfunctions are analyzed based on the characteristics of the sensor's measurement data, and can be classified into six categories: deviation, gain, drift, accuracy degradation, jamming, and outliers (sensor jumps).

[0004] In existing technologies, a common approach is to statistically describe existing historical data, determine a confidence interval by giving a confidence probability, and then identify and eliminate errors that exceed the confidence interval as gross errors. Commonly used methods for identifying gross errors include the 3σ criterion (Laida criterion), box plot method, Chauvenet criterion, Grubbs criterion, and refinement criteria. However, these methods make certain assumptions about the data distribution (such as normal distribution) and have relatively strict definitions of anomalies. They diagnose anomalies by using fixed static thresholds, ignoring the physical meaning of the actual monitored data, and thus have significant limitations. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for diagnosing abnormal monitoring data by combining prior physical information, which can make full use of the physical characteristics of historical monitoring data, thereby effectively guiding the abnormal data diagnosis process and providing reliable data support for railway monitoring and operation safety, in order to solve at least one of the technical problems existing in the above-mentioned background art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a method for diagnosing anomaly monitoring data that combines prior physical information, comprising:

[0008] Correlation calculations were performed on historical monitoring data to obtain sensor combinations that showed a high degree of correlation.

[0009] For highly correlated sensor combinations, a correlation prior hypothesis is proposed, namely that data anomalies are accidental events and it is impossible for multiple sensors to malfunction simultaneously. Based on the correlation prior hypothesis, the correlation between multiple sensors is calculated in real time with a fixed time window, and anomalies are judged based on the calculation results.

[0010] For sensors that do not have a correlation, analyze the typical characteristics of their historical monitoring data and establish an observation regression model based on prior physical information;

[0011] By combining Bayesian theory and observational regression models, prior assumptions are made about the number and location of outliers within the observation area. The highest probability combination of values ​​is obtained through Markov chain sampling to determine the number and location of outliers.

[0012] As a further limitation of the first aspect of the present invention, historical monitoring data is collected and recorded according to the sensor serial number as {s1,s2,....s}. n};

[0013] Starting from s1, calculate the correlation between each sensor and all other sensor observations, and record it as {(s1,s2),(s1,s3),......(s1,s...} n ),(s2,s1),(s2,s3),......(s n ,s n-1 )}, retain the combination of each sensor with the highest correlation to other sensors, and record it as {(s1,s... i ),(s2,s j ),......(s n ,s k )};

[0014] Incorporating highly correlated combinations into a highly correlated sensor suite is equivalent to {(s)} i ,s j )......} corr Sensors that do not have a high degree of correlation are recorded as {s} k ......} incorr .

[0015] As a further limitation of the first aspect of the invention, for highly correlated sensor combinations {(s) i ,sj )......} corr The relevant prior hypotheses are proposed, including: monitoring data anomalies are accidental events, and it is impossible for multiple sensors to malfunction simultaneously; based on prior calculations, {(s i ,s j )......} corr The sensor array exhibits a high correlation; a significant decrease in correlation indicates an anomaly in one of the sensors. Based on prior assumptions, the system calculates {(s)} in real-time within a fixed time window. i ,s j )......} corr The correlation between two sensor combinations is used to determine if one of the sensors in the combination is abnormal when the correlation does not meet the assumption of high correlation.

[0016] As a further limitation of the first aspect of the invention, for sensors that do not possess high correlation {s k ......} incorr By conducting time and frequency domain analyses on its historical data, its periodic characteristics are clarified, and an observational regression model based on prior physical information is established.

[0017]

[0018] Where ε represents the regression error, ω q , a represents the frequency and phase information of a specific periodic component, respectively. q The regression coefficients under the corresponding periodic components are M, where M is the number of components that have a significant impact on the observed value Y.

[0019] Based on the frequency domain signal, determine its main wavelength component ω q The number and values ​​are determined, and to avoid solving for phase information, the above equation is transformed into the sum of sine and cosine components using trigonometric functions:

[0020]

[0021] Among them, the wavelength component was determined by prior analysis, and the initial values ​​of other error terms and coefficients were set to 1. Subsequently, the initial parameters were dynamically improved based on the measured data to make the regression model better fit the observed Y.

[0022] As a further limitation of the first aspect of the present invention, based on a priori regression model, the monitoring data of total length N is calculated in its subset Y. i:j ={Y i ,Y i+1 ,......Y j The likelihood probability on} (1≤i≤j≤N), i.e.:

[0023] f(Yi:j )=f(Y i ,Y i+1 ,…,Y j-1 ,Y j |X)

[0024] The actual observed value of the monitoring data is exactly the predicted value Y. i:j The joint probability; assume there are K change points, and record the change point positions as C. K ={c1,c2,......c k Given the prior conditions of the number and location of the variable points, the likelihood probability is:

[0025] P k (Y 1:j )=f(Y1,Y2,…,Y j |K=k,C K ,X)

[0026] When the observed monitoring data shows K state changes, and the corresponding change time is C K Under the premise that the actual observed value is exactly the predicted value Y i:j The joint probability.

[0027] As a further limitation of the first aspect of the present invention, by combining Bayesian inference theory, the number of variable points K and the location of the variable points C are obtained. K Posterior probability, and its most likely combination is determined through repeated sampling:

[0028]

[0029] In the formula, f(Y) 1:N |K=k,C K That is, P k (Y 1:j ), f(K=k), f(C) K |K=k) represent prior estimates of the number of change points and the probability density of change point locations, respectively. By sampling through a Markov chain, the value of K with the highest probability is obtained. When K increases compared to the previous observation time, the latest observation location is considered to be a newly detected change point, that is, a sensor anomaly has occurred at this time.

[0030] Secondly, the present invention provides an anomaly monitoring data diagnosis system that combines prior physical information, comprising:

[0031] The calculation module is used to perform correlation calculations on historical monitoring data to obtain sensor combinations that are highly correlated.

[0032] The judgment module is used to propose a correlation prior hypothesis for highly correlated sensor combinations, namely that data anomalies are accidental events and it is impossible for multiple sensors to be abnormal at the same time; based on the correlation prior hypothesis, the correlation between multiple sensors is calculated in real time within a fixed time window, and anomalies are judged based on the calculation results.

[0033] The module is used to analyze the typical characteristics of historical monitoring data of sensors that have no correlation and to establish an observation regression model based on prior physical information.

[0034] The determination module combines Bayesian theory with observational regression models to make prior assumptions about the number and location of outliers within the observation area. It then uses Markov chain sampling to obtain the most probable combination of values ​​to determine the number and location of outliers.

[0035] Thirdly, the present invention provides a non-transitory computer-readable storage medium for storing computer instructions, which, when executed by a processor, implement the anomaly monitoring data diagnosis method combining prior physical information as described in the first aspect.

[0036] Fourthly, the present invention provides a computer device including a memory and a processor, the processor and the memory communicating with each other, the memory storing program instructions executable by the processor, and the processor calling the program instructions to execute the anomaly monitoring data diagnosis method combining prior physical information as described in the first aspect.

[0037] Fifthly, the present invention provides an electronic device, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the anomaly monitoring data diagnosis method combining prior physical information as described in the first aspect.

[0038] The beneficial effects of this invention are as follows: It uses the correlation coefficient method to monitor anomalies in sensor monitoring data, which is different from traditional anomaly detection methods. It can calculate the diagnostic results of change points of each sensor in real time. It combines prior physical information to obtain a regression observation model, dynamically adjusts the model parameters with actual observation values, and combines Bayesian inference theory to obtain change point information within the observation range through repeated sampling. It has high detection accuracy, low computational load, and can quantify the uncertainty of the number and location of change points.

[0039] The advantages of additional aspects of the invention will be set forth more clearly in the following description or will be learned by practice of the invention. Attached Figure Description

[0040] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is a flowchart of the abnormal detection data diagnosis method according to an embodiment of the present invention.

[0042] Figure 2 This is a graph showing the linear fitting effect of the highly correlated sensor described in an embodiment of the present invention.

[0043] Figure 3 This is a diagram showing the linear fitting effect of a non-highly correlated sensor according to an embodiment of the present invention.

[0044] Figure 4 This is a graph showing the correlation coefficient calculation results according to an embodiment of the present invention. Figure 4 (a) This shows the fluctuations of the three correlation coefficients—Spearman, Kendall, and the maximum mutual information coefficient (MIC)—over time; Figure 4 (b) shows the fluctuation of frequency with correlation coefficient.

[0045] Figure 5 This is a diagram illustrating the anomaly diagnosis effect based on sensor correlation as described in an embodiment of the present invention.

[0046] Figure 6 This is a diagram illustrating the effect of local anomaly diagnosis based on sensor correlation as described in an embodiment of the present invention.

[0047] Figure 7 This is a graph showing the frequency domain component analysis results of the sensor monitoring data described in an embodiment of the present invention.

[0048] Figure 8 This is a diagram illustrating the effect of monitoring data anomaly identification based on Bayesian theory as described in an embodiment of the present invention.

[0049] in, Figure 8 (a) is a graph showing the monitoring results of the temperature; Figure 8 (b) is the posterior probability monitoring plot. Detailed Implementation

[0050] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0051] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0052] It should also be understood that terms such as those defined in general dictionaries should be understood to have meanings consistent with their meanings in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless defined as here.

[0053] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, and / or groups thereof.

[0054] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0055] To facilitate understanding of the present invention, the present invention will be further explained and described below with reference to the accompanying drawings and specific embodiments. However, the specific embodiments do not constitute a limitation on the embodiments of the present invention.

[0056] Those skilled in the art should understand that the accompanying drawings are merely schematic diagrams of embodiments, and the components in the drawings are not necessarily essential for implementing the present invention.

[0057] Example 1

[0058] In this embodiment 1, an anomaly monitoring data diagnosis system combining prior physical information is first provided, including: a calculation module, used to calculate the correlation of historical monitoring data to obtain sensor combinations with high correlation; a judgment module, used to propose a prior correlation hypothesis for highly correlated sensor combinations, namely, that data anomalies are accidental events and it is impossible for multiple sensors to be abnormal at the same time; based on the prior correlation hypothesis, the correlation between multiple sensors is calculated in real time within a fixed time window, and anomalies are judged based on the calculation results; a construction module, used to analyze the typical characteristics of historical monitoring data of sensors without correlation and establish an observation regression model based on prior physical information; and a determination module, used to combine Bayesian theory and the observation regression model to make prior assumptions about the number and location of anomalies in the observation area, and obtain the most probable combination of values ​​through Markov chain sampling to determine the number and location of anomalies.

[0059] In this embodiment, the above-described system is used to implement an anomaly monitoring data diagnosis method that combines prior physical information, including the following process steps:

[0060] S1. Perform correlation calculations on historical monitoring data to obtain sensor combinations that are highly correlated;

[0061] S2. For highly correlated sensor combinations, a correlation prior hypothesis is proposed, namely that data anomalies are accidental events and it is impossible for multiple sensors to malfunction simultaneously.

[0062] S3. Based on the assumption of S2, calculate the correlation between multiple sensors in real time with a fixed time window, and judge the anomalies based on the calculation results;

[0063] S4. For sensors that do not have a correlation, analyze the typical characteristics of their historical monitoring data and establish an observation regression model based on prior physical information;

[0064] S5. Combining Bayesian theory and observational regression models, prior assumptions are made about the number and location of outliers within the observation area. The combination of the most probable values ​​is obtained through Markov chain sampling to determine the number and location of outliers.

[0065] Step S1 collects historical monitoring data from this monitoring system or a similar monitoring system, and records it according to the sensor serial number as {s1, s2, ... s n}:

[0066] Starting from s1, calculate the correlation between each sensor and all other sensor observations, and record it as {(s1,s2),(s1,s3),......(s1,s...} n ),(s2,s1),(s2,s3),......(s n ,s n-1)}, retain the combination of each sensor with the highest correlation to other sensors, and record it as {(s1,s... i ),(s2,s j ),......(s n ,s k In this case, i, j, ... can be repeated.

[0067] Subsequently, combinations that satisfy high correlation (correlation coefficient greater than 0.6) are included in the highly correlated sensor combination, namely {(s i ,s j )......} corr Sensors that do not have a high degree of correlation are recorded as {s} k ......} incorr .

[0068] In S2, for highly correlated sensor combinations {(s) i ,s j )......} corr Make prior assumptions about the relevance, including:

[0069] (1) Abnormal monitoring data is an accidental event; it is impossible for multiple sensors to malfunction simultaneously.

[0070] (2) Based on prior calculations, {(s i ,s j )......} corr The sensor array exhibits a high degree of correlation; when the correlation decreases significantly, it indicates that a particular sensor has malfunctioned.

[0071] In S3, based on the prior assumptions of S2, {(s)} is calculated online in real time within a fixed time window (e.g., 24 hours). i ,s j )......} corr The correlation between two sensors in a given sensor combination is considered to be abnormal if it does not meet the assumption of high correlation (i.e., the correlation coefficient is less than 0.6). This process is based on sensor correlation information for diagnosis and has sufficient physical significance.

[0072] In step S4, for sensors that do not have a high correlation {s k ......} incorr By conducting time and frequency domain analyses on its historical data, its periodic characteristics are clarified, and an observational regression model based on prior physical information is established.

[0073]

[0074] Where ε represents the regression error, ω q, a represents the frequency and phase information of a specific periodic component, respectively. q The regression coefficients under the corresponding periodic components are M, where M is the number of components that have a significant impact on the observed value Y.

[0075] Based on the frequency domain signal, determine its main wavelength component ω q The number and values ​​are determined, and to avoid solving for phase information, the above equation is transformed into the sum of sine and cosine components using trigonometric functions:

[0076]

[0077] In this model, the wavelength components were determined through prior analysis, while other error terms and coefficients were initially set to 1. Subsequently, the initial parameters were dynamically improved based on the measured data to better fit the observed sequence Y. The observational regression model derived from the time and frequency domain analysis of the measured data is a comprehensive standard paradigm that considers the sensor monitoring data period, error, and regression coefficients. The regression model transformed using the phase shift property of trigonometric functions is more conducive to fitting the measured data.

[0078] In S5, based on the prior regression model established in S4, the monitoring data of total length N is calculated in its subset Y. i:j ={Y i ,Y i+1 ,......Y j The likelihood probability on} (1≤i≤j≤N), i.e.:

[0079] f ( Y i:j ) =f ( Y i ,Y i+1 ,…,Y j-1 ,Y j |X )

[0080] This means that the actual observed value of the monitoring data is exactly the predicted value Y. i:j The joint probability. Assume there are K changing points, and record the positions of these changing points as C. K ={c1,c2,......c k Given the prior conditions of the number and location of the variable points, the likelihood probability is:

[0081] P k (Y 1:j )=f ( Y1,Y2,…,Y j |K=k,C K ,X)

[0082] This means that when the observed and monitored data undergoes K state changes, and the corresponding change time is C... K Under the premise that the actual observed value is exactly the predicted value Y i:j The joint probability can be obtained by iterative calculation according to the chain reaction rule.

[0083] Based on this, and combined with Bayesian inference theory, the number of change points K and the location of change points C are further obtained. K The posterior probability is used to determine the most probable combination through repeated sampling. The specific formula is:

[0084]

[0085] In the formula, f(Y) 1:N |K=k,C K That is, P k (Y 1:j ), f(K=k), f(C) K |K=k) represents the prior estimates of the number of change points and the probability density of change point locations, respectively. The value of K with the highest probability is obtained through Markov chain sampling. When K increases compared to the previous observation time, the latest observation location is considered a newly detected change point, indicating a sensor anomaly.

[0086] In summary, combining S3 and S5 respectively, for {(s i ,s j )......} corr With {s k ......} incorr Real-time anomaly detection using two types of sensors can meet the anomaly data diagnosis needs of all sensing devices in the monitoring system. The data algorithm described in this embodiment is simple and has high accuracy, providing a reliable diagnostic method for processing anomaly data in railway monitoring.

[0087] Example 2

[0088] like Figure 1 As shown in the figure, this embodiment 2 provides a method for diagnosing abnormal monitoring data by combining prior physical information, specifically including the following steps:

[0089] S1. Calculate the correlation of historical monitoring data to obtain sensor combinations with high correlation. In a specific example, historical monitoring data of temperature sensors and track temperature sensors at corresponding locations of a certain railway line station were collected. The correlation of each sensor with the observation data of all other sensors was calculated, and the combination with the highest correlation of each sensor with other sensors was retained. Subsequently, combinations that meet the high correlation (correlation coefficient greater than 0.6) are included in the highly correlated sensor combinations for processing in steps S2-S3, while sensors that do not have high correlation are included in the non-highly correlated sensor combinations for processing in steps S4-S5.

[0090] S2. For highly correlated sensor combinations, a prior correlation hypothesis is made, namely, that data anomalies are accidental events and it is impossible for multiple sensors to malfunction simultaneously. In a specific instance, there are clearly cases where different sensor combinations exhibit different correlations, such as... Figure 2 As shown, in a highly correlated sensor, the sensor monitoring data points are densely distributed around the fitted curve, with an R² score of 0.993, and the x and y axis data show extremely strong correlation; for example... Figure 3 As shown, in a certain non-highly correlated sensor, the sensor monitoring data points are discretely distributed in four places on the fitted curve. Although there is a certain positive correlation, the correlation is very low.

[0091] S3. Based on the S2 assumption, calculate the correlation between multiple sensors in real time within a fixed time window (e.g., 24 hours), and determine anomalies based on the calculation results. In a specific example, based on the prior assumption of S2, calculate the correlation between two highly correlated sensor combinations in real time within a 24-hour time window, such as... Figure 4 The figure shows the fluctuations of the Spearman, Kendall, and maximum mutual information (MIC) correlation coefficients over time. The Spearman and Kendall correlation coefficients remained above 0.8 for most of the time, indicating a strong positive correlation between air temperature and rail temperature.

[0092] During real-time computation, if the high correlation assumption is not met (i.e., the correlation coefficient is less than 0.6), it is considered that one of the sensors in the sensor array is malfunctioning. The data anomaly diagnosis results derived from the three correlation coefficients are as follows: Figure 5 As shown, for the abnormal distribution of these three correlation coefficients, the Spearman and Kendall correlation coefficients are more concentrated, while the maximum mutual information coefficient is more dispersed.

[0093] Local anomaly diagnosis calculated using measured temperature, such as Figure 6As shown, this includes both Spearman and Kendall correlation coefficients. It can be seen that using the Kendall correlation coefficient is more effective in diagnosing anomalies in monitoring data, and it can sensitively diagnose abnormal data. This method can effectively utilize the correlation between highly correlated sensors to identify and process abnormal data, providing strong data support for railway operation safety.

[0094] S4. For sensors with no correlation, analyze the typical characteristics of their historical monitoring data and establish an observational regression model based on prior physical information. In a specific example, the establishment and adjustment of the observational regression model are shown in the following steps:

[0095] (1) Perform time and frequency domain physical information analysis on historical monitoring data of sensors that do not have a high degree of correlation, clarify their periodic characteristics, and establish an observation regression model;

[0096] (2) By Figure 7 The frequency domain distribution shows that the frequency peak of the temperature change signal is relatively concentrated at 1 day (24 hours). Therefore, its main wavelength component is determined to be 1, and then the observation regression model is transformed by using trigonometric function relationships to represent the sum of sine and cosine components.

[0097] (3) Subsequently, the initial parameters were dynamically improved based on the measured data to make the regression model fit the sequencing better.

[0098] S5. Combining Bayesian theory and the observational regression model, prior assumptions are made regarding the number and location of outliers within the observation area. Markov chain sampling is used to obtain the most probable combination of values ​​to determine the number and location of outliers. Based on the prior regression model established in S4, the steps for determining the change point information in the monitoring data are as follows:

[0099] (1) Calculate the likelihood probability of monitoring data with a total length of N on its subset window, that is, the joint probability that the actual observed value of the monitoring data is exactly the predicted value;

[0100] (2) The joint probability that the actual observed value is exactly the predicted value is obtained by iterative calculation based on the chain reaction rule;

[0101] (3) Finally, by combining Bayesian inference theory, we further obtain the number of change points and the posterior probability of the change point location, and determine the most likely combination by repeated sampling. By combining Markov chain repeated sampling, we obtain the value with the highest probability of the number of change points.

[0102] In a specific example, taking the rail temperature monitored by a certain sensor as an example, a dynamic analysis is performed on the changes in the measured data of the sensor within the observation period, and the posterior probability of a sudden state change at each observation time is calculated. The results are as follows. Figure 8As shown in the image, the black line represents the monitoring data, the red line represents the regression results of the prior model, and the blue scatter plots represent the posterior probability that each observation is exactly a change point. The magnified image shows a significant difference in the development trend of rail axial force before and after the peak of the posterior probability at the change point. Using the change point monitoring method, when the number of change points obtained from the observation regression model and repeated sampling increases compared to the previous observation time, abnormal data is diagnosed, the sampling window continues to move forward, and the diagnosis of all monitoring data is completed.

[0103] Example 3

[0104] This embodiment 3 provides a non-transitory computer-readable storage medium for storing computer instructions. When these computer instructions are executed by a processor, they implement the anomaly monitoring data diagnosis method combining prior physical information as described above. The method includes:

[0105] Correlation calculations were performed on historical monitoring data to obtain sensor combinations that showed a high degree of correlation.

[0106] For highly correlated sensor combinations, a correlation prior hypothesis is proposed, namely that data anomalies are accidental events and it is impossible for multiple sensors to malfunction simultaneously. Based on the correlation prior hypothesis, the correlation between multiple sensors is calculated in real time with a fixed time window, and anomalies are judged based on the calculation results.

[0107] For sensors that do not have a correlation, analyze the typical characteristics of their historical monitoring data and establish an observation regression model based on prior physical information;

[0108] By combining Bayesian theory and observational regression models, prior assumptions are made about the number and location of outliers within the observation area. The highest probability combination of values ​​is obtained through Markov chain sampling to determine the number and location of outliers.

[0109] Example 4

[0110] This embodiment 4 provides a computer device, including a memory and a processor, wherein the processor and the memory communicate with each other, and the memory stores program instructions executable by the processor. The processor calls the program instructions to execute the anomaly monitoring data diagnosis method combining prior physical information as described above, the method including:

[0111] Correlation calculations were performed on historical monitoring data to obtain sensor combinations that showed a high degree of correlation.

[0112] For highly correlated sensor combinations, a correlation prior hypothesis is proposed, namely that data anomalies are accidental events and it is impossible for multiple sensors to malfunction simultaneously. Based on the correlation prior hypothesis, the correlation between multiple sensors is calculated in real time with a fixed time window, and anomalies are judged based on the calculation results.

[0113] For sensors that do not have a correlation, analyze the typical characteristics of their historical monitoring data and establish an observation regression model based on prior physical information;

[0114] By combining Bayesian theory and observational regression models, prior assumptions are made about the number and location of outliers within the observation area. The highest probability combination of values ​​is obtained through Markov chain sampling to determine the number and location of outliers.

[0115] Example 5

[0116] This embodiment 5 provides an electronic device, including: a processor, a memory, and a computer program; wherein, the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to cause the electronic device to execute instructions for implementing the anomaly monitoring data diagnosis method combining prior physical information as described above. The method includes:

[0117] Correlation calculations were performed on historical monitoring data to obtain sensor combinations that showed a high degree of correlation.

[0118] For highly correlated sensor combinations, a correlation prior hypothesis is proposed, namely that data anomalies are accidental events and it is impossible for multiple sensors to malfunction simultaneously. Based on the correlation prior hypothesis, the correlation between multiple sensors is calculated in real time with a fixed time window, and anomalies are judged based on the calculation results.

[0119] For sensors that do not have a correlation, analyze the typical characteristics of their historical monitoring data and establish an observation regression model based on prior physical information;

[0120] By combining Bayesian theory and observational regression models, prior assumptions are made about the number and location of outliers within the observation area. The highest probability combination of values ​​is obtained through Markov chain sampling to determine the number and location of outliers.

[0121] In summary, the anomaly monitoring data diagnosis method combining prior physical information described in this invention first calculates the correlation of historical monitoring data to obtain sensor combinations with high correlation. Then, for these highly correlated sensor combinations, a prior correlation hypothesis is proposed, namely that data anomalies are accidental events and it is impossible for multiple sensors to malfunction simultaneously. Based on this hypothesis, the correlation between multiple sensors is calculated in real time within a fixed time window (e.g., 24 hours), and anomalies are determined based on the calculation results. For sensors without correlation, typical characteristics of their historical monitoring data are analyzed, and an observation regression model based on prior physical information is established. Finally, combining Bayesian theory and the prior regression model, prior hypotheses are made regarding the number and location of anomaly points within the observation area. Markov chain sampling is used to obtain the combination of the most probable values ​​to determine the number and location of anomaly points. This invention proposes prior correlation hypotheses and prior physical regression models based on the prior physical information of monitoring data, avoiding the strong dependence of traditional anomaly detection algorithms on data characteristics and distribution, thereby achieving rapid and accurate capture of anomaly monitoring data and providing strong assurance for the data reliability of railway monitoring systems.

[0122] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0123] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0124] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0125] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment, whereby a series of operational steps are performed to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0126] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that, based on the technical solutions disclosed in the present invention, various modifications or variations that can be made by those skilled in the art without creative effort should be included within the scope of protection of the present invention.

Claims

1. An abnormal monitoring data diagnostic method incorporating prior physical information, characterized by, The method comprises the following steps: correlation calculation is performed on historical monitoring data to obtain a sensor combination with high correlation; a correlation prior hypothesis is proposed for the sensor combination with high correlation, that is, data anomaly is a casual event and multiple sensors cannot be abnormal at the same time; based on the correlation prior hypothesis, correlation between multiple sensors is calculated in real time in a fixed time window, and an anomaly is determined according to the calculation result; for sensors without correlation, typical characteristics of historical monitoring data are analyzed, and an observation regression model based on prior physical information is established; Collect historical monitoring data and record it according to the sensor serial number. ;since Begin by calculating the correlation between each sensor's observations and those of all other sensors, and record the results. The combination of each sensor with the highest correlation to other sensors is retained and recorded as follows: Incorporating highly correlated combinations into highly correlated sensor combinations is equivalent to... Sensors that do not have a high degree of correlation record as For sensors that do not have a high correlation By conducting time- and frequency-domain analysis on its historical data, its periodic characteristics were clarified, and an observational regression model based on prior physical information was established. ； wherein, represents the regression error, , respectively represent the frequency and phase information of a specific periodic component, is the regression coefficient under the corresponding periodic component, and M is the number of components that have a significant impact on the observation value Y. determining its main wavelength components from the frequency domain signal The number and values are determined and, to avoid solving the phase information, the above equation is transformed into a sum of cosine components using trigonometric functions: ； a prior hypothesis is made on the number and position of abnormal points in an observation area by combining Bayesian theory and the observation regression model, and the combination with the highest possibility is obtained through Markov chain sampling to determine the number and position of abnormal points; wherein Based on the prior regression model, the likelihood probability of the monitoring data of total length N on its subset of length k is calculated, i.e.: Based on the prior regression model, the likelihood probability of the monitoring data of total length N on its subset of length k is calculated, i.e.: ； The actual observations of the monitoring data are taken as the predicted values The joint probability of the observations; suppose there are K change points in the data, and record the change point positions as Then, the likelihood probability under the prior condition of the number and position of change points is: ； When the observation monitoring data has K state changes, and the corresponding change time is , the joint probability that the actual observation value is exactly the prediction value is obtained. Combining the Bayesian inference theory, the number of change points K and the change point positions are obtained posterior probabilities, and determine their most likely combination by repeated sampling: ； wherein i.e. , , respectively represent the prior estimation of the number of change points and the probability density of the change point position, and the K most likely values are obtained by Markov chain sampling. When K increases compared with the previous observation time, it is considered that the latest observation position is a newly detected change point, i.e. a sensor anomaly occurs at this time.

2. The method of claim 1, wherein the method further comprises: For the existence of highly correlated sensor combination , the correlation prior hypothesis is proposed, including: monitoring data anomalies are accidental events, and it is impossible for multiple sensors to be abnormal at the same time; through prior calculation, , there is a high correlation between the sensor combination, and when the correlation obviously decreases, it means that an abnormal sensor appears; based on the prior hypothesis, the correlation between , when it does not meet the highly correlated assumption, it is considered that an abnormal sensor appears in the sensor combination.

3. An abnormal monitoring data diagnostic system incorporating prior physical information based on the method of claim 1 or 2, characterized by, wherein the wavelength component is determined by prior analysis, and other error terms and coefficient initial values are 1, and then the initial parameters are dynamically improved based on measured data, so that the regression model better fits the observation sequence Y; The method comprises the following steps: a calculation module is configured to perform correlation calculation on historical monitoring data to obtain a sensor combination with high correlation; a judgment module is configured to propose a correlation prior hypothesis for the sensor combination with high correlation, that is, data anomaly is a casual event and multiple sensors cannot be abnormal at the same time; based on the correlation prior hypothesis, correlation between multiple sensors is calculated in real time in a fixed time window, and an anomaly is determined according to the calculation result; a construction module is configured to analyze typical characteristics of historical monitoring data for sensors without correlation, and establish an observation regression model based on prior physical information; 4. A non-transitory computer-readable storage medium, comprising, a determination module is configured to make a prior hypothesis on the number and position of abnormal points in an observation area by combining Bayesian theory and the observation regression model, and obtain the combination with the highest possibility through Markov chain sampling to determine the number and position of abnormal points.

5. A computer device, comprising: The non-transitory computer readable storage medium is used to store computer instructions, which are executed by a processor to implement the abnormal monitoring data diagnosis method combining prior physical information according to claim 1 or 2.

6. An electronic device, comprising: The electronic device comprises a memory and a processor, the processor and the memory are in communication with each other, the memory stores program instructions executed by the processor, and the processor invokes the program instructions to execute the abnormal monitoring data diagnosis method combining prior physical information according to claim 1 or 2. The electronic device comprises: a processor, a memory and a computer program; wherein the processor is connected with the memory, and the computer program is stored in the memory; when the electronic device is running, the processor executes the computer program stored in the memory to make the electronic device execute instructions for implementing the abnormal monitoring data diagnosis method combining prior physical information according to claim 1 or 2.

Images