Turnout anomaly detection method, device, equipment and medium

Abstract

The invention provides a turnout anomaly detection method, device and equipment and a medium, and relates to the technical field of turnout system fault detection, and the method comprises the steps: obtaining multi-dimensional time series data of turnout equipment; the multi-dimensional time sequence data is used for representing the operation conditions of the turnout equipment at different moments collected by the plurality of sensors; performing time domain and frequency domain analysis on the time sequence data, and constructing a feature vector; performing weighted summation on each feature of each sample in the feature vector by adopting a multi-dimensional weighted quartile algorithm to obtain a total abnormal score of each sample; the total abnormal score is used for representing the overall operation state of the turnout equipment; comparing the total anomaly score with a current dynamic threshold value to obtain an anomaly detection result; the current dynamic threshold is determined from historical statistical data and a sensitivity control factor. The scheme can adapt to turnout operation environment and working condition changes in real time, the false alarm rate and the missing report rate under complex working conditions are remarkably reduced, and the accuracy of anomaly detection is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault detection of turnout systems, in particular to a turnout anomaly detection method, device, equipment and medium. BACKGROUND

[0002] With the rapid development of rail transit technology, as a key switching device in the rail transit line, the turnout system undertakes the important function of guiding the train to turn from one track to another, and the stability of its working state is the core prerequisite for ensuring the safe and punctual operation of the train. The turnout system has a complex structure, covering multiple mechanical components such as switch machines, points, stock rails, frog switches, and sliding bed plates, and relies on electrical control circuits to achieve action driving and state feedback. In the long-term service process, affected by multiple factors such as train impact vibration, environmental temperature and humidity changes, component wear and aging, and electrical contact wear, various abnormal faults are prone to occur, such as poor adhesion of the point and the stock rail, switch machine jamming, electrical circuit wire breakage, and sliding bed plate rusting and jamming. In order to ensure the safe operation of the rail transit system, it is particularly important to study the anomaly detection of the turnout system.

[0003] At present, in the related technology, sensors are used to collect data, and the data is subjected to simple threshold judgment to realize anomaly detection. However, for the turnout system under complex working conditions, this scheme is relatively one-sided and single, and is prone to false positives or false negatives, resulting in low accuracy of anomaly detection. SUMMARY

[0004] The present application provides a turnout anomaly detection method, device, equipment and medium.

[0005] The first aspect of the present application provides a turnout anomaly detection method, which comprises: Obtaining multi-dimensional time series data of a turnout device; the multi-dimensional time series data is used to represent the running conditions of the turnout device at different times collected by multiple sensors; Performing time domain and frequency domain analysis on the time series data to construct a feature vector; Using a multi-dimensional weighted quartile distance algorithm, the features of each sample in the feature vector are weighted and summed to obtain a total anomaly score of each sample; the total anomaly score is used to represent the overall running state of the turnout device; Comparing the total anomaly score with a current dynamic threshold to obtain an anomaly detection result; the current dynamic threshold is determined according to historical statistical data and a sensitivity control factor.

[0006] In an optional embodiment of the present application, the time domain and frequency domain analysis of the time series data to construct a feature vector comprises: performing time domain feature extraction on the time series data to obtain time domain features; the time domain features comprise at least one of the following: mean value, standard deviation, maximum value, minimum value, skewness and peak factor; performing Fourier transform processing on the time series data to extract frequency domain features; the frequency domain features comprise at least one of the following: main frequency information, frequency spectrum energy and frequency band energy; performing combined processing on the time domain features and the frequency domain features to construct the feature vector.

[0007] In an optional embodiment of the present application, a multi-dimensional weighted quartile deviation algorithm is used to perform weighted summation on each feature of each sample in the feature vector to obtain a total anomaly score of each sample, comprising: calculating an anomaly score of each feature for each sample in the feature vector; determining a weighting factor of each feature; performing weighted summation processing on the anomaly score and the weighting factor of each feature to obtain a total anomaly score of each sample.

[0008] In an optional embodiment of the present application, calculating an anomaly score of each feature comprises: determining a first quartile and a third quartile of the feature; the first quartile is a value corresponding to a 25% position in the feature data points, and the third quartile is a value corresponding to a 75% position in the feature data points; calculating a quartile deviation according to the first quartile and the third quartile; determining an anomaly interval based on the first quartile, the third quartile and the quartile deviation; obtaining an anomaly score of a corresponding data point under the feature based on a degree of deviation of each data point under the feature from the anomaly interval.

[0009] In an optional embodiment of the present application, determining a weighting factor of each feature comprises: calculating a standard deviation value of each feature for each sample in the feature vector; performing standardization processing on the standard deviation value to obtain a standardized standard deviation value, and taking the standardized standard deviation value as the weighting factor of the feature; the standardization processing is a ratio processing of a standard deviation value of a single feature to a sum of standard deviation values of all features.

[0010] In an optional embodiment of the present application, comparing the total anomaly score with a current dynamic threshold to obtain an anomaly detection result, comprising: determining that the turnout device is in an abnormal state when the total abnormal score is greater than the current dynamic threshold value; determining that the turnout device is in a normal state when the total abnormal score is not greater than the current dynamic threshold value.

[0011] In an optional embodiment of the present application, the method further comprises: obtaining historical statistical data, and determining a historical score mean and a historical score standard deviation according to an abnormal score of the historical statistical data; determining the current dynamic threshold value based on the historical score mean, the historical score standard deviation, and a sensitivity control factor.

[0012] A second aspect of the embodiments of the present application provides a turnout abnormality detection device, comprising: an obtaining module configured to obtain multi-dimensional time series data of a turnout device; the multi-dimensional time series data is used to represent running conditions of the turnout device at different time instants collected by multiple sensors; a constructing module configured to perform time domain and frequency domain analysis on the time series data, and construct a feature vector; a processing module configured to perform weighted summation on each feature of each sample in the feature vector by using a multi-dimensional weighted quartile range algorithm, and construct a total abnormal score of the each sample; the total abnormal score is used to represent an overall running state of the turnout device; an abnormality detection module configured to compare the total abnormal score with a current dynamic threshold value, and obtain an abnormality detection result; the current dynamic threshold value is determined according to historical statistical data and a sensitivity control factor.

[0013] A third aspect of the embodiments of the present application provides a computer device, comprising a memory and a processor; the memory stores a computer program; and the processor implements steps of any one of the above methods when executing the computer program.

[0014] A fourth aspect of the embodiments of the present application provides a computer readable storage medium, which stores a computer program; and the computer program implements steps of any one of the above methods when executed by a processor.

[0015] The method is applied to a turnout anomaly detection system, and the method comprises: acquiring multi-dimensional time sequence data of a turnout device; the multi-dimensional time sequence data is used to represent the running conditions of the turnout device at different time points collected by multiple sensors; time domain and frequency domain analysis is performed on the time sequence data to construct a feature vector; a multi-dimensional weighted quartile range algorithm is used to perform weighted summation on each feature of each sample in the feature vector to obtain a total anomaly score of each sample; the total anomaly score is used to represent the overall running state of the turnout device; the total anomaly score is compared with a current dynamic threshold to obtain an anomaly detection result; the current dynamic threshold is determined according to historical statistical data and a sensitivity control factor. Compared with the prior art, the technical solution in the application can comprehensively capture the running state information of the turnout device at different time points by acquiring multi-dimensional running data of the turnout device collected by multiple sensors synchronously, avoids the problem that the running conditions are not completely described due to single data dimension, and lays a data foundation for subsequent accurate anomaly detection; time domain and frequency domain analysis is performed on the time sequence data to construct a feature vector, which breaks through the limitation of simple threshold judgment of original data in related technologies, deeply excavates deep features of the data, can capture potential anomaly signals that are difficult to identify by traditional methods, and improves the recognition ability of complex fault modes; the multi-dimensional weighted quartile range algorithm is used to distribute a weighting factor according to a feature standard deviation standardization result and perform weighted summation on each feature anomaly score to obtain a total anomaly score, solves the problem that related technologies do not consider feature contribution differences, makes features more sensitive to anomalies play a greater role, accurately quantifies the sample anomaly degree, and improves the accuracy of representing the overall running state of the turnout; and then, the current dynamic threshold is determined by combining historical anomaly score statistical data and a sensitivity control factor, the total anomaly score is compared with the current dynamic threshold to obtain a detection result, the dynamic threshold is set to adapt to defects of working condition changes, can adapt to changes of the turnout running environment and working conditions in real time, significantly reduces the false positive rate and the false negative rate under complex working conditions, and improves the accuracy of anomaly detection. BRIEF DESCRIPTION OF DRAWINGS

[0016] The drawings described herein are used to provide further understanding of the application, constitute a part of the application, the illustrative embodiments of the application and the description thereof are used to explain the application, and do not constitute improper limitations on the application. In the drawings: Figure 1 A structural schematic diagram of a computer device provided for an embodiment of the application; Figure 2 A flowchart of a turnout anomaly detection method provided for an embodiment of the application; Figure 3 A time sequence data schematic diagram of an A-phase current provided for an embodiment of the application; Figure 4 A time sequence data schematic diagram of a B-phase current provided for an embodiment of the application; Figure 5 A time series data schematic diagram of the C-phase current provided for an embodiment of the present application is shown in FIG. 1; Figure 6 A time series data schematic diagram of the power provided for an embodiment of the present application is shown in FIG. 2; Figure 7 A flowchart of a turnout anomaly detection method provided for another embodiment of the present application is shown in FIG. 3; Figure 8 A flowchart of a method for obtaining a total anomaly score of each sample provided for an embodiment of the present application is shown in FIG. 4; Figure 9 A structural schematic diagram of a turnout anomaly detection device provided for an embodiment of the present application is shown in FIG. 5. DETAILED DESCRIPTION

[0017] In the process of implementing the present application, the inventors found that the traditional turnout anomaly detection scheme is relatively one-sided, and is prone to false positives or false negatives, resulting in low anomaly detection accuracy.

[0018] In order to make the technical solutions and advantages in the embodiments of the present application clearer, the exemplary embodiments of the present application will be further described in detail below with reference to the drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0019] As mentioned in the background, one way in the related art is to use manual inspection to check the turnout equipment, but manual inspection has problems such as long periodicity, easy to overlook, slow reaction, etc. Especially in a high-load and high-frequency transportation environment, it is impossible to ensure the immediate detection and accurate judgment of the turnout equipment. Another way is to use a sensor-based monitoring scheme to perform simple threshold judgment on the data to realize anomaly detection. However, for a turnout system under complex working conditions, this scheme is relatively one-sided, and is prone to false positives or false negatives, resulting in low anomaly detection accuracy.

[0020] Based on the above defects, the application provides a turnout anomaly detection method. Compared with the related art, the technical scheme in the application can comprehensively capture the running state information of the turnout equipment at different times by acquiring multi-dimensional running data of the turnout equipment synchronously collected by multiple sensors, avoid the problem that the running condition is not completely described due to single data dimension, and lay a data foundation for subsequent accurate anomaly detection; time domain and frequency domain analysis is performed on the time series data to construct a feature vector, thereby breaking through the limitation of the related art that only relies on simple threshold judgment of the original data, deeply mining deep features of the data, capturing potential abnormal signals that are difficult to identify by traditional methods, and improving the recognition ability of complex fault modes; a multi-dimensional weighted quartile range algorithm is used to allocate a weighting factor according to a feature standard deviation standardization result and to weight and sum each feature anomaly score to obtain a total anomaly score, thereby solving the problem that the related art does not consider feature contribution differences, making features that are more sensitive to anomalies play a greater role, accurately quantifying the degree of sample anomalies, and improving the accuracy of representing the overall running state of the turnout; and then, a current dynamic threshold is determined in combination with historical anomaly score statistical data and a sensitivity control factor, the total anomaly score is compared with the dynamic threshold to obtain a detection result, and by setting the dynamic threshold to adapt to the defects of working condition changes, the turnout running environment and working condition changes can be adapted in real time, the false positive rate and the false negative rate under complex working conditions are significantly reduced, and the accuracy of anomaly detection is improved.

[0021] In the embodiments of the application, the scheme can be applied to a computer device, which is implemented by using various computer languages, for example, an object-oriented program design language Java and an interpreted script language JavaScript.

[0022] Please refer to Figure 1 The embodiments of the application provide a structural schematic diagram of an example computer device. As shown in the figure, Figure 1 The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium may, for example, be a disk. The non-volatile storage medium stores files (which can be files to be processed or processed files), an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement a turnout anomaly detection method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device, or an external keyboard, touchpad or mouse, etc.

[0023] Please refer to Figure 2 The following embodiments take the above computer device as an execution subject, and take the turnout anomaly detection method provided by the embodiments of the present application as an example to perform data processing. The turnout anomaly detection method provided by the embodiments of the present application includes the following steps 201-204: Step 201, acquiring multi-dimensional time series data of the turnout device; the multi-dimensional time series data is used to represent the running conditions of the turnout device at different time points collected by multiple sensors.

[0024] It should be noted that the above multi-dimensional time series data refers to a multi-dimensional continuous data set of the running conditions of the turnout device at different time points collected by the sensors. The sensors can include current sensors, voltage sensors or power sensors. Different sensors collect different data, and each sensor is used to provide one-dimensional time series data, such as voltage, current, power, etc. By using multiple sensors to collect the running data of the turnout device, corresponding time series data is obtained.

[0025] For example, the A-phase current, B-phase current and C-phase current of the turnout device are collected by using the current sensor, and the power information of the turnout device is measured by using the power sensor, thereby forming the multi-dimensional time series data. The data collected by each sensor can be displayed as a set of time series data. For example, for a sample data set, each sample includes data of T time points, and each time point data can include four dimensions, i.e., A-phase current, B-phase current, C-phase current and power. That is, the data set format of the multi-dimensional time series data can be [N, T, D]. Wherein, N represents the number of samples, T represents the number of time points, and D=4, which corresponds to the four data dimensions of A-phase current, B-phase current, C-phase current and power. Different phase currents reflect different current information latitudes of the turnout device, and the schematic diagram of the time series data of each dimension can be seen from Figures 3-6 , Figure 3 and Figure 4 are the time series data schematic diagrams of A-phase current and B-phase current, respectively, Figure 5 and Figure 6 are the time series data schematic diagrams of C-phase current and power, respectively.

[0026] In this step, the key dimensions such as current and power are covered by multiple sensors, which can break through the limitation of traditional single sensor that can only monitor local parameters, and comprehensively capture the state changes of the turnout at different stages such as starting, action and stable operation. For example, abnormal fluctuation of A-phase current may reflect a switch machine coil fault, and sudden power rise may correspond to problems such as point rail blocking, and multi-dimensional data combination can avoid misjudgment of the running state due to parameter loss. The continuity of time series can retain the dynamic change characteristics of the equipment running state, rather than just capturing parameters at a certain static moment, which enables subsequent analysis to trace the time node and evolution process of the abnormality, such as identifying intermittent and gradual potential faults (such as slow increase of action resistance caused by bedplate wear) by analyzing the coordinated change trend of current and power in a certain period, providing richer and more accurate original data support for subsequent feature extraction and abnormality judgment, and solving the one-sidedness problem of traditional detection caused by single data dimension and lack of dynamic information from the source.

[0027] Step 202, time domain and frequency domain analysis is performed on the time series data to construct a feature vector.

[0028] It can be understood that the feature vector is a multi-dimensional data vector for comprehensively representing the essential properties of the turnout equipment running state, which is formed by combining various features obtained by time domain analysis and frequency domain analysis on the multi-dimensional time series data of the turnout equipment.

[0029] Specifically, after obtaining the time series data, time domain analysis and frequency domain analysis are performed on the time series data of each sensor to extract features of different latitudes. Among them, time domain analysis focuses on the statistical characteristics of data in the time dimension, and captures the concentration trend, dispersion degree and distribution form of the turnout equipment running parameters by calculating indicators such as mean, standard deviation, maximum value, minimum value, skewness and peak factor. For example, taking the time series data as the current signal, the mean of the current signal can reflect the average load of the equipment running, the standard deviation can reflect the severity of parameter fluctuation, and the peak factor can effectively identify abnormal signals such as transient impact. These time domain features are directly related to the stable running state of the turnout system, such as whether the switch machine action is smooth, whether there is instantaneous current surge caused by blocking, etc.

[0030] Frequency domain analysis converts time series data into frequency domain through Fourier transform to extract features such as main frequency, spectral energy and specific band energy, and reveals the periodicity and frequency components hidden in the data. Mechanical vibration of the turnout equipment and electromagnetic interference of electrical components will form specific frequency characteristics in the signal, such as gear wear of the switch machine may cause energy enhancement of specific frequency components, and motor fault may cause main frequency shift. Time domain features and frequency domain features are combined and processed to construct a feature vector, which is used to comprehensively reflect the running state of the turnout equipment.

[0031] The feature vectors in this step can comprehensively characterize the essential attributes of the turnout equipment's operating status from different dimensions. They not only preserve the dynamic changes of parameters over time, but also uncover the potential patterns of signals in the frequency domain, laying the foundation for accurate identification of complex fault modes and overcoming the information loss problem caused by relying solely on single-dimensional features in traditional methods.

[0032] Step 203: Using the multidimensional weighted interquartile range algorithm, the features of each sample in the feature vector are weighted and summed to obtain the total anomaly score of each sample; the total anomaly score is used to characterize the overall operating status of the turnout equipment.

[0033] It should be noted that this application employs an improved interquartile range (IQR) method to calculate the anomaly score for each sample. Unlike the traditional IQR method, this application assigns a corresponding weighting factor to each feature in the feature vector, ensuring that different features are calculated according to their respective weighting factors during anomaly detection. This means that the contribution of different features to the final score is reasonably adjusted, thereby greatly enhancing the robustness and accuracy of the anomaly detection method.

[0034] Specifically, the anomaly score is calculated individually for each feature in the feature vector to accurately capture anomaly signals in a single feature dimension. Simultaneously, the improved four-way distance algorithm introduces a weighting mechanism, using the standardized standard deviation of each feature as a weighting factor. Features with greater volatility and higher sensitivity to anomalies receive higher weights, ensuring that key anomaly signals are fully reflected in the total score. This avoids the problem of anomalies being masked or mis-amplified due to equal contributions from all features in traditional methods.

[0035] In this embodiment, the total anomaly score, obtained by weighted summation of the anomaly scores of each feature for each sample, can comprehensively reflect the overall operating status of the turnout equipment. Unlike traditional single-feature anomaly scores, which can only reflect the anomalies of local parameters, the total anomaly score in this application achieves a global assessment of the equipment's operating status by weighted fusion of multi-dimensional feature information.

[0036] In this step, the total anomaly score is calculated by using a multidimensional weighted interquartile range algorithm, which can accurately quantify the degree of anomaly in the turnout equipment. This quantification method not only retains the differentiated contribution of different features to anomaly detection, but also realizes the organic integration of multidimensional information, providing an intuitive and comprehensive quantitative basis for subsequent anomaly judgment and effectively improving the accuracy of identifying complex fault modes.

[0037] Step 204: Compare the total anomaly score with the current dynamic threshold to obtain the anomaly detection result; the current dynamic threshold is determined based on historical statistical data and sensitivity control factors.

[0038] It can be understood that in the turnout anomaly detection scene, the traditional fixed threshold determination method has obvious limitations. Once the threshold is set, it will not change for a long time, and it cannot adapt to the differences in normal data distribution caused by environmental changes (such as low temperature leading to increased friction of the frog, high temperature causing parts to expand) and running state fluctuations (such as high frequency of action during peak hours and low load operation during off-peak hours). It is easy to misjudge normal fluctuations as anomalies (false positives), or miss early minor anomalies due to high threshold (false negatives). Therefore, the dynamic threshold determination method introduced in the present application has the core of using the abnormal score distribution of historical data as the core basis for threshold adjustment, dynamically adjusting the current threshold, so as to adapt to the turnout system under different environments and different running states, thereby improving the accuracy of anomaly detection.

[0039] It should be noted that the above total abnormal score is a quantitative value obtained by weighting and summing the feature abnormal scores of each sample in the turnout device feature vector through the multi-dimensional weighted quartile range algorithm. Its core function is to comprehensively represent the overall running state of the turnout device corresponding to a single sample. The higher the score, the more significant the degree of deviation of the device from the normal running state, which can intuitively reflect whether the device has potential failure and the severity of the failure, and provide a core quantitative basis for subsequent anomaly determination.

[0040] The above current dynamic threshold is a dynamic determination standard for determining whether the total abnormal score is "abnormal". Unlike the traditional fixed threshold, its value will be adjusted in real time according to the running condition changes of the turnout device. The determination of this threshold depends on two key elements: one is historical statistical data, that is, by analyzing the total abnormal scores of the turnout device during normal operation in the past, calculating the mean, standard deviation and other statistical indicators of the historical scores, and using them as the basis for threshold setting; the second is the sensitivity control factor, usually represented by k, which is a manually adjustable parameter. The current dynamic threshold is determined by the historical score mean, the historical score standard deviation and the sensitivity control factor.

[0041] After obtaining the total abnormal score of a single sample, the total abnormal score of the single sample is compared with the current dynamic threshold to obtain an anomaly detection result. The anomaly detection result can be divided into two categories: "normal" and "abnormal". When the total abnormal score is not higher than the current dynamic threshold, it is determined that the turnout device is in a normal running state and has no abnormal risk. When the total abnormal score is higher than the current dynamic threshold, it is determined that the turnout device has an anomaly, i.e. it is in an abnormal state and may have a potential failure, which needs to be further investigated or trigger an alarm. The core value is to provide clear device state determination results for maintenance personnel to guide subsequent maintenance actions.

[0042] Optionally, when it is determined that the turnout device is in an abnormal state, an alarm prompt information can be generated and an alarm prompt can be performed, for example, the alarm can be performed in a sound manner, the alarm can be performed in an image display manner, or the alarm can be performed in an information push prompt manner, and the embodiment does not limit the prompt manner of the abnormal alarm in any way.

[0043] Exemplarily, referring to FIG. 6, Figure 7 As shown in FIG. 6, the multi-dimensional time series data reflecting the operation state of the turnout device is collected by the current sensor and the power sensor, the time series data can be represented in the form of a current curve or a power curve, and time domain and frequency domain feature extraction is performed on the time series data to obtain time domain features and frequency domain features. The time domain features and the frequency domain features are combined to construct a feature vector. For each feature of each sample in the feature vector, the standard deviation value of each feature is calculated, and according to the standard deviation value, the weighted factor of each feature, i.e., the feature weight value, is calculated. Then, the improved IQR method is used to calculate the abnormal score of each feature. The abnormal scores of the features are weighted and summed to obtain the total abnormal score of each sample. Further, the current dynamic threshold is determined according to the historical statistical data and the sensitivity control factor, the total abnormal score is compared with the current dynamic threshold, and whether the turnout device is abnormal is determined according to the comparison result, so as to detect and identify the potential fault and other abnormal conditions of the turnout device.

[0044] By using the improved interquartile range (IQR) method and combining the multi-dimensional feature weighting and dynamic threshold determination technology, the potential abnormalities and faults can be more accurately and sensitively identified, thereby improving the safety and stability of the turnout system. Not only can the false positives and false negatives be effectively reduced, but the detection sensitivity can also be flexibly adjusted in actual operation to adapt to the needs of different working conditions and ensure the continuous and stable operation of the turnout device.

[0045] Compared with related technologies, the technical solution in the present application can comprehensively capture the running state information of the turnout device at different times by acquiring multi-dimensional running data of the turnout device synchronously collected by multiple sensors, avoid the problem of incomplete description of the running condition caused by single data dimension, lay a data foundation for subsequent accurate anomaly detection, perform time domain and frequency domain analysis on the time series data, construct a feature vector, break through the limitation of simple threshold judgment of original data in related technologies, deeply mine deep features of the data, capture potential abnormal signals that are difficult to identify by traditional methods, and improve the recognition ability of complex fault modes; the multi-dimensional weighted quartile range algorithm is used to allocate a weighted factor according to the feature standard deviation standardization result and to weight and sum the feature anomaly scores to obtain a total anomaly score, thereby solving the problem that related technologies do not consider the feature contribution difference, making the features more sensitive to anomalies play a greater role, accurately quantifying the sample anomaly degree, and improving the accuracy of representing the overall running state of the turnout; further, the current dynamic threshold is determined by combining historical anomaly score statistical data and a sensitivity control factor, the total anomaly score is compared with the dynamic threshold to obtain a detection result, and by setting the dynamic threshold to adapt to the defects of the working condition change, the turnout running environment and the working condition change can be adapted in real time, the false positive rate and the false negative rate under complex working conditions are significantly reduced, and the accuracy of anomaly detection is improved.

[0046] In an optional embodiment of the present application, a specific implementation manner of performing time domain and frequency domain analysis on the time series data and constructing a feature vector is also provided, including the following method steps: The time series data is subjected to time domain feature extraction to obtain time domain features; the time domain features include at least one of the following: mean, standard deviation, maximum value, minimum value, skewness and peak factor; the time series data is subjected to Fourier transform processing to extract frequency domain features; the frequency domain features include at least one of the following: dominant frequency information, spectral energy and frequency band energy; the time domain features and the frequency domain features are combined and processed to construct a feature vector.

[0047] The above time domain features are features extracted after statistical analysis of multi-dimensional time series data of the turnout device on the time axis, such as mean, standard deviation, maximum value, etc., which can directly reflect the concentration trend, dispersion degree and other time dimension running state laws of the data. The frequency domain features are features extracted after the multi-dimensional time series data of the turnout device is converted to the frequency domain through Fourier transform, such as dominant frequency, spectral energy, etc., which can reveal the periodicity and frequency level running state features hidden in the data.

[0048] Specifically, after obtaining the multi-dimensional time series data, the mean of the data can be calculated, the mean refers to the average value of the data, representing the central position of the time series data, for example, the mean of the current signal can reflect the average power load in the switch machine driving process, if the mean is long-term high, it implies that the switch machine has the problem of increased mechanical resistance (such as abnormal friction between the point rail and the basic rail), which provides a basis for judging whether the equipment is in an overload operation state. The mean can be represented by the following formula: ; Wherein, X(t) is the data of the tth time point in the time series data, and T is the number of time points.

[0049] The standard deviation (STD) is used to represent the fluctuation range of the data, reflecting the dispersion degree of the data. For example, if the standard deviation of the power signal suddenly increases, it means that the energy consumption fluctuation of the turnout operation is intensified, which may be an early signal of abnormality such as bedplate jamming and unstable switch machine action, and the abnormal fluctuation trend of the parameter can be captured in time. It can be represented by the following formula: ; Wherein, is the mean, X(t) is the data of the tth time point in the time series data, and T is the number of time points.

[0050] The maximum value (Max) and the minimum value (Min) are used to reflect the extreme value in the time series data, which specifically represents the peak state and the valley state in the equipment operation. For example, if the current maximum value exceeds the upper limit of the rated current of the equipment, it may be caused by the serious fault such as switch machine locked rotor; and if the current minimum value is too low, it may imply the problem of poor circuit contact, and the combination of the two can quickly identify whether the parameter is out of the extreme value range of the normal operation of the equipment. It can be represented by the following formula: ; Wherein, X(T) is the data of the Tth time point in the time series data.

[0051] Skewness reflects the stability of the running state by quantifying the asymmetry of the data distribution. If the skewness is close to 0, it means that the data is symmetrically distributed around the mean, and the equipment is running stably; if the skewness is positive, it means that there is a long tail on the right side (high value side) of the data, which may exist intermittent high value anomaly (such as occasional instantaneous current surge); if the skewness is negative, it means that there is a long tail on the left side (low value side) of the data, which may exist persistent low value anomaly (such as persistent low current), which can mine potential asymmetric anomalies from the data distribution form. It can be represented by the following formula: ; Wherein, is the mean value, X(t) is the data at the tth time point in the time series data, T is the number of time points, is the standard deviation.

[0052] Peak-to-Peak is used to represent the maximum fluctuation amplitude of the time series, for example, if the peak-to-peak of the power signal suddenly increases during the turnout action process, it means that the fluctuation span of the power in a short time has become larger, which may be caused by abnormality such as obstruction of switch machine action, load mutation, etc. It can quickly judge whether there is an abnormal situation of violent fluctuation in the equipment operation. It can be expressed by the following formula: ; Where Max is the maximum value, and Min is the minimum value.

[0053] After determining the time domain features, the time series data is processed by Fourier transform to convert the data from time domain to frequency domain, and the frequency domain features are extracted. These time domain features can reflect the periodicity and frequency characteristics in the signal, and can include dominant frequency, spectral energy and band energy. Among them, the dominant frequency is the frequency corresponding to the maximum amplitude in the frequency domain, the spectral energy is used to reflect the frequency energy distribution value of the whole signal, and the band energy is the frequency spectrum energy value in a specific frequency band range; After obtaining the frequency domain features, the extracted time domain features and frequency domain features are combined and processed to obtain a feature vector [N, M]. Where N represents the number of samples, and M is the number of feature dimensions. Through these feature vectors, the running state of the turnout equipment can be fully reflected.

[0054] Exemplarily, the time domain features contained therein, such as mean value and standard deviation, and the frequency domain features, such as dominant frequency and spectral energy, can be extracted first, and then these features from different analysis dimensions are arranged in a predetermined order to form a one-dimensional vector with a length of M. The one-dimensional vectors of all N samples are stacked in rows to finally obtain a feature matrix with a dimension of [N, M].

[0055] In this embodiment, by constructing the feature vector, the dynamic change rule of the parameter reflected by the time domain feature is retained, and the periodic fault signal revealed by the frequency domain feature is also included, so that the feature vector can fully depict the essential attributes of the turnout running state from two dimensions of time and frequency, provide input data containing multi-dimensional and multi-level information for subsequent anomaly detection algorithm, overcome the limitations of one-sidedness of single domain feature information, and improve the characterization ability of the feature to complex fault modes.

[0056] In an optional embodiment of the present application, a specific implementation of using a multi-dimensional weighted interquartile range algorithm to perform weighted summation on each feature of each sample in the feature vector to obtain the total anomaly score of each sample is provided, please refer to Figure 8 The method comprises the following steps: Step 301, for each feature of each sample in the feature vector, calculate the anomaly score of each feature.

[0057] Step 302, determine the weighting factor of each feature.

[0058] Step 303, perform weighted summation processing on the anomaly score and the weighting factor of each feature to obtain the total anomaly score of each sample.

[0059] In this embodiment, the precision and sensitivity of detection are improved by using a multi-dimensional weighted interquartile range (IQR) method combined with dynamic threshold determination. The core idea is to assign different weights to different features to more accurately assess the degree of abnormality. In order to achieve accurate monitoring of the turnout system, the anomaly degree of each sample is evaluated according to the collected time series data, and finally early warning of potential faults is realized. In the traditional anomaly detection method, statistical methods or machine learning models are usually used to analyze data and determine whether the data is outside the normal range. The multi-dimensional weighted IQR anomaly score calculation method proposed in the present application is an improvement based on the classic IQR method. By introducing the feature weighting and dynamic threshold mechanism, the detection accuracy is improved, and the adaptability to complex environments and variable fault modes is enhanced.

[0060] It can be understood that the interquartile range is a common statistical tool for measuring the dispersion of data. Its calculation method is the difference between the third quartile Q3 and the first quartile Q1 of the data, representing the distribution range of the middle 50% of the data. According to the interquartile range, the abnormal interval is determined, and then the degree of deviation of the data points from the abnormal interval is determined to determine the anomaly score of each data.

[0061] After determining the anomaly score of each feature, the weighting factor of each feature is determined, which can reflect the importance of the feature in the overall anomaly detection. The weighting factor is set according to the importance of each feature in anomaly detection. In order to reasonably assign different weights to each feature, the weighting factor can be calculated according to the data volatility and contribution to anomaly detection of each feature. For each sample, the anomaly scores and weighting factors of each feature are weighted and averaged to obtain the total anomaly score of each sample, which can be represented by the following formula: ; where w i is the weighting factor of the i-th feature in the feature vector, Si is the anomaly score of the ith feature, and M is the number of feature dimensions.

[0062] The anomaly score is used to evaluate the anomaly degree of each sample, which not only solves the limitation of traditional IQR in processing multi-dimensional data, but also improves the identification accuracy and environmental adaptability of complex fault of the turnout, and realizes early warning of potential fault.

[0063] In this embodiment, the anomaly score of each sample for each feature in the feature vector is first calculated, then the weighting factor of each feature is determined, and finally the total anomaly score is obtained by weighting and summing the anomaly scores and the weighting factors. This process can accurately capture the abnormal signals of each feature dimension, and at the same time, through the weighting mechanism, the contribution of the features more sensitive to anomalies is highlighted, avoiding the abnormal information being covered or misamplified due to the equal effect of each feature, so as to more comprehensively and accurately quantify the overall anomaly degree of each sample, significantly improving the accuracy of the evaluation of the running state of the turnout device, and providing a reliable quantitative basis for subsequent anomaly detection.

[0064] In an optional embodiment of the present application, the anomaly score of each feature is calculated, comprising: determining the first quartile and the third quartile of the feature; the first quartile is the value corresponding to the 25% position of the feature data points, and the third quartile is the value corresponding to the 75% position of the feature data points; calculating the interquartile range according to the first quartile and the third quartile; determining the abnormal interval based on the first quartile, the third quartile and the interquartile range; obtaining the anomaly score of the corresponding data point under the feature based on the degree of deviation of each data point under the feature from the abnormal interval.

[0065] Specifically, for each feature in the feature vector, its anomaly score can be calculated separately; first, the first quartile Q1 of the feature is determined, which is the value corresponding to the 25% position of the feature data, i.e. the lower quartile, and the third quartile Q3 is determined, which is the value corresponding to the 75% position of the feature data, i.e. the upper quartile; then the third quartile Q3 is subtracted from the first quartile Q1 to calculate the interquartile range (IQR), which is represented by the formula IQR=Q3-Q1, i.e. the width of the middle 50% range of data. Further, the abnormal interval is determined by the formula [Q1-1.5×IQR,Q3+1.5×IQR]; the anomaly score of the corresponding data point under the feature is obtained according to the degree of deviation of each data point from the abnormal interval. The higher the anomaly score of the data point, the higher the anomaly degree of the data point; the lower the anomaly score of the data point, the lower the anomaly degree of the data point.

[0066] It should be noted that when the traditional IQR is used for anomaly detection, the normal data interval is directly taken as [Q1-1.5*IQR, Q3+1.5*IQR], and data outside the interval is determined as abnormal, but this method is only applicable to one-dimensional data and does not consider the difference in the contribution of different features to anomaly detection, which is prone to misjudgment in the multi-dimensional data scenario of the turnout.

[0067] The multi-dimensional weighted IQR method in the present application makes key improvements on the basis of the classical IQR for the detection requirements of the turnout: on the one hand, for each dimension of the feature vector, the IQR is used to calculate the anomaly score of each dimension, ensuring that the anomaly signal of each feature can be preliminarily captured; on the other hand, a feature weighting mechanism is introduced, the standard deviation of each feature (reflecting the volatility of the feature, and the feature with greater volatility is more sensitive to anomalies) is standardized to obtain a weighting factor, so that the anomaly score of the key feature accounts for a higher proportion in the total score, avoiding interference from secondary features. Combined with a dynamic threshold (adjusted according to historical anomaly score statistical data, suitable for different working conditions of the turnout), the final total anomaly score is obtained by weighted summation.

[0068] In the present embodiment, the first quartile (25% position value) and the third quartile (75% position value) of the feature data are first determined, then the interquartile range is calculated and the abnormal interval is determined, and finally the feature anomaly score is obtained according to the degree of deviation of the data points from the abnormal interval. This process can accurately define the normal range based on the distribution of the feature data itself, effectively filter out normal fluctuations in the data, and accurately capture the anomaly signal of the feature dimension by quantifying the degree of deviation, avoiding misjudgment caused by the traditional fixed threshold deviating from the data distribution. It provides a reliable anomaly basis for each feature dimension for subsequent total anomaly score calculation, and improves the accuracy of anomaly detection of the turnout equipment.

[0069] In an optional embodiment of the present application, for each feature of each sample in the feature vector, the standard deviation value of each feature is calculated, including: The standard deviation value is standardized to obtain a standardized standard deviation value, and the standardized standard deviation value is taken as the weighting factor of the feature. The standardization processing is to compare the standard deviation value of a single feature with the sum of the standard deviation values of all features.

[0070] It can be understood that in order to reasonably assign different weights to each feature, the weighting factor can be calculated according to the data volatility and contribution to anomaly detection of each feature. One possible implementation is to measure the volatility of the feature by standard deviation, and use the volatility to determine the weighting factor.

[0071] Specifically, after obtaining each feature in the feature vector, the standard deviation value STD of each feature can be calculated, which reflects the fluctuation degree of the feature in the time series, and the greater the fluctuation, the greater the contribution of the feature in anomaly detection. Then the standard deviation value of each feature is normalized, that is, the standard deviation of a single feature is divided by the sum of the standard deviations of all features to obtain the normalized standard deviation value, which can be represented by the following formula: ; wherein, is the normalized standard deviation value of the jth feature, which reflects the dispersion degree of the feature data, represents the sum of the standard deviations of all n features, and j is the standard deviation value of the jth feature. By this formula, the standard deviation of each feature is divided by the sum of the standard deviations of all features, and the result is the weighting factor of the feature in the overall feature set. This means that the larger the standard deviation (i.e., the higher the data dispersion and the stronger the volatility), the larger the weighting factor, and in subsequent scenarios involving weighted calculation (such as weighted summation of anomaly scores), the feature will be given a higher weight, thereby highlighting its impact on anomaly detection results.

[0072] The normalized standard deviation value can be directly used as the weighting factor of each feature in the weighting step of anomaly score calculation, reflecting the contribution of the fluctuation of the feature to anomaly detection. It is represented by the following formula: ; wherein, j is the weighting factor of the jth feature.

[0073] In this embodiment, the standard deviation value of each feature is first calculated to reflect its data dispersion, and then the single feature standard deviation is normalized by dividing it by the sum of the standard deviations of all features to obtain the weighting factor. This operation can assign different weights according to the fluctuation characteristics of each feature, allowing features with greater fluctuations and greater sensitivity to anomalies to play a greater role in subsequent calculations (such as weighted summation of anomaly scores), effectively highlighting the anomaly information of key features, avoiding interference caused by differences in dimensions or fluctuation amplitudes between features, and improving the accuracy and relevance of turnout equipment anomaly detection.

[0074] In an optional embodiment of the present application, a specific implementation of determining the current dynamic threshold is also provided, including: obtaining historical statistical data, determining the historical score mean and the historical score standard deviation according to the anomaly scores of the historical statistical data; and determining the current dynamic threshold based on the historical score mean, the historical score standard deviation, and the sensitivity control factor.

[0075] It should be noted that the above historical statistical data refers to the set of all total abnormal scores and derived statistical indicators calculated by the multi-dimensional weighted interquartile range algorithm during the past normal operation period of the turnout equipment. In addition to the mean and standard deviation used to calculate the current dynamic threshold, it can also reflect the abnormal score distribution law of the equipment in normal operation, provide data support for the adaptive adjustment of the dynamic threshold, ensure that the threshold setting always fits the normal state benchmark of long-term operation of the equipment, and avoid the disconnection of the threshold from the actual normal range due to changes in working conditions.

[0076] In the traditional scheme, the static threshold is fixed and cannot adapt to the scenario where the data distribution changes over time and working conditions. The dynamic threshold determination of the present application allows the abnormal detection threshold to be dynamically adjusted based on the statistical characteristics of historical data, achieving adaptive detection.

[0077] Specifically, the abnormal scores in the historical statistical data generated during the past normal operation period of the turnout equipment are first statistically analyzed, and the historical score mean and historical score standard deviation are calculated. The historical score mean is used to reflect the score benchmark in the normal state, and the historical score standard deviation is used to reflect the normal fluctuation range. Subsequently, based on these two key statistical quantities, combined with the adjustable sensitivity control factor (k value), the current dynamic threshold is calculated by the following formula: ; wherein k is the sensitivity control factor, is the historical score mean, is the historical score standard deviation. k can be flexibly adjusted according to the requirement of abnormal detection sensitivity, such as increasing the value of k to increase the threshold and reduce the detection sensitivity to reduce false positives; reducing the value of k reduces the threshold and increases the detection sensitivity to reduce false negatives, thereby adapting to the detection requirements in different operating scenarios.

[0078] When new data arrives, the total abnormal score of the data sample is calculated by the IQR calculation method, which is used to reflect the overall operation state of the turnout equipment, and the total abnormal score is compared with the current dynamic threshold. If the total abnormal score exceeds the threshold, it is determined that the turnout equipment is in an abnormal state, and further triggers an alarm or performs fault prediction.

[0079] The adjustment mechanism in the embodiment can adaptively optimize the threshold value with the change of the turnout operating environment and state: for example, in winter, the normal fluctuation range of the turnout operating current is expanded, the standard deviation of the historical score is increased accordingly, and the dynamic threshold value is increased to avoid misjudgment of the normal current fluctuation caused by low temperature as an anomaly; when the turnout enters the peak operating period of high-frequency operation, the dynamic threshold value can be reduced by appropriately reducing the sensitivity control factor k, and the capture sensitivity to slight anomalies (such as slow increase of the operating resistance caused by initial wear of the switch machine) is improved. Ultimately, the dynamic threshold value determination method breaks the rigid restriction of the fixed threshold value, so that the anomaly detection standard is always accurately matched with the actual operating conditions of the turnout, fundamentally reducing false positives and false negatives in complex scenarios, and further ensuring the accuracy and reliability of the anomaly detection results.

[0080] In the embodiment, the current dynamic threshold value is determined by obtaining historical statistical data and determining the mean and standard deviation of the anomaly score, and then combining the sensitivity control factor, which can adapt the anomaly detection standard to the dynamic change of the data distribution, avoid the problem of missed detection or misjudgment caused by the change of data distribution over time and operating conditions, and thus maintain high anomaly detection accuracy and adaptability in different scenarios.

[0081] It should be understood that although each step in the flowchart is displayed in sequence according to the direction of the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, there is no strict order limitation for the execution of these steps, and these steps can be executed in other orders. Moreover, at least part of the steps in the figure can include multiple sub-steps or multiple stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in rotation or alternation with at least part of other steps or sub-steps or stages of other steps.

[0082] In another embodiment provided in the present application, a turnout anomaly detection device is also provided, please refer to Figure 9 As shown in the figure, the device comprises: The acquisition module 810 is configured to acquire multi-dimensional time series data of the turnout device; the multi-dimensional time series data is used to represent the operating conditions of the turnout device collected by multiple sensors at different times; The construction module 820 is configured to perform time domain and frequency domain analysis on the time series data to construct a feature vector; The processing module 830 is configured to use a multi-dimensional weighted quartile distance algorithm to perform weighted summation on each feature of each sample in the feature vector to construct a total anomaly score of each sample; the total anomaly score is used to represent the overall operating state of the turnout device; The anomaly detection module 840 is configured to compare the total anomaly score with a current dynamic threshold to obtain an anomaly detection result, and the current dynamic threshold is determined according to historical statistical data and a sensitivity control factor.

[0083] Optionally, the construction module 820 is specifically configured to: perform time domain feature extraction on the time series data to obtain time domain features, and the time domain features include at least one of the following: mean value, standard deviation, maximum value, minimum value, skewness, and peak factor; perform Fourier transform processing on the time series data to extract frequency domain features, and the frequency domain features include at least one of the following: main frequency information, frequency spectrum energy, and frequency band energy; perform combination processing on the time domain features and the frequency domain features to construct a feature vector.

[0084] Optionally, the processing module 830 is specifically configured to: calculate an anomaly score of each feature of each sample in the feature vector; determine a weighting factor of each feature; perform weighted summation processing on the anomaly scores of the features and the weighting factors to obtain a total anomaly score of each sample.

[0085] Optionally, the processing module 830 is further configured to: determine a first quartile and a third quartile of the feature, the first quartile being a value corresponding to a 25% position in feature data points, and the third quartile being a value corresponding to a 75% position in the feature data points; calculate a quartile range according to the first quartile and the third quartile; determine an anomaly interval based on the first quartile, the third quartile, and the quartile range; obtain an anomaly score of each data point under the feature based on a degree of deviation of each data point under the feature from the anomaly interval.

[0086] Optionally, the processing module 830 is further configured to: calculate a standard deviation value of each feature of each sample in the feature vector; perform standardization processing on the standard deviation value to obtain a standardized standard deviation value, and use the standardized standard deviation value as the weighting factor of the feature; the standardization processing is to compare the standard deviation value of a single feature with a sum of standard deviation values of all features.

[0087] Optionally, the anomaly detection module 840 is specifically configured to: determine that the turnout device is in an abnormal state when the total anomaly score is greater than the current dynamic threshold; determine that the turnout device is in a normal state when the total anomaly score is not greater than the current dynamic threshold.

[0088] Optionally, the device further has the functions of: acquiring historical statistical data, determining a historical score mean and a historical score standard deviation according to an abnormal score of the historical statistical data; determining the current dynamic threshold based on the historical score mean, the historical score standard deviation, and a sensitivity control factor.

[0089] The specific limitations of the above-described turnout anomaly detection device can be referred to the limitations of the turnout anomaly detection method described above, which will not be repeated here. Each module in the above-described turnout anomaly detection device can be realized by software, hardware, or a combination thereof, in whole or in part. The above-described modules can be embedded in or independent of the processor in the computer device in hardware form, or can be stored in the memory in the computer device in software form, so as to be called and executed by the processor to perform the operations corresponding to each module.

[0090] In an embodiment, a computer device is provided, and an internal structure diagram of the computer device can be as shown in Figure 1 The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is used to store data. The network interface of the computer device is used to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement a turnout anomaly detection method as described above. The computer program includes a memory and a processor, and the memory stores a computer program. When the processor executes the computer program, it implements any step in the above-described turnout anomaly detection method.

[0091] In an embodiment, a computer-readable storage medium is provided, and the computer-readable storage medium stores a computer program. When the computer program is executed by a processor, it can implement any step in the above-described turnout anomaly detection method.

[0092] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code.

[0093] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 one or more flowcharts and / or blocks

[0094] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks Figure 1 one or more flowcharts and / or blocks

[0095] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks ​ one or more flowcharts and / or blocks

[0096] While the preferred embodiments of the application have been described, additional variations and modifications can be employed by those skilled in the art. Therefore, the appended claims intend to cover all such modifications and variations as fall within the true spirit and scope of the application. Further, the appended claims can be construed to cover all alternatives falling within the equivalent of the groups of claims.

[0097] It will be apparent to those skilled in the art that various modifications and variations can be made to the present application without departing from the spirit or scope of the application. Thus, it is intended that the present application cover modifications and variations of this application provided they come within the scope of the appended claims and their equivalents.

Claims

1. A method for detecting turnout anomalies, characterized in that, The turnout anomaly detection method includes: Acquire multi-dimensional time series data of the turnout equipment; the multi-dimensional time series data is used to characterize the operating status of the turnout equipment at different times, as collected by multiple sensors; Perform time-domain and frequency-domain analysis on the time-series data to construct feature vectors; A multidimensional weighted interquartile range algorithm is used to sum the weighted features of each sample in the feature vector to obtain the total anomaly score for each sample; the total anomaly score is used to characterize the overall operating status of the turnout equipment. The total anomaly score is compared with the current dynamic threshold to obtain the anomaly detection result; the current dynamic threshold is determined based on historical statistical data and sensitivity control factors.

2. The method according to claim 1, characterized in that, The time series data is analyzed in both the time domain and frequency domain to construct feature vectors, including: The time series data is subjected to time-domain feature extraction to obtain time-domain features; the time-domain features include at least one of the following: mean, standard deviation, maximum value, minimum value, skewness, and kurtosis factor; The time series data is subjected to Fourier transform processing to extract frequency domain features; the frequency domain features include at least one of the following: main frequency information, spectral energy, and frequency band energy; The time-domain features and the frequency-domain features are combined to construct the feature vector.

3. The method according to claim 1, characterized in that, A multidimensional weighted interquartile range (IIR) algorithm is used to sum the weighted features of each sample in the feature vector to obtain the total anomaly score for each sample, including: For each feature of each sample in the feature vector, calculate the anomaly score for each feature; Determine the weighting factor for each of the aforementioned features; The anomaly scores of each feature and the weighting factor are weighted and summed to obtain the total anomaly score for each sample.

4. The method according to claim 3, characterized in that, Calculating the anomaly score for each of the aforementioned features includes: Determine the first quartile and the third quartile of the feature; the first quartile is the value corresponding to the 25th percentile of the feature data points, and the third quartile is the value corresponding to the 75th percentile of the feature data points; Calculate the interquartile range based on the first quartile and the third quartile; Based on the first quartile, the third quartile, and the interquartile range, an abnormal interval is determined; Based on the degree to which each data point deviates from the abnormal interval under the aforementioned feature, an anomaly score is obtained for the corresponding data point under the aforementioned feature.

5. The method according to claim 3, characterized in that, Determining the weighting factor for each of the aforementioned features includes: For each feature of each sample in the feature vector, calculate the standard deviation of each feature; The standard deviation is standardized to obtain a standardized standard deviation, which is then used as a weighting factor for the feature. The standardization process involves comparing the standard deviation of a single feature with the sum of the standard deviations of all features.

6. The method according to claim 1, characterized in that, The total anomaly score is compared with the current dynamic threshold to obtain the anomaly detection result, including: When the total anomaly score is greater than the current dynamic threshold, the turnout equipment is determined to be in an abnormal state. When the total anomaly score is not greater than the current dynamic threshold, the turnout equipment is determined to be in normal condition.

7. The method according to claim 1, characterized in that, The method further includes: Obtain historical statistical data, and determine the historical score mean and historical score standard deviation based on the abnormal scores in the historical statistical data; The current dynamic threshold is determined based on the historical score mean, the historical score standard deviation, and the sensitivity control factor.

8. A turnout anomaly detection device, characterized in that, The turnout anomaly detection device includes: The acquisition module is used to acquire multi-dimensional time series data of the turnout equipment; the multi-dimensional time series data is used to characterize the operating status of the turnout equipment at different times, collected by multiple sensors. The construction module is used to perform time-domain and frequency-domain analysis on the time series data and construct feature vectors; The processing module is used to employ a multidimensional weighted interquartile range algorithm to perform weighted summation on each feature of each sample in the feature vector, thereby constructing a total anomaly score for each sample; the total anomaly score is used to characterize the overall operating status of the turnout equipment. An anomaly detection module is used to compare the total anomaly score with the current dynamic threshold to obtain an anomaly detection result; the current dynamic threshold is determined based on historical statistical data and a sensitivity control factor.

9. A computer device, comprising: A memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images