An elevator abnormal behavior recognition monitoring method and system for feature extraction

Abstract

The application discloses an elevator abnormal behavior recognition monitoring method and system for feature extraction, and relates to the technical field of elevators, and the technical solution points of the application include the following steps: classifying and processing elevator full-scene operation data to obtain a scene data group, constructing a multi-dimensional feature condition set based on the scene data group, and performing feature optimization processing on the multi-dimensional feature condition set to generate core feature condition factors corresponding to the scene data group; encoding the core feature condition factors according to a dynamic coding rule to obtain feature condition identifiers corresponding to the scene data group; obtaining a hierarchical feature verification system according to the hierarchical verification relationship between the feature condition identifiers and the feature condition identifiers corresponding to the scene data group, determining node condition feature values of the hierarchical feature verification system, and extracting path tracing information of each feature condition identifier in the hierarchical feature verification system; and the effect is to effectively support the intelligent safety management of elevators.

Description

Technical Field

[0001] This invention relates to the field of elevator technology, and more specifically, to a method and system for identifying and monitoring abnormal elevator behavior based on feature extraction. Background Technology

[0002] Traditional elevator monitoring methods often rely on single sensor data or simple numerical threshold judgments. For example, they may use the rated value of the load sensor to determine overload or the extreme value of the speed sensor to identify speeding. A single data dimension cannot cover the complex operating scenarios of elevators, such as door jamming or hidden abnormal vibrations of the motor. These issues are difficult to detect in a timely manner with a single data point, which can easily lead to missed detection of hidden faults. Simple threshold judgments lack consideration for data correlation. Elevator operation is a process of interconnection between various systems. For example, there is a logical correlation between the door opening and closing status and load changes, but traditional methods do not incorporate such correlations into the judgment system. This often leads to misjudgments of abnormalities due to instantaneous data fluctuations or failure to locate faults due to hidden correlations between systems.

[0003] With the diversification of elevator usage scenarios (such as high customer traffic in shopping malls and aging equipment in old residential areas), the volume and complexity of elevator operation data have increased significantly. Traditional monitoring methods are unable to efficiently process multi-source data, and are unable to extract key features from massive amounts of data or build a unified status identification system. This results in low efficiency of real-time monitoring, and lacks a clear tracing path after a fault occurs. Maintenance personnel often need to investigate one by one, which is time-consuming and labor-intensive. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for identifying and monitoring abnormal elevator behavior based on feature extraction.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for identifying and monitoring abnormal elevator behavior through feature extraction, comprising the following steps:

[0007] The elevator operation data of the whole scene is classified and processed to obtain the scene data group. Based on the scene data group, a multi-dimensional feature status set is constructed. The multi-dimensional feature status set is subjected to feature optimization processing to generate the core feature status factor corresponding to the scene data group.

[0008] The core feature status factors are encoded according to the dynamic coding rules to obtain the feature status identifiers corresponding to the scene data group;

[0009] According to the hierarchical verification relationship between the feature condition identifiers and the feature condition identifier corresponding to the scene data set, a hierarchical feature verification system is obtained, the node condition feature values of the hierarchical feature verification system are determined, and path tracing information of each feature condition identifier in the hierarchical feature verification system is extracted;

[0010] Real-time scene data of the elevator is obtained, a real-time multi-dimensional feature set is obtained according to the real-time scene data of the elevator, real-time core feature factors are generated after the real-time multi-dimensional feature set is pre-set feature screened and optimized, and real-time feature identifiers are obtained by encoding the real-time core feature factors according to a dynamic coding rule;

[0011] The real-time feature identifiers, the feature condition identifiers, the path tracing information, the node feature values and the node condition feature values are processed to determine whether the elevator has abnormal behavior.

[0012] Preferably, the real-time feature identifiers, the feature condition identifiers, the path tracing information, the node feature values and the node condition feature values are processed to determine whether the elevator has abnormal behavior, specifically including the following steps:

[0013] The real-time feature identifiers are compared with the feature condition identifiers to obtain a first determination result of real-time operation of the elevator;

[0014] A real-time hierarchical feature verification system is constructed based on the real-time feature identifiers and the path tracing information, and a second determination result of real-time operation of the elevator is obtained according to the node feature values and the node condition feature values of the real-time hierarchical feature verification system;

[0015] If the first determination result and the second determination result are both abnormal, it is determined that the elevator has abnormal behavior.

[0016] Preferably, the elevator full-scene operation data is classified to obtain a scene data set, and a multi-dimensional feature condition set is constructed based on the scene data set, specifically including the following steps:

[0017] A scene classification standard is set; wherein the scene classification standard includes elevator operation state classification and data acquisition type classification;

[0018] The elevator full-scene operation data is divided into scene data sets according to the scene classification standard;

[0019] The time domain features, frequency domain features and statistical features of the data in the scene data sets are extracted;

[0020] The multi-dimensional feature condition set is constructed according to the time domain features, frequency domain features and statistical features.

[0021] Preferably, the core feature condition factors are encoded according to a dynamic coding rule to obtain feature condition identifiers corresponding to the scene data sets, specifically including the following steps:

[0022] a dynamic coding rule is set based on the number of dimensions of the core feature condition factor and the range of eigenvalues, wherein the dynamic coding rule comprises an eigenvalue segmentation mapping rule and a dimension priority coding rule;

[0023] Each dimension eigenvalue of the core feature condition factor is converted into a coding segment according to the eigenvalue segmentation mapping rule, and the coding segments are sequentially spliced to generate a feature condition identifier corresponding to the scene data group according to the dimension priority coding rule.

[0024] Preferably, the following steps are further included:

[0025] The acquisition time stamp and data source identifier of the data in each type of scene data group are obtained;

[0026] The time sequence correlation attribute of the data in the scene data group is determined based on the acquisition time stamp, and the interactive dependency relationship between different scene data groups is determined through the data source identifier;

[0027] The direct correlation relationship between the feature condition identifiers is determined according to the interactive dependency relationship of the scene data group; the indirect verification relationship between the feature condition identifiers is established after judging the logical dependency of the core feature condition factor in the scene data group; and the hierarchical verification relationship between the feature condition identifiers is obtained based on the direct correlation relationship and the indirect verification relationship.

[0028] Preferably, the node condition eigenvalue of the hierarchical feature verification system is determined, specifically including the following steps:

[0029] The correlation strength of each node with the superior and inferior nodes in the hierarchical feature verification system is judged, and the influence weight of the node in the system is determined according to the correlation strength;

[0030] The elevator operation basic state information associated with the feature condition identifier of the node is extracted, and the stability degree of the elevator operation basic state information is judged;

[0031] The influence weight and the state stability degree are integrated to form the basic parameter of the node condition;

[0032] The basic parameter is verified to obtain a verification basic parameter;

[0033] The verification basic parameter is derived through the state attribute correlation to obtain the node condition eigenvalue.

[0034] Preferably, the path tracing information of each feature condition identifier in the hierarchical feature verification system is extracted, specifically including the following steps:

[0035] The node attribution of each feature condition identifier in the hierarchical feature verification system is determined, and the pre-association of each feature condition identifier in the hierarchical feature verification system is obtained;

[0036] The tracking feature condition identifier identifies the evolution track of the corresponding elevator operation state;

[0037] The indirect association link between each feature condition identifier is determined, the conduction path between each feature condition identifier is determined according to the indirect association link, and the state influence range of each feature condition identifier is determined;

[0038] The path tracking information is formed by integrating the node attribution, the pre-association, the evolution track, the conduction path and the state influence range.

[0039] Preferably, the real-time feature identifier is compared with the feature condition identifier to obtain a first determination result of the real-time operation of the elevator, which specifically includes the following steps:

[0040] The similarity between the real-time feature identifier and the feature condition identifier is calculated;

[0041] If the similarity is greater than or equal to a preset similarity threshold, it is determined that the first determination result is normal;

[0042] If the similarity is less than the preset similarity threshold, it is determined that the first determination result is abnormal.

[0043] Preferably, a second determination result of the real-time operation of the elevator is obtained according to the node feature value and the node condition feature value of the real-time hierarchical feature verification system, which specifically includes the following steps:

[0044] If the node condition feature value of the real-time hierarchical feature verification system is consistent with the node condition feature value, it is determined that the second determination result is normal;

[0045] If the node condition feature value of the real-time hierarchical feature verification system is inconsistent with the node condition feature value, it is determined that the second determination result is abnormal.

[0046] An elevator abnormal behavior identification monitoring system for feature extraction, comprising:

[0047] The classification module: classifies the elevator full-scene operation data to obtain a scene data group, constructs a multi-dimensional feature condition set based on the scene data group, and optimizes the features of the multi-dimensional feature condition set to generate core feature condition factors corresponding to the scene data group;

[0048] The first processing module: encodes the core feature condition factors according to a dynamic coding rule to obtain feature condition identifiers corresponding to the scene data group;

[0049] The extraction module: obtains a hierarchical feature verification system according to the hierarchical verification relationship between the feature condition identifiers and the feature condition identifiers corresponding to the scene data group, determines the node condition feature values of the hierarchical feature verification system, and extracts the path tracking information of each feature condition identifier in the hierarchical feature verification system;

[0050] The second processing module: obtain real-time scene data of the elevator, obtain a real-time multi-dimensional feature set according to the real-time scene data of the elevator, generate a real-time core feature factor after pre-set feature screening optimization is performed on the real-time multi-dimensional feature set, and perform coding processing on the real-time core feature factor according to a dynamic coding rule to obtain a real-time feature identifier;

[0051] The output module: after processing the real-time feature identifier, the feature condition identifier, the path tracing information, the node feature value and the node condition feature value, it is determined whether the elevator has an abnormal behavior.

[0052] Compared with the prior art, the present application has the following beneficial effects:

[0053] The present application realizes comprehensive coverage of the running state of the elevator through full-scene data classification and multi-dimensional feature construction. The efficiency and accuracy of abnormality identification are improved through feature optimization and dynamic coding rules. The multi-dimensional features are optimized to generate core feature condition factors, and the dynamic coding rules convert abstract feature data into structured feature condition identifiers. The matching of the coding identifier is more intuitive and efficient, and the preliminary verification of real-time data and normal state can be completed in a short time. Combined with the linkage relationship between systems, such as the abnormality of the driving system, the state of the nodes such as speed and door machine is associated, avoiding misjudgment caused by single feature fluctuation. The path tracing information can locate the source node, transmission path and influence range after the abnormality occurs, helping the maintenance personnel to troubleshoot the fault, thereby shortening the maintenance response time. The similarity between the real-time feature identifier and the historical identifier is used to complete the preliminary determination, and the node feature value of the real-time hierarchical system is used for secondary verification, which not only ensures the timeliness of real-time monitoring, but also avoids the loopholes of single determination standard, such as the instantaneous load fluctuation of the elevator that may trigger preliminary abnormality, but the secondary node feature value verification can exclude misjudgment combined with the running stability, making the abnormality identification result more reliable, which can effectively support the intelligent safety management of the elevator and reduce the safety hidden danger. BRIEF DESCRIPTION OF DRAWINGS

[0054] Fig. 1 A feature extraction elevator abnormal behavior identification monitoring method step schematic diagram is proposed for the present application;

[0055] Fig. 2 A feature extraction elevator abnormal behavior identification monitoring system module schematic diagram is proposed for the present application. DETAILED DESCRIPTION

[0056] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification.

[0057] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details set forth in this description. In other instances, well-known methods, procedures, components, and circuits have not been described in detail so as not to unnecessarily obscure aspects of the present application.

[0058] It is also noted that, as used herein, "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one implementation of the present application. The appearances of the phrase "in one embodiment" or "an embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.

[0059] Referring to Figs. 1-2 as shown.

[0060] Embodiments further illustrate a feature extraction elevator abnormal behavior recognition monitoring method and system proposed by the present application.

[0061] A feature extraction elevator abnormal behavior recognition monitoring method, the method comprising the following steps:

[0062] Classifying the elevator full-scene running data to obtain a scene data group, and constructing a multi-dimensional feature condition set based on the scene data group, specifically comprising the following steps:

[0063] Setting a scene classification standard; wherein the scene classification standard comprises an elevator running state classification and a data acquisition type classification;

[0064] Dividing the elevator full-scene running data into the scene data group according to the scene classification standard;

[0065] Extracting time domain features, frequency domain features and statistical features of the data in the scene data group;

[0066] Constructing a multi-dimensional feature condition set according to the time domain features, the frequency domain features and the statistical features.

[0067] Setting a scene classification standard, the scene classification standard comprises an elevator running state classification and a data acquisition type classification, the elevator running state classification divides the running state of the elevator into uplink, downlink and stop waiting for elevator state; the data acquisition type classification is load data of a load sensor, speed data of a speed sensor and opening and closing door data of a door machine system. According to the scene classification standard, all running data of the elevator full-scene is divided into corresponding scene data groups, for example, all data collected by the speed sensor during the uplink process of the elevator is classified into uplink speed scene data group, and the opening and closing door data of the door machine during the stop waiting for elevator state is classified into waiting for elevator opening and closing door scene data group.

[0068] The time domain features, the frequency domain features and the statistical features are extracted from each scene data group. The time domain features are developed around the change law of data over time, and the time domain features include peak value, mean value, variance, root mean square, rising edge slope and falling edge slope. The peak value is the maximum value of data in a certain period of time, such as the maximum value of the speed in the uplink speed scene data group. The mean value is the average level of data in a certain period of time, such as the average value of the uplink speed in ten minutes. The mean value is equal to the sum of all speed data in the time period divided by the number of data collection. The variance is used to reflect the dispersion degree of data, and the variance is equal to the square sum of the difference between each speed data and the mean value divided by the number of data. The root mean square is the square sum of each speed data divided by the number of data, and then the square root is taken. The rising edge slope reflects the rate of data from the minimum value to the maximum value. The rising edge slope is equal to the difference between the peak value and the valley value divided by the time of the rising process. The falling edge slope is the rate of data from the peak value to the valley value. The falling edge slope is equal to the difference between the peak value and the valley value divided by the time of the falling process.

[0069] The frequency domain features include the main frequency amplitude, the main frequency and the harmonic amplitude. The frequency domain features need to be obtained by converting the time domain data to the frequency domain through Fourier transform. The main frequency amplitude is the amplitude corresponding to the highest energy frequency in the frequency spectrum, such as the vibration amplitude corresponding to the highest energy frequency after the vibration data of the elevator is transformed. The main frequency is the highest energy frequency value. The harmonic amplitude is the amplitude corresponding to the integer multiple frequency of the main frequency, and the harmonic frequency is the integer multiple frequency of the main frequency. These features can capture periodic abnormalities, such as the wear of the motor bearing which can cause a significant increase in the amplitude of a certain harmonic.

[0070] The statistical features include cumulative value, frequency and quantile. The cumulative value is the sum of data in a certain period of time, such as the total number of starts and stops of the elevator in a day. The frequency is the number of times a certain specific data appears, such as the number of times the load exceeds the rated value. The quantile is the value corresponding to the proportion of data after sorting, such as the median (50th quantile) which can reflect the intermediate level of data, and the 25th quantile and the 75th quantile which can reflect the distribution range of data.

[0071] The time domain features, the frequency domain features and the statistical features are integrated to construct a multi-dimensional feature condition set. The multi-dimensional feature condition set completely covers various state information of the elevator running under the corresponding scene, and provides basic feature support for subsequent identification of abnormal behavior.

[0072] The multi-dimensional feature condition set is subjected to feature optimization processing to generate core feature condition factors corresponding to the scene data group.

[0073] Since the multi-dimensional feature condition set contains time domain features, frequency domain features and statistical features, some features may have information redundancy or low recognition of elevator state changes, and therefore need to be optimized. The feature optimization process usually combines feature importance evaluation and redundancy judgment, such as judging the importance by calculating the correlation degree of each feature and the abnormal state of the elevator, the higher the correlation degree, the higher the priority of the feature; at the same time, the correlation between different features is judged, if the correlation between two features is too high, it means that the overlap degree of information carried by them is large, then the more representative one is retained.

[0074] Taking the multi-dimensional feature condition set of the elevator uplink scene as an example, the multi-dimensional feature condition set contains the peak mean variance root mean square of uplink speed, the main frequency amplitude main frequency frequency of vibration data, and the cumulative value quantile feature of load data. When optimizing, the importance of each feature is evaluated: for example, the root mean square of speed can comprehensively reflect the fluctuation degree of speed, and has a high correlation with the smoothness of elevator operation; the main frequency amplitude of vibration is directly related to the running state of the motor and other components, and is also important. If the mean value of speed has a high correlation with the root mean square, and the recognition degree of the root mean square is stronger, the mean value feature can be removed.

[0075] After such optimization, the original multi-dimensional feature condition set is simplified to a combination of core features, which are the core feature condition factors corresponding to the scene data group. For example, the core feature condition factors of the uplink scene retain the root mean square of speed, the main frequency amplitude of vibration, and the 75th quantile of load, which are key features. They not only cover the key dimensions of elevator operation, but also avoid information redundancy, and can more efficiently support subsequent encoding and abnormality judgment.

[0076] According to the dynamic coding rule, the core feature condition factor is coded to obtain the feature condition identifier corresponding to the scene data group, which includes the following steps:

[0077] The dynamic coding rule is set based on the number of dimensions of the core feature condition factor and the range of feature values; wherein the dynamic coding rule includes a feature value segmentation mapping rule and a dimension priority coding rule;

[0078] The dimension feature values of the core feature condition factor are converted into coding segments according to the feature value segmentation mapping rule, and the coding segments are sequentially spliced to generate the feature condition identifier corresponding to the scene data group according to the dimension priority coding rule.

[0079] Firstly, the corresponding dynamic coding rules need to be set based on the number of dimensions of the core feature condition factor and the range of characteristic values of each dimension. The dynamic coding rules include characteristic value segmentation mapping rules and dimension priority coding rules. Taking the core feature condition factor of the elevator uplink scene as an example, assuming that the core feature condition factor includes the root mean square of speed, the amplitude of the vibration main frequency, and the 75th percentile of the load. First, determine the range of characteristic values of each dimension, such as the normal range of the root mean square of speed is 0.2 to 0.8 m / s, the normal range of the amplitude of the vibration main frequency is 0.1 to 0.5 g, and the normal range of the 75th percentile of the load is 80 to 120 kg. Then set the characteristic value segmentation mapping rule, divide the characteristic value range of each dimension into several intervals, each interval corresponds to a fixed coding segment, such as the root mean square of speed 0.2-0.4 corresponds to coding 1, 0.4-0.6 corresponds to coding 2, 0.6-0.8 corresponds to coding 3; the amplitude of the vibration main frequency 0.1-0.2 corresponds to coding A, 0.2-0.4 corresponds to coding B, 0.4-0.5 corresponds to coding C; the 75th percentile of the load 80-100 corresponds to coding X, 100-120 corresponds to coding Y. Then set the dimension priority coding rule, according to the importance of each feature dimension to the elevator abnormality identification, such as the amplitude of the vibration main frequency is directly related to the component failure, so the priority is the highest, the root mean square of speed is the second, and the 75th percentile of the load is the lowest, the corresponding priority order is the amplitude of the vibration main frequency, the root mean square of speed, and the 75th percentile of the load.

[0080] Convert the characteristic values of each dimension of the core feature condition factor to coding segments according to the characteristic value segmentation mapping rule. For example, the root mean square of speed of the core feature condition factor in a certain uplink scene is 0.5 m / s, corresponding to coding 2; the amplitude of the vibration main frequency is 0.3 g, corresponding to coding B; the 75th percentile of the load is 90 kg, corresponding to coding X. According to the priority order of the amplitude of the vibration main frequency, the root mean square of speed, and the 75th percentile of the load, the corresponding coding segments B, 2, and X are sequentially spliced to generate the feature condition identifier B2X corresponding to the scene data set. This identifier contains the state information of each core feature, and also reflects the importance level of the feature through priority ordering, providing a structured basis for subsequent association verification between feature condition identifiers.

[0081] Obtain the collection time stamp and data source identifier of the data in each type of scene data set;

[0082] Determine the time sequence association attribute of the data in the scene data set based on the collection time stamp, and determine the interaction dependency relationship between different scene data sets through the data source identifier;

[0083] Determine the direct association relationship between the feature condition identifiers according to the interactive dependency relationship of the scene data groups; determine the indirect verification relationship between the feature condition identifiers after judging the logical dependency of the core feature condition factors in the scene data groups; and obtain the hierarchical verification relationship between the feature condition identifiers based on the direct association relationship and the indirect verification relationship.

[0084] First, the acquisition time stamp and data source identifier of the data in each type of scene data group are obtained. The acquisition time stamp is the acquisition time corresponding to each piece of data, such as the acquisition time stamp of a certain speed data in the uplink speed scene data group being a specific time, minute, and second; the data source identifier is used to distinguish which sensor or system the data comes from, such as the source identifier of the speed data being a speed sensor and the source identifier of the vibration data being a vibration sensor.

[0085] The time sequence association attribute of the data in the scene data group is determined based on the acquisition time stamp, such as the speed data arranged in the uplink speed scene data group according to the time stamp, which can reflect the continuous state change of the elevator from starting to stable uplink to deceleration. The order of these data in the time dimension is the time sequence association attribute. The interactive dependency relationship between different scene data groups can be determined through the data source identifier, such as the source identifier of the door machine opening and closing scene data group being the door machine system and the source identifier of the load scene data group being the load sensor. When the door machine system is in the opening state, the load data usually fluctuates, which constitutes the interactive dependency relationship between the two scene data groups.

[0086] The direct association relationship between the feature condition identifiers is determined according to the interactive dependency relationship of the scene data groups. For example, the feature condition identifier corresponding to the door machine opening and closing scene and the feature condition identifier corresponding to the load scene form a direct association relationship due to the interactive dependency of the door machine and the load. At the same time, the logical dependency of the core feature condition factors in the scene data group is judged, such as the change of the root mean square of the speed in the uplink scene depending on the running state of the drive system, and the state of the drive system corresponding to another core feature condition factor, thereby establishing the indirect verification relationship between the feature condition identifiers.

[0087] The hierarchical verification relationship between the feature condition identifiers is obtained based on the direct association relationship and the indirect verification relationship. The feature condition identifier corresponding to the drive system is at a higher level, the feature condition identifier corresponding to the root mean square of the speed is at a lower level, and the two form an indirect verification hierarchical relationship; and the feature condition identifier corresponding to the door machine opening and closing and the feature condition identifier corresponding to the load are at the same level, forming a direct association hierarchical relationship, which together constitute the hierarchical verification system between the feature condition identifiers, providing an associated logical basis for subsequent anomaly checking.

[0088] The hierarchical feature verification system is obtained according to the hierarchical verification relationship between the feature condition identifiers and the feature condition identifiers corresponding to the scene data groups.

[0089] The hierarchical verification relationship between different feature condition identifiers is determined, and each scene data set corresponds to a unique feature condition identifier. Now these elements need to be combined to build a hierarchical feature verification system. Taking the elevator running scene as an example, suppose there are feature condition identifiers corresponding to the drive system scene, the door machine scene, the speed scene, and the load scene, among which the feature condition identifier corresponding to the drive system is at the top of the system because it is the core system affecting the elevator running; the feature condition identifiers corresponding to the speed scene and the door machine scene are at the middle layer, the identifier of the speed scene has an indirect hierarchical relationship with the identifier of the drive system, and the identifier of the door machine scene has a direct hierarchical relationship with the identifier of the drive system; the feature condition identifier corresponding to the load scene is at the lower layer, and has a direct hierarchical relationship with the identifier of the door machine.

[0090] Arrange these feature condition identifiers according to the determined hierarchical verification relationship, the top layer is the feature condition identifier of the drive system, the middle layer arranges the feature condition identifiers of the speed and the door machine in turn, and the lower layer is the feature condition identifier of the load, and the association path between each identifier is also determined, such as the drive system identifier can verify the speed identifier, and the door machine identifier can associate the load identifier. In this way, a system containing scene feature condition identifiers and having clear hierarchical and verification relationships between each identifier is formed, that is, a hierarchical feature verification system. The hierarchical feature verification system completely covers the key scenes of elevator running, and the feature condition identifiers at each level are associated and verified with each other, and then in real-time monitoring, by comparing the matching degree of the identifier corresponding to the real-time data with the identifier in the system and whether the verification relationship of each level identifier is normal, it is determined whether the elevator running state is abnormal.

[0091] Determine the node condition feature value of the hierarchical feature verification system, specifically including the following steps:

[0092] Determine the influence weight of the node in the system according to the association strength of each node in the hierarchical feature verification system and the upper and lower nodes;

[0093] Extract the elevator running basic state information associated with the feature condition identifier corresponding to the node, and determine the stability degree of the elevator running basic state information;

[0094] Integrate the influence weight and the state stability degree to form the basic parameter of the node condition;

[0095] The basic parameter is verified to obtain a verification basic parameter;

[0096] The verification basic parameter is associated with the state attribute to derive the node condition feature value.

[0097] First, the association strength of each node and the upper and lower nodes in the hierarchical feature verification system is judged. The association strength is usually determined according to the frequency and influence degree of data interaction between nodes, for example, the interaction frequency between the driving system node and the speed node is high, and the driving system directly affects the speed state, so the association strength of the two is high.

[0098] The frequency of data interaction between nodes is counted, and the number of synchronous updates of the corresponding feature data of the upper and lower nodes is counted in the historical running data, for example, the data of the driving system node and the speed node. If the number of synchronous updates per unit time is more, it means that the interaction between the two is more frequent, and the basic score of the association strength is higher. It is judged whether the state change of the upper node will directly cause the state fluctuation of the lower node, or whether the abnormality of the lower node will inversely affect the operation of the upper node. For example, the power change of the driving system node directly leads to the speed value change of the speed node, and this direct causal influence will increase the association strength. If the state change of a node indirectly affects another node through multiple intermediate nodes, the association strength is relatively low. The data interaction frequency is converted into a score between 0 and 0.5, and the more frequent the interaction, the higher the score. The directness of state influence is converted into a score between 0 and 0.5, and the more direct the influence, the higher the score. The sum of the two gives the association strength score of the node and the upper and lower nodes. The closer the score is to 1, the higher the association strength.

[0099] Taking the driving system node and the speed node as an example, if the number of synchronous updates of the two nodes per unit time accounts for 90% of the total collection times, the corresponding interaction frequency score is 0.45. At the same time, the state change of the driving system directly leads to the speed change, and the corresponding influence directness score is 0.5, so the association strength of the two is 0.45+0.5=0.95, which belongs to very high association strength. If the synchronous update times of the door machine node and the load node account for 60%, the corresponding interaction frequency score is 0.3, and the state change of the door machine needs to indirectly affect the load through the passenger, the corresponding influence directness score is 0.2, so the association strength of the two is 0.3+0.2=0.5, which belongs to medium association strength.

[0100] According to the association strength, the influence weight of the node in the system is determined, and the weight is normalized, the formula is: influence weight = association strength of the node / sum of association strengths of all nodes in the system, to ensure that the sum of the influence weights of all nodes is 1. If the hierarchical system is taken as an example, the association strength of the driving system node is 0.4, the association strength of the speed node is 0.3, the association strength of the door machine node is 0.2, and the association strength of the load node is 0.1. The influence weight of the driving system node is 0.4 / (0.4+0.3+0.2+0.1)=0.4.

[0101] Then, the elevator running basic state information associated with the feature status identifier of the node is extracted, such as the stable range of the motor speed associated with the drive system node, the power fluctuation interval and the like, and then the stability degree of the basic state information is judged. The stability degree can be calculated by the fluctuation coefficient of the historical normal running data, and the stability degree = 1-(standard deviation of historical data / mean value of historical data). The value is closer to 1, indicating that the state is more stable. Assuming that the historical mean value of the motor speed of the drive system is 1500 revolutions / minute, and the standard deviation is 30 revolutions / minute, then the stability degree = 1-(30 / 1500) = 0.98.

[0102] The basic parameters of the node status are formed by integrating the influence weight and the stability degree, and the basic parameter = influence weight x stability degree. Taking the drive system node as an example, the basic parameter = 0.4 x 0.98 = 0.392.

[0103] The basic parameter is compared with the standard basic parameter of the node in the historical normal running, and if it is within ±5% of the standard parameter, the basic parameter is directly used as the verification basic parameter; if it is out of range, it is corrected by the formula. The verification basic parameter = basic parameter x (standard basic parameter / current basic parameter), so that the verification value fits the normal state. Assuming that the standard basic parameter of the drive system node is 0.4, and the current basic parameter 0.392 is within ±5%, the verification basic parameter is 0.392.

[0104] The node status feature value is derived by associating the verification basic parameter with the state attribute, and the state association logic of the node and the upper and lower nodes is combined, such as the verification basic parameter of the drive system node needs to be associated with the state of the speed node. If the stability degree of the speed node is also in the normal interval, the node status feature value = verification basic parameter x average stability degree of upper and lower nodes. Assuming that the stability degree of the speed node is 0.95, then the status feature value of the drive system node = 0.392 x 0.95 ≈ 0.372.

[0105] The path tracing information of each feature status identifier in the hierarchical feature verification system is extracted, including the following steps:

[0106] The node attribution of each feature status identifier in the hierarchical feature verification system is determined, and the pre-associated of each feature status identifier in the hierarchical feature verification system is obtained;

[0107] The evolution track of the elevator running state corresponding to the feature status identifier is tracked;

[0108] The indirect association link between each feature status identifier is judged, the conduction path between each feature status identifier is determined according to the indirect association link, and the state influence range of each feature status identifier is defined;

[0109] The node attribution, the front association, the evolution track, the conduction path and the state influence range are integrated to form the path tracing information.

[0110] Firstly, the node attribution of each feature status identifier in the hierarchical feature verification system is determined, that is, it is determined which hierarchical node in the system corresponds to each feature status identifier, and the front association of each feature status identifier is obtained, that is, which feature status identifiers are the directly upstream associated nodes of the identifier. For example, the feature status identifier of the speed scene belongs to the middle layer node, and its front association is the driving system feature status identifier of the top layer; the feature status identifier of the load scene belongs to the lower layer node, and its front association is the door machine feature status identifier of the middle layer.

[0111] Then, the evolution track of the corresponding elevator running state of the feature status identifier is tracked, that is, the change process of the identifier corresponding to the elevator state in the historical operation is recorded. For example, the door machine feature status identifier corresponds to the opening and closing state, and the complete state change time sequence from normal opening and closing to jam and then to normal recovery needs to be recorded, and these time sequence information collectively constitutes the evolution track.

[0112] The indirect association link between each feature status identifier is determined, that is, the association relationship established through intermediate nodes, and then the conduction path is determined according to these links, that is, the transmission order in the case of state abnormality. For example, the abnormality of the driving system feature status identifier will be transmitted to the speed feature status identifier first, and then indirectly affect the door machine identifier through the speed identifier, and the link of the driving system, the speed and the door machine is the indirect association link, and the corresponding transmission order is the conduction path. At the same time, the state influence range of each feature status identifier is defined: for example, the abnormality of the driving system identifier affects the states corresponding to the speed, the door machine and the load identifiers, so its influence range covers these three identifiers; and the abnormality of the load identifier will only affect the load state corresponding to itself, and the influence range is limited to itself.

[0113] The node attribution, the front association, the evolution track, the conduction path and the state influence range are integrated to form the complete path tracing information. Taking the driving system feature status identifier as an example, its path tracing information includes: belonging to the top layer node, no front association (because it is the top layer node), the evolution track is that the rotating speed is stable at 1500 revolutions per minute in the recent three operations, the conduction path is the driving system, the speed, the door machine and the load, and the state influence range covers all nodes of the middle layer and the lower layer. When the elevator appears abnormal, the source node of the abnormality, the transmission path of the abnormality and other nodes affected by the abnormality are quickly located by querying the path tracing information, so as to provide the direction for fault troubleshooting.

[0114] Obtaining elevator real-time scene data, obtaining a real-time multi-dimensional feature set according to the elevator real-time scene data, generating a real-time core feature factor after pre-set feature screening optimization on the real-time multi-dimensional feature set, and obtaining a real-time feature identifier by encoding the real-time core feature factor according to a dynamic encoding rule;

[0115] Firstly, real-time scene data of the elevator is obtained, which is collected by sensors and systems in the process of elevator operation, such as uplink speed data uploaded by a speed sensor in real time, motor vibration data collected by a vibration sensor in real time, and car load data recorded by a load sensor in real time. These data correspond to different real-time scene data groups according to scene classification standards, such as speed data in the uplink process being classified into real-time uplink speed scene data group.

[0116] Time domain features, frequency domain features and statistical features are extracted from these real-time scene data groups to construct a real-time multi-dimensional feature set. Taking the real-time uplink speed scene data group as an example, the time domain features include the peak value, mean value, variance, root mean square, rising edge slope and falling edge slope of the real-time speed; the frequency domain features need to be converted from the time domain data of the real-time speed to the frequency domain by Fourier transform to obtain the main frequency amplitude, main frequency and harmonic amplitude; the statistical features include the cumulative value, frequency and quantile of the real-time speed. These features are integrated to form the real-time multi-dimensional feature set corresponding to the scene.

[0117] The real-time multi-dimensional feature set is screened and optimized according to pre-set features. The screening rule is consistent with the feature optimization logic of historical data, that is, features with high correlation degree and low information redundancy are retained. For example, in the multi-dimensional feature set of the real-time uplink speed scene, if the recognition degree of the speed root mean square is higher than that of the mean value, the mean value feature is removed, and the core features of the speed root mean square, the vibration main frequency amplitude and the 75th quantile of the load are retained to generate the real-time core feature factor.

[0118] The real-time core feature factor is encoded according to the set dynamic encoding rule. Taking the speed root mean square, the vibration main frequency amplitude and the 75th quantile of the load included in the real-time core feature factor as an example, the real-time speed root mean square (such as 0.5 m / s) corresponds to code 2, the real-time vibration main frequency amplitude (such as 0.3 g) corresponds to code B, and the real-time load 75th quantile (such as 90 kg) corresponds to code X according to the feature value segmentation mapping rule; according to the dimension priority encoding rule (vibration main frequency amplitude > speed root mean square > 75th quantile of load), the code fragments B, 2 and X are sequentially spliced to obtain the real-time feature identifier B2X, which can be directly used for subsequent comparison with the historical feature status identifier.

[0119] The real-time feature identifier, the feature status identifier, the path tracing information, the node feature value and the node status feature value are processed to determine whether the elevator has abnormal behavior, which includes the following steps:

[0120] The real-time feature identifier is compared with the feature condition identifier to obtain a first determination result of the real-time operation of the elevator, specifically including the following steps:

[0121] The similarity between the real-time feature identifier and the feature condition identifier is calculated.

[0122] If the similarity is greater than or equal to a preset similarity threshold, it is determined that the first determination result is normal.

[0123] If the similarity is less than the preset similarity threshold, it is determined that the first determination result is abnormal.

[0124] First, the similarity between the real-time feature identifier and the feature condition identifier needs to be calculated. The similarity is based on the matching degree of the encoded segments of the two, similarity = number of identical encoded segments in the real-time feature identifier and the feature condition identifier ÷ total number of encoded segments. Assuming that the feature condition identifier is B2X (composed of three segments of vibration main frequency amplitude encoding B, speed root mean square encoding 2, and load 75th percentile encoding X), and the real-time feature identifier is B3X (the vibration main frequency amplitude encoding is still B, the speed root mean square encoding is changed to 3, and the load 75th percentile encoding is still X), the number of identical encoded segments is 2, and the total number of encoded segments is 3, then similarity = 2 ÷ 3 ≈ 0.67.

[0125] Then, a preset similarity threshold needs to be set. The preset similarity threshold is usually determined according to the matching of the identifier when the elevator is normally running. For example, according to historical data statistics, the similarity of the identifier under normal state is mostly not less than 0.8, so the preset similarity threshold is set to 0.8.

[0126] According to the comparison result of the similarity and the threshold, the first determination result is obtained: if the similarity is greater than or equal to the preset similarity threshold, it means that the real-time identifier and the normal state identifier have a high matching degree, and the first determination result is determined to be normal; if the similarity is less than the preset similarity threshold, it means that there is a significant difference between the real-time identifier and the normal identifier, and the first determination result is determined to be abnormal.

[0127] For example, the similarity 0.67 is less than the preset threshold 0.8, so the first determination result is abnormal; if the real-time feature identifier is B2X, which is completely consistent with the feature condition identifier, the number of identical encoded segments is 3, and the similarity = 3 ÷ 3 = 1, which is greater than the threshold 0.8, so the first determination result is normal.

[0128] Based on the real-time feature identifier and the path tracing information, a real-time hierarchical feature verification system is constructed.

[0129] Firstly, the path tracing information needs to be called to determine the logical relationship of each feature condition identifier, such as node ownership, pre-association, and transmission path. Taking the previous hierarchical system as an example, the path tracing information records that the drive system identifier belongs to the top layer, the pre-association is empty, the speed identifier belongs to the middle layer, the pre-association is the drive system identifier, the door machine identifier belongs to the middle layer, the pre-association is the drive system identifier, and the load identifier belongs to the lower layer, the pre-association is the door machine identifier, and it also clearly defines the transmission paths of the drive system, speed, door machine, and load.

[0130] The real-time feature identifier is mapped to the node structure in the path tracing information, such as the real-time feature identifier containing the encoding of the drive system, the encoding of the speed, the encoding of the door machine, and the encoding of the load. According to the node ownership in the path tracing information, these real-time identifiers are respectively mapped to the top layer, middle layer, and lower layer nodes. Based on the pre-association and transmission path of the path tracing information, the hierarchical relationship between the real-time identifiers is established, such as the real-time drive system identifier as the top layer node, associated with the real-time speed identifier and door machine identifier in the middle layer, and the real-time door machine identifier associated with the real-time load identifier in the lower layer, which replicates the same structure as the historical hierarchical feature verification system.

[0131] By constructing the real-time hierarchical feature verification system, it not only contains the real-time feature identifier corresponding to the current running state of the elevator, but also preserves the same hierarchical association logic as the historical system. For example, when the real-time drive system identifier appears an abnormal code, through the transmission path in the system, it is checked whether the associated real-time speed identifier and door machine identifier also appear corresponding abnormalities, providing a structured verification framework for subsequent comparison of node condition feature values.

[0132] According to the node feature values of the real-time hierarchical feature verification system and the node condition feature values, a second determination result of the real-time running of the elevator is obtained, which includes the following steps:

[0133] If the node condition feature values of the real-time hierarchical feature verification system and the node condition feature values are consistent, it is determined that the second determination result is normal;

[0134] If the node condition feature values of the real-time hierarchical feature verification system and the node condition feature values are inconsistent, it is determined that the second determination result is abnormal;

[0135] If the first determination result and the second determination result are both abnormal, it is determined that the elevator has an abnormal behavior.

[0136] The node condition feature values of each node in the historical hierarchical feature verification system need to be obtained, which are the baseline values calculated based on the normal running data of the elevator. Taking the drive system node as an example, its historical node condition feature value is 0.372, which is a comprehensive value of the influence weight and state stability of the node.

[0137] The node condition characteristic value of each node in the real-time hierarchical feature verification system is obtained according to the same calculation method as the historical system. For example, the influence weight of the real-time driving system node is still 0.4, but the state stability degree is changed to 0.9 due to the fluctuation of the real-time running data, so the real-time node condition characteristic value = 0.4 x 0.9 = 0.36, and the final real-time node condition characteristic value is obtained after adaptive verification.

[0138] The node condition characteristic value of the real-time hierarchical feature verification system is compared with the corresponding node condition characteristic value of the historical system: if the values are consistent, it means that the state of the real-time node is consistent with the normal baseline state, and the second determination result is normal; if the values are inconsistent, it means that the real-time node state deviates from the normal baseline, and the second determination result is abnormal.

[0139] Taking the driving system node as an example, if the historical node condition characteristic value is 0.372 and the real-time node condition characteristic value is 0.36, there is a difference between the two, and the second determination result is abnormal; if the real-time node condition characteristic value is also 0.372, which is consistent with the historical value, the second determination result is normal.

[0140] Only when both results are abnormal is it determined that the elevator has abnormal behavior. For example, the first determination result is abnormal because the similarity between the real-time feature identifier and the historical identifier is insufficient, and the second determination result is abnormal because the real-time node condition characteristic value is inconsistent with the historical value, at which time it is determined that the elevator is abnormal; if either of the results is normal, it is not determined to be abnormal, thereby avoiding misjudgment caused by single determination and improving the accuracy of abnormal identification.

[0141] If both the first determination result and the second determination result are normal, it means that the matching degree of the real-time feature identifier and the historical feature condition identifier meets the requirements, and the node condition characteristic value of the real-time hierarchical feature verification system is also consistent with the normal baseline, at which time it is determined that the elevator is in a normal running state, and no additional intervention is required, only the real-time data needs to be continuously recorded, and the normal real-time monitoring needs to be maintained.

[0142] If only the first determination result is abnormal and the second determination result is normal, it means that there is a certain difference between the real-time feature identifier and the historical identifier, but the node condition characteristic value of the real-time hierarchical system still fits the normal baseline, which is usually a temporary fluctuation of real-time data, such as a slight overloading of the elevator for a short time but soon recovered. At this time, a temporary data tracking mechanism needs to be started to continuously collect real-time data under this scene, and the similarity of the real-time feature identifier is repeatedly calculated within a preset time period (such as 5 minutes): if the subsequent similarity rises above the threshold, it is determined to be a temporary fluctuation and the normal monitoring is restored; if the subsequent similarity continues to be below the threshold, the second determination process needs to be triggered again to verify the node condition characteristic value.

[0143] If only the second determination result is abnormal and the first determination result is normal, it indicates that the real-time feature identifier has a high matching degree with the historical identifier, but the node condition characteristic value of the real-time hierarchical system deviates from the normal benchmark. In this case, the node association logic has a hidden abnormality, for example, the association strength between the driving system and the speed node slightly decreases, but the feature identifier does not reflect it. At this time, the path tracing information needs to be called to check the transmission path and the influence range of the abnormal node, and the real-time running basic state information corresponding to the node is extracted. Compare the fluctuation range of the historical data: if the basic state information is still in the normal interval, it is determined that the node characteristic value is slightly offset, and it is marked as an observed state and continuously monitored; if the basic state information is outside the normal interval, the similarity of the feature identifier needs to be re-verified in combination with the first determination result, and if necessary, manual inspection is triggered.

[0144] An elevator abnormal behavior recognition monitoring system for feature extraction, comprising:

[0145] A classification module: classifying the elevator full-scene running data to obtain a scene data group, constructing a multi-dimensional feature condition set based on the scene data group, and performing feature optimization processing on the multi-dimensional feature condition set to generate core feature condition factors corresponding to the scene data group;

[0146] A first processing module: encoding the core feature condition factors according to a dynamic coding rule to obtain feature condition identifiers corresponding to the scene data group;

[0147] An extraction module: obtaining a hierarchical feature verification system according to the hierarchical verification relationship between the feature condition identifiers and the feature condition identifiers corresponding to the scene data group, determining node condition characteristic values of the hierarchical feature verification system, and extracting path tracing information of each feature condition identifier in the hierarchical feature verification system;

[0148] A second processing module: obtaining real-time scene data of the elevator, obtaining real-time multi-dimensional feature sets according to the real-time scene data of the elevator, performing pre-set feature screening optimization on the real-time multi-dimensional feature sets to generate real-time core feature factors, and encoding the real-time core feature factors according to a dynamic coding rule to obtain real-time feature identifiers;

[0149] An output module: processing the real-time feature identifiers, feature condition identifiers, path tracing information, node characteristic values and node condition characteristic values to determine whether the elevator has abnormal behavior.

[0150] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purposes of the embodiments according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0151] Through the description of the above embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0152] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An elevator abnormal behavior recognition monitoring method of feature extraction, characterized by, The method comprises the following steps: The full-scene operation data of the elevator is classified to obtain a scene data group, and a multi-dimensional feature condition set is constructed based on the scene data group, specifically comprising the following steps: Set a scene classification standard; wherein the scene classification standard comprises an elevator operation state classification and a data acquisition type classification; Divide the full-scene operation data of the elevator into a scene data group according to the scene classification standard; Extract the time domain features, frequency domain features and statistical features of the data in the scene data group; Construct a multi-dimensional feature condition set according to the time domain features, frequency domain features and statistical features; Optimize the multi-dimensional feature condition set to generate a core feature condition factor corresponding to the scene data group; Encode the core feature condition factor according to a dynamic encoding rule to obtain a feature condition identifier corresponding to the scene data group, specifically comprising the following steps: Set a dynamic encoding rule based on the number of dimensions of the core feature condition factor and the feature value range; wherein the dynamic encoding rule includes a feature value segmentation mapping rule and a dimension priority encoding rule; Convert the feature value of each dimension of the core feature condition factor into an encoding segment according to the feature value segmentation mapping rule, and combine the dimension priority encoding rule to sequentially splice the encoding segment to generate a feature condition identifier corresponding to the scene data group; Obtain the acquisition time stamp and data source identifier of the data in each type of scene data group; Determine the time sequence association attribute of the data in the scene data group based on the acquisition time stamp, and determine the interaction dependency relationship between different scene data groups through the data source identifier; Determine the direct association relationship between the feature condition identifiers according to the interaction dependency relationship of the scene data group; after judging the logical dependency of the core feature condition factor in the scene data group, establish the indirect verification relationship between the feature condition identifiers; obtain the hierarchical verification relationship between the feature condition identifiers based on the direct association relationship and the indirect verification relationship; Obtain a hierarchical feature verification system according to the hierarchical verification relationship between the feature condition identifiers and the feature condition identifier corresponding to the scene data group, and determine the node condition feature value of the hierarchical feature verification system, specifically comprising the following steps: Determine the association strength of each node and the upper and lower nodes in the hierarchical feature verification system, and determine the influence weight of the node in the system according to the association strength; Extract the elevator operation basic state information associated with the feature condition identifier of the node, and judge the stability degree of the elevator operation basic state information; Integrate the influence weight and the state stability degree to form the basic parameter of the node condition; Verify the basic parameter to obtain a verified basic parameter; Derive the node condition feature value through the state attribute association of the verified basic parameter; Extract the path tracing information of each feature condition identifier in the hierarchical feature verification system; Obtain real-time scene data of the elevator, obtain a real-time multi-dimensional feature set according to the real-time scene data of the elevator, generate a real-time core feature factor after pre-set feature screening optimization of the real-time multi-dimensional feature set, and encode the real-time core feature factor according to the dynamic encoding rule to obtain a real-time feature identifier; The real-time feature identifier, the feature condition identifier, the path tracking information, the node feature value and the node condition feature value are processed to determine whether the elevator has abnormal behavior, specifically including the following steps: The real-time feature identifier is compared with the feature condition identifier to obtain a first determination result of the real-time running of the elevator, specifically including the following steps: The similarity of the real-time feature identifier and the feature condition identifier is calculated; If the similarity is greater than or equal to a preset similarity threshold, it is determined that the first determination result is normal; If the similarity is less than the preset similarity threshold, it is determined that the first determination result is abnormal; A real-time hierarchical feature verification system is constructed based on the real-time feature identifier and the path tracking information, and a second determination result of the real-time running of the elevator is obtained according to the node feature value and the node condition feature value of the real-time hierarchical feature verification system, specifically including the following steps: If the node condition feature value of the real-time hierarchical feature verification system is consistent with the node condition feature value, it is determined that the second determination result is normal; If the node condition feature value of the real-time hierarchical feature verification system is inconsistent with the node condition feature value, it is determined that the second determination result is abnormal; If the first determination result and the second determination result are both abnormal, it is determined that the elevator has abnormal behavior.

2. The method of claim 1, wherein the method further comprises: The path tracking information of each feature condition identifier in the hierarchical feature verification system is extracted, specifically including the following steps: The node attribution of each feature condition identifier in the hierarchical feature verification system is determined, and the pre-association of each feature condition identifier in the hierarchical feature verification system is obtained; The evolution track of the corresponding elevator running state of the feature condition identifier is tracked; The indirect association link between each feature condition identifier is judged, the conduction path between each feature condition identifier is determined according to the indirect association link, and the state influence range of each feature condition identifier is defined; The node attribution, pre-association, evolution track, conduction path and state influence range are integrated to form the path tracking information.

3. An elevator abnormal behavior recognition monitoring system of feature extraction, applied to the elevator abnormal behavior recognition monitoring method of any one of claims 1 to 2, characterized in that, It includes: Classification module: classify the elevator full-scene running data to obtain scene data group, construct multi-dimensional feature condition set based on scene data group, and perform feature optimization processing on multi-dimensional feature condition set to generate core feature condition factor corresponding to scene data group; First processing module: encode the core feature condition factor according to dynamic coding rule to obtain feature condition identifier corresponding to scene data group; Extraction module: obtain hierarchical feature verification system according to hierarchical verification relationship between feature condition identifiers and feature condition identifier corresponding to scene data group, determine node condition feature value of hierarchical feature verification system, and extract path tracking information of each feature condition identifier in hierarchical feature verification system; Second processing module: obtain real-time scene data of elevator, obtain real-time multi-dimensional feature set according to real-time scene data of elevator, perform preset feature screening optimization on real-time multi-dimensional feature set to generate real-time core feature factor, and encode real-time core feature factor according to dynamic coding rule to obtain real-time feature identifier; Output module: process the real-time feature identifier, the feature condition identifier, the path tracking information, the node feature value and the node condition feature value to determine whether the elevator has abnormal behavior.

Images