Intelligent elevator state detection method

Abstract

The invention discloses an intelligent elevator state detection method. The method comprises the steps of elevator state data acquisition, change index determination, elevator state primary clustering, elevator state primary clustering optimization, elevator state secondary clustering refinement and elevator state detection. The invention belongs to the field of data processing, and particularly relates to an intelligent elevator state detection method.According to the scheme, income and cost are evaluated through change indexes, the elevator state feature processing quality and response speed are dynamically balanced by introducing a time delay attention degree weight, and when the system load is high or the response demand is emergency, the fault detection timeliness is preferentially guaranteed; the elevator state detection accuracy is improved; segmented chaotic mapping and a nonlinear search strategy are introduced, and the overall processing duration of elevator state detection is shortened. Performing secondary clustering optimization on a historical elevator state set in combination with representative vectors in clusters and time continuity, and introducing a time penalty term optimization merging strategy to reduce a false abnormal alarm rate; and the elevator state detection effect is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to an intelligent elevator state detection method. Background Art

[0002] The elevator operation detection method is a technical means to collect and analyze various status data of the elevator during operation, monitor the operation health of the elevator in real time, and perform fault warning or status evaluation based on the analysis results. However, the general elevator status detection method has the problem of comprehensive consideration of computational overhead and recognition accuracy when processing the characteristics of the elevator operation status, which is not conducive to the real-time warning of elevator faults and is prone to miss minor abnormalities and early signs of faults; the general elevator status detection method has the problem of occasional impact signals of elevator collisions and emergency stops causing large deviations in global parameter estimation, and misjudging the instantaneous noise of the elevator status. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent elevator state detection method. In view of the fact that the general elevator state detection method has the problem of comprehensive consideration of computational overhead and recognition accuracy when processing the elevator running state characteristics, which is not conducive to the real-time warning of elevator faults and is easy to miss minor abnormalities and early fault signs, this scheme evaluates the benefits and costs by changing the index, so as to achieve the optimal compromise between the accuracy and speed of fault feature extraction; and introduces the delay attention weight to dynamically balance the elevator state feature processing quality and response speed. When the system load is high or the response demand is urgent, it automatically tends to reduce the amount of calculation and give priority to ensuring the timeliness of fault detection; thereby improving the accuracy of elevator state detection; in view of the fact that the occasional impact signals of elevator collision and emergency stop in the general elevator state detection method cause a large deviation to the global parameter estimation, and misjudge the instantaneous noise of the elevator state, this scheme introduces segmented chaotic mapping and nonlinear search strategy to accelerate the optimization convergence speed and shorten the overall processing time of elevator state detection; combines the representative vector within the cluster with the time continuity to perform secondary clustering optimization of the historical elevator state set, introduces the time penalty term optimization merging strategy, reduces the state jump frequency, and reduces the false abnormal alarm rate; thereby improving the elevator state detection effect.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent elevator state detection method, which comprises the following steps:

[0005] Step S1: Elevator status data acquisition;

[0006] Step S2: determining a change indicator;

[0007] Step S3: initial clustering of elevator states;

[0008] Step S4: initial clustering optimization of elevator status;

[0009] Step S5: secondary clustering refinement of elevator status;

[0010] Step S6: Elevator status detection.

[0011] Furthermore, in step S1, the elevator status data acquisition is to record the sensor signal during the elevator operation according to a fixed time length. Split, additionally construct half-length windows and double long windows , and get the window set S; the window is the data segment obtained by dividing the continuously collected sensor signal into time segments; calculate the feature vector X for all windows; obtain the historical elevator state set through data preprocessing ; is the set of all feature vectors; and the historical elevator state set is labeled with the elevator operation state type, and the elevator operation state type is used as a data label, which is only used for cluster label selection.

[0012] Further, in step S2, the change index is determined based on the historical elevator state set. Run K-means clustering, and increase the number of clusters c from 1 to ; Calculate the variance percentage and ; Percentage of variance It is expressed as: ; where j is the cluster cable; is the L2 norm; is the sample set of the jth cluster; is a sample, i is the sample index; is the center of the jth cluster; is the mean vector of the historical elevator state set; sample is the data of the historical elevator state set; the ratio of the number of adjacent clusters is calculated using the formula: ; ;in, and They are the variance percentage rate and clustering running time ratio from cluster number n to cluster number n+1 respectively; and are the variance percentages of cluster number n and cluster number n+1 respectively; the value range of n is 1 to ; and The clustering running time of cluster number n and cluster number n+1 respectively; the delay attention weight is introduced, and the change index is finally calculated , the formula used is: ; ;in, is the delayed attention weight; is the maximum indicator weight; is the current clustering running time; is the maximum clustering running time.

[0013] Further, in step S3, the initial clustering of the elevator status is completed when the change index converges. That is, the optimal number of clusters of the historical elevator state set; the clustering result when K-means clustering converges on the historical elevator state set based on the optimal number of clusters is used as the initial clustering of the elevator state, and step S5 is executed; if it reaches If the time-varying index does not converge, then execute step S4; if , then the variation index converges, is the convergence judgment threshold.

[0014] Furthermore, in step S4, the initial clustering optimization of the elevator status specifically includes the following steps:

[0015] Step S41: Initialize and optimize individual positions: establish a parameter search space based on the maximum number of clusters, the maximum indicator weight, and the convergence judgment threshold, with a total of three dimensions; define the parameters set based on the individual position to run K-means clustering on the historical elevator state set As the individual fitness value; initialize and optimize the individual position; introduce the piecewise linear chaotic map, which is expressed as: ; Where x(·) is the chaotic sequence value; b is the number of chaotic mappings; p is the segmentation threshold; The initial value of is a random value between 0 and 1;

[0016] Step S42: Sort by fitness value: Sort by fitness value from small to large, and mark the first three individuals as , and , use g to traverse; the rest of the individuals are traversed by o; using a nonlinear decreasing factor a, the formula used is: ; ; Where m is the current optimization number; M is the maximum optimization number; is the random coefficient vector of the optimization individual g; is a random vector; is the Hadamard product;

[0017] Step S43: Update individual position: The formula used is: ; ; And based on the change index, the first three fitness values ​​are updated, expressed as: ;in, is the candidate position of the optimizing individual g for the optimizing individual o; and They are respectively to optimize the positions of individual g and individual o; is the updated position of individual o; and The definitions are the same, they are all random vectors, but they are independent of each other; , and Optimize individual , and For optimizing the candidate position of individual o; It is the updated position of the optimized individual g;

[0018] Step S44: Optimization iteration: set the index threshold. When there is an individual fitness value lower than the index threshold, the optimization iteration ends. The maximum number of clusters, the maximum index weight and the convergence judgment threshold are re-determined based on the individual position, and the K-means clustering is re-run on the historical elevator state set. The optimal number of clusters is determined according to the change index, and the initial clustering result of the elevator state is obtained. Execute step S5; if the maximum number of optimization times is reached, return to execute the initial optimization of the individual position; otherwise, return to execute the fitness sorting.

[0019] Furthermore, in step S5, the elevator status secondary clustering refinement is to use the cluster of the initial clustering of the elevator status as the initial division, arrange them in chronological order to obtain a cluster sequence, the chronological order refers to the timestamps of the samples in each cluster, and select the earliest sample timestamp in the cluster to represent the order of appearance of the cluster; for each cluster, the mean vector is calculated. and center sample ; Evaluate cluster quality and , the formula used is: ; ; Where Y is the feature vector of any sample in cluster C; Select the representative vector , the formula used is: ; Then add the time continuity penalty term to define the optimization distance , the formula used is: ; Among them, I and J are two adjacent clusters; and are the representative vectors of cluster I and cluster J respectively; is the time penalty coefficient; is the timestamp of the last sample assignment in cluster I; is the timestamp of the first sample assignment in cluster J; T is the timestamp of the total data assignment; the adjacent clusters with the smallest optimization distance are iteratively merged until the predetermined number of clusters is merged to obtain the final clustering result of the historical elevator state set.

[0020] Furthermore, in step S6, the elevator state detection is based on the final clustering result of the historical elevator state set, and the label of the cluster center in each cluster is selected as the cluster label; the elevator state data is collected in real time and allocated to the corresponding cluster based on the Euclidean distance, and the cluster label of the cluster where the real-time collected elevator state data is located is used as the elevator state detection result.

[0021] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0022] (1) In view of the problem that general elevator status detection methods have a comprehensive consideration of computational overhead and recognition accuracy when processing elevator operation status characteristics, which is not conducive to real-time early warning of elevator faults and is prone to missing minor abnormalities and early signs of faults, this scheme evaluates the benefits and costs by changing indicators, thereby achieving the optimal compromise between the accuracy and speed of fault feature extraction; and introduces a delay attention weight to dynamically balance the elevator status feature processing quality and response speed. When the system load is high or the response demand is urgent, it automatically tends to reduce the amount of calculation and give priority to ensuring the timeliness of fault detection, thereby improving the accuracy of elevator status detection.

[0023] (2) In view of the problem that the occasional impact signals of elevator collision and emergency stop in general elevator state detection methods cause large deviations in global parameter estimation and misjudge the instantaneous noise of elevator state, this scheme introduces piecewise chaotic mapping and nonlinear search strategy to accelerate the optimization convergence speed and shorten the overall processing time of elevator state detection; it combines the representative vector within the cluster with time continuity to perform secondary clustering optimization of the historical elevator state set, introduces time penalty term optimization merging strategy, reduces the state jump frequency, and reduces the false abnormal alarm rate; thereby improving the elevator state detection effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of a flow chart of an intelligent elevator state detection method provided by the present invention;

[0025] Figure 2 It is a schematic diagram of the process of step S4.

[0026] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0027] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0028] In the description of the present invention, it is necessary to understand that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the referred system or element must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as a limitation on the present invention.

[0029] Example 1, see Figure 1 The present invention provides an intelligent elevator state detection method, which comprises the following steps:

[0030] Step S1: Elevator status data acquisition: collect elevator status data during elevator operation and construct a historical elevator status set;

[0031] Step S2: Determine the change index: Run K-means clustering on the historical elevator state set, and define the change index of the historical elevator state set cluster;

[0032] Step S3: Initial clustering of elevator status: judging whether the change indicator converges according to the change indicator, if the change indicator converges, determining the optimal number of clusters and completing the initial clustering, and executing step S5; if the change indicator does not converge, executing step S4;

[0033] Step S4: initial clustering optimization of elevator status: introduce chaotic mapping and individual position update mechanism to optimize initial clustering, obtain the initial clustering result of elevator status, and execute step S5;

[0034] Step S5: Secondary clustering refinement of elevator status: Based on the initial clustering, the elevator status is clustered again in combination with the representative vector within the cluster and the time continuity penalty;

[0035] Step S6: Elevator status detection: Based on the final clustering result obtained by the secondary refinement clustering of the elevator status, online detection of the elevator operation status is realized.

[0036] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the elevator status data acquisition is to record the sensor signal during the elevator operation according to a fixed time length. Split, additionally construct half-length windows and double long windows , the half-length window captures the short-term events of door opening and closing; the double-length window reflects the overall trend of the entire operation, and the window set S is obtained; the window is the data segment obtained by dividing the continuously collected sensor signal into time segments; the feature vector X is calculated for all windows, and each feature vector includes one or more dimensions of acceleration mean, variance, speed mean and maximum, motor current peak, door opening and closing times, acceleration mutation rate and current root mean square value; the historical elevator state set is obtained through data preprocessing ; is the set of all feature vectors; the historical elevator state set is labeled with the elevator operation state type, and the elevator operation state type is used as a data label, which is only used for cluster label selection; the acceleration mutation rate J is expressed as: ; Current RMS value It is expressed as: ; Where t is the time index in the window; a(·) is the acceleration value at the corresponding moment; is the motor current; the elevator operating status types include normal operation, slight abnormality, moderate abnormality and severe abnormality; in order to avoid misjudgment in the actual labeling process, or due to the complexity of the elevator operating environment, some states are difficult to accurately define. The same sample vector will have different operating states in different environments. For example, the same vector in the descending state, ascending state, and rapid change state will have different operating states, but a single label cannot be used for judgment. Therefore, based on semi-supervised learning, clustering is performed through the characteristics and internal structure of the data itself, and hidden patterns and features in the data are mined to have a more comprehensive understanding of the elevator's operating status; and the parameters are adjusted through a unique optimization method to ensure accuracy while improving processing time.

[0037] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the change index is determined to be a historical elevator state set. Run K-means clustering, and increase the number of clusters c from 1 to ; Calculate the variance percentage and ; Percentage of variance It is expressed as: ; where j is the cluster cable; is the sample set of the jth cluster; is the L2 norm; is a sample, i is the sample index; is the center of the jth cluster; is the mean vector of the historical elevator state set; sample is the data of the historical elevator state set; the ratio of the number of adjacent clusters is calculated using the formula: ; ;in, and They are the variance percentage rate and clustering running time ratio from cluster number n to cluster number n+1; the value range of n is 1 to ; and are the variance percentages of cluster number n and cluster number n+1, respectively; and They are the clustering running time of cluster number n and cluster number n+1 respectively; in order to balance the computing resource occupation and response speed, the delay attention weight is introduced, and the final calculation change index , the formula used is: ; ;in, is the delayed attention weight; is the maximum indicator weight; is the current clustering running time; is the maximum clustering running time; by comparing the clustering quality and computational cost of the historical elevator state set, the best compromise is found to ensure that neither the elevator state is over-segmented nor the elevator state is missed due to too few clusters.

[0038] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, the initial clustering of elevator status is completed when the change index converges. That is, the optimal number of clusters of the historical elevator state set; the clustering result when K-means clustering converges on the historical elevator state set based on the optimal number of clusters is used as the initial clustering of the elevator state, and step S5 is executed; if it reaches If the time-varying index does not converge, then execute step S4; if , then the variation index converges, is the convergence judgment threshold.

[0039] By performing the above operations, in view of the problem that the general elevator status detection method has a comprehensive consideration of computational overhead and recognition accuracy when processing the elevator operation status characteristics, which is not conducive to the real-time warning of elevator faults and is prone to missing minor abnormalities and early fault signs, this solution evaluates the benefits and costs by changing indicators, so as to achieve the optimal compromise between the accuracy and speed of fault feature extraction; and introduces the delay attention weight to dynamically balance the elevator status feature processing quality and response speed. When the system load is high or the response demand is urgent, it automatically tends to reduce the amount of calculation and give priority to ensuring the timeliness of fault detection; thereby improving the accuracy of elevator status detection.

[0040] Example 5, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S4, the initial clustering optimization of elevator status specifically includes the following steps:

[0041] Step S41: Initialize and optimize individual positions: establish a parameter search space based on the maximum number of clusters, the maximum indicator weight, and the convergence judgment threshold, with a total of three dimensions; define the parameters set based on the individual position to run K-means clustering on the historical elevator state set As the individual fitness value; initialize and optimize the individual position; introduce piecewise linear chaotic mapping to speed up the initial clustering optimization of the elevator state; the formula used is: ; Where x(·) is the chaotic sequence value; b is the number of chaotic mappings; p is the segmentation threshold; The initial value of is a random value between 0 and 1; the initialization optimization individual position is expressed as: ;in, is the initial position of the kth dimension of the uth optimized individual; and are the upper and lower limits of the kth dimension respectively;

[0042] Step S42: Sort by fitness value: Sort by fitness value from small to large, and mark the first three individuals as , and , use g to traverse; the rest of the individuals are traversed by o; using a nonlinear decreasing factor a, the formula used is: ; ; Where m is the current optimization number; M is the maximum optimization number; is the random coefficient vector of the optimization individual g; is a random vector, each component is independently and uniformly sampled in [0,1]; is the Hadamard product;

[0043] Step S43: Update individual position: The formula used is: ; ; And based on the change index, the first three fitness values ​​are updated, expressed as: ;in, is the candidate position of the optimizing individual g for the optimizing individual o; and They are respectively to optimize the positions of individual g and individual o; is the updated position of individual o; and The definitions are the same, they are all random vectors, but they are independent of each other; , and Optimize individual , and For optimizing the candidate position of individual o; It is the updated position of the optimized individual g;

[0044] Step S44: Optimization iteration: set the index threshold. When there is an individual fitness value lower than the index threshold, the optimization iteration ends. Based on the individual position, the maximum number of clusters, the maximum index weight and the convergence judgment threshold are re-determined, and the K-means clustering is re-run on the historical elevator state set. The optimal number of clusters is determined according to the change index, and the initial clustering result of the elevator state is obtained. Execute step S5; if the maximum number of optimization times is reached, return to execute the initial optimization of the individual position; otherwise, return to execute the fitness sorting; sudden shocks or signal jitters are inevitable in the operation of the elevator. This optimization strategy can avoid the overall offset caused by a single abnormal sample, and is more stable and reliable.

[0045] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the secondary clustering refinement of the elevator status is to use the cluster of the initial clustering of the elevator status as the initial division, and arrange them in time order to obtain a cluster sequence. The time order refers to the timestamps of the samples in each cluster. The earliest sample timestamp in the cluster is selected to represent the order of appearance of the cluster; the mean vector is calculated for each cluster. and center sample ; Evaluate cluster quality and , the formula used is: ; ; Where Y is the feature vector of any sample in cluster C; Select the representative vector , the formula used is: ; Then add the time continuity penalty term to define the optimization distance , the formula used is: ; Among them, I and J are two adjacent clusters; and are the representative vectors of cluster I and cluster J respectively; is the time penalty coefficient; is the timestamp of the last sample assignment in cluster I; is the timestamp assigned to the first sample in cluster J; T is the timestamp assigned to the total data; iteratively merge and optimize the adjacent clusters with the smallest distance until the number of clusters is 4, and the final clustering result of the historical elevator state set is obtained; in view of the possible over-segmentation of the short-term door switch or the same motion state in the initial clustering, the merger is used to ensure that each elevator state is expressed continuously and completely, reduce the state jump frequency, and reduce the misjudgment of the instantaneous false abnormal state by the operation and maintenance personnel.

[0046] By performing the above operations, in order to solve the problem that the occasional impact signals of elevator collision and emergency stop in general elevator state detection methods cause large deviations in global parameter estimation and misjudge the instantaneous noise of elevator state, this scheme introduces piecewise chaotic mapping and nonlinear search strategy to accelerate the optimization convergence speed and shorten the overall processing time of elevator state detection; combines the representative vector within the cluster with time continuity to perform secondary clustering optimization of the historical elevator state set, introduces time penalty term optimization merging strategy, reduces the state jump frequency, and reduces the false abnormal alarm rate; thereby improving the elevator state detection effect.

[0047] Embodiment 7, see Figure 1 This embodiment is based on the above embodiment. In step S6, the elevator state detection is based on the final clustering result of the historical elevator state set, and the label of the cluster center in each cluster is selected as the cluster label; the elevator state data is collected in real time and allocated to the corresponding cluster based on the Euclidean distance, and the cluster label of the cluster where the real-time collected elevator state data is located is used as the elevator state detection result; at this time, the elevator state data that has been detected is incorporated into the historical elevator state set, and the clustering result is updated at a fixed period.

[0048] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0049] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0050] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent elevator state detection method, characterized in that: The method comprises the following steps: Step S1: Elevator status data acquisition: collect elevator status data during elevator operation and construct a historical elevator status set; Step S2: Determine the change index: Run K-means clustering on the historical elevator state set, and define the change index of the historical elevator state set cluster; Step S3: Initial clustering of elevator status: judging whether the change indicator converges according to the change indicator, if the change indicator converges, determining the optimal number of clusters and completing the initial clustering, and executing step S5; if the change indicator does not converge, executing step S4; Step S4: initial clustering optimization of elevator status: introduce chaotic mapping and individual position update mechanism to optimize initial clustering, obtain the initial clustering result of elevator status, and execute step S5; Step S5: Secondary clustering refinement of elevator status: Based on the initial clustering, the elevator status is clustered again in a refined manner by combining the representative vectors within the cluster and the time continuity penalty; Step S6: Elevator status detection: Based on the final clustering result obtained by the secondary refinement clustering of the elevator status, online detection of the elevator operation status is realized.

2. An intelligent elevator state detection method according to claim 1, characterized in that: In step S2, the change index is determined based on the historical elevator status set. Run K-means clustering, and increase the number of clusters c from 1 to ; Calculate the variance percentage and clustering running time after each cluster addition ; Percentage of variance It is expressed as: ; where j is the cluster cable; is the sample set of the jth cluster; is the L2 norm; is a sample, i is the sample index; is the center of the jth cluster; is the mean vector of the historical elevator state set; sample is the data of the historical elevator state set; the ratio of the number of adjacent clusters is calculated using the formula: ; ;in, and They are the variance percentage rate and clustering running time ratio from cluster number n to cluster number n+1; the value range of n is 1 to ; and are the variance percentages of cluster number n and cluster number n+1, respectively; and The clustering running time of cluster number n and cluster number n+1 respectively; the delay attention weight is introduced, and the change index is finally calculated , the formula used is: ; ;in, is the delayed attention weight; is the maximum indicator weight; is the current clustering execution time; is the maximum clustering execution time.

3. An intelligent elevator state detection method according to claim 2, characterized in that: In step S3, the initial clustering of the elevator status is completed when the change index converges. That is, the optimal number of clusters of the historical elevator state set; the clustering result when K-means clustering converges on the historical elevator state set based on the optimal number of clusters is used as the initial clustering of the elevator state, and step S5 is executed; If reached If the time-varying index does not converge, then execute step S4; if , then the variation index converges, is the convergence judgment threshold.

4. An intelligent elevator state detection method according to claim 3, characterized in that: In step S4, the initial clustering optimization of the elevator status specifically includes the following steps: Step S41: Initialize and optimize individual positions: establish a parameter search space based on the maximum number of clusters, the maximum indicator weight, and the convergence judgment threshold, with a total of three dimensions; define the parameters set based on the individual position to run K-means clustering on the historical elevator state set As the individual fitness value; initialize and optimize the individual position; introduce the piecewise linear chaotic map, which is expressed as: ; Where x(·) is the chaotic sequence value; b is the number of chaotic mappings; p is the segmentation threshold; The initial value of is a random value between 0 and 1; the initialization optimization individual position is expressed as: ;in, is the initial position of the kth dimension of the uth optimized individual; and are the upper and lower limits of the kth dimension respectively; Step S42: Sort by fitness value: Sort by fitness value from small to large, and mark the first three individuals as , and , use g to traverse; the rest of the individuals are traversed by o; using a nonlinear decreasing factor a, the formula used is: ; ; Where m is the current optimization number; M is the maximum optimization number; is the random coefficient vector of the optimization individual g; is a random vector; is the Hadamard product; Step S43: Update individual position: The formula used is: ; ; And based on the change index, the first three fitness values ​​are updated, expressed as: ;in, is the candidate position of the optimizing individual g for the optimizing individual o; and They are respectively to optimize the positions of individual g and individual o; is the updated position of individual o; and The definitions are the same, they are all random vectors, but they are independent of each other; , and Optimize individual , and For optimizing the candidate position of individual o; It is the updated position of the optimized individual g; Step S44: Optimization iteration: set the index threshold. When there is an individual fitness value lower than the index threshold, the optimization iteration ends. The maximum number of clusters, the maximum index weight and the convergence judgment threshold are re-determined based on the individual position, and the K-means clustering is re-run on the historical elevator state set. The optimal number of clusters is determined according to the change index, and the initial clustering result of the elevator state is obtained. Execute step S5; if the maximum number of optimization times is reached, return to execute the initial optimization of the individual position; otherwise, return to execute the fitness sorting.

5. An intelligent elevator state detection method according to claim 4, characterized in that: In step S5, the secondary clustering refinement of the elevator status is to use the cluster of the initial clustering of the elevator status as the initial division, and arrange them in chronological order to obtain a cluster sequence. The chronological order refers to the timestamps of the samples in each cluster, and the earliest sample timestamp in the cluster is selected to represent the order of appearance of the cluster; the mean vector is calculated for each cluster and center sample ; Evaluate cluster quality and , the formula used is: ; ; Where Y is the feature vector of any sample in cluster C; Select the representative vector , the formula used is: ; Then add the time continuity penalty term to define the optimization distance , the formula used is: ; Among them, I and J are two adjacent clusters; and are the representative vectors of cluster I and cluster J respectively; is the time penalty coefficient; is the timestamp of the last sample assignment in cluster I; is the timestamp of the first sample assignment in cluster J; T is the timestamp of the total data assignment; the adjacent clusters with the smallest optimization distance are iteratively merged until the predetermined number of clusters is merged to obtain the final clustering result of the historical elevator state set.

6. An intelligent elevator state detection method according to claim 1, characterized in that: In step S1, the elevator status data acquisition is to record the sensor signal during the elevator operation according to a fixed time length. Split, additionally construct half-length windows and double long windows , and get the window set S; the window is the data segment obtained by dividing the continuously collected sensor signal into time segments; calculate the feature vector X for all windows; obtain the historical elevator state set through data preprocessing ; is the set of all feature vectors X; and the historical elevator state set is labeled with the elevator operation state type, and the elevator operation state type is used as the data label, which is only used for cluster label selection.

7. An intelligent elevator state detection method according to claim 1, characterized in that: In step S6, the elevator state detection is based on the final clustering result of the historical elevator state set, and the label of the cluster center in each cluster is selected as the cluster label; The elevator status data is collected in real time and allocated to the corresponding cluster based on the Euclidean distance. The cluster label of the cluster where the real-time collected elevator status data is located is used as the elevator status detection result.

Images