WO2022105266A1 - 电梯故障预测方法、系统、装置、计算机设备和存储介质 - Google Patents

电梯故障预测方法、系统、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2022105266A1
WO2022105266A1 PCT/CN2021/105895 CN2021105895W WO2022105266A1 WO 2022105266 A1 WO2022105266 A1 WO 2022105266A1 CN 2021105895 W CN2021105895 W CN 2021105895W WO 2022105266 A1 WO2022105266 A1 WO 2022105266A1
Authority
WO
WIPO (PCT)
Prior art keywords
fault
data
sample
target
historical
Prior art date
Application number
PCT/CN2021/105895
Other languages
English (en)
French (fr)
Inventor
江荣钿
李良
庄旭强
李骞
李志武
伍乃骐
Original Assignee
日立楼宇技术（广州）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 日立楼宇技术（广州）有限公司 filed Critical 日立楼宇技术（广州）有限公司
Publication of WO2022105266A1 publication Critical patent/WO2022105266A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/20—Administration of product repair or maintenance
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y20/00—Information sensed or collected by the things
    • G16Y20/20—Information sensed or collected by the things relating to the thing itself
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y40/00—IoT characterised by the purpose of the information processing
    • G16Y40/20—Analytics; Diagnosis
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y40/00—IoT characterised by the purpose of the information processing
    • G16Y40/40—Maintenance of things
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16Y—INFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y40/00—IoT characterised by the purpose of the information processing
    • G16Y40/50—Safety; Security of things, users, data or systems
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02B—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO BUILDINGS, e.g. HOUSING, HOUSE APPLIANCES OR RELATED END-USER APPLICATIONS
    • Y02B50/00—Energy efficient technologies in elevators, escalators and moving walkways, e.g. energy saving or recuperation technologies

Definitions

  • the present application relates to the technical field of elevator fault handling, and in particular, to an elevator fault prediction method, system, device, computer equipment and storage medium.
  • elevators have gradually become necessary equipment in residential buildings, office buildings, shopping malls and other buildings.
  • the safety of elevators directly affects the life safety of passengers.
  • Exclusion is the focus of elevator research.
  • the specific process and threshold of the central server determine whether the elevator is faulty.
  • the main control board can perform fault diagnosis without relying on the central server, due to the limitation of the storage capacity and computing power of edge devices, the main control board cannot cache a large amount of historical data, nor can Very complicated logic analysis, lack of repeated learning and reproduction of a large amount of data, so that the main control board can only run relatively simple judgment logic.
  • the second diagnosis method can be adopted, which transmits data to the server, and can satisfy various logical judgments through offline statistics and analysis, but this method needs to rely on the central server and has low timeliness.
  • a method for predicting an elevator failure comprising:
  • the historical fault data includes fault data of multiple fault types
  • the edge server is used to obtain the status data and fault data of the elevator, determine the current sample feature parameter according to the status data and the fault data, and determine the current sample feature parameter according to the current sample feature parameter With the fault association table, the probability of occurrence of any fault type is determined.
  • the method before inputting the sample data into the fault prediction model, the method further includes:
  • the fault codes of each of the target fault types are respectively used as target fault codes, and the sample data is divided into positive samples and negative samples; the positive samples indicate that the fault codes are samples of the target fault codes, and the Negative samples indicate that the fault code is not the sample of the target fault code;
  • the determining of the target fault and multiple target fault types under the target fault includes:
  • the sample characteristic parameter includes a fault parameter
  • the fault parameter includes a fault code and the number of the fault code
  • the correlation analysis is performed on the sample characteristic parameters and the target fault code, and a fault correlation table is obtained, including:
  • the number of each fault code and the target fault code are correlated and analyzed by the chi-square test method, and the fault corresponding to the target fault code and the target fault code is obtained.
  • the correlation weight of the number of each DTC before the type occurs.
  • the sample characteristic parameter further includes a state parameter, and the state parameter includes an instantaneous index and an index value of the instantaneous index;
  • the correlation analysis is performed on the sample characteristic parameters and the target fault code to obtain a fault correlation table, further comprising:
  • the index value of each of the instantaneous indicators is discretized into at least two categories, and according to the categories obtained by discretization and the index value of each of the instantaneous indicators, the index value of each instantaneous indicator and the target fault are analyzed by the chi-square test method.
  • the correlation analysis is performed on the code to obtain the correlation weight of the index value of each instantaneous index before the occurrence of the target fault code and the fault type corresponding to the target fault code.
  • An elevator fault prediction system the system includes: an edge server and a center server, the edge server communicates with the center server through a network, wherein,
  • the central server is used to acquire historical elevator fault data and historical status data corresponding to the historical fault data as sample data; the historical fault data includes fault data of multiple fault types; input the sample data into fault data In the prediction model, a fault correlation table between the sample characteristic parameters of the sample data and each fault type is obtained; the fault correlation table records the correlation weight of each sample characteristic parameter and the fault type when the fault type occurs; sending the fault association table to the edge server;
  • the edge server is used to obtain the status data and fault data of the elevator, determine the current sample characteristic parameter according to the status data and the fault data, and determine any fault according to the current sample characteristic parameter and the fault association table probability of occurrence of the type.
  • the edge server is further configured to obtain the life index of the elevator operation and the index value of the life index, and according to the index value of the life index, the current sample characteristic parameter and the fault association table, Determine the probability of occurrence of any failure type.
  • An elevator fault prediction device includes:
  • a data acquisition module for acquiring historical elevator fault data and historical status data corresponding to the historical fault data as sample data;
  • the historical fault data includes fault data of multiple fault types;
  • an association table generation module configured to input the sample data into a fault prediction model, and obtain a fault association table between the sample characteristic parameters of the sample data and each fault type; the fault association table records that when the fault type occurs, Correlation weight between each sample feature parameter and the fault type;
  • the association table sending module is used to send the fault association table to the edge server; the edge server is used to obtain the status data and fault data of the elevator, and determine the current sample characteristic parameter according to the status data and the fault data, The probability of occurrence of any fault type is determined according to the current sample characteristic parameter and the fault association table.
  • a computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:
  • the historical fault data includes fault data of multiple fault types
  • the edge server is used to obtain the status data and fault data of the elevator, determine the current sample feature parameter according to the status data and the fault data, and determine the current sample feature parameter according to the current sample feature parameter With the fault association table, the probability of occurrence of any fault type is determined.
  • the historical fault data includes fault data of multiple fault types
  • the edge server is used to obtain the status data and fault data of the elevator, determine the current sample feature parameter according to the status data and the fault data, and determine the current sample feature parameter according to the current sample feature parameter With the fault association table, the probability of occurrence of any fault type is determined.
  • the above-mentioned elevator fault prediction method, system, device, computer equipment and storage medium by carrying out model training in the fault prediction model according to the elevator historical fault data and corresponding historical state data in the central server, the sample characteristic parameters of the sample data and the corresponding historical state data are obtained.
  • the fault correlation table of each fault type is used to record the correlation weight of each sample characteristic parameter and the fault type when each fault type occurs.
  • the central server sends the fault association table to the edge server, so that the edge server obtains the fault data and status data of the elevator, and determines the current sample characteristic parameters according to the fault data and status data, and can directly use the sample characteristic parameters and faults.
  • the association table determines the probability of occurrence of any fault type and realizes the prediction of elevator faults.
  • the storage and logical operation of elevator operation data can be realized in the edge server, fault prediction can be performed, complex on-site calculation and on-site storage of a large amount of data can be realized, and the performance of the elevator main control board is not limited, and there is no need to store the acquired data.
  • the data is returned to the central server, and storage and operation can be realized at the edge nodes close to the elevator. While ensuring the improvement of the computing power of the elevator, the delay is reduced, and it has high timeliness.
  • the data is distributed and stored, and the logic operation runs divergently.
  • the central server only needs to perform model training, which greatly reduces the pressure on the central server, and does not need to purchase a large number of central servers to achieve large-scale edge storage and computing, and does not require large-capacity storage in the elevator.
  • Units and processing chips, programs are directly deployed on the edge cloud provided by the operator, the capacity and performance can be improved, and the program improvement and upgrade are also easier.
  • FIG. 1 is a scene diagram of a traditional elevator fault diagnosis method in one embodiment
  • Fig. 2 is the application scene diagram of the elevator fault prediction method in one embodiment
  • FIG. 3 is a schematic flowchart of an elevator fault prediction method in one embodiment
  • FIG. 4 is a schematic flowchart of steps for generating a fault association table in one embodiment
  • Fig. 5 is the flow chart of sample characteristic parameter correlation analysis in one embodiment
  • FIG. 6 is a schematic diagram of a sample feature generation form in one embodiment
  • FIG. 7 is a schematic diagram of an elevator failure prediction system in one embodiment
  • FIG. 8 is a structural block diagram of an elevator fault prediction device in one embodiment
  • Figure 9 is a diagram of the internal structure of a computer device in one embodiment.
  • the elevator fault prediction method provided by the present application can be applied to the application environment shown in FIG. 2 .
  • the edge server 202 is deployed in the machine room close to the elevator side, and the sample data is input into the fault prediction model in the central server 204 according to the historical elevator fault data and the corresponding historical status data as sample data.
  • a fault correlation table between the sample characteristic parameters of the sample data and each fault type is obtained, and the correlation weight of each sample characteristic parameter and the fault type is recorded when each fault type occurs through the fault correlation table.
  • the fault association table is sent to the edge server 202 by the central server 204, so that the edge server 202 can obtain the fault data and status data of the elevator, and determine the current sample feature parameters according to the fault data and status data.
  • the parameter and fault correlation table determines the probability of any fault type, and realizes the prediction of elevator faults.
  • the storage and logical operation of elevator operation data can be realized in the edge server 202, and fault diagnosis can be realized without returning the acquired data to the center server 204, and the operation logic is sunk to the edge site or the place close to the edge, Reduce communication IO, simplify program upgrades, do not need to face a large number of devices, and only face edge nodes of the cloud, thereby increasing computing power and reducing latency.
  • the edge server 202 communicates with the central server 204 through the network.
  • the edge server 202 and the central server 204 can be implemented by an independent server or a server cluster composed of multiple servers.
  • a method for predicting an elevator fault is provided, which is described by taking the method applied to the central server 204 in FIG. 2 as an example, including the following steps:
  • Step S302 obtain the historical fault data of the elevator and the historical status data corresponding to the historical fault data as sample data; the historical fault data includes fault data of multiple fault types.
  • the historical fault data represents the relevant data of each elevator failure.
  • the historical fault data may include elevator model, fault occurrence time, fault type/fault description information, fault code, fault level, risk level, etc. As shown in Table 1 below, it is an example of historical fault data.
  • the historical status data represents the running status data of the elevator before the fault.
  • the historical status data can be the time from the door opening command to the car door switch disconnection, the time from the door opening command to the door lock disconnection, the average operating current of door opening, and the average door closing time. Working current, door opening time, etc.
  • correlation analysis can be performed based on the elevator model, that is, for each elevator model, corresponding historical fault data and historical status data are obtained respectively, and a corresponding fault correlation table is generated for each elevator model.
  • the sample data threshold can also be set.
  • the sample data threshold can be 500.
  • the various faults under the ladder type are calculated.
  • Correlation results of the type of fault code with other fault or status data If it is lower than the sample data threshold, the number of samples is too small, the results of the correlation analysis are greatly affected by noise, and the accuracy of the analysis results will also be affected. Therefore, the sample data of this elevator model may not be processed.
  • the method further includes: determining the target fault and multiple target fault types under the target fault, and obtaining data from the historical fault data and historical fault data.
  • target historical fault data and target historical status data corresponding to the target fault and target fault type are filtered out as sample data.
  • the target fault represents a fault that needs to be predicted, and in this application, the target fault may include a stop fault and a door machine fault.
  • the target fault After the target fault is selected, multiple fault types corresponding to the target fault are determined, and each fault type has a corresponding fault code. Therefore, the faults related to the target fault can be filtered from the historical fault data according to the fault code.
  • the historical fault data corresponding to the type of fault code is used as the target fault data, and then the target status data corresponding to the target fault data is filtered out from the historical status data, and the target fault data and target status data are used as sample data.
  • door machine faults can be determined by matching fault codes.
  • the stop record For the elevator stop fault, you can use the stop record to find out the historical fault data with the fault level A/B/C within the set time (such as 5 minutes) before and after each fault stop, and select the fault record with OrderNO. , as the target fault for the parking fault prediction.
  • the sample data after obtaining the sample data, it is also necessary to filter the sample data to filter out the sample data that meets the preset conditions. More specifically, the data in the sample data whose state parameters do not meet the parameter threshold requirements can be filtered, for example, the data with a speed greater than 2000 or a floor code value greater than 200 can be filtered out. And/or, filter data in the sample data whose failure time is less than a time threshold, for example, determine whether the failure is resolved within 5 minutes, and if so, filter out the failure. And/or, filter the fault data that has been in the fault state in the sample data, for example, determine whether there is other fault data within half an hour before the fault, the value of the IsMaintained field is 1, and if so, the fault is filtered.
  • the sample data whose difference between the set parameters before and after the preset time period does not meet the difference threshold.
  • a set number of times such as 3000 times
  • the difference between RunTotalTime is greater than the set time (such as 24 ⁇ 3600 seconds)
  • the difference between DoorTimes is greater than the second set number of times (such as 5000 times)
  • the data is filtered out.
  • the data that is not a real fault is filtered out, and the obtained filtered data is valid sample data, so that when the filtered sample data is input into the fault prediction model for correlation analysis, the reliability of the analysis results can be improved. accuracy.
  • Step S304 input the sample data into the fault prediction model, and obtain a fault correlation table between the sample characteristic parameters of the sample data and each fault type; the fault correlation table records the correlation weight of each sample characteristic parameter and the fault type when the fault type occurs.
  • the fault codes of each target fault type under the target fault to be predicted can be used as the target fault code in turn, and the sample data can be divided into faults
  • the code is the positive sample of the target fault code and the fault code is not the negative sample of the target fault code.
  • the fault code is the target fault code; 2.
  • the fault code is not the target fault code, then the distribution of the obtained target fault code is a two-category
  • the method of chi-square test can be used, and the p-value, degree of freedom and test value of the chi-square distribution can be used to comprehensively judge the degree of correlation between each sample characteristic parameter and the target fault code.
  • Step S306 sending the fault association table to the edge server; the edge server is used to obtain the status data and fault data of the elevator, determine the current sample characteristic parameters according to the status data and the fault data, and determine any parameter according to the current sample characteristic parameters and the fault association table.
  • the probability of occurrence of a failure type is used to obtain the status data and fault data of the elevator, determine the current sample characteristic parameters according to the status data and the fault data, and determine any parameter according to the current sample characteristic parameters and the fault association table.
  • machine learning is performed in the central server through the fault prediction model, and after obtaining the fault correlation table between the sample feature parameters and each fault type, the fault correlation table can be sent to the edge server, so that the edge server can obtain it in real time.
  • the current sample characteristic parameters are determined according to the status data and fault data, and the probability of occurrence of each fault type under the current sample characteristic parameters is determined according to the correlation weight of each sample characteristic parameter when each fault type occurs. .
  • the model training is performed on the fault prediction model according to the historical elevator fault data and the corresponding historical state data in the central server, so as to obtain the sample characteristic parameters of the sample data and the fault correlation table of each fault type, through
  • the fault correlation table records the correlation weight of each sample characteristic parameter and the fault type when each fault type occurs.
  • the central server sends the fault association table to the edge server, so that the edge server obtains the fault data and status data of the elevator, and determines the current sample characteristic parameters according to the fault data and status data, and can directly use the sample characteristic parameters and faults.
  • the association table determines the probability of occurrence of any fault type and realizes the prediction of elevator faults.
  • the storage and logical operation of elevator operation data can be realized in the edge server, fault prediction can be performed, complex on-site calculation and on-site storage of a large amount of data can be realized, and the performance of the elevator main control board is not limited, and there is no need to store the acquired data.
  • the data is returned to the central server, and storage and operation can be realized at the edge nodes close to the elevator. While ensuring the improvement of the computing power of the elevator, the delay is reduced, and it has high timeliness.
  • the data is distributed and stored, and the logic operation runs divergently.
  • the central server only needs to perform model training, which greatly reduces the pressure on the central server, and does not need to purchase a large number of central servers to achieve large-scale edge storage and computing, and does not require large-capacity storage in the elevator.
  • Units and processing chips, programs are directly deployed on the edge cloud provided by the operator, the capacity and performance can be improved, and the program improvement and upgrade are also easier.
  • step S304 specifically includes:
  • Step S402 respectively taking the fault codes of each target fault type as the target fault codes, and dividing the sample data into positive samples and negative samples; the positive samples represent the samples of the target fault codes, and the negative samples represent that the fault codes are not. is a sample of the target fault code;
  • Step S404 Determine a plurality of sample features representing the sample data, obtain the sample feature parameters of each sample feature statistically from the positive sample and the negative sample, and perform correlation analysis on the sample feature parameters and target fault codes to obtain a fault correlation table.
  • the sample characteristic parameters include fault parameters and state parameters, and the fault parameters include fault codes and the number of fault codes.
  • the correlation analysis of the sample data includes the correlation analysis of the fault part and the correlation analysis of the status part.
  • the fault part correlation analysis refers to analyzing when a certain fault X occurs, other faults that occur within a certain period of time before it occurs, such as Y, whether it occurs, how many times it occurs, and the correlation with the fault X.
  • the state partial correlation analysis refers to analyzing the statistical results of various states within a certain period of time before the occurrence of a certain fault X, such as the mean/variance of the elevator door opening and closing time 0-5 minutes before the occurrence of the fault X, etc. , the correlation with this fault X.
  • the fault code of the target fault type under the elevator model can be regarded as the target fault code in turn, and recorded as the combination form of "elevator model a-fault code X", and the "positive sample” corresponding to the combination can be generated. and "negative samples”.
  • the positive sample means that the elevator model is equal to a
  • the fault code is the sample of the target fault code X
  • the negative sample is the sample of the elevator model equal to a
  • the fault code is not equal to the target fault code X.
  • the first column of data in the table indicates that the target The number of faults f2 is divided into three levels.
  • the data in the second column of the table represent statistics obtained from the positive samples. Within 30 minutes before the occurrence of the target fault f2, the faults 30 occurred 0 times, 1 time and greater than 1 respectively. The number of occurrences of the next three situations.
  • the third column of data indicates the statistics obtained from the negative samples. Within 30 minutes before the occurrence of the non-target fault, the fault 30 occurred 0 times, 1 time and more than 1 time, respectively. the number of occurrences.
  • the fifth row and the sixth column represent the statistical total value of the corresponding column or row.
  • the fault association table can be expressed as "elevator model-fault code-sample characteristic parameter association table", recording each combination of "elevator model-fault code”, the characteristic parameters of each sample are sorted according to the correlation from high to low , the name and relevance weight of each sample feature parameter. Further, the fault association table can be periodically updated periodically to adapt to the state change of the elevator.
  • FIG. 5 it is a flowchart of the correlation analysis of sample feature parameters.
  • the historical data analysis stage that is, in the central server, by obtaining historical fault data and historical status data as historical samples (ie sample data), the sample data is divided into Correlation analysis is performed on positive samples and negative samples to obtain a fault code-ladder-parameter (ie, sample feature parameter) correlation table, and record the name of each sample feature parameter and the correlation weight with the target fault.
  • the real-time data query stage that is, in the edge server, when predicting whether a certain fault code is likely to occur, the predicted time, trapezoid type, and fault code can be input, and the prediction time of the elevator can be obtained from the sensors or counters installed on the elevator.
  • the sample data is divided into positive samples and negative samples by sequentially taking each fault code as the target fault code, and the sample characteristic parameters of the positive sample and the negative sample are counted respectively, and finally the chi-square test is passed.
  • the method analyzes the correlation between each target fault code and the sample characteristic parameters, and obtains a fault correlation table, so that the edge server can predict the fault according to the fault correlation table.
  • determining the target fault and multiple target fault types under the target fault includes: acquiring the number of each fault code in the historical fault data; filtering out the number of each fault code exceeding a set number of times threshold The fault code corresponding to the fault code is used as the target fault type.
  • numerical statistics can be performed on all fault codes in the sample data to obtain the number of faults corresponding to each fault code, and set a threshold for the number of fault codes, which can be 50 times. If the number of a certain fault code is less than the threshold, the possibility of the correlation analysis result being affected by noise is also very high. This fault code can be ignored, and the number of occurrences exceeding the set number is filtered out from each fault code.
  • the fault code of the threshold value, and the fault type corresponding to the fault code is used as the target fault type.
  • each fault code in the sample data is filtered according to the number of each fault code, and the fault type corresponding to the fault code exceeding the set number of times threshold is used as the target fault type, and the occurrence times less than the set number of times are excluded.
  • the fault code of the threshold value can reduce the influence of noise on the correlation analysis results and improve the accuracy of the correlation analysis results between the sample characteristic parameters and the target fault.
  • step S404 correlation analysis is performed on the sample characteristic parameters and the target fault code to obtain a fault correlation table, which specifically includes:
  • the number of fault codes can be used as a sample characteristic parameter to perform correlation analysis with the target fault. Since the chi-square test method requires that all parameters should be categorical, it is necessary to discretize the sample feature parameters first. Since the number of occurrences of any fault (A, B...X, Y) within 30 minutes before the occurrence of the target fault X is mostly 0, and a few are not 0, the data is highly sparse. The number of failures before all target failures can be divided into three categories by manual division: 0 times, 1 time, and >1 time.
  • the correlation analysis is performed between the number of each fault code in different categories and the target fault code, and the number of each fault code in different categories is obtained.
  • the weight of the correlation between the number and the target fault code so that after obtaining the fault correlation table, the edge server can obtain the number of each fault code in different categories through the fault data obtained in real time, so as to determine the probability of the target fault.
  • the correlation analysis is performed on the sample characteristic parameter and the target fault code to obtain a fault correlation table, which further includes:
  • Determine a plurality of instantaneous indicators perform statistical processing on the historical state data according to the preset time window length, and obtain the indicator value of each instantaneous indicator; discretize the indicator value of each instantaneous indicator into at least two categories, according to the category and the index value of each instantaneous index, through the chi-square test method, the index value of each instantaneous index and the target fault code are correlated and analyzed, and the target fault code and the fault type corresponding to the target fault code are obtained. Relevance weight.
  • the instantaneous index can reflect the status of various components of the elevator in a short period of time before running.
  • the instantaneous index can be the 40D pull-in and release time, the working current when opening and closing the door, and whether the elevator has reported other types of malfunction, etc.
  • the features of each sample can be generated from dynamically changing data.
  • the value of the instantaneous indicator itself and the change ratio in different time periods may be an important basis for judging whether the fault is real. Therefore, the instantaneous indicator data can be represented by multiple time windows. Since the goal of failure prediction is to predict whether the target failure predicted by the elevator will occur in the next maintenance cycle, it can be generated based on the maintenance plan of a certain elevator (maintenance plan date - 3 days) ---> (maintenance plan +14 days), and determine whether there is a target fault within the time window, and the fault code when the fault exists.
  • the index value of the instantaneous index in each time window is obtained, and each instantaneous index in the sample feature can be further discretized. Since most of the instantaneous indicators are continuous values (time, current magnitude, etc.), different from the number of failures in the sample feature parameters, they are all highly sparse integer values. Therefore, the discretization method based on percentile can be used to discretize each instantaneous index.
  • the system will find the value of the statistical indicator in each row, sort it by size, and find 20%, 40% , 60% and 80% quantiles, based on these 4 quantiles, all the "mean value of elevator door opening time from 0 to 5 minutes" are divided into 5 categories, and given the corresponding index of the category to which they belong. Since some instantaneous indicators may have a large number of identical values (such as 0), in this case, there may be multiple quantiles that are actually one value. At this time, the categories with the same value can be automatically merged, that is, the number of categories obtained will be less than 5. This merging is allowed to be done automatically, provided it is ensured that each instantaneous indicator is divided into at least two categories.
  • the cumulative indicators of the elevator within 24 hours before the target failure occurs such as the number of operations, the total number of 15B movements, the elevator model, the installation time, etc.
  • This type of data is generally likely to have a low direct correlation with failure effectiveness when given alone.
  • life indicators such as the pull-in and release time of 40G
  • the mean and variance always being 0, which has no effect on fault filtering.
  • life indicators recorded It is the random state of the elevator during a certain operation, such as the elevator door opening and closing, the number of times the light curtain is reopened, etc., which has a low correlation with whether a fault may occur. Therefore, the life indicators in these two cases can be eliminated, the finally obtained life indicators can be used as the target life indicators, and the mean value of the target life indicators in each time window can be calculated as the sample characteristic parameters.
  • the index value of each instantaneous index is statistically obtained from the historical state data, and the index value of each instantaneous index is discretized into at least two categories, and the index value of each instantaneous index in different
  • the correlation analysis is performed between the index value under the category and the target fault code, and the correlation weight between the index value of each instantaneous index under different categories and the target fault code is obtained, so that the edge server can obtain the fault correlation table in real time.
  • the data statistics are used to obtain the index values of each instantaneous index under different categories, thereby determining the probability of the target failure.
  • steps in the flowcharts of FIGS. 3-5 are shown in sequence according to the arrows, these steps are not necessarily executed in the sequence shown by the arrows. Unless explicitly stated herein, the execution of these steps is not strictly limited to the order, and these steps may be performed in other orders. Moreover, at least a part of the steps in FIGS. 3-5 may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed and completed at the same time, but may be executed at different times. These sub-steps or stages are not necessarily completed at the same time. The order of execution of the steps is not necessarily sequential, but may be performed alternately or alternately with other steps or at least a part of sub-steps or stages of other steps.
  • an elevator failure prediction system including: an edge server and a center server, the edge server and the center server communicate through a network, wherein,
  • the central server is used to obtain the historical fault data of the elevator and the historical status data corresponding to the historical fault data as sample data; the historical fault data includes fault data of multiple fault types; the sample data is input into the fault prediction model to obtain the sample data
  • the fault correlation table between the sample characteristic parameters of the data and each fault type; the fault correlation table records the correlation weight of each sample characteristic parameter and the fault type when the fault type occurs; the fault correlation table is sent to the edge server;
  • the edge server is used to obtain the status data and fault data of the elevator, determine the current sample characteristic parameter according to the status data and the fault data, and determine the probability of occurrence of any fault type according to the current sample characteristic parameter and the fault association table.
  • the system further includes sub-area nodes, and the sub-area nodes communicate with the edge server and the central server respectively through a network.
  • the sub-area node is used to perform data statistics on elevator status data collected by multiple edge servers associated with it, and send the data to the central server.
  • the edge server is further configured to obtain the life index of the elevator operation and the index value of the life index, and determine the occurrence of any fault type according to the index value of the life index, the current sample characteristic parameters and the fault association table. probability.
  • an elevator fault prediction device including: a data acquisition module 802, an association table generation module 804, and an association table sending module 806, wherein:
  • the data acquisition module 802 is used to acquire the historical fault data of the elevator and the historical state data corresponding to the historical fault data, as sample data; the historical fault data includes the fault data of multiple fault types;
  • the association table generation module 804 is used to input the sample data into the fault prediction model, and obtain the sample characteristic parameters of the sample data and the fault association table of each fault type; the fault association table records that when the fault type occurs, each sample characteristic parameter and the fault type The relevance weight of ;
  • the association table sending module 806 is used to send the fault association table to the edge server; the edge server is used to obtain the status data and fault data of the elevator, determine the current sample feature parameters according to the status data and the fault data, and determine the current sample feature parameters according to the current sample feature parameters and fault data. Correlation table to determine the probability of occurrence of any fault type.
  • the above-mentioned data acquisition module 802 is further configured to determine the target fault and multiple target fault types under the target fault, and filter out the target fault and target fault types corresponding to the historical fault data and historical status data.
  • Target historical fault data and target historical status data as sample data;
  • the above-mentioned association table generating module 804 is further configured to use the fault codes of each target fault type as the target fault codes respectively, and divide the sample data into positive samples and negative samples; the positive samples indicate that the fault codes are the samples of the target fault codes, Negative samples represent samples whose fault codes are not the target fault codes; determine multiple sample features that characterize sample data, obtain the sample feature parameters of each sample feature from the positive samples and negative samples, and compare the sample feature parameters with the target fault. Correlation analysis is performed on the code to obtain a fault correlation table.
  • the above-mentioned association table generation module 804 is further configured to obtain the number of each fault code in the historical fault data; filter out the fault code whose occurrence number exceeds the set number of times threshold from each fault code, and select the fault code from the fault code.
  • the corresponding fault type is used as the target fault type.
  • the above-mentioned association table generating module 804 is further configured to divide the number of fault codes into at least two categories; according to the number of each fault code and the categories obtained by the division, the chi-square test method is used to analyze each fault code.
  • the number of codes and the target fault code are correlated and analyzed, and the correlation weight of the number of each fault code before the occurrence of the target fault code and the fault type corresponding to the target fault code is obtained.
  • the above-mentioned association table generating module 804 is further configured to determine a plurality of instantaneous indicators, perform statistical processing on the historical state data according to the preset time window length, and obtain the indicator value of each instantaneous indicator;
  • the index values of the s The correlation weight of the index values of each instantaneous index before the fault type corresponding to the target fault code occurs.
  • the elevator failure prediction device of the present application corresponds to the elevator failure prediction method of the present application, and the technical features and beneficial effects described in the embodiments of the elevator failure prediction method are applicable to the implementation of the elevator failure prediction device.
  • the description in the method embodiment of the present application which will not be repeated here, but is hereby declared.
  • each module in the above-mentioned elevator fault prediction device may be implemented in whole or in part by software, hardware and combinations thereof.
  • the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, and can also be stored in the memory in the computer device in the form of software, so that the processor can call and execute the corresponding operations of the above-mentioned modules.
  • a computer device is provided, and the computer device may be a server, and its internal structure diagram may be as shown in FIG. 9 .
  • the computer device includes a processor, memory, and a network interface connected by a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium, an internal memory.
  • the nonvolatile storage medium stores an operating system, a computer program, and a database.
  • the internal memory provides an environment for the execution of the operating system and computer programs in the non-volatile storage medium.
  • the database of the computer equipment is used to store the data generated during the elevator failure prediction process.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer program when executed by a processor, implements a method of elevator failure prediction.
  • FIG. 9 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer equipment to which the solution of the present application is applied. Include more or fewer components than shown in the figures, or combine certain components, or have a different arrangement of components.
  • a computer device including a memory and a processor, where a computer program is stored in the memory, and the processor implements the steps in the foregoing method embodiments when the processor executes the computer program.
  • a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, implements the steps in the foregoing method embodiments.
  • Non-volatile memory may include read-only memory (Read-Only Memory, ROM), magnetic tape, floppy disk, flash memory, or optical memory, and the like.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • the RAM may be in various forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

Abstract

本申请涉及一种电梯故障预测方法、系统、装置、计算机设备和存储介质。所述方法包括：获取电梯历史故障数据以及与所述历史故障数据对应的历史状态数据，作为样本数据；所述历史故障数据包括多个故障类型的故障数据；将所述样本数据输入故障预测模型中，得到所述样本数据的样本特征参数与各个故障类型的故障关联表；将所述故障关联表发送至边缘服务器；所述边缘服务器，用于获取电梯的状态数据和故障数据，根据所述状态数据与所述故障数据确定当前样本特征参数，根据所述当前样本特征参数与所述故障关联表，确定任一故障类型发生的概率。采用本方法能够在保证电梯端计算能力提升的同时，降低了时延，具有较高的时效性。

Description

电梯故障预测方法、系统、装置、计算机设备和存储介质 技术领域
本申请涉及电梯故障处理技术领域，特别是涉及一种电梯故障预测方法、系统、装置、计算机设备和存储介质。
背景技术
近年来，随着人们生活水平的提高，电梯逐渐成为小区、写字楼、商场等建筑中的必备设备，电梯的安全直接影响乘梯人员的生命安全，因此，及时发现电梯可能发生的故障并提前排除，是电梯研究的重点。传统的电梯故障诊断方式主要有两种，如图1所示，一种为主控板根据电梯运行的实时数据判断故障是否发生，另一种为主控板将运行数据传输到中心服务器，由中心服务器的特定流程和阈值判断电梯是否故障。
然而，第一种诊断方式虽然由主控板便可进行故障诊断，无需依赖于中心服务器，但基于边缘设备的存储能力和计算能力的限制，主控板不能缓存大量的历史数据，也不能进行很复杂的逻辑分析，缺乏大量数据的反复学习和复现，导致主控板只能运行较为简单的判断逻辑。为了实现复杂的计算分析，可采用第二种诊断方式，将数据传输到服务器，通过离线统计和分析，可以满足多种逻辑判断，但该方法需依赖于中心服务器，且时效性较低。
因此，传统的故障诊断方式存在无法兼容诊断的时效性和逻辑判断能力的问题。
发明内容
基于此，有必要针对上述故障诊断方式存在无法兼容诊断的时效性和逻辑判断能力的技术问题，提供一种电梯故障预测方法、系统、装置、计算机设备和存储介质。
一种电梯故障预测方法，所述方法包括：
获取电梯历史故障数据以及与所述历史故障数据对应的历史状态数据，作为样本数据；所述历史故障数据包括多个故障类型的故障数据；
将所述样本数据输入故障预测模型中，得到所述样本数据的样本特征参数与各个故障类型的故障关联表；所述故障关联表记录有所述故障类型发生时，各个样本特征参数与所述故障类型的关联性权重；
将所述故障关联表发送至边缘服务器；所述边缘服务器，用于获取电梯的状态数据和故障数据，根据所述状态数据与所述故障数据确定当前样本特征参数，根据所述当前样本特征参数与所述故障关联表，确 定任一故障类型发生的概率。
在其中一个实施例中，在将所述样本数据输入故障预测模型中之前，还包括：
确定目标故障以及目标故障下的多个目标故障类型，从所述历史故障数据和所述历史状态数据中，筛选出与所述目标故障和所述目标故障类型对应的目标历史故障数据和目标历史状态数据，作为样本数据；
所述将所述样本数据输入故障预测模型中，得到所述样本数据的样本特征参数与各个故障类型的故障关联表，包括：
分别将各个所述目标故障类型的故障码作为目标故障码，将所述样本数据分为正例样本和负例样本；所述正例样本表示故障码为所述目标故障码的样本，所述负例样本表示故障码不为所述目标故障码的样本；
确定多个表征所述样本数据的样本特征，从所述正例样本和所述负例样本中统计得到各个样本特征的样本特征参数，对所述样本特征参数与所述目标故障码进行关联分析，得到所述故障关联表。
在其中一个实施例中，所述确定目标故障以及目标故障下的多个目标故障类型，包括：
获取所述历史故障数据中各个故障码的个数；
从各个故障码中筛选出发生次数超出设定次数阈值的故障码，将该故障码对应的故障类型作为目标故障类型。
在其中一个实施例中，所述样本特征参数包括故障参数，所述故障参数包括故障码和故障码的个数；
所述对所述样本特征参数与所述目标故障码进行关联分析，得到故障关联表，包括：
将所述故障码的个数划分为至少两个类别；
根据各个故障码的个数和划分得到的类别，通过卡方测试方法对各个故障码的个数与所述目标故障码进行关联分析，得到所述目标故障码与所述目标故障码对应的故障类型发生前各个故障码的个数的关联性权重。
在其中一个实施例中，所述样本特征参数还包括状态参数，所述状态参数包括瞬时指标和瞬时指标的指标值；
所述对所述样本特征参数与所述目标故障码进行关联分析，得到故障关联表，还包括：
确定多个瞬时指标，按照预设的时间窗长度，对所述历史状态数据进行统计处理，得到各个瞬时指标的指标值；
将各个所述瞬时指标的指标值离散化为至少两个类别，根据离散化得到的类别和各个所述瞬时指标的 指标值，通过卡方测试方法对各个瞬时指标的指标值与所述目标故障码进行关联分析，得到所述目标故障码与所述目标故障码对应的故障类型发生前各个瞬时指标的指标值的关联性权重。
一种电梯故障预测系统，所述系统包括：边缘服务器和中心服务器，所述边缘服务器与所述中心服务器通过网络进行通信，其中，
所述中心服务器，用于获取电梯历史故障数据和与所述历史故障数据对应的历史状态数据，作为样本数据；所述历史故障数据包括多个故障类型的故障数据；将所述样本数据输入故障预测模型中，得到所述样本数据的样本特征参数与各个故障类型的故障关联表；所述故障关联表记录有所述故障类型发生时，各个样本特征参数与所述故障类型的关联性权重；将所述故障关联表发送至边缘服务器；
所述边缘服务器，用于获取电梯的状态数据和故障数据，根据所述状态数据与所述故障数据确定当前样本特征参数，根据所述当前样本特征参数与所述故障关联表，确定任一故障类型发生的概率。
在其中一个实施例中，所述边缘服务器，还用于获取电梯运行的寿命指标和寿命指标的指标值，根据所述寿命指标的指标值、所述当前样本特征参数和所述故障关联表，确定任一故障类型发生的概率。
一种电梯故障预测装置，所述装置包括：
数据获取模块，用于获取电梯历史故障数据和与所述历史故障数据对应的历史状态数据，作为样本数据；所述历史故障数据包括多个故障类型的故障数据；
关联表生成模块，用于将所述样本数据输入故障预测模型中，得到所述样本数据的样本特征参数与各个故障类型的故障关联表；所述故障关联表记录有所述故障类型发生时，各个样本特征参数与所述故障类型的关联性权重；
关联表发送模块，用于将所述故障关联表发送至边缘服务器；所述边缘服务器，用于获取电梯的状态数据和故障数据，根据所述状态数据与所述故障数据确定当前样本特征参数，根据所述当前样本特征参数与所述故障关联表，确定任一故障类型发生的概率。
一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，所述处理器执行所述计算机程序时实现以下步骤：
获取电梯历史故障数据以及与所述历史故障数据对应的历史状态数据，作为样本数据；所述历史故障数据包括多个故障类型的故障数据；
将所述样本数据输入故障预测模型中，得到所述样本数据的样本特征参数与各个故障类型的故障关联表；所述故障关联表记录有所述故障类型发生时，各个样本特征参数与所述故障类型的关联性权重；
将所述故障关联表发送至边缘服务器；所述边缘服务器，用于获取电梯的状态数据和故障数据，根据所述状态数据与所述故障数据确定当前样本特征参数，根据所述当前样本特征参数与所述故障关联表，确定任一故障类型发生的概率。
一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被处理器执行时实现以下步骤：
获取电梯历史故障数据以及与所述历史故障数据对应的历史状态数据，作为样本数据；所述历史故障数据包括多个故障类型的故障数据；
将所述样本数据输入故障预测模型中，得到所述样本数据的样本特征参数与各个故障类型的故障关联表；所述故障关联表记录有所述故障类型发生时，各个样本特征参数与所述故障类型的关联性权重；
将所述故障关联表发送至边缘服务器；所述边缘服务器，用于获取电梯的状态数据和故障数据，根据所述状态数据与所述故障数据确定当前样本特征参数，根据所述当前样本特征参数与所述故障关联表，确定任一故障类型发生的概率。
上述电梯故障预测方法、系统、装置、计算机设备和存储介质，通过在中心服务器中根据电梯历史故障数据和对应的历史状态数据，对故障预测模型中进行模型训练，得到样本数据的样本特征参数与各个故障类型的故障关联表，通过故障关联表记录各个故障类型发生时，各个样本特征参数与故障类型的关联性权重。由中心服务器将该故障关联表发送至边缘服务器上，使得边缘服务器在获取电梯的故障数据和状态数据，根据该故障数据、状态数据确定当前样本特征参数后，可直接根据该样本特征参数与故障关联表确定任一故障类型发生的概率，实现电梯故障的预测。由此在边缘服务器中便可实现对电梯运行数据的存储和逻辑运算，进行故障预测，实现复杂的现场计算和大量数据的现场存储，不受电梯主控板性能的限制，无需再将获取的数据返回给中心服务器，在接近电梯的边缘节点可以实现存储和运算，在保证电梯端计算能力提升的同时，降低了时延，具有较高的时效性，把数据分布存储，逻辑运算发散运行，中心服务器只需进行模型训练即可，极大地减少了中心服务器的压力，且不需要购置大量的中心服务器即可实现大规模的边缘端存储和计算，也不需要在电梯中配备大容量的存储单元和处理芯片，程序直接在运营商提供的边缘云上部署，容量和性能都能得到提升，而且程序的改进、升级也更容易。
附图说明
图1为一个实施例中传统电梯故障诊断方法的场景图；
图2为一个实施例中电梯故障预测方法的应用场景图；
图3为一个实施例中电梯故障预测方法的流程示意图；
图4为一个实施例中故障关联表生成步骤的流程示意图；
图5为一个实施例中样本特征参数关联分析的流程图；
图6为一个实施例中样本特征生成形式的示意图；
图7为一个实施例中电梯故障预测系统的示意图；
图8为一个实施例中电梯故障预测装置的结构框图；
图9为一个实施例中计算机设备的内部结构图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行进一步详细说明。应当理解，此处描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
本申请提供的电梯故障预测方法，可以应用于如图2所示的应用环境中。通过采用5G边缘云计算技术，在靠近电梯侧的机房部署边缘服务器202，通过在中心服务器204中根据电梯历史故障数据和对应的历史状态数据，作为样本数据，将样本数据输入故障预测模型中进行模型训练，得到样本数据的样本特征参数与各个故障类型的故障关联表，通过故障关联表记录各个故障类型发生时，各个样本特征参数与故障类型的关联性权重。由中心服务器204将该故障关联表发送至边缘服务器202上，使得边缘服务器202在获取电梯的故障数据和状态数据，根据该故障数据、状态数据确定当前样本特征参数后，可直接根据该样本特征参数与故障关联表确定任一故障类型发生的概率，实现电梯故障的预测。由此在边缘服务器202中便可实现对电梯运行数据的存储和逻辑运算，实现故障诊断，无需再将获取的数据返回给中心服务器204，将运算逻辑下沉到边缘现场或接近边缘的地方，减少通信IO，程序升级简化，不需要面对数量庞大的设备，只面向云的边缘节点，从而增加了计算能力，降低了时延。其中，边缘服务器202与中心服务器204通过网络进行通信。边缘服务器202和中心服务器204可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
在一个实施例中，如图3所示，提供了一种电梯故障预测方法，以该方法应用于图2中的中心服务器204为例进行说明，包括以下步骤：
步骤S302，获取电梯历史故障数据以及与历史故障数据对应的历史状态数据，作为样本数据；历史 故障数据包括多个故障类型的故障数据。
其中，历史故障数据表示电梯每次故障时的相关数据，例如，历史故障数据可包括电梯型号、故障发生时间、故障类型/故障描述信息、故障码、故障等级、风险等级等。如下表1所示，为一个历史故障数据的示例说明。
表1
Figure PCTCN2021105895-appb-000001
其中，历史状态数据表示电梯故障前的运行状态数据，例如，历史状态数据可以为电梯开门指令到轿门开关断开的时间、开门指令到门锁断开的时间、开门平均工作电流、关门平均工作电流、开门时间等。
实际应用中，可基于电梯型号进行关联性分析，即针对每一种电梯型号，分别获取对应的历史故障数据和历史状态数据，为每一种电梯型号分别生成对应的故障关联表。由于需要基于电梯型号进行关联性分析，因此还可设置样本数据阈值，样本数据阈值可以为500，当某一电梯型号的所有样本数据超过该样本数据阈值时，才计算该梯型下各种故障类型的故障码与其他故障或状态数据的关联性结果。如果低于该样本数据阈值，则样本数过少，关联性分析的结果受到噪声的影响较大，分析结果的准确性也将收到影响，因此，可不处理这种电梯型号的样本数据。
进一步地，在一个实施例中，在获取电梯历史故障数据以及与历史故障数据对应的历史状态数据后，还包括：确定目标故障以及目标故障下的多个目标故障类型，从历史故障数据和历史状态数据中，筛选出与目标故障和目标故障类型对应的目标历史故障数据和目标历史状态数据，作为样本数据。
其中，目标故障表示需要进行预测的故障，本申请中目标故障可包括停梯故障和门机故障。
具体地，在选定目标故障后，确定与目标故障对应的多个故障类型，每个故障类型具有对应的故障码，因此，可根据故障码从历史故障数据中筛选出与目标故障下的故障类型的故障码对应的历史故障数据，作为目标故障数据，进而从历史状态数据中筛选出与目标故障数据对应的目标状态数据，将该目标故障数据和目标状态数据，作为样本数据。
实际应用中，门机故障可采用故障码匹配的方式进行确定。对于停梯故障，则可以利用停梯记录找出每次故障停梯前后设定时间(如5分钟)内的，故障等级为A/B/C类的历史故障数据，选择有OrderNO 的故障记录，作为停梯故障预测的目标故障。
进一步地，在得到样本数据后，还需要对样本数据进行过滤处理，筛选出符合预设条件的样本数据。更具体地，可以对样本数据中状态参数不符合参数阈值要求的数据进行过滤处理，例如，将速度大于2000或楼层代码数值大于200的数据过滤掉。和/或，对样本数据中故障时间小于时间阈值的数据进行过滤处理，例如，判断故障是否在5分钟内被解除，若是，则将该故障过滤掉。和/或，对样本数据中一直处于故障状态的故障数据进行过滤处理，例如，判断故障前半小时内，是否有其它故障数据的IsMaintained字段数值为1，如果有的话，则该故障被过滤。和/或，对设定参数在预设时间段前后的差值不符合差值阈值的样本数据进行过滤处理，例如，通过对比当天前后几条数据包中的状态数据，将满足RunTimes相差大于第一设定次数(如3000次)、RunTotalTime相差大于设定时间(如24×3600秒)、DoorTimes相差大于第二设定次数(如5000次)中的任一个条件的数据过滤掉。
通过上述过滤处理，将并非真正故障的数据过滤掉，得到的过滤后的数据便为有效样本数据，以便于后续将过滤后的样本数据输入故障预测模型进行相关性分析时，可提高分析结果的准确性。
步骤S304，将样本数据输入故障预测模型中，得到样本数据的样本特征参数与各个故障类型的故障关联表；故障关联表记录有故障类型发生时，各个样本特征参数与故障类型的关联性权重。
具体实现中，在确定待预测的目标故障和目标故障下的多个目标故障类型后，可依次将待预测的目标故障下各个目标故障类型的故障码作为目标故障码，将样本数据分为故障码为该目标故障码的正例样本和故障码非该目标故障码的负例样本。确定多个用于表征样本数据的样本特征，并从正例样本和负例样本中统计得到各个样本特征的特征参数，对该样本特征参数和目标故障码进行关联分析，得到各个故障类型发生时，样本特征参数与该故障类型的关联性权重，记录为故障关联表。
由于在确定一个目标故障码后，所有的样本数据可以分为两种：1、故障码是目标故障码；2、故障码不是目标故障码，则得到的目标故障码的分布状况为一个二类别的分布，为了表征该目标故障码的分布与其它样本特征参数的相关程度，需要选择一个合适的基准值。故可采用卡方测试的方法，利用卡方分布的p值，自由度及测试值来综合进行各个样本特征参数与目标故障码相关性程度的判断。
步骤S306，将故障关联表发送至边缘服务器；边缘服务器，用于获取电梯的状态数据和故障数据，根据状态数据与故障数据确定当前样本特征参数，根据当前样本特征参数与故障关联表，确定任一故障类型发生的概率。
具体实现中，在中心服务器中通过故障预测模型进行机器学习，得到样本特征参数与各个故障类型的故障关联表后，可将该故障关联表发送至边缘服务器，由此，边缘服务器即可实时获取电梯的状态数据 和故障数据后，根据状态数据与故障数据确定当前样本特征参数，根据各个故障类型发生时，各个样本特征参数的关联性权重，确定当前样本特征参数下，各个故障类型发生的概率。
上述电梯故障预测方法中，通过在中心服务器中根据电梯历史故障数据和对应的历史状态数据，对故障预测模型中进行模型训练，得到样本数据的样本特征参数与各个故障类型的故障关联表，通过故障关联表记录各个故障类型发生时，各个样本特征参数与故障类型的关联性权重。由中心服务器将该故障关联表发送至边缘服务器上，使得边缘服务器在获取电梯的故障数据和状态数据，根据该故障数据、状态数据确定当前样本特征参数后，可直接根据该样本特征参数与故障关联表确定任一故障类型发生的概率，实现电梯故障的预测。由此在边缘服务器中便可实现对电梯运行数据的存储和逻辑运算，进行故障预测，实现复杂的现场计算和大量数据的现场存储，不受电梯主控板性能的限制，无需再将获取的数据返回给中心服务器，在接近电梯的边缘节点可以实现存储和运算，在保证电梯端计算能力提升的同时，降低了时延，具有较高的时效性，把数据分布存储，逻辑运算发散运行，中心服务器只需进行模型训练即可，极大地减少了中心服务器的压力，且不需要购置大量的中心服务器即可实现大规模的边缘端存储和计算，也不需要在电梯中配备大容量的存储单元和处理芯片，程序直接在运营商提供的边缘云上部署，容量和性能都能得到提升，而且程序的改进、升级也更容易。
在一个实施例中，如图4所示，上述步骤S304具体包括：
步骤S402，分别将各个目标故障类型的故障码作为目标故障码，将样本数据分为正例样本和负例样本；正例样本表示故障码为目标故障码的样本，负例样本表示故障码不为目标故障码的样本；
步骤S404，确定多个表征样本数据的样本特征，从正例样本和负例样本中统计得到各个样本特征的样本特征参数，对样本特征参数与目标故障码进行关联分析，得到故障关联表。
其中，样本特征参数包括故障参数和状态参数，故障参数包括故障码和故障码的个数。
具体实现中，对将样本数据的关联分析包括故障部分关联分析和状态部分关联分析。其中，故障部分关联分析指的是分析当某种故障X发生时，其发生前一定时间段内发生的其它故障，例如Y，是否发生，发生多少次，与该故障X的相关性。状态部分关联分析指的是分析当某种故障X发生时，其发生前一定时间段内的各种状态的统计结果，例如故障X发生之前0-5分钟的电梯开关门时间的均值/方差等，与该故障X的相关性。因此可首先基于电梯型号，依次将该电梯型号下的目标故障类型的故障码作为目标故障码，记为“电梯型号a-故障码X”组合形式，生成该组合所对应的“正例样本”和“负例样本”。则正例样本表示电梯型号等于a，故障码为目标故障码X的样本，而负例样本为电梯型号等于a，故障码不等于目标 故障码X的样本。之后，统计各个样本特征的样本特征参数，通过卡方测试方法对样本特征参数与目标故障码进行关联性分析，得到故障关联表。
例如，以对故障次数与目标故障码进行关联性分析为例，如下表2所示，通过卡方测试确定故障Y与目标故障X发生的关联性的示例，表中第一列数据表示将目标故障f2的个数划分的三个等级，表中第二列的数据分别表示从正例样本中统计得到的，目标故障f2发生前30分钟内，故障30分别出现0次、1次和大于1次三种情况出现的次数，第三列数据中分别表示从负例样本中统计得到的，非目标故障发生前30分钟内，故障30分别发生了0次、1次和大于1次三种情况出现的次数。第五行和第六列则表示对应列或行的统计总数值。
表2
Figure PCTCN2021105895-appb-000002
该表格中，按照故障是否目标故障(f2),以及故障前30分中内出现过故障码为“30”的故障的次数进行了联合统计。其中故障码“30”出现的次数划分为3档，及0次，1次和大于等于1次。首先做出待检验假设：“故障30在前30分钟内出现的次数，与故障是否为目标故障f2不相关”。在此假设基础上，两者应该是统计独立的，即目标故障/非目标故障的比例在0，1，>1三个档位上应该基本一致，都等于33465/768964约0.0435的比例。接下来计算实际统计结果与假设的不一致程度：
对于“0-目标故障”的格内，按照统计独立的假设，其数值应为801535*(33465/803149)＝33397条。其与实际数据的偏差计算为：(33423-33397) 2/33397＝0.02；利用相同方法计算”0-非故障目标”的偏差为0.0008；“1-目标故障”的偏差为3.857；“1-非目标故障”的偏差为0.1155；“>1-目标故障”的偏差为6.919；“1-非目标故障”的偏差为0.302，因此总偏差为10.158。由于故障30在前30分钟内个数共分3类，而目标故障/非目标故障共分两类，因此自由度为(3-1)*(2-1)＝2。查找自由度为2时的卡方分布，可以得到10.158对应的p值为0.006。这意味着，当统计独立的假设成立时，只有0.006的几率出现该统计结果，因此可以拒绝该假设，而判定故障30和f2的分布具有相关性，并可根据表2中的统计结果计算故障30与目标故障f2的关联性权重。类似地，可通过目标故障发生前的状态参数的统计结果，获取状态参数与目标故 障的关联性权重，依次类推，可得到各个样本特征参数与各个目标故障的故障关联表。
其中，故障关联表可表示为“电梯型号-故障码-样本特征参数关联表”，记录每一种“电梯型号-故障码”的组合条件下，各个样本特征参数按照关联性从高到低排序，各样本特征参数的名称和关联性权重。进一步地，该故障关联表可定时进行周期性更新，以适应电梯的状态变化。
参考图5，为样本特征参数关联分析的流程图，在历史数据分析阶段，即在中心服务器中，通过获取历史故障数据和历史状态数据，作为历史样本(即样本数据)，将样本数据划分为正例样本和负例样本进行关联分析，得到故障码-梯型-参数(即样本特征参数)关联表，记录各个样本特征参数的名称和与目标故障的关联性权重。在实时数据查询阶段，即在边缘服务器中，在对某一故障码是否可能发生进行预测时，可输入预测时间、梯型、故障码，从电梯上安装的传感器或计数器等获取电梯在该预测时间前一段时间的状态数据和故障数据，从该状态数据和故障数据，统计出当前的样本特征参数，根据当前的样本特征参数和故障码-梯型-参数(即样本特征参数)关联表，确定所预测的故障码发生的概率。
本实施例中，通过依次将各个故障码作为目标故障码，来将样本数据分为正例样本和负例样本，并分别统计正例样本和负例样本的样本特征参数，最后通过卡方测试方法对各个目标故障码与样本特征参数的关联性进行分析，得到故障关联表，以便于边缘服务器可根据该故障关联表进行故障预测。
在一个实施例中，确定目标故障以及目标故障下的多个目标故障类型，包括：获取所述历史故障数据中各个故障码的个数；从各个故障码中筛选出个数超出设定次数阈值的故障码，将该故障码对应的故障类型作为目标故障类型。
具体实现中，可针对每种电梯型号，对其样本数据中的所有故障码进行数值统计，得到各个故障码对应故障的发生次数，并设置故障码发生次数的阈值，该阈值可以为50次。如果某一故障码的个数少于该阈值，则关联性分析结果受到噪声影响的可能性也会非常大，可不处理这种故障码，而从各故障码中筛选出发生次数超过设定次数阈值的故障码，将该故障码对应的故障类型作为目标故障类型。
本实施例中，通过各个故障码的个数对样本数据中的各个故障码进行过筛选，将超出设定次数阈值的故障码对应的故障类型作为目标故障类型，而剔除发生次数小于设定次数阈值的故障码，以减少噪声对关联性分析结果的影响，提高样本特征参数与目标故障关联性分析结果的准确性。
在一个实施例中，上述步骤S404中，对样本特征参数与目标故障码进行关联分析，得到故障关联表，具体包括：
将故障码的个数划分为至少两个类别；根据各个故障码的个数和划分得到的类别，通过卡方测试方法对各个故障码的个数与目标故障码进行关联分析，得到目标故障码与目标故障码对应的故障类型发生前各个故障码的个数的关联性权重。
具体实现中，在确定目标故障类型和对应的故障码后，可将故障码的个数作为样本特征参数，与目标故障进行关联性分析。由于卡方测试方法要求所有参数应该为类别性，因此，需要先对样本特征参数进行离散化处理。由于目标故障X发生前30分钟内，任意一种故障(A、B……X、Y)的发生次数大都为0，少部分不为0，因此该数据是高度稀疏的。可采用人工划分的方式将所有的目标故障发生前的故障次数划分为三个类别：0次，1次，>1次。完成类别划分后，即可利用卡方测试方法计算目标故障和目标故障发生前30分钟内各种故障种类的次数的相关性。由于所有的故障均划分为三个类，因此计算结果的自由度固定为(3-1)*(2-1)＝2，不同的故障种类之间的结果可以通过输出的pvalue值进行相关性强弱的比较。
由于故障码的种类数众多，而绝大部分故障码出现的次数非常有限，对于这些故障码，计算他们和目标故障的相关性一方面没有必要，另一方面计算出的相关性结果准确度也存疑。因此在故障部分关联分析的时候可只选取出现频度前n位(例如，前30位)的故障码来进行计算，以提高关联性分析的效率。
本实施例中，通过将故障码的个数离散化为至少两个类别，对各个故障码在不同类别下的个数与目标故障码进行关联性分析，得到各个故障码在不同类别下的个数与目标故障码的关联性权重，以便于边缘服务器在获取故障关联表后，可通过实时获取的故障数据统计得到各个故障码在不同类别下的个数，由此确定目标故障发生的概率。
在一个实施例中，上述步骤S404中，对样本特征参数与目标故障码进行关联分析，得到故障关联表，还包括：
确定多个瞬时指标，按照预设的时间窗长度，对历史状态数据进行统计处理，得到各个瞬时指标的指标值；将各个瞬时指标的指标值离散化为至少两个类别，根据离散化得到的类别和各个瞬时指标的指标值，通过卡方测试方法对各个瞬时指标的指标值与目标故障码进行关联分析，得到目标故障码与目标故障码对应的故障类型发生前各个瞬时指标的指标值的关联性权重。
其中，瞬时指标能够反映电梯在运行之前一个短时间内各种部件的状态，例如，瞬时指标可以为40D吸合、释放时间，开关门时的工作电流大小，电梯在之前是否上报过其他种类的故障等。
具体实现中，由于进行样本特征参数关联分析的目的是为故障时状态汇总提供基于相关性的排序，因此可以动态变化的数据来生成每个样本的特征。其中，瞬时指标自身的取值，以及不同时间段内的变化 比例，都可能是用来判断故障是否真实的重要依据，因此，瞬时指标数据可采用多时间窗来进行特征的表示。由于故障预测的目标是预测在下个保养周期内是否会出现电梯所预测的目标故障，因此可基于某台电梯的保养计划，生成(保养计划日期-3天)--->(保养计划+14天)的时间窗，并判断该时间窗内是否存在目标故障，以及故障存在时的故障码。
例如，取故障发生前2个小时的瞬时指标数据，以5分钟为时间窗长度，收集每个5分钟窗内电梯门机运行数据指标的平均值、方差、最大值和最小值等。如果在时间窗内不存在瞬时指标数据则统一置0。
由此，得到瞬时指标在各个时间窗内的指标值，进一步可对样本特征中的各个瞬时指标进行离散化。由于瞬时指标中大部分都是连续值(时间，电流大小等)，不同于样本特征参数中的故障次数，均是高度稀疏的整数值。因此，可采用基于百分位的离散化方法，对各瞬时指标进行离散化。例如，若要将0-5分钟电梯开门时间的均值进行离散化到5个类别内，则系统会找到每一行中该统计指标的数值，并将其按照大小排序后，找到20％、40％、60％和80％的分位点，基于这4个分位点将所有的“0-5分钟电梯开门时间的均值”分成5个类别，并给予其所属类别对应的index。由于某些瞬时指标可能存在大量的相同数值(例如0)，这种情况下，有可能存在多个分位点实际上是一个数值。此时可将数值相同的类别自动进行合并，也就是得到的类别数会少于5个。在确保每个瞬时指标被分成至少两个类别的前提下，允许这种合并自动完成。
另外，除了瞬时指标对目标故障具有影响外，还存在数据变化比较慢或静态的寿命指标数据对目标故障的影响。例如，目标故障发生前24小时内电梯的各项累计指标，例如运行次数，15B总动作次数，电梯型号，安装时长等。这类数据通常来说单独给出时与故障有效性的直接相关性可能较低。但是可能能够为动态变化的数据提供额外的补充信息，特别是在一些树算法中可能提供有效的信息增益。因此，最终生成的样本特征生成形式可表示为如图6所示，图中的触发型状态数据表示瞬时指标数据，累积型状态数据可表示寿命指标数据。
实际应用中，经过统计后，发现部分寿命指标(如40G的吸合释放时间)从未发生过变化，导致均值、方差等始终为0，对于故障过滤没有任何作用，此外还有一些寿命指标记录的是电梯某次运行时的随机状态，例如电梯开关门楼层，光幕重新打开次数等，与是否可能发生故障关联性较低。因此，可将这两种情况下的寿命指标剔除，最后得到的寿命指标作为目标寿命指标，计算目标寿命指标在每个时间窗内的均值等作为样本特征参数。
本实施例中，通过按照预设的时间窗长度，从历史状态数据中统计得到各个瞬时指标的指标值，并将各个瞬时指标的指标值离散化为至少两个类别，对各个瞬时指标在不同类别下的指标值与目标故障码进 行关联性分析，得到各个瞬时指标在不同类别下的指标值与目标故障码的关联性权重，以便于边缘服务器在获取故障关联表后，可通过实时获取的数据统计得到各个瞬时指标在不同类别下的指标值，由此确定目标故障发生的概率。
应该理解的是，虽然图3-5的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，这些步骤可以以其它的顺序执行。而且，图3-5中的至少一部分步骤可以包括多个子步骤或者多个阶段，这些子步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，这些子步骤或者阶段的执行顺序也不必然是依次进行，而是可以与其它步骤或者其它步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
在一个实施例中，提供了一种电梯故障预测系统，包括：边缘服务器和中心服务器，边缘服务器与中心服务器通过网络进行通信，其中，
所述中心服务器，用于获取电梯历史故障数据和与历史故障数据对应的历史状态数据，作为样本数据；历史故障数据包括多个故障类型的故障数据；将样本数据输入故障预测模型中，得到样本数据的样本特征参数与各个故障类型的故障关联表；故障关联表记录有故障类型发生时，各个样本特征参数与故障类型的关联性权重；将故障关联表发送至边缘服务器；
所述边缘服务器，用于获取电梯的状态数据和故障数据，根据状态数据与故障数据确定当前样本特征参数，根据当前样本特征参数与故障关联表，确定任一故障类型发生的概率。
在一个实施例中，如图7所示，所述系统还包括分区域节点，所述分区域节点分别与边缘服务器和中心服务器通过网络进行通信。所述分区域节点，用于对与其关联的多个边缘服务器采集的电梯状态数据进行数据统计，发送给中心服务器。
在一个实施例中，所述边缘服务器，还用于获取电梯运行的寿命指标和寿命指标的指标值，根据寿命指标的指标值、当前样本特征参数和故障关联表，确定任一故障类型发生的概率。
在一个实施例中，如图8所示，提供了一种电梯故障预测装置，包括：数据获取模块802、关联表生成模块804和关联表发送模块806，其中：
数据获取模块802，用于获取电梯历史故障数据和与历史故障数据对应的历史状态数据，作为样本数 据；历史故障数据包括多个故障类型的故障数据；
关联表生成模块804，用于将样本数据输入故障预测模型中，得到样本数据的样本特征参数与各个故障类型的故障关联表；故障关联表记录有故障类型发生时，各个样本特征参数与故障类型的关联性权重；
关联表发送模块806，用于将故障关联表发送至边缘服务器；边缘服务器，用于获取电梯的状态数据和故障数据，根据状态数据与故障数据确定当前样本特征参数，根据当前样本特征参数与故障关联表，确定任一故障类型发生的概率。
在一个实施例中，上述数据获取模块802，还用于确定目标故障以及目标故障下的多个目标故障类型，从历史故障数据和历史状态数据中，筛选出与目标故障和目标故障类型对应的目标历史故障数据和目标历史状态数据，作为样本数据；
上述关联表生成模块804，还用于分别将各个目标故障类型的故障码作为目标故障码，将样本数据分为正例样本和负例样本；正例样本表示故障码为目标故障码的样本，负例样本表示故障码不为目标故障码的样本；确定多个表征样本数据的样本特征，从正例样本和负例样本中统计得到各个样本特征的样本特征参数，对样本特征参数与目标故障码进行关联分析，得到故障关联表。
在一个实施例中，上述关联表生成模块804，还用于获取历史故障数据中各个故障码的个数；从各个故障码中筛选出发生次数超出设定次数阈值的故障码，将该故障码对应的故障类型作为目标故障类型。
在一个实施例中，上述关联表生成模块804，还用于将故障码的个数划分为至少两个类别；根据各个故障码的个数和划分得到的类别，通过卡方测试方法对各个故障码的个数与目标故障码进行关联分析，得到目标故障码与目标故障码对应的故障类型发生前各个故障码的个数的关联性权重。
在一个实施例中，上述关联表生成模块804，还用于确定多个瞬时指标，按照预设的时间窗长度，对历史状态数据进行统计处理，得到各个瞬时指标的指标值；将各个瞬时指标的指标值离散化为至少两个类别，根据离散化得到的类别和各个瞬时指标的指标值，通过卡方测试方法对各个瞬时指标的指标值与目标故障码进行关联分析，得到目标故障码与目标故障码对应的故障类型发生前各个瞬时指标的指标值的关联性权重。
需要说明的是，本申请的电梯故障预测装置与本申请的电梯故障预测方法一一对应，在上述电梯故障预测方法的实施例阐述的技术特征及其有益效果均适用于电梯故障预测装置的实施例中，具体内容可参见本申请方法实施例中的叙述，此处不再赘述，特此声明。
此外，上述电梯故障预测装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储 器中，以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中，提供了一种计算机设备，该计算机设备可以是服务器，其内部结构图可以如图9所示。该计算机设备包括通过系统总线连接的处理器、存储器和网络接口。其中，该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机程序和数据库。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该计算机设备的数据库用于存储电梯故障预测过程中产生的数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机程序被处理器执行时以实现一种电梯故障预测方法。
本领域技术人员可以理解，图9中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定，具体的计算机设备可以包括比图中所示更多或更少的部件，或者组合某些部件，或者具有不同的部件布置。
在一个实施例中，还提供了一种计算机设备，包括存储器和处理器，存储器中存储有计算机程序，该处理器执行计算机程序时实现上述各方法实施例中的步骤。
在一个实施例中，提供了一种计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现上述各方法实施例中的步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的计算机程序可存储于一非易失性计算机可读取存储介质中，该计算机程序在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和易失性存储器中的至少一种。非易失性存储器可包括只读存储器(Read-Only Memory，ROM)、磁带、软盘、闪存或光存储器等。易失性存储器可包括随机存取存储器(Random Access Memory，RAM)或外部高速缓冲存储器。作为说明而非局限，RAM可以是多种形式，比如静态随机存取存储器(Static Random Access Memory，SRAM)或动态随机存取存储器(Dynamic Random Access Memory，DRAM)等。
以上实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对发明专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (10)

  1. 一种电梯故障预测方法，其特征在于，所述方法包括：
    获取电梯历史故障数据以及与所述历史故障数据对应的历史状态数据，作为样本数据；所述历史故障数据包括多个故障类型的故障数据；
    将所述样本数据输入故障预测模型中，得到所述样本数据的样本特征参数与各个故障类型的故障关联表；所述故障关联表记录有所述故障类型发生时，各个样本特征参数与所述故障类型的关联性权重；
    将所述故障关联表发送至边缘服务器；所述边缘服务器，用于获取电梯的状态数据和故障数据，根据所述状态数据与所述故障数据确定当前样本特征参数，根据所述当前样本特征参数与所述故障关联表，确定任一故障类型发生的概率。
  2. 根据权利要求1所述的方法，其特征在于，在将所述样本数据输入故障预测模型中之前，还包括：
    确定目标故障以及目标故障下的多个目标故障类型，从所述历史故障数据和所述历史状态数据中，筛选出与所述目标故障和所述目标故障类型对应的目标历史故障数据和目标历史状态数据，作为样本数据；
    所述将所述样本数据输入故障预测模型中，得到所述样本数据的样本特征参数与各个故障类型的故障关联表，包括：
    分别将各个所述目标故障类型的故障码作为目标故障码，将所述样本数据分为正例样本和负例样本；所述正例样本表示故障码为所述目标故障码的样本，所述负例样本表示故障码不为所述目标故障码的样本；
    确定多个表征所述样本数据的样本特征，从所述正例样本和所述负例样本中统计得到各个样本特征的样本特征参数，对所述样本特征参数与所述目标故障码进行关联分析，得到所述故障关联表。
  3. 根据权利要求2所述的方法，其特征在于，所述确定目标故障以及目标故障下的多个目标故障类型，包括：
    获取所述历史故障数据中各个故障码的个数；
    从各个故障码中筛选出发生次数超出设定次数阈值的故障码，将该故障码对应的故障类型作为目标故障类型。
  4. 根据权利要求2所述的方法，其特征在于，所述样本特征参数包括故障参数，所述故障参数包括故障码和故障码的个数；
    所述对所述样本特征参数与所述目标故障码进行关联分析，得到故障关联表，包括：
    将所述故障码的个数划分为至少两个类别；
    根据各个故障码的个数和划分得到的类别，通过卡方测试方法对各个故障码的个数与所述目标故障码进行关联分析，得到所述目标故障码与所述目标故障码对应的故障类型发生前各个故障码的个数的关联性权重。
  5. 根据权利要求2所述的方法，其特征在于，所述样本特征参数还包括状态参数，所述状态参数包括瞬时指标和瞬时指标的指标值；
    所述对所述样本特征参数与所述目标故障码进行关联分析，得到故障关联表，还包括：
    确定多个瞬时指标，按照预设的时间窗长度，对所述历史状态数据进行统计处理，得到各个瞬时指标的指标值；
    将各个所述瞬时指标的指标值离散化为至少两个类别，根据离散化得到的类别和各个所述瞬时指标的指标值，通过卡方测试方法对各个瞬时指标的指标值与所述目标故障码进行关联分析，得到所述目标故障码与所述目标故障码对应的故障类型发生前各个瞬时指标的指标值的关联性权重。
  6. 一种电梯故障预测系统，其特征在于，所述系统包括：边缘服务器和中心服务器，所述边缘服务器与所述中心服务器通过网络进行通信，其中，
    所述中心服务器，用于获取电梯历史故障数据和与所述历史故障数据对应的历史状态数据，作为样本数据；所述历史故障数据包括多个故障类型的故障数据；将所述样本数据输入故障预测模型中，得到所述样本数据的样本特征参数与各个故障类型的故障关联表；所述故障关联表记录有所述故障类型发生时，各个样本特征参数与所述故障类型的关联性权重；将所述故障关联表发送至边缘服务器；
    所述边缘服务器，用于获取电梯的状态数据和故障数据，根据所述状态数据与所述故障数据确定当前样本特征参数，根据所述当前样本特征参数与所述故障关联表，确定任一故障类型发生的概率。
  7. 根据权利要求6所述的系统，其特征在于，所述边缘服务器，还用于获取电梯运行的寿命指标和寿命指标的指标值，根据所述寿命指标的指标值、所述当前样本特征参数和所述故障关联表，确定任一故障类型发生的概率。
  8. 一种电梯故障预测装置，其特征在于，所述装置包括：
    数据获取模块，用于获取电梯历史故障数据和与所述历史故障数据对应的历史状态数据，作为样本数据；所述历史故障数据包括多个故障类型的故障数据；
    关联表生成模块，用于将所述样本数据输入故障预测模型中，得到所述样本数据的样本特征参数与各 个故障类型的故障关联表；所述故障关联表记录有所述故障类型发生时，各个样本特征参数与所述故障类型的关联性权重；
    关联表发送模块，用于将所述故障关联表发送至边缘服务器；所述边缘服务器，用于获取电梯的状态数据和故障数据，根据所述状态数据与所述故障数据确定当前样本特征参数，根据所述当前样本特征参数与所述故障关联表，确定任一故障类型发生的概率。
  9. 一种计算机设备，包括存储器和处理器，所述存储器存储有计算机程序，其特征在于，所述处理器执行所述计算机程序时实现权利要求1至5中任一项所述方法的步骤。
  10. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于，所述计算机程序被处理器执行时实现权利要求1至5中任一项所述的方法的步骤。
PCT/CN2021/105895 2020-11-17 2021-07-13 电梯故障预测方法、系统、装置、计算机设备和存储介质 WO2022105266A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011284955.6A CN112365066B (zh) 2020-11-17 2020-11-17 电梯故障预测方法、系统、装置、计算机设备和存储介质
CN202011284955.6 2020-11-17

Publications (1)

Publication Number Publication Date
WO2022105266A1 true WO2022105266A1 (zh) 2022-05-27

Family

ID=74515056

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/105895 WO2022105266A1 (zh) 2020-11-17 2021-07-13 电梯故障预测方法、系统、装置、计算机设备和存储介质

Country Status (3)

Country Link
CN (1) CN112365066B (zh)
AU (1) AU2021107643A4 (zh)
WO (1) WO2022105266A1 (zh)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115037599A (zh) * 2022-06-13 2022-09-09 中国电信股份有限公司 一种通信网络故障预警方法、装置、设备及介质
CN115130702A (zh) * 2022-09-02 2022-09-30 山东汇泓纺织科技有限公司 一种基于大数据分析的纺织机故障预测系统
CN115576724A (zh) * 2022-09-19 2023-01-06 成都飞机工业（集团）有限责任公司 Piu子系统的故障隔离方法、装置、设备、介质及产品
CN115659812A (zh) * 2022-10-29 2023-01-31 思维实创(哈尔滨)科技有限公司 基于城轨iscs的电扶梯寿命预测方法、系统、设备及介质
CN115980531A (zh) * 2023-03-16 2023-04-18 江苏大全长江电器股份有限公司 一种特定环境下的gis开关柜质量检测方法及系统
CN117193088A (zh) * 2023-09-22 2023-12-08 珠海臻图信息技术有限公司 一种工业设备监控方法、装置以及服务器
CN117370848A (zh) * 2023-12-08 2024-01-09 深圳市明心数智科技有限公司 设备故障预测方法、装置、计算机设备及存储介质
CN117193088B (zh) * 2023-09-22 2024-04-26 珠海臻图信息技术有限公司 一种工业设备监控方法、装置以及服务器

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP6898280B2 (ja) * 2018-08-31 2021-07-07 ファナック株式会社 知識作成システム
CN112365066B (zh) * 2020-11-17 2023-05-02 日立楼宇技术（广州）有限公司 电梯故障预测方法、系统、装置、计算机设备和存储介质
CN113158362B (zh) * 2021-03-18 2023-11-28 浙江理工大学 一种物理失效和数据驱动融合的电梯剩余寿命预测方法
CN113819575B (zh) * 2021-08-18 2023-03-21 青岛海尔空调器有限总公司 用于空调的控制方法、装置和服务器
CN113651202B (zh) * 2021-08-19 2023-02-17 广州广日电梯工业有限公司 电梯异常的智能识别方法以及智能识别装置
CN115314401B (zh) * 2022-06-30 2024-03-15 中铁第四勘察设计院集团有限公司 接触网状态监测方法、装置、电子设备及存储介质
CN117172746A (zh) * 2023-09-05 2023-12-05 泰州润丽工程科技有限公司 智能解决故障的楼宇管理系统
CN117172431B (zh) * 2023-11-03 2024-01-19 山东锦都食品有限公司 一种食品加工器械设备管理方法及系统
CN117608974A (zh) * 2024-01-22 2024-02-27 金品计算机科技(天津)有限公司 基于人工智能的服务器故障检测方法、装置、设备及介质

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160274551A1 (en) * 2015-03-18 2016-09-22 Accenture Global Services Limited Method and system for predicting equipment failure
US9535774B2 (en) * 2013-09-09 2017-01-03 International Business Machines Corporation Methods, apparatus and system for notification of predictable memory failure
CN107194053A (zh) * 2017-05-16 2017-09-22 歌拉瑞电梯股份有限公司 一种智能化电梯控制系统运行故障预测方法
CN109110608A (zh) * 2018-10-25 2019-01-01 歌拉瑞电梯股份有限公司 一种基于大数据学习的电梯故障预测方法
CN110647539A (zh) * 2019-09-26 2020-01-03 汉纳森（厦门）数据股份有限公司 一种用于车辆故障的预测方法和系统
CN111638989A (zh) * 2020-04-15 2020-09-08 北京三快在线科技有限公司 故障诊断方法、装置、存储介质及设备
CN112365066A (zh) * 2020-11-17 2021-02-12 日立楼宇技术（广州）有限公司 电梯故障预测方法、系统、装置、计算机设备和存储介质

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107886168B (zh) * 2017-11-07 2018-11-09 歌拉瑞电梯股份有限公司 一种使用多层感知器神经网络进行电梯故障识别的方法
CN108693868B (zh) * 2018-05-25 2021-06-11 深圳市轱辘车联数据技术有限公司 故障预测模型训练的方法、车辆故障预测的方法及装置
CN109634828A (zh) * 2018-12-17 2019-04-16 浪潮电子信息产业股份有限公司 故障预测方法、装置、设备及存储介质
CN109607344B (zh) * 2018-12-20 2020-04-14 华北水利水电大学 一种基于神经网络垂直电梯故障预测系统及方法
CN110929934A (zh) * 2019-11-22 2020-03-27 深圳市通用互联科技有限责任公司 设备故障预测方法、装置、计算机设备和存储介质
CN111422718B (zh) * 2020-05-29 2021-09-17 中国科学院福建物质结构研究所 一种电梯物联边缘计算系统及方法
CN111864706A (zh) * 2020-08-19 2020-10-30 剑科云智(深圳)科技有限公司 一种配电网的故障预警和继电保护系统

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9535774B2 (en) * 2013-09-09 2017-01-03 International Business Machines Corporation Methods, apparatus and system for notification of predictable memory failure
US20160274551A1 (en) * 2015-03-18 2016-09-22 Accenture Global Services Limited Method and system for predicting equipment failure
CN107194053A (zh) * 2017-05-16 2017-09-22 歌拉瑞电梯股份有限公司 一种智能化电梯控制系统运行故障预测方法
CN109110608A (zh) * 2018-10-25 2019-01-01 歌拉瑞电梯股份有限公司 一种基于大数据学习的电梯故障预测方法
CN110647539A (zh) * 2019-09-26 2020-01-03 汉纳森（厦门）数据股份有限公司 一种用于车辆故障的预测方法和系统
CN111638989A (zh) * 2020-04-15 2020-09-08 北京三快在线科技有限公司 故障诊断方法、装置、存储介质及设备
CN112365066A (zh) * 2020-11-17 2021-02-12 日立楼宇技术（广州）有限公司 电梯故障预测方法、系统、装置、计算机设备和存储介质

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115037599A (zh) * 2022-06-13 2022-09-09 中国电信股份有限公司 一种通信网络故障预警方法、装置、设备及介质
CN115130702A (zh) * 2022-09-02 2022-09-30 山东汇泓纺织科技有限公司 一种基于大数据分析的纺织机故障预测系统
CN115576724A (zh) * 2022-09-19 2023-01-06 成都飞机工业（集团）有限责任公司 Piu子系统的故障隔离方法、装置、设备、介质及产品
CN115576724B (zh) * 2022-09-19 2024-04-12 成都飞机工业（集团）有限责任公司 Piu子系统的故障隔离方法、装置、设备、介质及产品
CN115659812A (zh) * 2022-10-29 2023-01-31 思维实创(哈尔滨)科技有限公司 基于城轨iscs的电扶梯寿命预测方法、系统、设备及介质
CN115980531A (zh) * 2023-03-16 2023-04-18 江苏大全长江电器股份有限公司 一种特定环境下的gis开关柜质量检测方法及系统
CN117193088A (zh) * 2023-09-22 2023-12-08 珠海臻图信息技术有限公司 一种工业设备监控方法、装置以及服务器
CN117193088B (zh) * 2023-09-22 2024-04-26 珠海臻图信息技术有限公司 一种工业设备监控方法、装置以及服务器
CN117370848A (zh) * 2023-12-08 2024-01-09 深圳市明心数智科技有限公司 设备故障预测方法、装置、计算机设备及存储介质
CN117370848B (zh) * 2023-12-08 2024-04-02 深圳市明心数智科技有限公司 设备故障预测方法、装置、计算机设备及存储介质

Also Published As

Publication number Publication date
CN112365066B (zh) 2023-05-02
AU2021107643A4 (en) 2022-01-20
CN112365066A (zh) 2021-02-12

Similar Documents

Publication Publication Date Title
WO2022105266A1 (zh) 电梯故障预测方法、系统、装置、计算机设备和存储介质
US11748227B2 (en) Proactive information technology infrastructure management
CN103761173A (zh) 一种基于日志的计算机系统故障诊断方法及装置
CN111412579B (zh) 一种基于大数据的空调机组故障类型诊断方法和系统
KR102432284B1 (ko) It관리대상의 이벤트 알람이나 장애 문제를 실시간 자동으로 조치하는 시스템 및 그 운용방법
CN103049365B (zh) 信息与应用资源运行状态监控及评价方法
CN109034580B (zh) 一种基于大数据分析的信息系统整体健康度评估方法
CN110188015B (zh) 一种主机访问关系异常行为自适应检测装置及其监测方法
CN114267178A (zh) 一种车站的智能运营维护方法及装置
CN111123223A (zh) 用于雷达健康管理的通用开发平台、管理系统及方法
CN112415331A (zh) 基于多源故障信息的电网二次系统故障诊断方法
CN110580492A (zh) 一种基于小幅波动检测的轨道电路故障前兆发现方法
CN112231352B (zh) 一种多路阀失效数据管理及失效模式自动诊断系统
CN113313304A (zh) 一种基于大数据决策树的电网事故异常分析方法及系统
CN106649034B (zh) 一种可视化智能运维方法及平台
CN110415136B (zh) 一种电力调度自动化系统服务能力评估系统与方法
CN111614504A (zh) 基于时间序列和故障树分析的电网调控数据中心业务特性故障定位方法及系统
CN111679970A (zh) 机器人软件系统运行环境状态预测方法
CN112769619B (zh) 一种基于决策树的多分类网络故障预测方法
CN109558258B (zh) 一种分布式系统根源故障定位的方法及装置
CN111625525B (zh) 一种环境数据修复/填充方法及系统
CN114385403A (zh) 基于双层知识图谱架构的分布式协同故障诊断方法
CN112699048A (zh) 基于人工智能的程序故障处理方法、装置、设备及存储介质
CN116304663B (zh) 基于非平衡样本增强的列控车载设备健康状态管理装置
CN117657912B (zh) 一种工地施工升降机监测系统、方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21893418

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21893418

Country of ref document: EP

Kind code of ref document: A1