An elevator operation state evaluation method and system based on multi-source heterogeneous data

Abstract

The present application relates to elevator operation state evaluation technical field, specifically to a kind of elevator operation state evaluation method and system based on multi-source heterogeneous data, comprising the following steps: collection elevator car vertical acceleration signal, traction machine output torque signal, traction machine output current signal, motor encoder speed signal, elevator operation log time, stop floor record, operation frequency record and switch door frequency record, based on timestamp alignment each data, and according to elevator operation start, running, stop state segmentation data stream forms time period, generates synchronization multi-source operation data stream.The present application integrates elevator car vertical acceleration signal, traction machine output torque signal, motor encoder speed signal and other multi-source heterogeneous data, and generates synchronization data stream based on timestamp alignment and operation state segmentation, eliminates the feature misjudgment problem caused by data asynchronization or state mixture.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of elevator operation state evaluation, and in particular to an elevator operation state evaluation method and system based on multi-source heterogeneous data. BACKGROUND

[0002] The technical field of elevator operation state evaluation belongs to the intersection field of elevator safety monitoring and intelligent operation and maintenance, and its core goal is to realize the quantitative evaluation and abnormal early warning of the health degree of elevator operation through multi-dimensional data collection, dynamic feature analysis and risk modeling.

[0003] The prior art uses a single index threshold to judge abnormalities, for example, only monitors acceleration or current signal over-limit, ignores the difference in correlation between mechanical and electrical parameters, and is difficult to distinguish between normal working condition fluctuations and potential fault characteristics. At the same time, the existing risk evaluation model does not fuse external environmental data, for example, does not consider the influence of heavy rain on the friction coefficient of the elevator guide rail or the short-time high-frequency use scenario caused by large-scale activities, resulting in that the conventional range setting deviates from the working condition requirements. Therefore, improvement is needed. SUMMARY

[0004] The purpose of the present application is to solve the shortcomings in the prior art and to provide an elevator operation state evaluation method and system based on multi-source heterogeneous data.

[0005] In order to achieve the above-mentioned purpose, the present application adopts the following technical scheme, an elevator operation state evaluation method based on multi-source heterogeneous data, comprising the following steps:

[0006] Collecting elevator car vertical acceleration signals, traction machine output torque signals, traction machine output current signals, motor encoder speed signals, elevator operation log time, stop floor records, operation frequency records and switch door frequency records, aligning each data based on time stamp, and dividing data stream to form time period according to the start, running and stop state of elevator operation, generating synchronized multi-source operation data stream;

[0007] Based on the synchronized multi-source operation data stream, the ratio sequence of the instantaneous torque value and the acceleration value in the elevator start-stop process is calculated, the difference value sequence between the motor encoder speed signal and the speed estimation value calculated based on the traction machine output torque signal is monitored, the rapid change point of the car vertical acceleration signal amplitude is detected, and a preliminary signal deviation index set is obtained, based on the preliminary signal deviation index set, the abnormal signal points are screened and judged, and a running feature point set is obtained;

[0008] According to the synchronized multi-source operation data stream, the time period operation frequency is calculated to obtain a running basic statistic, based on the running basic statistic, whether the combination of each statistic value and external condition deviates from the conventional range is judged by comparing the weather conditions and large-scale activity information, and a context-related operation mode portrait is established;

[0009] According to the occurrence frequency of the abnormal signal points in the operation feature point set and the context-related operation mode image, an operation environment-related risk score value is calculated, the operation environment-related risk score value is compared with a preset attention alarm threshold value, and an elevator comprehensive operation state evaluation conclusion is generated.

[0010] Preferably, the acquisition step of the synchronized multi-source operation data stream is:

[0011] The vertical acceleration signal of the elevator car, the output torque signal of the traction machine, the output current signal of the traction machine, the motor encoder speed signal, the elevator operation log time, the stop floor record, the operation frequency record, and the switch door frequency record are collected, the timestamp field of each signal is extracted, the absolute value of the reference time difference between the timestamp of each signal and the elevator operation log time is calculated, the absolute value of the difference is defined as the initial transmission delay value, for the signals with a difference absolute value exceeding a preset threshold value, a cubic spline interpolation method is used to fill in the missing time points, and a multi-source signal time series data set with completely aligned timestamps is generated;

[0012] Based on the multi-source signal time series data set with completely aligned timestamps, the state switching markers in the elevator operation log time are analyzed, the starting time point of the start state, the duration interval of the running state, and the termination time point of the stop state are extracted, the data segment boundaries are divided according to the starting time point and the termination time point, the car vertical acceleration signal is cut into start, run, and stop segments according to the boundaries, the output torque signal of the traction machine, the output current signal of the traction machine, and the motor encoder speed signal are synchronously segmented into corresponding data segments, and a segmented multi-source signal data set with state labels is generated;

[0013] Based on the segmented multi-source signal data set with state labels, the floor number and timestamp mapping relationship in the stop floor record is extracted, the floor number is matched to the corresponding state interval according to the divided data segment boundaries, and the operation frequency record and the switch door frequency record are merged to the end of the data segment according to the timestamp, to generate a synchronized multi-source operation data stream.

[0014] Preferably, the acquisition step of the preliminary signal deviation index set is:

[0015] Based on the synchronized multi-source operation data stream, the instantaneous values of the torque signal and the acceleration signal in the elevator start-stop process are extracted, the data segments are divided according to the time window, and a start-stop process instantaneous torque-acceleration data set is generated;

[0016] According to the start-stop process instantaneous torque-acceleration data set, the torque-to-acceleration ratio at each time point is calculated , a torque-acceleration ratio sequence is generated, and the calculation formula is:

[0017] ;

[0018] wherein, is the torque value at the time point, is the acceleration value at the time point;

[0019] performing point-by-point difference calculation on the motor encoder recorded rotation speed signal and the speed estimation value derived from the traction machine output torque signal to obtain difference values , to generate a rotation speed-estimated speed difference value sequence, the formula being:

[0020] ;

[0021] based on the acceleration value , calculating the acceleration difference value between adjacent time points , wherein is the acceleration value at the adjacent time point, and the mutation threshold is set as , is the mean value of , is the standard deviation of , and all time points are marked as acceleration rapid change points, and the torque-acceleration ratio value sequence, the rotation speed-estimated speed difference value sequence, and the acceleration rapid change points are integrated to generate a preliminary signal deviation index set.

[0022] Preferably, the operation feature point set acquisition step is:

[0023] based on the preliminary signal deviation index set, extracting the preset threshold value of the torque-acceleration ratio value sequence, the preset threshold value of the rotation speed-estimated speed difference value sequence, and the acceleration rapid change point amplitude threshold value, defining the time window range as the interval of 250 milliseconds before and after the adjacent time points, generating multi-index threshold parameters and time window definition rules;

[0024] According to the multi-index threshold parameters and time window definition rules, each time point in the preliminary signal deviation index set is traversed, if the torque-acceleration ratio value corresponding to the current time point exceeds the preset threshold value, and there is a rotation speed-estimated speed difference value exceeding the preset threshold value within the 250 milliseconds window before and after the current time point, and the acceleration rapid change point amplitude within the window exceeds the threshold value, the current time point is marked as a candidate abnormal signal point, and a multi-index synchronous overrun time point set is generated;

[0025] Based on the multi-index synchronized ultra-limit time point set, candidate abnormal signal points with adjacent time intervals less than 300 milliseconds are merged, the time point with the largest torque-acceleration ratio deviation amplitude in each cluster is retained as a representative point after merging, and candidate points without adjacent clusters are removed to generate a running feature point set.

[0026] Preferably, the running basic statistic quantity acquisition step is:

[0027] Based on the elevator running log time in the synchronized multi-source running data stream, the complete time axis recorded from 0 o'clock to 24 o'clock in the log is extracted, the time axis is divided into 48 continuous intervals according to every 30 minutes as a statistical period, the running frequency record value and the door opening and closing frequency record value in each period are calculated in combination with the time stamp in the stop floor record, and a running-door opening and closing frequency data set after period division is generated;

[0028] According to the running-door opening and closing frequency data set after period division, the running frequency value of each period is calculated , and a period running frequency sequence is generated, and the calculation formula is:

[0029] ;

[0030] Among them, is the running frequency of the th period, is the door opening and closing frequency of the th period, is the period length;

[0031] Based on the period running frequency sequence, the sliding average value and the standard deviation of the running frequency values of consecutive 6 periods are calculated, the period with a sliding average value exceeding 2 times the standard deviation of the historical same period average value is marked as a high-frequency running interval, the running frequency values of all periods are integrated, and a running basic statistic quantity is generated.

[0032] Preferably, the context-related running mode image acquisition step is:

[0033] Based on the running basic statistic quantity, the running frequency value and the high-frequency running interval label of each statistical period are extracted, the rainfall and wind speed in the external input weather condition data and the activity scale and the number of participants in the large-scale activity information are associated, and are matched to the corresponding statistical period according to the time stamp, to generate an external condition-running statistic quantity association data set;

[0034] Based on the external conditions-operational statistics associated dataset, compare the mean and standard deviation of the operation frequency values ​​over the past three months under the same rainfall and activity scale classification conditions in the historical database. If the operation frequency value of the current period exceeds twice the standard deviation of the historical mean, then mark the current period as a statistical period data that deviates from the normal range and generate a statistical deviation label set.

[0035] Based on the statistical deviation marker set, the distribution density of abnormal signal points in the set of operational feature points during large-scale scenarios is calculated. The percentage of abnormal signal points in the total number of large-scale scenarios per unit time is calculated. Large-scale scenarios refer to rainfall levels of rainstorms or typhoons and large-scale events with a scale of tens of thousands of people. If the percentage exceeds a preset threshold, it is determined to be a high-density abnormal scenario. All high-density abnormal scenarios and statistical time period data that deviate from the normal range are integrated to generate a scenario-related operational mode profile.

[0036] Preferably, the steps for obtaining the associated risk score of the operating environment are as follows:

[0037] Based on the set of running feature points, the number of occurrences of abnormal signal points in each statistical period is extracted. Combined with the high-density abnormal situations in the scenario-related running mode profile, the ratio of the number of occurrences of abnormal signal points under high-density abnormal situations in each period to the total running time is calculated to generate an abnormal frequency-high-density scenario-related dataset.

[0038] Based on the anomaly frequency-high density context association dataset, the percentage deviation between the number of anomaly signal points and the historical average number of anomaly points is calculated. Periods with a deviation exceeding 20% ​​are defined as risk-sensitive periods, and a risk-sensitive period marker sequence is generated.

[0039] Based on the aforementioned risk-sensitive period marker sequence, the associated risk score of the operating environment is calculated using the following formula:

[0040] ;

[0041] in, Associating risk scores with the operating environment For the first The number of abnormal signal points occurring in each time period. For the first High-density abnormal situation influencing factors in each time period, here , This represents the number of abnormal points within the time period. This represents the total runtime of the time period.

[0042] Preferably, the steps for obtaining the comprehensive elevator operation status assessment conclusion are as follows:

[0043] Based on the operation environment associated risk score value, the operation environment associated risk score value of each statistical period is extracted, the preset attention alarm threshold is compared, the threshold level interval is defined, and the risk score data set with threshold interval is generated;

[0044] According to the risk score data set with threshold interval, the risk score value of each statistical period is traversed, if the score value exceeds the serious alarm threshold, it is marked as "emergency state", if it exceeds the moderate alarm threshold but does not reach the serious threshold, it is marked as "high risk state", if it exceeds the mild attention threshold but does not reach the moderate threshold, it is marked as "potential risk state", and the rest is marked as "normal state", and a risk level state marking set is generated;

[0045] Based on the risk level state marking set, the number of periods marked as "emergency state" or "high risk state" in the last 3 periods is counted, if the last 3 periods are all high risk or contain 1 emergency state, "immediate maintenance suggestion" conclusion is generated, and the elevator comprehensive operation state evaluation conclusion is obtained.

[0046] The application provides an elevator operation state evaluation system, comprising:

[0047] A data synchronization module collects car vertical acceleration signals, traction machine output torque signals, traction machine output current signals, motor encoder speed signals, operation log time, stop floor records, operation frequency records and switch door frequency records, aligns the collection time with a time stamp, divides the start state, the running state and the stop state into independent time periods, serially combines all signals in the divided time periods in time sequence, and establishes a multi-source synchronous data stream.

[0048] A signal offset determination module, based on the multi-source synchronous data stream, pairs the acceleration signals and torque signals of the start state and the stop state with corresponding values, then takes the ratio sequence, takes the numerical difference sequence of the motor encoder speed signal and the speed estimation value calculated from the traction machine output torque signal, and marks the mutation point position after performing amplitude increment operation between adjacent values in the sequence on the car vertical acceleration signal, filters the maximum value point in the ratio sequence, the absolute deviation point in the difference sequence and the time period where the mutation point position is located, and obtains an offset characteristic signal set.

[0049] An operation frequency statistical module, based on the multi-source synchronous data stream, counts the total number of daily operations, the number of changes in the stop floor per day, the duration of each operation and the total number of switch doors per day, extracts the numerical distribution interval within 24 hours per day, compares the difference between the distribution interval and the common distribution range in the historical operation period, and establishes a periodic operation frequency feature.

[0050] The situation feature description module calls the weather condition information of each statistical day and the large activity record in the region based on the cycle operation frequency feature, compares the total number of daily operation and the weather sunny and rainy state, combines the floor change number and the existence of the large activity, and interval combines the total number of daily opening and closing doors and the temperature level, respectively establishes three combined relationships of the operation number and the weather combination mode, the floor change and the activity combination mode, and the opening and closing door and the temperature combination mode, marks the combination type of the distribution exceeding the normal mode interval, collects the appearance frequency of the combination type and the statistical time interval, and generates a situation combination operation atlas.

[0051] The risk score module assigns a risk level to the combination type with the number of offsets higher than the standard deviation interval in the combined relationship based on the signal offset points marked in the offset feature signal set, and obtains a risk distribution score result.

[0052] Compared with the prior art, the advantages and positive effects of the present application are that:

[0053] The present application integrates elevator car vertical acceleration signals, traction machine output torque signals, motor encoder speed signals and other multi-source heterogeneous data, and generates a synchronized data stream based on timestamp alignment and operation state segmentation, eliminating feature misjudgment problems caused by data asynchrony or state mixing. By calculating the ratio sequence of the torque and acceleration during the start and stop of the elevator, the difference value sequence of the speed signal and the torque derived speed, and combining the acceleration signal amplitude rapid change point detection, a multi-dimensional deviation index is constructed from the perspective of mechanical dynamic response and electrical control correlation, improving the sensitivity and anti-interference ability of abnormal signal identification. The situation correlation analysis of weather conditions and large activity information is introduced, and a dynamic matching model of external environment and operation statistics is established, so that the risk score value can reflect the comprehensive influence under complex working conditions. The multi-level comparison mechanism of risk score and preset threshold combines discrete abnormal points and continuous high-density situations, realizes hierarchical evaluation from local anomaly to systematic risk, reduces the false alarm rate of single threshold judgment, and provides priority ranking basis for maintenance decision. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 The present application is a step schematic diagram. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the present application and do not limit the present application.

[0056] Please refer to Figure 1 The present application provides a technical scheme, a kind of elevator operation state evaluation method based on multi-source heterogeneous data, comprising the following steps:

[0057] collecting an elevator car vertical acceleration signal, a hoisting machine output torque signal, a hoisting machine output current signal, a motor encoder rotation speed signal, an elevator operation log time, a stop floor record, a number of operation records, and a number of door opening and closing records, aligning each data based on a time stamp, and dividing a data stream into time periods according to a start, operation, and stop state of the elevator operation to generate a synchronized multi-source operation data stream;

[0058] Based on the synchronized multi-source operation data stream, a ratio sequence of instantaneous torque values and acceleration values in the elevator start-stop process is calculated, a difference value sequence between the motor encoder rotation speed signal and a speed estimation value calculated based on the hoisting machine output torque signal is monitored, a rapid change point of the car vertical acceleration signal amplitude is detected, a preliminary signal deviation index set is obtained, abnormal signal points are screened and determined based on the preliminary signal deviation index set, and a running feature point set is obtained;

[0059] According to the synchronized multi-source operation data stream, a time period operation frequency is calculated to obtain a running basic statistic, and based on the running basic statistic, whether the values of each statistic deviate from the conventional range in combination with external conditions is judged by comparing weather conditions and large event information to establish a context-related operation mode portrait;

[0060] According to the occurrence frequency of the abnormal signal points in the running feature point set and the context-related operation mode portrait, a running environment-related risk score value is calculated, the running environment-related risk score value is compared with a preset attention alarm threshold value to judge, and an elevator comprehensive operation state evaluation conclusion is generated.

[0061] The acquisition steps of the synchronized multi-source operation data stream are as follows:

[0062] collecting an elevator car vertical acceleration signal, a hoisting machine output torque signal, a hoisting machine output current signal, a motor encoder rotation speed signal, an elevator operation log time, a stop floor record, a number of operation records, and a number of door opening and closing records, extracting a time stamp field of each signal, calculating an absolute value of a reference time difference between the time stamp of each signal and the elevator operation log time, defining the absolute value of the difference as an initial transmission delay value, filling in missing time points for signals with an absolute value of the difference exceeding a preset threshold value using a cubic spline interpolation method, and generating a multi-source signal time series data set with completely aligned time stamps;

[0063] Based on the timestamp complete alignment of multi-source signal timing data set, the state switching mark in the elevator operation log time is parsed, the starting time point of the starting state, the duration interval of the running state and the termination time point of the stopping state are extracted, the data segment boundary is divided according to the starting time point and the termination time point, the car vertical acceleration signal is cut into the starting segment, the running segment and the stopping segment according to the boundary, the traction machine output torque signal, the traction machine output current signal and the motor encoder speed signal are synchronously segmented into the corresponding data segment, and the segmented multi-source signal data set with state label is generated;

[0064] Based on the segmented multi-source signal data set with state label, the floor number and timestamp mapping relationship in the stop floor record is extracted, the floor number is matched into the corresponding state interval according to the divided data segment boundary, the running number record and the switch door number record are merged to the end of the data segment according to the timestamp, and the synchronized multi-source operation data stream is generated.

[0065] Specifically, the acquisition of the synchronized multi-source operation data stream is based on the collection of multiple original signals of the elevator, specifically including the three-axis acceleration signal of the car collected by the MEMS acceleration sensor installed on the top or bottom of the car (focusing on the vertical direction, unit: m / s²), the output torque signal collected by the torque sensor installed on the main shaft or output shaft of the reduction box (unit: Nm), the three-phase current signal collected by the phase current sensor installed on the motor input end through the current transformer (unit: A), and the motor speed signal output by the rotary encoder coaxially installed with the traction motor (unit: r / min), and the running log timestamp (accurate to milliseconds) containing the running state, fault code and other information, the stop floor event and timestamp recorded by the controller, the running number counter value and the corresponding timestamp, and the door opening and closing event number and timestamp recorded by the door machine system are read from the elevator control system log database. First, the timestamp field in each signal data record is extracted, for example, a traction torque signal record is read as {Timestamp: 1678886401500, Torque: 120.5}, and the timestamp is 1678886401500 milliseconds, and a elevator operation log record is read as {LogTime: 1678886401450, Event: 'Start closing door'}, and the timestamp is 1678886401450 milliseconds. Calculate the absolute value of the timestamp reference time difference value of the two, that is 50 milliseconds, which is the initial transmission delay value of the torque signal sampling point relative to the log record, for each sampling point of all collected signals (acceleration, torque, current, speed), the difference is calculated with the closest elevator operation log timestamp in time, a preset transmission delay threshold is set, which is determined based on statistical analysis of historical data transmission delay, for example, the mean and standard deviation of signal delay time in the past 24 hours are calculated, the mean is 40 milliseconds and the standard deviation is 15 milliseconds, then the preset threshold can be set to the mean plus three times the standard deviation, that is milliseconds, if the absolute value of the difference calculated by a signal sampling point, for example 88 milliseconds, exceeds the threshold of 85 milliseconds, it is considered that the signal point has significant delay or timestamp anomaly and needs to be processed, for the signal points whose difference exceeds the threshold and the time point missing problem of the data sequence caused by them, a cubic spline interpolation method is used to fill in, specifically, the invalid time points marked due to delay exceeding and the completely missing time points caused by network packet loss and other reasons are identified, the nearest valid data points before and after them are found, based on the values and timestamps of these valid data points, a piecewise cubic polynomial is used to fit the data curve, ensuring the continuity of the function value, first derivative and second derivative at the interpolation point, so as to calculate the estimated signal value at the missing time point, for example, if the acceleration data is missing at time point T, the nearest valid point before T (T1, A1) and the nearest valid point after T (T2, A2) are found, as well as their respective adjacent points, a cubic spline function is constructed, and the estimated acceleration value A interp corresponding to time point T is solved, all signals (acceleration, torque, current, speed) are processed in this way to make their timestamp sequence strictly aligned to the unified reference time axis (usually taking the operation log time as the reference, or selecting one of the high-frequency signals as the reference to align other signals), and a multi-source signal time series data set with completely aligned timestamps is generated.

[0066] Based on the multi-source signal time series data set with completely aligned timestamps, the operation log content recorded by the elevator control system is further analyzed. These log entries contain key information describing the state transition of the elevator, such as the text labels "elevator starts running", "elevator arrives at target floor and starts deceleration", "elevator stops at landing", "door opening command issued", "door closing action completed", etc. By identifying these specific state switching marker texts or corresponding state codes, the accurate timestamps accompanying these markers are extracted to determine the key time nodes of the elevator running cycle. Specifically, find the log entry marked as "start" or similar, record its timestamp as the start time point of the start state , find the log entry marked as "stop", "landing in place" or similar, record its timestamp as the termination time point of the stop state , define the time interval from Start with a short period of time (e.g., based on experience, set at 1.5 seconds, i.e.) arrive The acceleration phase is defined as starting from a certain point in time before the stop signal appears (e.g., set according to the typical deceleration time). )arrive To stop the deceleration phase, then the acceleration phase has ended ( ) and the start of the deceleration phase ( The time interval between these points is the duration of the running state, and the starting time point is determined accordingly. and termination time point (and intermediate dividing points) and Using these boundary points as data segments, the car's vertical acceleration signal sequence, aligned with the timestamps, is cut. For example, if a complete operating cycle starts at timestamp 1678886405000 and stops completely at timestamp 1678886415000, then the data range for the start segment is [1678886405000, 1678886406500], the data range for the running segment is (1678886406500, 1678886413000), and the data range for the stop segment is [1678886413000, 1678886415000]. Simultaneously, identical boundaries are used. The timestamps [1678886405000, 1678886406500, 1678886413000, 1678886415000] are used to synchronously segment the traction machine output torque signal sequence, traction machine output current signal sequence, and motor encoder speed signal sequence with the timestamps aligned. This ensures that each signal is segmented into a start-up segment, a running segment, and a stop segment corresponding to the same time range. Finally, each segmented data segment (whether it is an acceleration, torque, current, or speed signal segment) is assigned a corresponding status label, such as "start," "run," or "stop," generating a segmented multi-source signal dataset with status labels.

[0067] Based on the segmented multi-source signal dataset with state labels, further integrate the discrete event information, first extract the stop floor information recorded by the elevator controller, which usually contains the floor number (such as the physical floor number "5" or the logical number) and the timestamp of the stop event at this floor (for example, the door opening to position timestamp 1678886415100), analyze this timestamp 1678886415100, and match it with the data segment boundary divided in the previous step to determine which state interval of which running cycle the stop event belongs to. Since the stop occurs at the end of the running, the floor number "5" will be associated with the stop segment corresponding to the timestamp 1678886415000 or immediately after it, establishing the mapping relationship between the floor number and the corresponding running cycle state interval. Then, process the running number record and the door opening and closing number record. These records may be cumulative values or single event records. If it is a single event record (such as a running completion event with timestamp 1678886415050; a door opening and closing cycle completion event with timestamp 1678886418000), according to its timestamp, it will be merged into the last data segment (usually the stop segment or the door opening and closing action segment) of the corresponding running cycle. For example, the running completion event timestamp 1678886415050 should belong to the stop segment of the running cycle that has just ended, and the door opening and closing number record 1678886418000 should also belong to the stop segment associated with the subsequent door operation time. If the record is a cumulative value of a certain time point (for example, the total number of runs is 1025 at 1678886400000, and the total number of runs is 1026 at 1678886420000), the difference between adjacent record points is calculated to determine the number of occurrences in this time period (1026-1025=1 run), and this run is attributed to the end of the stop segment of the running cycle that occurs in this time period [1678886400000, 1678886420000]. The extracted and mapped floor number, as well as the running number and door opening and closing number information merged by timestamp, are attached to the metadata of the corresponding data segment to form a structured data stream, where each entry represents a complete running cycle segment with state labels (start, run, stop), and additional context information (stop floor, running and door opening and closing count within the cycle) is generated. Synchronized multi-source running data stream.

[0068] The preliminary signal deviation index set acquisition step is:

[0069] Based on the synchronized multi-source running data stream, extract the instantaneous values of the torque signal and the instantaneous values of the acceleration signal during the elevator start-stop process, divide the data segments according to the time window, and generate the start-stop process instantaneous torque-acceleration data set;

[0070] According to the start-stop process instantaneous torque-acceleration data set, the torque-to-acceleration ratio at each time point is calculated , a torque-acceleration ratio sequence is generated, and the calculation formula is:

[0071] ;

[0072] Among them, is the torque value at the th time point, is the acceleration value at the th time point;

[0073] Based on the torque-acceleration ratio sequence, the rotational speed signal recorded by the motor encoder and the speed estimation value derived from the traction machine output torque signal are calculated point by point, and the difference value is obtained, and a rotational speed-estimated speed difference value sequence is generated, and the formula is:

[0074] ;

[0075] Based on the acceleration value , the acceleration difference value of adjacent time points is calculated, where is the acceleration value of adjacent time points, and the mutation threshold is set to , is the mean value of , and is the standard deviation of , and all time points are marked as acceleration rapid change points. The torque-acceleration ratio sequence, the rotational speed-estimated speed difference value sequence, and the acceleration rapid change points are integrated to generate a preliminary signal deviation index set.

[0076] Specifically, based on the synchronized multi-source operation data stream, the instantaneous values of the traction machine output torque signal and the instantaneous values of the car vertical acceleration signal in the starting and stopping dynamic processes of the elevator are extracted, all data segments marked as "start segment" and "stop segment" are screened out, and the time and corresponding torque and acceleration values in these segments are paired, for example, for a start segment, the data may be [(t1, M1, A1), (t2, M2, A2),..., (tn, Mn, An)], all start and stop segment data points are combined to form a set, denoted as the start-stop process instantaneous torque-acceleration data set, then, according to the data set, the torque-to-acceleration ratio at each time point is calculated , and the formula is applied for calculation, where is the torque value measured at time point is the acceleration value measured at the same time point , it is necessary to handle the acceleration near zero to avoid division by zero errors, for example, a very small positive number , such as 0.01 m / s², can be set , when , the ratio is not calculated for this point or marked as invalid, for example, at the start instant, , the torque Nm, and the acceleration m / s², then the ratio Nm / (m / s²) is calculated, this calculation is repeated for all valid time points of the start and stop segments, generating a time series of torque-acceleration ratios , then, based on the complete synchronized data stream (containing the run segment), the speed signal directly recorded by the motor encoder (unit: m / s, converted from r / min, for example, by , where is the r / min speed, is the traction wheel diameter, such as 0.5 m, is the reduction ratio) and the speed estimate calculated using the traction machine output torque signal through the motor model are compared point by point, the difference between the speed estimate and the encoder measured speed , where is a simplification factor, for each time point , the absolute value of the difference between the two is calculated , for example, at time point , the encoder measures a speed m / s, the model estimates a speed m / s, then the difference value m / s, generating a sequence of speed-estimated speed difference values , in addition, using the acceleration signal (also complete data stream), the absolute value of the acceleration difference between adjacent time points is calculated , for example, at time point , the acceleration m / s², at the next time point , the acceleration m / s², then m / s², after calculating the value of the entire sequence, it is necessary to set an acceleration jump threshold to identify rapid change points, this threshold is defined as ,in yes The mean of the sequence over a representative period (e.g., the past 24 hours). It is the standard deviation of the sequence, and the specific calculation process is as follows: collect, for example, all data from the past 24 hours... The value is obtained by calculating its arithmetic mean. Calculate its standard deviation to obtain For example, calculated m / s², m / s², then the mutation threshold m / s², all calculated Compare the value with this threshold, if m / s², then the time point (That is, the point in time after the change occurs) is marked as the point of rapid change in acceleration. Finally, the generated torque-acceleration ratio sequence is... Rotational speed-estimated speed difference value sequence This, along with the set of timestamps marked as points of rapid acceleration change, generates a preliminary set of signal deviation indicators.

[0077] formula illustrate:

[0078] This formula is used to calculate the instantaneous torque-acceleration ratio during elevator start-up or stop. Representative at the The torque-acceleration ratio at each time point. By using a torque sensor at a specific time point The instantaneous value of the output torque of the traction machine was collected. At the same time point, through the accelerometer Instantaneous values ​​of the car's vertical acceleration collected. (Subscript) This indicates the index of the time point corresponding to this value. The calculation logic of this formula is to divide the instantaneous driving force (proportional to torque) by the generated instantaneous acceleration to obtain a quantity that reflects the combined effect of system inertia and load. Its physical meaning is approximately equivalent to the system's equivalent moment of inertia or mass (depending on the relationship between specific forces and motion). This ratio is calculated to monitor the consistency between the driving force and motion response of the elevator during dynamic processes.

[0079] Example: For instance, at a certain point in the startup process. The collected torque Nm, and the acceleration collected simultaneously m / s². Substitute into the formula to calculate:

[0080] ;

[0081] This calculation result Nm / (m / s²) is part of the preliminary signal deviation index set.

[0082] Formula Explanation:

[0083] This formula is used to calculate the absolute difference between the actual measured rotational speed of the motor encoder at any time point and the estimated speed based on the model (for example, using the traction machine output torque signal) . represents the speed difference value at the th time point. is the instantaneous speed of the car directly measured by the motor encoder, and the superscript indicates that it is the th point in the sequence. is the estimated speed calculated using other signals (mainly torque signal) and motor / system model, and the superscript also indicates that it is the th point in the sequence. The absolute value symbol indicates that the non-negative value of the difference between the two is taken. The calculation logic of this formula is to quantify the deviation between the actual measured speed and the model predicted speed, which can reflect the motor control performance, model accuracy or encoder itself problem.

[0084] Example: for example, at a time point during operation, the encoder measures the speed m / s. Based on the torque value at that time and a simplified model , for example Nm, the estimated speed is m / s. Put into the formula calculation:

[0085] ;

[0086] This calculation result m / s is another part of the preliminary signal deviation index set.

[0087] Formula and Explanation:

[0088] Calculate the absolute change of acceleration between the two adjacent time points and , that is, the absolute value of the difference of acceleration, which can be regarded as a discrete approximation of the acceleration change rate (jerk or jerk). is the time point acceleration value at the previous time point, is the acceleration value at the previous time point. This calculation is used to detect sudden changes in acceleration during operation, which may indicate unstable operation, vibration or impact.

[0089] The threshold value that determines whether the acceleration change is "fast" or "abrupt" is defined. is the abrupt threshold value. is the arithmetic mean of all calculated values in history (e.g. in the past day). is the standard deviation of these values. This formula adopts the "3 sigma principle" in statistics, i.e. if the data roughly obeys the normal distribution, about 99.7% of the data will fall within the range of the mean plus or minus three times the standard deviation, and points outside this range are considered statistical outliers or small probability events. The logic here is that if the acceleration change exceeds the usual fluctuation range (defined by ), it is considered a significant, fast change point worthy of attention.

[0090] Example: suppose the historical data has calculated the acceleration difference mean m / s², and the standard deviation m / s². Then the threshold value m / s². Now calculate the acceleration difference value at the current time point . If m / s², m / s², then m / s². Because , the time point is marked as a fast acceleration change point. This marked information is also part of the preliminary signal deviation indicator set.

[0091] The benefits of the formula are:

[0092] By monitoring the changes in driving force and motion response, it helps to identify load abnormalities or changes in mechanical resistance; By comparing the measured speed with the model estimated speed, it helps to find control system deviations or sensor faults; In combination with , it objectively identifies jolts or impact events in operation through statistical methods, and these indicators together constitute a preliminary quantitative evaluation of signal deviation from normal behavior patterns.

[0093] The results show that the calculated ratio , difference value , acceleration difference (And whether the threshold is exceeded) are all specific values ​​that constitute the initial signal deviation index set. These numerical sequences and markers provide the basic data for subsequent identification of operational feature points.

[0094] The steps for obtaining the feature point set are as follows:

[0095] Based on the preliminary signal deviation index set, the preset thresholds for the torque-acceleration ratio sequence, the preset thresholds for the speed-estimated speed difference sequence, and the amplitude thresholds for rapid acceleration change points are extracted. The time window range is defined as an interval of 250 milliseconds before and after adjacent time points, and multi-index threshold parameters and time window definition rules are generated.

[0096] Based on the multi-index threshold parameters and time window definition rules, each time point in the preliminary signal deviation index set is traversed. If the torque-acceleration ratio corresponding to the current time point exceeds the preset threshold, and the speed-estimated speed difference exceeds the preset threshold within a 250-millisecond window before and after the current time point, and the amplitude of the rapid acceleration change point within the window exceeds the threshold, then the current time point is marked as a candidate abnormal signal point, and a set of multi-index synchronous over-limit time points is generated.

[0097] Based on the set of time points for synchronous over-limit of multiple indicators, candidate abnormal signal points with adjacent time intervals of less than 300 milliseconds are merged. After merging, the time point with the largest deviation of torque-acceleration ratio in each cluster is retained as the representative point, and isolated candidate points without adjacent clusters are removed to generate a set of operating feature points.

[0098] Specifically, based on the initial signal deviation index set, it is first necessary to set various thresholds for filtering operational characteristic points. These thresholds need to be determined with reference to the historical data distribution under normal operating conditions of the equipment, such as the torque-acceleration ratio sequence. preset threshold It can be set to the 95th percentile of the distribution of this ratio under normal operating conditions, for example, by analyzing historical normal data. The 95th percentile is 195 Nm / (m / s²), then set Rotational speed-estimated speed difference value sequence preset threshold The setting can be based on acceptable control accuracy or model error. For example, if the speed tracking error is required to be no more than 0.1 m / s, then set... m / s, amplitude threshold at points of rapid change in acceleration It is for those that have already passed Further filtering of rapidly changing points identified by threshold selection, used to focus on shocks with particularly large amplitudes, can be set to... 1.5 times the value or a fixed empirical value, for example, setting m / s², while defining the time window range as the current observation time point. The interval of 250 milliseconds before and after, i.e. The total duration is 500 milliseconds, and these thresholds ( The threshold parameters and time window rules (500ms) together constitute the multi-indicator threshold parameters and time window definition rules. Then, based on these rules, each time point in the initial signal deviation indicator set is traversed. Execute synchronization over-limit judgment logic: check the current time point. Torque-acceleration ratio Does it exceed its threshold? ,Right now If this condition is met, then further checks will be conducted on the following: Centered time window Does there exist at least one point in time within this period? The difference between its rotational speed and estimated speed Exceeding the threshold ,Right now If this condition is also met, then check the same time window again. Does the time point be marked as a point of rapid change in acceleration? (i.e., its corresponding) ), and the magnitude of the acceleration change It must also exceed the amplitude threshold. ,Right now Only when these three conditions ( Exceeded limits, existing within the window The current time point will only be updated when both the time limit is exceeded and the window memory changes rapidly at a point of significant acceleration. Points marked as candidate anomalous signals, for example, at time point Calculations yielded Nm / (m / s²), because If the first condition is met, then within the time window... An internal search revealed the time point. hour, m / s, because If the second condition is met, continue searching within the window and find the time point. The point is marked as a point of rapid change, and its magnitude of change is... m / s², because The third condition is met, therefore the time point is... Each point is marked as a candidate anomalous signal point. All time points that pass this triple-condition screening are collected to form a multi-indicator synchronous exceedance time point set. Then, based on this set, candidate points that are too close together are processed. A time interval threshold of 300 milliseconds is set, and the candidate points in the set are sorted by time. If two adjacent candidate points are found... and Time difference between If the time interval is less than 300 milliseconds, these are considered as a cluster of anomalous events. All adjacent points that meet this condition are merged into the same cluster. For each cluster, the torque-acceleration ratio of all candidate points within the cluster is compared. Choose that The time point with the largest value is taken as the unique representative point of the cluster and retained. Finally, all the remaining representative points after merging and filtering are checked. If a representative point is not merged into any cluster (i.e. there are no other candidate points within 300ms before and after it), it is considered an isolated point and is removed. All the representative points that are finally retained constitute the set of running feature points.

[0099] The steps for obtaining the basic statistics are as follows:

[0100] Based on the elevator operation log time in the synchronized multi-source operation data stream, the complete time axis from 0:00 to 24:00 of each day is extracted from the log. The time axis is divided into 48 continuous intervals with each 30-minute statistical period. Combined with the timestamps in the floor stop records, the number of operation records and the number of door opening and closing records in each period are counted to generate the operation-door opening and closing frequency dataset after the time period division.

[0101] Based on the time-segmented operation-door opening / closing frequency dataset, calculate the operation frequency value for each time segment. The frequency sequence of the time period is generated, and the calculation formula is as follows:

[0102] ;

[0103] in, For the first Number of runs in each time period For the first Number of times the door is opened and closed during a given time period The duration of the time period;

[0104] Based on the time period operation frequency sequence, the moving average and standard deviation of the operation frequency values ​​for 6 consecutive time periods are calculated. Time periods where the moving average exceeds twice the standard deviation of the historical average for the same period are marked as high-frequency operation intervals. The operation frequency values ​​of all time periods are integrated to generate basic operation statistics.

[0105] Specifically, based on the elevator operation log time and related counting records in the synchronized multi-source operation data stream, the complete time record for each day is first extracted from the log, covering the time range from 0:00:00 to 23:59:59. Then, the timeline of this whole day is divided into statistical periods of fixed length every 30 minutes, resulting in a total of 48 consecutive time intervals, numbered from j=1 (00:00:00-00:29:59) to j=48 (23:30:00-23:59:59). Combining the already merged stop floor records, operation count records, and door opening / closing count records in the synchronized data stream, the statistics are calculated for each 30-minute time interval. Total number of runs within and the total number of times the door is opened and closed. For example, for time period j=19 (09:00:00-09:29:59), the number of runs is accumulated by iterating through all the running cycle records within this time period and appending the count information at the end. Number of times the door is opened and closed These statistical results are then compiled into a dataset, where each entry contains a time period index j and the corresponding number of runs. and number of times the door is opened and closed The resulting time-segmented operation-door opening / closing frequency dataset is shown in Table 1 below.

[0106] Table 1. Examples of Operating and Door Opening / Closing Frequency During Different Time Periods

[0107]

[0108] As shown in Table 1, this table lists some 30-minute time periods and their corresponding operation and door opening / closing statistics. Based on this frequency dataset, the calculations for each time period are then performed. Comprehensive operating frequency value Using formula Perform calculations, where It is the first Number of runs during a time period It is the first Number of times the door is opened and closed during a given time period This refers to the duration of the time slot, which is fixed at 30 minutes. For example, to calculate the operating frequency for time period j=19: This calculation is performed per event / minute for all 48 time periods, generating a time period run frequency sequence containing 48 frequency values. Then, based on this frequency sequence, high-frequency operating intervals are identified using a sliding window method. The window size is set to 6 consecutive time periods (i.e., 3 hours). For each window (e.g., the window from time period j=18 to j=23), the operating frequency values ​​for these 6 time periods are calculated. arrive arithmetic mean and standard deviation At the same time, it is necessary to query the historical database to obtain the average operating frequency of the elevator during the same time period on the same day of the week (e.g., all historical Tuesdays from 08:30 to 11:30). and standard deviation The high-frequency judgment condition is set as the average frequency of the current sliding window. Whether it is greater than the historical average for the same period plus twice its standard deviation is used to determine... Whether it holds true, for example, for a window of time period 18 to 23, calculate the current... Query times / minute to get historical data for the same period times / minute times / minute, then compare and ,because Therefore, this time window (08:30-11:30) is marked as the high-frequency operating range, and the operating frequency values ​​for all time periods are... The high-frequency running interval marking information obtained from the sliding window comparison results is integrated together to generate basic running statistics.

[0109] formula illustrate:

[0110] This formula is used to calculate the first... The overall elevator operating frequency within a statistical period. Representative time period The operating frequency value is usually expressed in "times / minute" or "times / hour". During the time period The total number of elevator trips recorded internally. During the time period The total number of elevator door opening and closing cycles recorded internally. It is a time period The length of the time period is fixed at 30 minutes. The calculation logic of this formula is to add up the number of the two main elevator activities (running and opening / closing doors) within the time period, and then divide by the total duration of the time period to obtain an average activity rate per unit time, which is used to quantify the elevator usage intensity or busyness within that time period.

[0111] Example: For the time period j=19 (09:00:00-09:29:59), we know from Table 1 that... Second, Second. Length of period Minutes. Plug into the formula to calculate:

[0112] ;

[0113] The result of this calculation times per minute, is part of the running base statistics, indicating that during the period from 9 a.m. to 9:30 a.m., the elevator had an average of about 1.67 running or door opening and closing events per minute.

[0114] This result shows that the calculated frequency values for each period form a basic data sequence reflecting the daily usage pattern of the elevator, combined with the labels of the high-frequency intervals, together form the running base statistics, providing a basis for subsequent context-related analysis.

[0115] The steps for obtaining the context-related running mode image are:

[0116] Based on the running base statistics, extract the running frequency value of each statistical period and the high-frequency running interval label, correlate the rainfall and wind speed in the external input weather condition data and the activity size and number of participants in the large-scale activity information, and match them to the corresponding statistical period according to the time stamp, generating an external condition-running statistics correlation dataset;

[0117] According to the external condition-running statistics correlation dataset, compare the average and standard deviation of the running frequency value in the past three months under the same rainfall and same activity size classification conditions in the historical database, if the current period running frequency value exceeds twice the standard deviation of the historical average, mark the current period as a statistical period data deviating from the normal range, generate a statistics deviation label set;

[0118] Based on the statistics deviation label set, count the distribution density of abnormal signal points in the running feature point set in large-scale contexts, calculate the percentage of abnormal signal points in the total number of large-scale context running times per unit time, large-scale context refers to heavy rain or typhoon in rainfall level and large-scale activity with million-level in activity size, if the percentage exceeds the preset threshold, it is determined as a high-density abnormal context, integrate all high-density abnormal contexts and statistical period data deviating from the normal range, generate a context-related running mode image.

[0119] Specifically, based on the running frequency value of each statistical period (a total of 48) contained in the running base statistics Based on the high-frequency operating range markers and the previously generated set of operating feature points (including timestamps of anomalous signal points), we begin contextual correlation analysis. This initially requires introducing external environmental and event information data. For example, we obtain hourly rainfall data (in mm) and average wind speed data (in m / s) from meteorological service interfaces, and information about large-scale events from city management or event organizers, including event location, time, estimated scale (e.g., participant number tiers: <1000 people, 1000-5000 people, 5000-10000 people, >10000 people), and event details. The system aligns and matches these external data points with the elevator operation statistics for a 30-minute time period based on their recorded timestamps. If multiple external data records exist within a given time period, the average value or the most relevant value can be selected. For example, for time period j=30 (15:00-15:30), the system matches an average rainfall of 5mm and an average wind speed of 3m / s. It also finds a nearby venue hosting an exhibition with 8000 attendees. These external conditions (5mm rainfall, 3m / s wind speed, 8000 attendees) are then matched with the operation statistics for that time period (e.g.,...). This is used to correlate external conditions with runtime statistics (times / minute, not marked as high-frequency intervals). Then, this correlated dataset is compared with past runtime data stored in the historical database (e.g., data from the past three months) to filter records with the same or similar external conditions as the current time period. The similarity needs to be defined; for example, rainfall can be categorized (no rain, light rain, moderate rain, heavy rain, torrential rain), and activity scale can be categorized (as above). Then, all time periods in the historical database with rainfall belonging to the "light rain" level and activity scale belonging to the "5000-10000 people" category are searched, and the runtime frequency in these historical records is calculated. mean and standard deviation Then, the operating frequency value for the current time period. The judgment criteria are as follows: (Compare with the statistical range under the same historical context.) For example, a query of historical data shows that during light rain and with 5,000-10,000 people active, historical average Batch / minute, standard deviation times / minute, calculate the comparison threshold as: Because of the current So the current period j=30 is marked as a statistical period data deviating from the regular range, forming a statistical quantity deviation marking set, then the distribution density of the running feature points (abnormal signal points) under the specific "large-scale situation" needs to be analyzed, first define what is "large-scale situation", according to the text description, the large-scale situation includes that the rainfall level is judged as "heavy rain" (for example, hourly rainfall > 16mm) or "typhoon" (for example, average wind speed > 32.7m / s), and the activity scale is judged as "ten-thousand-person large-scale activity" (the number of participants > 10000), find out all the 30-minute statistical periods that meet any of the large-scale situation conditions, and count the total number of running feature points (abnormal signal points) that appear in these large-scale situation periods , and count the total number of elevator running times in these same periods , calculate the occurrence density of abnormal signal points under large-scale situation, that is, the proportion of abnormal points in unit running times: , set a density threshold , which is set according to risk tolerance, for example , which means that if more than 5 abnormal signal points appear in every 100 running times under large-scale situation, it is considered that the density is too high, if the calculated , it is determined that the current running state is in "high-density abnormal situation", finally, integrate all the period information marked as "high-density abnormal situation" and the "statistical quantity deviation marking set" generated before (that is, those period data whose running frequency is significantly higher under certain external conditions) to form the situation-related running mode portrait together.

[0120] The steps for obtaining the running environment-related risk score value are:

[0121] Based on the running feature point set, extract the number of abnormal signal points occurring in each statistical period, combine the high-density abnormal situation in the situation-related running mode portrait, and count the ratio of the number of abnormal signal points occurring in high-density abnormal situation to the total running time in each period, to generate an abnormal frequency-high-density situation-related data set;

[0122] According to the abnormal frequency-high-density situation-related data set, calculate the deviation percentage of the number of abnormal signal points from the historical average number of abnormal points, define the period with a deviation of more than 20% as a risk-sensitive period, and generate a risk-sensitive period marking sequence;

[0123] Based on the risk-sensitive period marking sequence, calculate the running environment-related risk score value, the calculation formula is:

[0124] ;

[0125] Wherein, is the running environment-related risk score value, For the first The number of abnormal signal points occurring in each time period. For the first High-density abnormal situation influencing factors in each time period, here , This represents the number of abnormal points within the time period. This represents the total runtime of the time period.

[0126] Specifically, based on the set of operational feature points generated in the previous steps (which records the timestamp of each abnormal signal occurrence) and the scenario-related operational pattern profile (which identifies high-density abnormal scenario periods), the first step is to statistically analyze each 30-minute statistical period. The total number of operational characteristic points (abnormal signal points) occurring within the timeframe is denoted as: Then, combining the information about "high-density anomalous situations" in the profile, the time periods marked as high-density anomalous situations were statistically analyzed. The number of times abnormal signal points occur within. (here equal (Because it's statistics for a specific time period k) and that time period Total runtime of internal elevators Total runtime It is obtained by accumulating the duration of the "running segment" of all running cycles within the 30-minute time period. For example, the time period k=35 (17:00-17:30) is marked as a high-density anomalous situation, during which a total of The total runtime for a given period is obtained by accumulating the runtime of each abnormal signal point. Seconds, this information, namely the number of outliers in each time period k, will be stored. Is it a high-density scenario? And if it is a high-density scenario, what is the number of outliers? and total runtime The data is then stored in association to form an anomaly frequency-high-density context association dataset. Next, based on this dataset, calculations are performed for each time period. Number of abnormal signal points The average number of outliers compared to the same historical period (e.g., the same day of the week and the same time period in the past month). The percentage deviation is calculated as follows: Set a risk sensitivity threshold, for example, 20%. Then the time period Periods marked as risk-sensitive, for example, period k=35. The average number of outliers in the same period in history Then the deviation ,because So the period k=35 is marked as a risk sensitive period, a sequence of all period risk sensitive marks is generated, finally, based on the sequence of risk sensitive period marks and the parameters calculated before, the running environment associated risk score value of each period is calculated , the formula is applied to calculate, where is the number of abnormal signal points of the th period, is the high density abnormal situation influence factor of the th period, and the calculation formula is , that is, the number of abnormal points in the period (equal to in the high density situation) divided by the total running time of the period , for the period of non-high density abnormal situation, the value of may be 0 or other benchmark value, here for example only in the period of high density situation , , seconds, the influence factor is calculated first, and then substituted into the risk score formula:

[0127] ;

[0128] Repeat this risk score calculation for all 48 periods , and get the running environment associated risk score value of each period .

[0129] The formula and explain:

[0130] Calculate the high density abnormal situation influence factor of the th period. represents the influence factor value. is the number of abnormal signal points occurring in the period (in particular in the period defined as a high density abnormal situation). is the total running time (seconds) of the elevator in the period . This factor quantifies the abnormal signal point occurrence rate per unit running time in a particular (high density) situation.

[0131] Calculate the running environment associated risk score value of the th period. is the final risk score. is the time period Total number of abnormal signal points within is the impact factor calculated above. This formula structure is similar to calculating some trigonometric value (or related to) of vector in two-dimensional space, but a more direct understanding is that it combines two risk indicators that can differ greatly in scale (one is the number , and one is the ratio ), and is normalized by the denominator. The denominator is the Euclidean length of the vector. When and are far from each other, the score will be closer to the smaller value (relative to its own scale), playing a balancing role. It ensures that the score value is bounded (theoretically near 0 to , depending on the specific numerical value, more accurately, the value range depends on the relative size and absolute value of and , but is always within a finite range). The formula aims to combine the total amount of abnormal occurrences and the intensity of occurrence in a specific context to obtain a comprehensive risk measure.

[0132] Example: Take time period k = 35 as an example, which is determined to be a high-density abnormal situation. number of abnormal points, number of abnormal points, total running time seconds.

[0133] First, calculate the impact factor :

[0134] ;

[0135] Then calculate the risk score :

[0136] ;

[0137] ;

[0138] The calculation result is the running environment-related risk score value of time period 35, which will be used for the final state assessment.

[0139] The benefit of the formula is that by combining the total number of abnormal events ( ) and its occurrence rate in a specific high-risk situation ( ), and using a normalization method, a comprehensive and scale-controllable evaluation of the risk level of a single time period is provided, which can more comprehensively reflect the potential risk, especially after considering external environmental pressure factors.

[0140] The result is a key indicator for quantifying the risk level of each period, and the value itself directly reflects the risk level, providing a direct basis for subsequent risk level division and evaluation conclusion.

[0141] The steps for obtaining the elevator comprehensive operation state evaluation conclusion are:

[0142] Based on the operation environment associated risk score value, the operation environment associated risk score value of each statistical period is extracted, the preset attention alarm threshold is compared, the threshold level interval is defined, and the risk score data set with threshold interval is generated;

[0143] According to the risk score data set with threshold interval, the risk score value of each statistical period is traversed, if the score value exceeds the serious alarm threshold, it is marked as "emergency state", if it exceeds the moderate alarm threshold but does not reach the serious threshold, it is marked as "high risk state", if it exceeds the mild attention threshold but does not reach the moderate threshold, it is marked as "potential risk state", and the rest is marked as "normal state", and a risk level state marking set is generated;

[0144] Based on the risk level state marking set, the number of periods marked as "emergency state" or "high risk state" in the continuous 3 periods is counted, if the continuous 3 periods are high risk or contain 1 emergency state, the "immediate maintenance suggestion" conclusion is generated, and the elevator comprehensive operation state evaluation conclusion is obtained.

[0145] Specifically, based on the operation environment associated risk score value of each statistical period , first of all, a set of risk score and state level corresponding rules need to be established, which requires pre-setting the attention, moderate alarm and serious alarm thresholds, the setting of these thresholds should be based on the analysis of the actual equipment state and failure record corresponding to different risk score values in the historical data, as well as industry safety standards and expert experience, for example, the mild attention threshold , the moderate alarm threshold , and the serious alarm threshold , these thresholds define the risk score level interval: [0, 0.02) is the normal interval, [0.02, 0.05) is the potential risk interval, [0.05, 0.10) is the high risk interval, and [0.10, +∞) is the emergency state interval, compare the risk score value of all periods with these thresholds to form a risk score data set with threshold interval, then according to this risk score data set with threshold interval, traverse the risk score value of each statistical period , and perform state determination and marking: if (For example If ), then that period is marked as "emergency state"; if (For example ,satisfy If ), then it is marked as "high-risk status"; if (For example, the previously calculated) ,satisfy If ), then it is marked as "potential risk status"; if (For example If any of the 48 time periods are marked as "high-risk," then the elevator is marked as "normal." After marking all 48 time periods, a set of labels containing the risk level status of each time period is generated. Finally, based on this set of risk level status labels, a comprehensive evaluation is performed to generate a final conclusion. A sliding window is used to check the status labels of three consecutive time periods (1.5 hours). If all three time periods are marked as "high-risk," or if at least one of the three time periods is marked as "emergency" (regardless of the status of the other two time periods), then the system determines that the current elevator status requires immediate attention and intervention, generating a conclusion of "immediate maintenance recommendation." For example, if the time... The states of segments 33, 34, and 35 are "high-risk state," "emergency state," and "high-risk state," respectively. Since "emergency state" is included, "immediate maintenance recommendation" is triggered. If segments 40, 41, and 42 are all "high-risk state," "immediate maintenance recommendation" is also triggered. If the risk state combination of three consecutive segments does not meet any of the above conditions, no immediate maintenance recommendation is generated. The comprehensive operational status assessment conclusion of the elevator may be given based on the risk level of the latest segment (such as "potential risk state") or the risk trend over a period of time (such as "recent risk level has increased").

[0146] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for evaluating elevator operating status based on multi-source heterogeneous data, characterized in that, Includes the following steps: Collect elevator car vertical acceleration signal, traction machine output torque signal, traction machine output current signal, motor encoder speed signal, elevator operation log time, floor stopping record, number of runs record, and number of door opening and closing record. Align the data based on timestamps, and divide the data stream into time periods according to the elevator's start, running, and stopped states to generate synchronized multi-source operation data streams. Based on the synchronized multi-source operation data stream, the ratio sequence of instantaneous torque value to acceleration value during elevator start-stop process is calculated, the difference sequence between the motor encoder speed signal and the speed estimate calculated based on the traction machine output torque signal is monitored, rapid change points of the car vertical acceleration signal amplitude are detected, a preliminary signal deviation index set is obtained, and based on the preliminary signal deviation index set, abnormal signal points are screened and judged to obtain a set of operation feature points. Based on the synchronized multi-source operation data stream, the operation frequency of the time period is calculated to obtain the basic operation statistics. Based on the basic operation statistics, the combination of each statistical value with external conditions is compared with the weather conditions and large-scale event information to determine whether it deviates from the normal range, and a scenario-related operation mode profile is established. Based on the frequency of occurrence of abnormal signal points within the set of operational feature points and the scenario-related operational mode profile, an operational environment-related risk score is calculated. The operational environment-related risk score is then compared with a preset alarm threshold to generate a comprehensive elevator operational status assessment conclusion.

2. The elevator operation status assessment method based on multi-source heterogeneous data according to claim 1, characterized in that, The steps for acquiring the synchronized multi-source running data stream are as follows: The system collects elevator car vertical acceleration signals, traction machine output torque signals, traction machine output current signals, motor encoder speed signals, elevator operation log time, floor stop records, number of runs records, and number of door opening and closing records. It extracts the timestamp field of each signal, calculates the absolute value of the time difference between each signal timestamp and the elevator operation log time, defines the absolute value of the difference as the initial transmission delay value, and uses cubic spline interpolation to fill in the missing time points for signals with absolute differences exceeding a preset threshold, generating a multi-source signal time series data set with fully aligned timestamps. Based on the multi-source signal time series data set with fully aligned timestamps, the state switching markers in the elevator operation log time are parsed, the start time point of the start state, the duration of the running state, and the end time point of the stop state are extracted, and the data segment boundaries are divided according to the start time point and the end time point. The car vertical acceleration signal is cut into the start segment, running segment, and stop segment according to the boundaries, and the traction machine output torque signal, traction machine output current signal, and motor encoder speed signal are synchronously segmented into the corresponding data segments to generate a segmented multi-source signal dataset with state labels. Based on the segmented multi-source signal dataset with status labels, the mapping relationship between floor numbers and timestamps in the floor stop records is extracted. The floor numbers are matched to the corresponding status intervals according to the boundaries of the divided data segments. The operation count records and door opening / closing count records are merged to the end of the data segment according to the timestamps to generate a synchronized multi-source operation data stream.

3. The elevator operation status assessment method based on multi-source heterogeneous data according to claim 1, characterized in that, The steps for obtaining the preliminary signal deviation index set are as follows: Based on the synchronized multi-source operation data stream, the instantaneous values ​​of torque signal and acceleration signal during the elevator start-up and stop process are extracted, and the data segments are divided according to the time window to generate an instantaneous torque-acceleration dataset for the start-up and stop process. Based on the instantaneous torque-acceleration dataset of the start-stop process, calculate the torque-acceleration ratio at each time point. Generate a torque-acceleration ratio sequence, calculated using the following formula: ； in, For the first Torque values ​​at each time point For the first Acceleration values ​​at each time point; Based on the torque-acceleration ratio sequence, the speed signal recorded by the motor encoder and the speed estimate derived from the output torque signal of the traction machine Perform point-by-point difference calculations to obtain the difference values. Generate a sequence of rotational speed-estimated speed difference values, using the following formula: ； Based on acceleration value Calculate the acceleration difference between adjacent time points. ,in The threshold for abrupt changes in acceleration values ​​at adjacent time points is set as follows: , for The mean, for Standard deviation, marking all The time points are the points of rapid acceleration change. By integrating the torque-acceleration ratio sequence, the speed-estimated speed difference sequence, and the points of rapid acceleration change, a preliminary set of signal deviation indicators is generated.

4. The elevator operation status assessment method based on multi-source heterogeneous data according to claim 1, characterized in that, The steps for obtaining the set of running feature points are as follows: Based on the preliminary signal deviation index set, preset thresholds for the torque-acceleration ratio sequence, preset thresholds for the speed-estimated speed difference sequence, and thresholds for the amplitude of rapid acceleration change points are extracted. The time window range is defined as an interval of 250 milliseconds before and after adjacent time points, and multi-index threshold parameters and time window definition rules are generated. Based on the multi-index threshold parameters and time window definition rules, each time point in the preliminary signal deviation index set is traversed. If the torque-acceleration ratio corresponding to the current time point exceeds the preset threshold, and the speed-estimated speed difference exceeds the preset threshold within a 250-millisecond window before and after the current time point, and the amplitude of the rapid acceleration change point within the window exceeds the threshold, then the current time point is marked as a candidate abnormal signal point, and a set of multi-index synchronous over-limit time points is generated. Based on the set of time points where the multi-index synchronous over-limit is obtained, candidate abnormal signal points with adjacent time intervals of less than 300 milliseconds are merged. After merging, the time point with the largest deviation of torque-acceleration ratio in each cluster is retained as the representative point, and isolated candidate points without adjacent clusters are removed to generate a set of running feature points.

5. The elevator operation status assessment method based on multi-source heterogeneous data according to claim 1, characterized in that, The steps for obtaining the basic operational statistics are as follows: Based on the elevator operation log time in the synchronized multi-source operation data stream, the complete time axis from 0:00 to 24:00 of each day is extracted from the log. The time axis is divided into 48 continuous intervals with each 30-minute statistical period. Combined with the timestamps in the floor stop records, the number of operation records and the number of door opening and closing records in each period are counted to generate the operation-door opening and closing frequency dataset after the time period division. Based on the operation-door opening / closing frequency dataset after the time period division, calculate the operation frequency value for each time period. The frequency sequence of the time period is generated, and the calculation formula is as follows: ； in, For the first Number of runs in each time period For the first Number of times the door is opened and closed during a given time period The duration of the time period; Based on the aforementioned time period frequency sequence, the moving average and standard deviation of the frequency values ​​for six consecutive time periods are calculated. Time periods where the moving average exceeds twice the standard deviation of the historical average for the same period are marked as high-frequency operating intervals. The frequency values ​​of all time periods are integrated to generate basic operating statistics.

6. The elevator operation status assessment method based on multi-source heterogeneous data according to claim 1, characterized in that, The steps for obtaining the context-related operation mode profile are as follows: Based on the aforementioned operational statistics, the operational frequency value and high-frequency operational interval markers for each statistical period are extracted. These are then associated with rainfall and wind speed from externally input weather data, as well as the scale and number of participants from large-scale event information. The data is then aligned and matched to the corresponding statistical period by timestamp to generate an external condition-operational statistics association dataset. Based on the external conditions-operational statistics associated dataset, compare the mean and standard deviation of the operation frequency values ​​over the past three months under the same rainfall and activity scale classification conditions in the historical database. If the operation frequency value of the current period exceeds twice the standard deviation of the historical mean, then mark the current period as a statistical period data that deviates from the normal range and generate a statistical deviation label set. Based on the statistical deviation marker set, the distribution density of abnormal signal points in the set of operational feature points during large-scale scenarios is calculated. The percentage of abnormal signal points in the total number of large-scale scenarios per unit time is calculated. Large-scale scenarios refer to rainfall levels of rainstorms or typhoons and large-scale events with a scale of tens of thousands of people. If the percentage exceeds a preset threshold, it is determined to be a high-density abnormal scenario. All high-density abnormal scenarios and statistical time period data that deviate from the normal range are integrated to generate a scenario-related operational mode profile.

7. The elevator operation status assessment method based on multi-source heterogeneous data according to claim 1, characterized in that, The steps for obtaining the associated risk score of the operating environment are as follows: Based on the set of running feature points, the number of occurrences of abnormal signal points in each statistical period is extracted. Combined with the high-density abnormal situations in the scenario-related running mode profile, the ratio of the number of occurrences of abnormal signal points under high-density abnormal situations in each period to the total running time is calculated to generate an abnormal frequency-high-density scenario-related dataset. Based on the anomaly frequency-high density context association dataset, the percentage deviation between the number of anomaly signal points and the historical average number of anomaly points is calculated. Periods with a deviation exceeding 20% ​​are defined as risk-sensitive periods, and a risk-sensitive period marker sequence is generated. Based on the aforementioned risk-sensitive period marker sequence, the associated risk score of the operating environment is calculated using the following formula: ； in, Associating risk scores with the operating environment For the first The number of abnormal signal points occurring in each time period. For the first High-density abnormal situation influencing factors in each time period, here , This represents the number of abnormal points within the time period. This represents the total runtime of the time period.

8. The elevator operation status assessment method based on multi-source heterogeneous data according to claim 1, characterized in that, The steps for obtaining the comprehensive elevator operation status assessment conclusion are as follows: Based on the associated risk score of the operating environment, the associated risk score of the operating environment for each statistical period is extracted, and the threshold level range is defined by comparing it with the preset attention alarm threshold, and a risk score dataset with threshold range is generated. Based on the risk score dataset with threshold intervals, the risk score values ​​for each statistical period are iterated. If the score value exceeds the severe alarm threshold, it is marked as "emergency state". If it exceeds the moderate alarm threshold but does not reach the severe threshold, it is marked as "high risk state". If it exceeds the mild concern threshold but does not reach the moderate threshold, it is marked as "potential risk state". The rest are marked as "normal state", and a risk level status label set is generated. Based on the risk level status marker set, the number of time periods marked as "emergency state" or "high-risk state" within three consecutive time periods is counted. If all three consecutive time periods are high-risk or contain one emergency state, an "immediate maintenance recommendation" conclusion is generated, and the comprehensive elevator operation status assessment conclusion is obtained.

9. The elevator operation status assessment system based on the elevator operation status assessment method according to any one of claims 1-8, characterized in that, include: The data synchronization module collects the car's vertical acceleration signal, traction machine output torque signal, traction machine output current signal, motor encoder speed signal, operation log time, floor stopping record, number of runs record, and number of door opening and closing record. It timestamps the collected time, divides the start state, running state, and stop state into independent time periods, and combines all signals in the divided time periods in chronological order to establish a multi-source synchronous data stream. The signal offset determination module, based on multi-source synchronous data stream, performs corresponding numerical pairing of acceleration and torque signals in the start and stop states and takes a ratio sequence. It takes a numerical difference sequence of the speed estimate calculated from the motor encoder speed signal and the speed estimate calculated from the torque output signal of the traction machine. It performs amplitude increment calculation between adjacent values ​​in the sequence of the car vertical acceleration signal and marks the abrupt change points. It filters the maximum value points in the ratio sequence, the absolute deviation points in the difference sequence, and the time periods where the abrupt change points are located to obtain the offset feature signal set. The frequency statistics module, based on multi-source synchronous data streams, counts the total number of daily operations, the number of daily changes in the number of floors stopped, the duration of each operation, and the total number of doors opened and closed per day. It extracts the numerical distribution range within a 24-hour period by day, compares the difference between the distribution range and the common distribution range in the historical operation cycle, and establishes the cycle operation frequency characteristics. The scenario feature characterization module, based on the frequency characteristics of periodic operation, calls the weather information and large-scale event records of each statistical day, compares the total number of daily operations with the weather conditions, combines the number of floor changes with the existence of large-scale events, and combines the total number of daily door openings with the temperature level into intervals. It establishes three types of combination relationships: operation frequency and weather combination pattern, floor change and event combination pattern, and door opening and closing and temperature combination pattern. It marks the combination types that exceed the normal pattern interval, summarizes the frequency of occurrence of combination types within the statistical time interval, and generates a scenario combination operation map. The risk scoring module assigns a risk level to combination types with a number of offsets exceeding the standard deviation interval based on the marked signal offset points in the offset feature signal set, thus obtaining the risk distribution score result.

Images