Abnormal people flow identification method and system based on elevator sensing data

Abstract

The invention relates to the technical field of elevator systems, and particularly discloses an abnormal people flow recognition method and system based on elevator sensing data, the method comprises the steps that current time period operation data of an elevator are obtained, and the current time period operation data comprise load, door opening and closing states, operation floors, operation time, operation acceleration and operation vibration data; preprocessing the running data of the elevator in the current time period to obtain the preprocessed running data in the current time period; establishing a personnel flow baseline model based on the operation data of the elevator in the historical time period; the preprocessed operation data of the current time period are compared with a people flow baseline model so as to recognize abnormal people flow events when the elevator operates in the current time period; and the abnormal people flow events of the elevator running in the current time period are graded, and an alarm is given according to the grading result of the abnormal people flow events. According to the invention, real-time identification of behaviors such as personnel peak, gathering and abnormal use can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of elevator systems, and more particularly to an abnormal passenger flow identification method based on elevator sensor data and an abnormal passenger flow identification system based on elevator sensor data. BACKGROUND

[0002] With the rapid development of smart communities and intelligent buildings, elevator systems, as the key link of vertical transportation in buildings, bear a large amount of passenger flow transmission tasks. Traditional monitoring of passenger flow relies on video monitoring or access control systems, but there are problems: (1) strong privacy sensitivity: video monitoring or face recognition is difficult to promote in residential buildings with high privacy requirements; (2) coarse data granularity: the access control system cannot reflect the flow trend and actual congestion degree between floors; (3) unable to respond at high frequency: event triggering delay, high deployment and maintenance cost.

[0003] In contrast, the weight, door control, position sensors and other data equipped in the elevator itself have high value for passenger flow analysis under the premise of protecting user privacy. However, existing systems are mostly used for basic statistics or operation state monitoring, and do not effectively tap their potential behavior identification capabilities. SUMMARY

[0004] In view of the above problems existing in the prior art, the present application provides an abnormal passenger flow identification method and system based on elevator sensor data, which can realize real-time identification of passenger peak, gathering, abnormal use and other behaviors.

[0005] As a first aspect of the present application, an abnormal passenger flow identification method based on elevator sensor data is provided, which comprises the following steps:

[0006] Step S1: obtaining current time period operation data of an elevator, wherein the current time period operation data of the elevator at least includes load, door opening and closing state, running floor, running time, running acceleration and running vibration data;

[0007] Step S2: preprocessing the current time period operation data of the elevator to obtain preprocessed current time period operation data;

[0008] Step S3: establishing a passenger flow baseline model based on elevator historical time period operation data;

[0009] Step S4: comparing the preprocessed current time period operation data with the passenger flow baseline model to identify abnormal passenger flow events of the elevator during the current time period operation;

[0010] Step S5: grading the abnormal passenger flow event of the elevator in the current time period, and issuing an alarm according to the grading result of the abnormal passenger flow event.

[0011] Further, in the step S2, further comprising:

[0012] The current time period operation data of the elevator is standardized, denoised and time series integrated to obtain the preprocessed current time period operation data.

[0013] Further, in the step S3, further comprising:

[0014] The historical time period operation data of the elevator is obtained, and the historical time period operation data of the elevator is standardized, denoised and time series integrated to obtain the preprocessed historical time period operation data; wherein the preprocessed historical time period operation data includes date type, time window, elevator ID, running floor ID and aggregation index, the aggregation index includes frequency type index, load type index, time type index and space behavior index, the frequency type index includes the average number of elevator up / down in the historical time period, the total number of elevator door opening and closing, the standard deviation / variance of the number of times and the number of times of the number of times, the load type index includes the average load of the elevator in the historical time period, the maximum load of the elevator, the standard deviation of the load and the load quantile, the time type index includes the total running time of the elevator in the historical time period, the single running time of the elevator, the average up time of the elevator, the average down time of the elevator and the average waiting time of the elevator, and the space behavior index includes the multi-lift same floor stop frequency in the historical time period and the main service floor of the elevator.

[0015] The preprocessed historical time period operation data is grouped according to the date type, time window, elevator ID and running floor ID to obtain the historical time period aggregation index under each group;

[0016] Based on the historical time period aggregation index under each group, the personnel flow baseline model is established.

[0017] Further, after the step S3, further comprising:

[0018] The latest elevator historical time period operation data is used to regularly update the personnel flow baseline model; wherein when updating the personnel flow baseline model, the known abnormal passenger flow event in the latest elevator historical time period operation data is excluded.

[0019] Further, in the step S4, further comprising:

[0020] comparing the pre-processed current time period operation data with the personnel flow baseline model, and identifying an abnormal personnel flow event of the elevator in the current time period operation based on a comparison result; wherein the identification manner comprises a frequency threshold judgment, a spatial aggregation identification, and a load change trend analysis;

[0021] (1) Frequency threshold judgment: monitoring the number of uses of the elevator in the current time period, and comparing it with the normal number of uses in the historical same period time in the personnel flow baseline model; if the current time period is a peak period, and the number of uses of the elevator in the current time period is greater than the normal number of uses in the personnel flow baseline model, it is identified that the elevator in the current time period has a peak congestion event; if the current time period is a non-peak period, and the number of uses of the elevator in the current time period is greater than the normal number of uses in the personnel flow baseline model, it is identified that the elevator in the current time period has an abnormal personnel flow event;

[0022] (2) Spatial aggregation identification: monitoring the behavior patterns of multiple elevators on the same floor or adjacent floors to determine whether there is an abnormal personnel aggregation event;

[0023] Continuously tracking the real-time stopping floor position and the opening and closing door events of multiple elevators; in the current time period, it is counted how many elevators stop at a certain floor at the same time, or how many "multi-elevator same floor stopping" events occur;

[0024] Comparing the number of elevators that stop at the current floor at the same time in the current time period with the normal number of elevators that stop at the same floor in the historical same period time in the personnel flow baseline model, if the number of elevators that stop at the current floor at the same time in the current time period is greater than the normal number of elevators in the personnel flow baseline model, it is identified that the elevator in the current time period has an abnormal personnel aggregation event;

[0025] (3) Load change trend analysis: analyzing the instantaneous load value and the load value change trend in a period of time of the elevator to reflect the actual congestion degree and flow dynamics of personnel;

[0026] In the continuous multiple operation cycles in the current time period, the elevator load value continuously exceeds 80% of the rated load, it is identified that the elevator in the current time period has a continuous high load event;

[0027] In the current time period, the instantaneous load value of the elevator has a sudden change, it is identified that the elevator in the current time period has a load abnormal sudden change event;

[0028] In the peak period of the current time period, the number of empty running of the elevator is higher than the set value, it is identified that the elevator in the current time period has a high-frequency empty running event;

[0029] In a non-peak period of the current time period, the elevator load data is higher than the set value, and it is identified that an abnormal personnel flow event of the elevator occurs in the current time period.

[0030] Further, in the spatial aggregation identification, further comprising:

[0031] The total load, total mileage and total running time of the entire elevator system of the building in the current time period are calculated, and when the total load, total mileage or total running time is far higher than the normal threshold in the historical same period in the personnel flow baseline model, a building-level peak abnormality is identified.

[0032] Further, in the grading of the abnormal passenger flow event of the elevator in the current time period, further comprising:

[0033] The abnormal passenger flow event of the elevator in the current time period is divided into a low-risk abnormal level, a medium-risk abnormal level and a high-risk abnormal level, wherein,

[0034] The type of the abnormal passenger flow event is divided into a peak period excessive congestion / high load event, a non-peak period abnormal active event, a regional personnel abnormal aggregation event and an abnormal load behavior event,

[0035] The detailed information of the abnormal passenger flow event includes: abnormal level, event type, occurrence time, involved elevator ID, involved floor and specific value of abnormal index.

[0036] Further, in the alarm according to the grading result of the abnormal passenger flow event, further comprising:

[0037] For the abnormal passenger flow event of the low-risk abnormal level, a first response suggestion is output; wherein the first response suggestion includes log recording, dashboard display, periodic report and no immediate notification is triggered;

[0038] For the abnormal passenger flow event of the medium-risk abnormal level, a second response suggestion is output; wherein the second response suggestion includes real-time notification, highlight display, suggestion for manual confirmation, optimization scheduling and preparation of emergency;

[0039] For the abnormal passenger flow event of the high-risk abnormal level, a third response suggestion is output; wherein the third response suggestion includes emergency notification, on-site verification, emergency scheduling, broadcast notification, external support and event record and analysis.

[0040] As a second aspect of the present application, an abnormal passenger flow identification system based on elevator sensor data is provided, comprising:

[0041] An acquisition module is configured to acquire current time period operation data of an elevator, wherein the current time period operation data of the elevator at least includes load, door opening and closing state, operation floor, operation time, operation acceleration and operation vibration data;

[0042] A preprocessing module is configured to preprocess the current time period operation data of the elevator to obtain preprocessed current time period operation data;

[0043] A model establishing module is configured to establish a personnel flow baseline model based on historical time period operation data of the elevator;

[0044] A comparison and identification module is configured to compare the preprocessed current time period operation data with the personnel flow baseline model to identify an abnormal people flow event of the elevator in a current time period;

[0045] A division and alarm module is configured to grade the abnormal people flow event of the elevator in the current time period and alarm according to the grading result of the abnormal people flow event.

[0046] Further, the abnormal people flow identification system based on elevator sensor data is deployed in an elevator controller, the elevator controller outputs alarm information of the abnormal people flow event to a smart community cloud platform through an edge computing gateway, and receives feedback information of the smart community cloud platform through the edge computing gateway.

[0047] The abnormal people flow identification method based on elevator sensor data provided by the application has the following beneficial effects:

[0048] (1) Privacy-friendly: does not rely on images, voice or facial information, and fully utilizes existing sensor data for behavior modeling;

[0049] (2) High frequency and real-time: based on sliding window and stream processing algorithm, realizes instant identification of behaviors such as personnel peak, gathering and abnormal use;

[0050] (3) Clear system architecture: decouples and combines data acquisition, preprocessing, behavior modeling, abnormal identification and alarm pushing, which is convenient for edge / cloud deployment;

[0051] (4) Strong scalability: suitable for multiple building scenarios such as residential buildings, office buildings, hospitals and shopping malls;

[0052] (5) The application also supports: multi-elevator collaborative analysis, different floor gathering behavior identification, event grading and intelligent response mechanism. BRIEF DESCRIPTION OF DRAWINGS

[0053] The accompanying drawings are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. The drawings are provided solely for purposes of illustration and should not be considered as a limitation of the application.

[0054] Figure 1 A flow chart of the abnormal flow recognition method based on elevator sensor data provided by the present application.

[0055] Figure 2 An abnormal event recognition flow chart provided by the present application.

[0056] Figure 3 A structure diagram of the abnormal flow recognition system based on elevator sensor data provided by the present application.

[0057] Figure 4 A practical deployment schematic diagram of the abnormal flow recognition system based on elevator sensor data provided by the present application. DETAILED DESCRIPTION

[0058] To further clarify the technical means and effects taken by the present application to achieve the predetermined inventive purpose, the specific implementation, structure, features and effects of the abnormal flow recognition method based on elevator sensor data provided by the present application are described in detail below in combination with the drawings and preferred embodiments. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0059] It should be noted that the terms "first", "second", and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0060] In the present embodiment, an abnormal flow recognition method based on elevator sensor data is provided, as shown in Figure 1 The abnormal flow recognition method based on elevator sensor data includes the following steps:

[0061] Step S1: obtaining current time period running data of the elevator, wherein the current time period running data of the elevator at least includes load, door opening and closing state, running floor, running time, running acceleration and running vibration data;

[0062] Step S2: preprocessing the current time period running data of the elevator to obtain preprocessed current time period running data;

[0063] Preferably, in the step S2, further comprising:

[0064] The current time period running data of the elevator is standardized, denoised and time series integrated to obtain the preprocessed current time period running data. Wherein, the time series integration is a process of aligning, synchronizing and resampling discrete, different frequency or different source event data and continuous running data according to time axis to form a unified, structured time series data set. This provides a unified data view for subsequent feature engineering and pattern recognition.

[0065] In the embodiment of the present application, the standardization of the load data includes:

[0066] (1) Min-max standardization: mapping the load data to the interval [0, 1].

[0067] X norm =(X-X min ) / (X max -X min )

[0068] For example, the maximum load X max of the elevator is 1000 kg, and the minimum load X min is 0 kg. A data point X of 500 kg will be standardized as (500-0) / (1000-0) = 0.5. This allows the comparison of elevator data with different upper limits of load.

[0069] (2) Z-score standardization / mean-variance standardization: converting data to a distribution with mean 0 and standard deviation 1.

[0070] X std =(X-μ) / σ

[0071] Wherein, μ is the mean of the load data X, and σ is the standard deviation of the load data X. This method is more robust in the presence of outliers, as it does not rely on the maximum and minimum values.

[0072] In the embodiment of the present application, the number of seconds of running time can be converted to minutes or standardized to total running time percentage, or Z-score standardization can also be used.

[0073] In embodiments of the application, the door open / close status: counts the number of door open / close events in a unit of time (e.g., number of door open / close events per 5 minutes), and then performs Z-score or Min-Max normalization on this number to compare with historical data or door open / close events in different time periods.

[0074] In embodiments of the application, the running floor itself is discrete ordinal data, usually does not need to be normalized, but if the "number of floors crossed" needs to be calculated, this derived feature might need to be normalized.

[0075] In embodiments of the application, the denoising of the load data includes:

[0076] (1) Moving average: apply a sliding window average on consecutive load readings to smooth out short-term fluctuations. For example, take the average of the last 5 seconds of load readings as the current load value.

[0077] (2) Median filter: similar to moving average, but use median instead of mean, less sensitive to abnormal spikes.

[0078] (3) Outlier detection and rejection / correction:

[0079] Statistical method: use Z-score or IQR (Interquartile Range) rule to identify load values that deviate too far from the mean / median. For example, if the elevator load suddenly jumps from 200 kg to 5000 kg (far beyond the maximum load) and then back to 200 kg, it is obviously a sensor transient fault or interference, which can be rejected or filled with the previous valid value.

[0080] Rule based on domain knowledge: for example, it is stipulated that the elevator load cannot change from full load to 0 in 1 second, unless there is an abnormal event (such as a fall, but that is not noise).

[0081] In embodiments of the application, the denoising of the door open / close status includes eliminating jitter: the sensor may have multiple rapid on / off signal jumps at the moment of door opening / closing. A minimum duration threshold can be set, for example, if the door status switches frequently within 100 milliseconds, the status during this period is considered unstable, and only the final stable status is recorded.

[0082] In embodiments of the application, the denoising of the running floor includes smoothing / logical verification: the elevator cannot jump non-adjacent floors in a short time, if there is a non-logical jump such as "1st floor -> 10th floor -> 2nd floor", the non-logical data in the middle may be noise. Logical judgment or interpolation repair can be used.

[0083] In embodiments of the application, the denoising of the running time includes deduplication and sorting: ensure that the timestamps are unique and sorted in ascending order, handle out-of-order or duplicate records caused by network latency or storage system.

[0084] In embodiments of the application, the time series integration of operational data includes:

[0085] (1) Data alignment and synchronization: The load, door opening / closing events, running floors, running time of the elevator can be reported by different subsystems or sensors, with their own timestamps. They need to be associated based on timestamps. For example, at a certain time point, record the load at that time, whether the door is open, the floor where the elevator is located. Among them, if the sampling frequencies of different data are different, methods such as nearest neighbor interpolation or linear interpolation can be used for alignment.

[0086] (2) Resampling and aggregation:

[0087] (21) Sensor data can have a very high sampling frequency (such as multiple times per second). In order to facilitate analysis and pattern recognition, it is usually necessary to aggregate it into a larger time window to form more meaningful indicators.

[0088] (22) Time window selection: According to the identification target (for example, is the peak period a minute-level or hour-level pattern), select the appropriate time window, such as 1 minute, 5 minutes, 15 minutes or 1 hour.

[0089] (23) Aggregation operation: In each time window, calculate various statistical quantities:

[0090] Load: average load, maximum load, load standard deviation, load percentage (percentage of maximum capacity).

[0091] Door opening / closing status: number of door openings, number of door closings, total number of door openings / closings in the window.

[0092] Running floors: number of different floors served by the elevator in the window, average stop floor, whether concentrated on a particular floor.

[0093] Running time: total running time in the window, empty running time, loaded running time.

[0094] Derived features: for example, "passenger turnover" per minute (estimated based on load and door opening / closing frequency), "service floor range" per minute, etc.

[0095] (3) Feature engineering: Extract higher-level features from raw data and aggregated data for behavior pattern modeling. For example:

[0096] Elevator usage frequency: the number of up and down trips of the elevator in a certain time window.

[0097] Single-use load: average load per elevator run.

[0098] Time distribution: frequency of stops at a particular floor in different time periods.

[0099] Uplink / Downlink Time Difference: Used to assess elevator load pressure.

[0100] Multi-Elevator Coordination Feature: Count of events where multiple elevators at the same floor stop simultaneously.

[0101] Note that the running time refers to the total time taken for an elevator to start from a certain floor, reach the target floor, and complete the opening and closing process (or only the time spent in the shaft during the running process, the specific definition needs to be unified). The uplink time refers to the time taken for an elevator to complete a service in the upward direction. The downlink time refers to the time taken for an elevator to complete a service in the downward direction. The start time, end time, start floor, end floor, and running direction (uplink / downlink) of each elevator trip are accurately recorded to calculate the uplink / downlink time difference. Load pressure not only refers to whether the load inside the elevator reaches the upper limit, but also refers to the "workload" or "congestion level" of the elevator system as a whole in a certain direction (uplink or downlink). When the load pressure is high, the elevator needs more time to serve a longer queue of calls or complete the run under full load.

[0102] Under normal circumstances, the uplink and downlink times of an elevator should be relatively close or have a certain, predictable difference (e.g., empty uplink and empty downlink) for the same trip (i.e., across the same number of floors). However, when the pattern of personnel flow changes, this difference becomes significant:

[0103] (1) Early peak (people flowing into the building): A large number of people enter the elevator from the lobby (low floor) and mainly move to high floors. At this time, the waiting time for uplink elevators will be longer, the load will be larger, and the number of stops will increase, resulting in a significant increase in the time taken for a single uplink. Downlink elevators may be relatively empty or only carry a few passengers.

[0104] (2) Late peak (people flowing out of the building): A large number of people leave the building from high floors and mainly move to the lobby (low floor). At this time, the waiting time for downlink elevators will be longer, the load will be larger, and the number of stops will increase, resulting in a significant increase in the time taken for a single downlink. Uplink elevators may be relatively empty.

[0105] (3) Midday peak (inter-floor flow): Personnel frequently flow between different floors (such as restaurant floors, conference room floors), which may result in high loads in both uplink and downlink directions, but the direction with greater pressure depends on the specific arrangement of lunch or meetings.

[0106] (4) Abnormal gathering / emergency event: for example, an emergency event (such as a fire evacuation) occurs on a certain floor, a large number of people rush to the elevator, which may cause the running time of the elevator in a single direction (downward) to be significantly prolonged, because the elevator needs to stop frequently and is overloaded. For example, abnormal visits at night may cause the running time of the upward elevator to be abnormally prolonged during the off-peak period.

[0107] Therefore, the "upward and downward time difference" can clearly reflect in which direction the current elevator is mainly under greater load pressure, thereby indirectly evaluating the directionality and intensity of the personnel flow.

[0108] Step S3: establishing a personnel flow baseline model based on the historical time period running data of the elevator;

[0109] Preferably, in the step S3, further comprising:

[0110] The historical time period running data of the elevator is obtained, and the historical time period running data of the elevator is standardized, denoised and time series integrated to obtain preprocessed historical time period running data; wherein the preprocessed historical time period running data comprises date type, time window, elevator ID, running floor ID and aggregation index, the aggregation index comprises frequency type index, load type index, time type index and space behavior index, the frequency type index comprises elevator average upward / downward total frequency, elevator total door opening / closing frequency, frequency standard deviation / variance and frequency quantile in the historical time period, the load type index comprises elevator average load, elevator maximum load, load standard deviation and load quantile in the historical time period, the time type index comprises elevator total running time, elevator single running time, elevator average upward time, elevator average downward time and elevator average waiting time in the historical time period, and the space behavior index comprises multi-elevator same floor stopping frequency and elevator main service floor in the historical time period.

[0111] Specifically, the time window is a time period of minute level or 5-minute level, the elevator ID is a unique identifier of each elevator, the running floor ID is a floor currently served by the elevator, and the aggregation index is a key feature calculated in each time window. The elevator total door opening / closing frequency is the total number of times of opening / closing the door of the elevator in the time window, the elevator average upward / downward total frequency is the number of times of upward / downward running of the elevator in the time window, the elevator total running time is the total running seconds of the elevator in the time window, the elevator single running time is the average duration of each running of the elevator in the time window, the elevator main service floor is the floor mainly stopped by the elevator in the time window, the elevator average load is the average load value (or the percentage of the load to the rated load) of the running of the elevator in the time window, and the elevator maximum load is the load peak value (the highest value of the historical load data) of the running of the elevator in the time window.

[0112] In embodiments of the present application, the average up / down frequency is the total number of up / down trips per 5 minutes (or 10 minutes) in the time window; the average door opening / closing frequency is the total number of door opening / closing per 5 minutes in the time window; the frequency standard deviation / variance is used to measure the volatility of the frequency; the frequency quantile, such as the 90th percentile (P90) or 95th percentile (P95), means that in normal circumstances, the running frequency of the elevator has a 90% or 95% probability of not exceeding this value, which is particularly important for identifying “high-frequency” abnormalities; the load standard deviation is used to measure the volatility of the load; the load quantile, such as the P90 or P95 load value, is used to determine whether it is in a high-load state; the histogram or probability density function of the load data reflects the common distribution of the load value; the average and standard deviation of the load percentage (relative to the rated load) can better reflect the degree of crowding. The average up-trip time of the elevator is the average time taken by the elevator to complete an up-trip; the average down-trip time of the elevator is the average time taken by the elevator to complete a down-trip; the average waiting time of the elevator is the average time from the arrival of the elevator at the call point to the opening of the door to pick up passengers; the up / down trip time difference of the elevator is the difference between the average up-trip time of the elevator and the average down-trip time of the elevator, reflecting the one-way load pressure. The multi-elevator same-floor stop frequency refers to the number of times two or more elevators stop at the same floor in the time window; the main service floor of the elevator refers to the floor that the elevator stops at most frequently and the corresponding frequency in the time window.

[0113] The preprocessed historical time period operation data is grouped according to “date type”, “time window”, “elevator ID” and “operation floor ID” to obtain historical time period aggregated indicators under each group; wherein (1) data points under the same combination of “(date type, time window, elevator ID)” in the past 30 days are counted (mean, median, standard deviation, quantile, etc. are calculated). For example, the average up-trip frequency, average load and their standard deviation or upper and lower limits of the No. 1 elevator in all morning peak (7:30-7:45) periods of weekdays are calculated. (2) The probability distribution of each indicator can be estimated using methods such as kernel density estimation (KDE), thereby more flexibly defining the “normal” range. (3) A moving average or exponentially weighted moving average can be used to update the baseline, enabling the model to slowly adapt to long-term trend changes (e.g., changes in building occupancy rate).

[0114] In embodiments of the present application, (1) data points under the same combination of “(date type, time window, elevator ID)” in the past 30 days are counted, wherein the average single up-trip time μ up and its standard deviation, the average single down-trip time μ down and its standard deviation, the average up / down trip time difference μ diff (μ diff = μup -μ down ) and its standard deviation; these averages can be further refined to "average time per floor crossed" to eliminate the effect of trip length. (2) Then collect the current time window's up and down run time data for this elevator. (3) Calculate the current average up time T up,current and current average down time T down,current . Calculate the current up vs down time difference Diff current (Diff current = T up,current - T down,current ). (4) Compare Diff current to the historical baseline μ diff . If Diff current significantly deviates from μ diff (e.g., exceeds μ diff + k*σ diff ), then it is deemed abnormal. The degree of deviation of T up,current from μ up and T down,current from μ down can be compared separately. If in the morning peak, ΔT up (current up time vs baseline) is much larger than ΔT down (current down time vs baseline), it indicates a huge up direction pressure. If in the evening peak, ΔT down is much larger than ΔT up , it indicates a huge down direction pressure.

[0115] The advantage of evaluating load pressure is that: (1) Dynamic: Run time directly reflects the "effort" and "efficiency" of the elevator in completing the task, which is more dynamic than simply load data. (2) Directional: Clearly reveals the main direction of current flow and which direction's elevator dispatch is more intense. (3) Beyond load: Even if the elevator is not fully loaded, but due to too many call requests or scattered landing floors, it may also lead to prolonged run time, thus reflecting high load pressure. (4) Auxiliary dispatch: Knowing which direction has high load pressure can guide the elevator dispatch system (if integrated) to optimize strategies, such as prioritizing elevator dispatch to bottom floors to serve up requests in the morning peak, or prioritizing high floor down requests in the evening peak. (5) Identify implicit congestion: Sometimes the load is not full, but if the run time is abnormally long, it means the waiting time is long, or the number of stops is high, which also represents a kind of congestion.

[0116] By analyzing the time difference between up and down, the system can more comprehensively and meticulously evaluate the running load pressure of the elevator, thus providing a powerful and insightful dimension for "peak abnormal flow pattern recognition".

[0117] Based on the historical time period aggregated indicators under each group, the personnel flow baseline model is established.

[0118] It should be noted that the personnel flow baseline model represents a multi-dimensional, periodic "fingerprint" of personnel flow that changes over time, date, floor, and elevator individual characteristics. This "fingerprint" represents the normal range and distribution of elevator operation indicators under specific circumstances. This personnel flow baseline model will serve as a reference for subsequent identification of whether the current behavior is abnormal.

[0119] In an embodiment of the present application, the elevator operation data of the last 30 consecutive days is selected, wherein the last 30 consecutive days can cover at least 4 complete week cycles, so as to capture the typical periodic patterns within the week (workdays vs. weekends) and within the day (morning peak, noon peak, evening peak, flat peak, night, etc.). At the same time, 30 days are short enough to reflect the current recent personnel flow state of the building, avoiding being disturbed by too far away data (which may be out of date).

[0120] It should be noted that before constructing the personnel flow baseline model, preliminary outlier detection and elimination is required for the elevator operation data of the last 30 consecutive days. Because if the historical data itself contains a large number of abnormalities (for example, a fire evacuation occurred in the building on a certain day), these abnormalities will pollute the definition of "normal" patterns. Simple statistical methods (such as Z-score>3 or far away from IQR range) can be used to identify and eliminate these historical abnormal data points.

[0121] Specifically, the "date type" is divided into weekdays, weekends, holidays, and special event days. Weekdays can distinguish the patterns from Monday to Friday; weekends can distinguish the patterns from Saturday and Sunday, and personnel flow is significantly reduced during weekends; holidays are optional, if the holiday data is sufficient, it can be modeled separately, otherwise it can be classified as weekend mode; special event days are optional, such as company team building day, property activity day.

[0122] Specifically, a day of 24 hours is divided into fixed length small time periods as "time windows", for example: every 5 minutes, every 10 minutes, every 15 minutes or every 30 minutes. It is crucial to choose the appropriate "time window" granularity, too rough will lose details, too fine will increase data volume and computational complexity, and may cause sparsity problems (too little data in some windows), for example: 08:00-08:05, 08:05-08:10, …, 22:55-23:00.

[0123] Specifically, the data of the last 30 days is analyzed to identify the daily and weekly cycle patterns of personnel flow.

[0124] Daily cycle pattern: For example, morning peak (7:30-9:30), lunch peak (11:30-13:00), evening peak (17:00-19:00) on weekdays, etc. The elevator operation characteristics (such as frequency, load, directionality) during these periods will be significantly higher than other periods.

[0125] Weekly cycle pattern: There are significant differences in personnel flow patterns between weekdays (Monday to Friday) and weekends / holidays. The model needs to distinguish between these "date types".

[0126] Specifically, for complex scenarios involving multiple buildings or different functional areas (such as office areas, residential areas, commercial areas), the model needs to establish independent baselines for each building or area, or introduce buildings / areas as features of the model.

[0127] Specifically, for a system with multiple elevators, each "elevator ID" usually has its own operation characteristics and service area. Therefore, the model needs to build a baseline independently for each elevator.

[0128] Specifically, some analyses may need to model the behavior of elevators for specific "operating floor IDs" (such as lobbies, main floors of companies).

[0129] Through the above detailed construction process, the personnel flow baseline model can provide a precise and multi-dimensional "normal" behavior baseline for anomaly identification, enabling the system to effectively distinguish between normal peak flow and abnormal flow patterns.

[0130] It should be noted that the personnel flow baseline model is not a single mathematical model, but a structured set of data or rules: (1) Look-up table / database: stores the "normal" average, standard deviation, upper limit and lower limit of each indicator at different times and different date types. (2) Statistical distribution model: for each indicator, the parameterized (such as mean and variance of Gaussian distribution) or non-parametric (such as histogram, quantile list) distribution characteristics may be stored.

[0131] In addition, for a system with multiple elevators, it is also necessary to capture the coordinated operation patterns between elevators to construct the personnel flow baseline model; (1) Floor aggregation coordination: under normal circumstances, if multiple elevators stop at the same floor at the same time, it usually means that there are a large number of people on the floor who need to take the elevator. The model learns the frequency and degree of this "normal aggregation". For example, during the morning peak of an office building, multiple elevators stopping at the lobby floor (1st floor) and picking up passengers is a normal phenomenon. (2) Load balancing pattern: under normal circumstances, the elevator dispatching system will try to balance the load of each elevator. The baseline model can learn the typical difference range of load and frequency between elevators during peak hours.

[0132] Preferably, the step S3 is followed by further comprising:

[0133] Although the people flow baseline model is built based on historical data, the real people flow baseline model can change slightly over time (e.g. building occupancy rate, company moving in or out). Therefore, it is usually supported that:

[0134] The people flow baseline model is updated regularly using the latest elevator historical time period operation data to ensure its timeliness; wherein, when updating the people flow baseline model, the known abnormal people flow events in the latest elevator historical time period operation data are excluded. Specifically, through techniques such as sliding window or exponential weighted average, the people flow baseline model can slowly adapt to new "normal" patterns, but it needs to be handled carefully to prevent abnormal events from being mislearned as normal.

[0135] Step S4: comparing the preprocessed current time period operation data with the people flow baseline model to identify abnormal people flow events of the elevator in the current time period;

[0136] Preferably, as shown in Figure 2 the step S4 further comprises:

[0137] comparing the preprocessed current time period operation data with the people flow baseline model, and identifying abnormal people flow events of the elevator in the current time period based on the comparison results; wherein, the identification method includes frequency threshold judgment, spatial aggregation identification and load change trend analysis;

[0138] (1) Frequency threshold judgment: monitor the number of times the elevator is used in the current time period, and compare it with the normal number of times in the historical same period in the people flow baseline model; if the current time period is a peak period, and the number of times the elevator is used in the current time period is greater than the normal number of times in the people flow baseline model, it is identified that the elevator has a peak congestion event in the current time period; if the current time period is a non-peak period, and the number of times the elevator is used in the current time period is greater than the normal number of times in the people flow baseline model, it is identified that the elevator has an abnormal people flow event in the current time period;

[0139] Specifically, peak congestion: in the peak period, the number of times the elevator is operated (up / down) and the number of times the door is opened and closed are significantly higher than the average level of the historical same period (same date type, same time window), even exceeding the historical same period extreme value (such as 95% or 99% quantile).

[0140] Specifically, abnormal activity: in a non-peak period (e.g. late at night, early in the morning), the frequency of elevator operation is much higher than its baseline level, which implies that there may be abnormal people flow.

[0141] In embodiments of the application, from the real-time data stream, the total number of up-and-down trips or the total number of door openings of a specific elevator in the current 5 minutes (or 10 minutes) is calculated. From the baseline model, the average number of trips μ of this elevator in the current time period (e.g., 23:00-23:05 on weekdays) is obtained freq and the standard deviation σ freq , or the 95% or 99% quantile P of it 95,freq If the current frequency > μ freq +k·σ freq (where k is an adjustable parameter, such as 2 or 3), or the current frequency > P 95,freq , it is determined that the frequency is abnormal.

[0142] The frequency threshold judgment can quickly capture sudden increases or sustained high usage of elevators, which may be due to overcrowding during peak hours, sudden events (such as the end of a meeting or an event), or abnormal personnel activities during non-normal periods.

[0143] (2) Spatial aggregation recognition: Monitor the behavior patterns of multiple elevators on the same floor or adjacent floors to determine whether there is an abnormal personnel aggregation event;

[0144] Continuously track the real-time stopping floor positions and door opening / closing events of multiple elevators; In the current time period, count how many elevators are simultaneously stopped at a certain floor, or how many "multiple elevator simultaneous stopping" events occur;

[0145] Compare the number of elevators simultaneously stopped at the current floor in the current time period with the number of normal elevators simultaneously stopped at the floor in the same period in the personnel flow baseline model. If the number of elevators simultaneously stopped at the current floor in the current time period is greater than the number of normal elevators in the personnel flow baseline model, an abnormal personnel aggregation event is identified for the elevators in the current time period;

[0146] Specifically, "multiple elevator simultaneous stopping" refers to multiple elevators simultaneously or frequently stopping at the same floor during an unexpected time period, accompanied by personnel access (load change); "single floor high frequency" refers to a specific floor being frequently accessed by multiple elevators or serving as the main floor for passenger boarding and alighting during an unexpected time period, while other floors are relatively calm; "concentrated aggregation" refers to an unexpected number of people gathering at a specific floor or area. This aggregation may be normal (such as early morning peak lobby gathering) or abnormal (such as abnormal activity on a specific floor during non-peak hours).

[0147] Simply looking at the load or frequency of a single elevator may not be enough to fully capture the complex anomaly pattern of "concentrated gathering". An elevator full of people could be a few people carrying heavy loads. An elevator running frequently could be a normal peak period or someone going up and down frequently. However, when multiple elevators exhibit simultaneous, high-frequency, and load-changing stopping behavior at the same floor, it strongly suggests that there is a large amount of abnormal personnel flow demand at that floor, forming a "concentrated gathering" of personnel. Spatial gathering recognition can discover abnormal gathering of personnel in a specific area of a building (such as a floor or the floor where a conference room is located), such as unauthorized activities, large-scale personnel gathering, emergency evacuation, etc.

[0148] Therefore, the present application not only focuses on the operation of a single elevator, but also considers the entire elevator system of a building as a whole, and judges whether there is an abnormal "concentrated gathering" of personnel by analyzing their coordinated behavior in space (the same floor) and time (simultaneously).

[0149] The advantages of this coordinated judgment are: (1) high confidence: multiple independent sensors (elevators) simultaneously send signals, confirming the same phenomenon, greatly improving the accuracy and reliability of identification. (2) Spatial positioning: can accurately identify the specific floor where the personnel gathering occurs. (3) Distinguish between normal and abnormal: for example, during the morning rush hour, multiple elevators simultaneously stopping at the lobby (1st floor) is a normal phenomenon, and the system will learn this through the baseline model. But if this pattern occurs at a certain floor (especially in non-public areas such as equipment floors, storage rooms, unauthorized floors) at night or off-peak hours, it is immediately determined to be abnormal. (4) Early warning: concentrated gathering is often an early sign of some sudden events (such as unauthorized activities, initial stage of emergency evacuation, illegal intrusion).

[0150] In an embodiment of the present application, the following key data of all elevators in the building are obtained in real time: elevator ID, current floor, door opening / closing status (whether the door is open, closed, or opening / closing), timestamp (accurate to seconds), and load data (to confirm whether there is personnel entering or leaving); Step 1: At any given floor F, detect how many elevators report that they are at floor F and their doors are open or have just closed within the current time window T. Step 2: Further confirm whether the load data of these elevators has a significant increase / decrease when they stop at floor F to confirm that there is indeed personnel entering or leaving (rather than just passing by or empty elevator stopping). Step 3: Count the number of elevators N that meet the above conditions. Step 4: Compare N with the preset "multi-elevator simultaneous stay threshold" (for example, N >= 2 or N >= 3).

[0151] In embodiments of the present application, a historical baseline of "multi-elevator same floor dwell" needs to be established for each (date type, time window, floor ID) combination. This baseline will record how many times or how many elevators will simultaneously dwell on this floor in this time period on average. For example: during weekday morning rush hour (8:00-8:30), the historical baseline of the lobby (1st floor) of an office building may show that 3-5 elevators will simultaneously dwell on the 1st floor and pick up passengers every 5 minutes on average, which is normal. But if the historical baseline of a non-public floor (such as the 7th floor, usually no one works) is 0-0.1 times / hour (i.e. almost never happens) during the night (23:00-05:00), if it is detected that 2 elevators are simultaneously dwelling on the 7th floor and there is a load change, this is a high-confidence anomaly.

[0152] In embodiments of the present application, if the frequency or number of elevators of the current detected multi-elevator simultaneous dwell is significantly higher than the historical baseline of this floor in this time period (for example, exceeding the baseline P99 or a certain absolute threshold), the "concentrated gathering" anomaly event is triggered.

[0153] By combining the coordinated behavior of multiple elevators, the present application can identify more complex and representative "concentrated gathering" anomaly patterns, thereby providing more accurate and insightful building passenger flow management capabilities. (1) Accurate identification of peak congestion: during peak hours, if a large number of elevators simultaneously dwell on the lobby or main exit floor and the load is saturated, it can be accurately judged as a regional serious congestion. (2) Nighttime abnormal gathering: during non-operating hours, if the system detects that two or more elevators are simultaneously dwelling on a specific floor and there is a load change, an immediate warning can be given to prevent illegal intrusion or abnormal activities. (3) Emergency response: in the event of a fire or earthquake, if the elevator data stream shows that multiple elevators are simultaneously dwelling on a high floor, it may indicate that personnel are waiting on that floor for evacuation, or that the elevator dispatching system has failed, requiring immediate manual intervention. (4) Building activity monitoring: if a floor that usually has little passenger flow (such as the floor where the conference room or activity center is located) suddenly has multiple elevators simultaneously dwell at a certain time period, it may indicate that a large-scale activity is taking place or ending, helping the property to adjust the service in a timely manner.

[0154] Preferably, in the spatial gathering identification, further comprising:

[0155] The total load, total running mileage and total running time of the entire elevator system of the building in the current time period are calculated, and when the total load, total running mileage or total running time far exceeds the normal threshold in the historical same period of time in the personnel flow baseline model, a building-level peak anomaly is identified.

[0156] Anomalies of a single elevator can be only local issues, or not even real anomalies (e.g. a temporary malfunction of a certain elevator, causing other elevators to be more loaded). With the help of coordination, the system can: (1) improve the accuracy of identification: reduce false positives and false negatives, distinguish individual elevator anomalies from overall building passenger flow anomalies. (2) identify more complex anomaly patterns: some anomaly patterns can only be manifested in the coordinated behavior of multiple elevators (e.g. concentrated gathering). (3) provide macroscopic situation: from "single point" problems to "global" problems, providing managers with more decision-making valuable information.

[0157] Specifically, (1) simultaneous stop detection: count how many elevators stop at a certain floor (especially the lobby, or a certain meeting / activity floor) at the same time within a certain time window (e.g. 3 minutes). (2) high-frequency access: if a certain floor is accessed by multiple elevators at a high frequency during off-peak hours, and accompanied by changes in load, it may indicate abnormal gathering. Example: at night, multiple elevators frequently and for a long time stop at a non-residential or non-office floor (such as equipment layer, storage layer), and the load is not empty, which may indicate illegal intrusion or suspicious activity. While during the day, if multiple elevators simultaneously serve the same floor at an unexpected time (such as off-peak hours) and the load is high, it may be an early sign of an emergency meeting or emergency evacuation.

[0158] It should be understood that (1) if a group of elevators (e.g. all up-going elevators) simultaneously experience a surge in load at a certain point in time, it is much more indicative of the congestion of the lobby area than the full load of a single elevator. (2) During the evening peak, all down-going elevators are running at a very high frequency and high load, which is normal. But if all elevators suddenly start running at a high frequency during off-peak hours, it is worth paying attention to. (3) The system will evaluate the load distribution between elevators. If one of the elevators is running at a low load for a long time, while other elevators are consistently overloaded, it may indicate a problem with the dispatching system or an abnormal passenger flow in a specific area.

[0159] It should be noted that the building-level people flow dynamic trend map is constructed, wherein the building-level people flow dynamic trend map comprises: (1) overall load and congestion index: showing the average load percentage of the entire building current elevator system, and the congestion index compared with the historical baseline. For example, it is displayed as "building overall congestion: high (30% higher than the baseline)". It can be a color-coded dashboard or progress bar. (2) Real-time hot floor / area: visually highlight the floors with the most frequent elevator activity, the most concentrated personnel, or the most densely packed elevator stops. For example, in the floor diagram, if there are a large number of elevators simultaneously stopping at the 3rd and 4th floors, and the load is large, these two floors will be marked as "high activity" or "abnormal gathering point". (3) Main people flow direction and intensity: indicating whether the current people flow is mainly upward or downward, and its intensity. For example, use stream arrows or numerical values to represent "current main upward flow, intensity 80% (peak)". This helps managers understand whether personnel are entering, leaving, or moving horizontally within the building. (4) Overview of abnormal events: summarize the current building-level abnormal events (such as "overall peak congestion", "abnormal gathering on X floor at night"), and show their level and duration. Provide quick links to view detailed information of these abnormalities. (5) Elevator group running status: display the overall running status of each elevator group (e.g. office area elevators, residential area elevators), including average waiting time (if estimable), average empty rate, etc., to help evaluate elevator dispatching efficiency.

[0160] Through "coordinated judgment of multiple elevator data" and "construction of building-level people flow dynamic trend map", the system realizes intelligent upgrading from "single point" to "global". This not only can more accurately and comprehensively identify potential abnormal flow patterns, but also can provide managers of property, security, operation, etc. with: (1) Global view: real-time understanding of the overall building personnel flow health status. (2) Precise early warning: timely discovery of large-scale gathering, sudden congestion and other building-level risks, rather than just single elevator problems. (3) Efficient decision-making: assisting managers to quickly locate problem areas and take targeted measures (such as increasing security, guiding people flow, adjusting elevator dispatching strategy), thereby improving building management efficiency and safety. (4) Emergency response: in emergency situations (such as fire evacuation), the building-level people flow trend map can quickly provide information on personnel distribution and evacuation bottlenecks to assist command.

[0161] (3) Load change trend analysis: analyzing the instantaneous load value and load value change trend in a period of time to reflect the actual congestion degree and flow dynamics of personnel;

[0162] In the current time period, the instantaneous load value of the elevator has a sudden change (abnormal, large instantaneous increase or decrease), and it is identified that the elevator has a load abnormal mutation event in the current time period; this may be related to the rapid influx or evacuation of a large number of people;

[0163] In the current time period, the instantaneous load value of the elevator has a sudden change (abnormal, large instantaneous increase or decrease), and it is identified that the elevator has a load abnormal mutation event in the current time period; this may be related to the rapid influx or evacuation of a large number of people;

[0164] In the peak period of the current time period, the number of empty running of the elevator is higher than the set value, and it is identified that the elevator has a high-frequency empty running event in the current time period; this indicates that the scheduling system is inefficient or has a fault.

[0165] In the non-peak period of the current time period, the load data of the elevator is higher than the set value, even close to full load, and it is identified that the elevator has an abnormal passenger flow event in the current time period. This implies that a large number of people travel at an abnormal time period.

[0166] In the embodiment of the present application, the average or median of the elevator load in the past N minutes is calculated, and compared with the baseline. A threshold of the percentage of load to rated load (for example, 90%) is set, and when it exceeds this threshold for X consecutive minutes, an alarm is triggered. The rate of change of load in unit time is calculated Detecting abnormally steep rises or falls.

[0167] Load change trend analysis can accurately assess the congestion and load pressure of the elevator, and find the real "peak" or "abnormal flow", not just simple flow fluctuations.

[0168] Specifically, as shown in the figure, receive elevator real-time data → data standardization and feature extraction → compare with historical baseline model → judge whether it exceeds the frequency / load threshold? - If yes: record abnormal event segment → send alarm + write log - If not: continue to monitor the next period. Figure 2 Step S5: grading the abnormal passenger flow event of the elevator running in the current time period, and alarming according to the grading result of the abnormal passenger flow event.

[0169] Preferably, the grading of the abnormal passenger flow event of the elevator running in the current time period further comprises:

[0170] dividing the abnormal passenger flow event of the elevator running in the current time period into a low-risk abnormal level, a medium-risk abnormal level and a high-risk abnormal level, wherein,

[0171]

[0172] ​Low-risk anomaly: (1) Only slightly higher frequency than baseline, but normal load. (2) Short duration of anomaly, or small scope of impact (e.g. only one elevator briefly overloading). (3) May be a trend signal in normal fluctuations, rather than an immediate threat. For example, an elevator has a slightly higher frequency than average (e.g. between P75-P85 of baseline) during a flat peak period, but normal load. For example, an infrequent, multi-elevator simultaneous landing on a floor occurs during a night period, but with low load.

[0173] Medium-risk anomaly: (1) High frequency, significantly deviating from historical baseline, reaching a pre-set warning threshold (e.g. exceeding P90 or P95 of baseline), and slightly higher load, but not sustained. (2) Longer duration of anomaly, or moderate scope of impact (e.g. sustained high load on a group of elevators). (3) May lead to decreased service efficiency, poor user experience, or indicate a potential, attention- demanding situation. For example, during an expected peak period, the average load of several elevators consistently exceeds 80% of rated capacity for more than 5 minutes. For example, during a non-peak period (e.g. night), the frequency of elevators in a building consistently significantly exceeds baseline, but load is not necessarily high (possibly due to multiple people entering and exiting frequently). For example, after a meeting on a conference floor, multiple elevators stay on that floor for a long time, but evacuate slowly.

[0174] High-risk anomaly: Consistently high frequency AND consistently high load AND occurring during non-normal time AND accompanied by multiple elevators simultaneously stopping on the same floor. Or, during a night period, an unexpected spatial concentration occurs, even if frequency and load are not particularly high, it can be marked as high risk. High risk may directly threaten personal safety, cause significant efficiency loss, or indicate illegal / emergency events. For example, multiple elevators are in an extremely crowded state (load close to or exceeding rated capacity) for a long time (e.g. more than 10 minutes), causing passenger safety risks. For example, during a night period (late night or early morning), elevators in a building have high frequency and high load, accompanied by multiple elevators stopping on the same floor, suggesting an abnormal concentration of people at night. For example, all elevators simultaneously run upwards / downwards with extremely high frequency and load, which occurs during a non-fire alarm or evacuation drill period, possibly indicating a sudden event. For example, elevators stay on an abnormal floor for a long time and have an abnormal load, or frequently move on a floor that should not be stopped (e.g. equipment floor, top floor machine room).

[0175] Through this multi-dimensional cross-judgment, false positives can be significantly reduced, improving the accuracy and reliability of anomaly identification, thereby more effectively supporting management decisions.

[0176] It should be understood that when the operating state data of the elevator (such as the load, frequency, floor stop situation, etc.) deviates significantly from the historical normal mode of the elevator in the same time period within a certain time period, it is identified as an "abnormal event". These "abnormal events" are no longer original, discrete sensor data points, but are high-level alarm information with clear meaning after intelligent analysis and context understanding. Specifically, a "abnormal event" can include but is not limited to the following types, which directly correspond to the judgment logic of "abnormal identification":

[0177] The types of the abnormal passenger flow event include peak over-crowding / high-load event, non-peak abnormal activity event, regional personnel abnormal gathering event, and abnormal load behavior event;

[0178] (1) Peak over-crowding / high-load event: frequency threshold judgment (operation frequency far exceeds baseline) + load change trend analysis (continuous high load, long time close to or reach full load). Among them, in the expected peak period (such as morning and evening peak), the carrying capacity and operation frequency of the elevator far exceed the historical normal level, indicating that the personnel are crowded, which may cause long waiting time, low service efficiency, and even potential safety hazards.

[0179] (2) Non-peak abnormal activity event: frequency threshold judgment (non-peak period, such as night or early morning, abnormal increase of elevator operation frequency). Among them, in the time period when personnel flow is usually sparse, the elevator is frequently operated, which may indicate unauthorized entry, illegal gathering, goods moving, or safety conditions that need attention.

[0180] (3) Regional personnel abnormal gathering event: spatial gathering recognition (multiple elevators appear abnormal, high-frequency stop or long stay at the same floor or adjacent floors, and accompanied by load change). Among them, a large number of people concentrate in a certain floor or area of the building, which may be related to sudden events (such as fire evacuation), unauthorized gatherings, or some form of "occupation" behavior.

[0181] (4) Abnormal load behavior event: load change trend analysis (elevator load suddenly fluctuates greatly under unexpected circumstances, is lower than normal load for a long time, or shows frequent operation when empty, etc.). Among them, it may indicate large-scale goods moving (exceeding normal load) or problems with the elevator dispatching system (high empty rate).

[0182] The detailed information of the abnormal passenger flow event includes: abnormal level (low, medium, high risk), event type (such as "peak congestion", "night abnormal gathering"), occurrence time (start time of abnormal event), duration (how long the abnormal state lasts), elevator ID involved (which elevator or elevators are involved in the abnormality), floor involved (if it is a spatial gathering, which floor), specific value of abnormal index (such as current frequency, average load, etc.), baseline reference value (corresponding normal historical baseline value), and processing status (initially "unprocessed", and then can be updated to "viewed", "solved", etc.).

[0183] Preferably, in the alerting according to the classification result of the abnormal passenger flow event, the alerting further includes:

[0184] For the abnormal passenger flow event of the low-risk abnormal level, a first response suggestion is output; wherein the first response suggestion includes log recording, dashboard display, periodic report, and no immediate notification is triggered;

[0185] It should be noted that (1) log recording: detailed record of the time, type, involved elevator / floor, and current index of the event occurrence. (2) Dashboard display: displayed on the dashboard of the management terminal in a non-intrusive manner (such as low saturation color), without triggering immediate alarm. (3) Periodic report: such events are included in daily / weekly operation reports for management personnel to review regularly. (4) No immediate notification is triggered: usually no notification is pushed to key personnel immediately to avoid "alarm fatigue".

[0186] For the abnormal passenger flow event of the medium-risk abnormal level, a second response suggestion is output; wherein the second response suggestion includes real-time notification, highlighting, suggested manual confirmation, optimized scheduling, and emergency preparation;

[0187] It should be noted that (1) Real-time notification: push notifications to property operation personnel, elevator maintenance personnel, or floor managers through SMS, email, or management APP. (2) Highlighting: highlight the abnormal area and elevator on the building situation map of the management terminal (such as bright yellow alarm). (3) Suggested manual confirmation: suggest relevant personnel to verify through existing means (such as video monitoring if allowed and necessary; or patrol personnel). (4) Optimized scheduling: if the system integrates elevator scheduling, it can suggest temporary adjustment of elevator scheduling strategy (for example, prioritize handling of crowded floors, increase elevator allocation during peak hours). (5) Emergency preparation: remind relevant personnel to pay attention to the development of the situation and prepare for possible escalation.

[0188] For the abnormal passenger flow event of the high-risk abnormal level, a third response suggestion is output; wherein the third response suggestion includes emergency notification, on-site verification, emergency scheduling, broadcast notification, external support, and event record and analysis.

[0189] Note that, (1) Emergency notification: immediately notify the security team, building manager, emergency response team through multiple channels (phone call, high-priority APP push, sound and light alarm). (2) On-site verification: immediately send security personnel to the scene to verify the situation. (3) Emergency dispatch: If there is a central dispatch system, forcibly adjust the elevator dispatch strategy (such as sending all elevators to the lobby for evacuation, limiting access to some floors, and forcibly shutting down some elevators to relieve pressure). (4) Broadcast notification: Suggest issuing a building broadcast to guide personnel evacuation, use stairs or inform of precautions. (5) External support: In extreme cases (such as persistent heavy congestion, suspected illegal activities), suggest notifying local police, fire department or medical institutions. (6) Event record and analysis: Record all alarm information and disposal process in detail as the basis for post-mortem and system optimization.

[0190] In the embodiments of the present application, the alarm implementation includes: (1) Configurable threshold: The threshold of alarm level should be configurable, allowing managers to adjust according to building characteristics and tolerance. (2) Notification channel: The module should support multiple notification channels, including SMS, email, mobile APP push, desktop client pop-up, sound and light alarm, etc. (3) Log system: Complete event log system, recording the trigger time, type, level, detailed data, processing status, processing personnel, etc. of each abnormal event. (4) Visual presentation: Provide intuitive alarm list, historical trend chart and abnormal point marking on building floor plan on the management terminal.

[0191] The present application classifies abnormal crowd events according to their severity, impact range and duration, and provides corresponding response suggestions for different levels, and finally pushes them to managers in real time, and records detailed logs. This greatly improves the practicality and intelligence level of the system, and converts raw abnormal data into actionable and meaningful decision-making information.

[0192] Note that when an abnormal crowd event is identified, it does not immediately give a single "abnormal" or "normal" judgment, but rather scores or evaluates according to one or more of the following dimensions, and finally combines these scores to determine the final level of the warning (e.g. prompt, warning, alarm).

[0193] (1) Event frequency dimension: refers to the number of times or density of the same type or similar characteristics of abnormal events occurring within a certain observation time window. The event frequency dimension is calculated by counting method and density analysis;

[0194] Counting method: In a sliding time window (such as the last 10 minutes, 1 hour), count the number of times a certain type of abnormality (such as "a certain elevator overload") is triggered.

[0195] Density Analysis: The frequency or percentage of abnormal indicators (such as running frequency, load exceeding) of a certain elevator or floor exceeding the threshold value within a certain time period.

[0196] For example, low frequency: a certain elevator has only one short load exceeding in the past 2 hours, which may be classified as "prompt level". Medium frequency: a certain elevator has 3 consecutive load exceedings in the past 15 minutes, which may be classified as "warning level". High frequency: a certain elevator has load exceeding every minute in the past 5 minutes, and each exceeding is close to the threshold. It may be immediately classified as "alarm level".

[0197] (2) Duration dimension: refers to the length of time from the start of the abnormal event being identified to the recovery of the normal state. Among them, the duration dimension is calculated by timer and threshold judgment;

[0198] Timer: start timing from the first identification of the abnormality, until all related indicators return to the normal baseline range.

[0199] Threshold judgment: set different levels of duration threshold.

[0200] For example, short duration: the load exceeding of a certain elevator lasted less than 30 seconds and returned to normal, which may be classified as "prompt level". Medium duration: the elevator congestion in a certain area lasted for 5 minutes, which may be classified as "warning level". Long duration: the entire building has abnormally high load and high frequency of operation during off-peak hours, which lasts for more than 15 minutes, which may be classified as "alarm level".

[0201] (3) Impact floor number dimension: refers to the range or number of floors directly or indirectly affected by the abnormal event. Among them, the impact floor number dimension is calculated by counting method and region division;

[0202] Counting method: count how many floors are identified as abnormal active, abnormal aggregation or elevator frequent stop in the current abnormal event.

[0203] Region division: can be further refined to which region is affected (such as only residential area, only office area, or cross-region).

[0204] For example, single floor impact: only a conference room floor has elevator aggregation after the meeting ends, but other floors are normal, which may be classified as "warning level". Multi-floor impact: during the morning peak of the entire office building, all floors from the lobby to the 10th floor have high load on the elevator and crowded people flow, which may be classified as "alarm level". Whole building impact: fire evacuation occurs, all elevator data shows abnormal downflow, and involves all floors, which is the highest level of "alarm level".

[0205] The present application will assign weights to each dimension, or use a decision matrix / rule set to integrate the final alarm level. The rule engine is as follows:

[0206] If (event frequency = high) AND (duration = long) AND (number of affected floors = many) -> high-risk alarm;

[0207] If (event frequency = medium) OR (duration = medium) -> medium-risk warning;

[0208] If (event frequency = low) AND (duration = short) AND (number of affected floors = single) -> low-risk prompt;

[0209] The present application can also assign a score to each degree (such as low, medium, high) in each dimension, then weight and sum all the scores of the dimensions to get a total risk score, and then map the score range to different alarm levels.

[0210] The significance and value of the warning result grading in the present application are as follows: (1) priority management: helping managers distinguish between urgent and non-urgent among numerous alarms and prioritize high-risk events; (2) resource allocation: different levels of alarms correspond to different response resources and teams (for example, the prompt level may only require automatic recording by the system, the warning level notifies the on-duty personnel, and the alarm level requires the security team to respond urgently); (3) reducing alarm fatigue: avoiding sending notifications of the same emergency level for all abnormalities, reducing false positives and unnecessary interference; (4) optimizing decision-making: the alarm level and its underlying dimension information provide a more comprehensive context for managers, helping them understand the nature and severity of the event more quickly and accurately, and thus make more effective response decisions.

[0211] Through detailed analysis and intelligent grading of the three key dimensions, the present application makes the alarm system more intelligent and user-friendly, and better supports the safe operation and management of the building.

[0212] As another embodiment of the present application, an abnormal passenger flow recognition system based on elevator sensor data is provided, as shown in Figure 3 The abnormal passenger flow recognition system based on elevator sensor data includes:

[0213] An acquisition module is configured to acquire current time period operation data of an elevator, wherein the current time period operation data of the elevator at least includes load, door opening and closing state, running floor, running time, running acceleration, and running vibration data;

[0214] A preprocessing module is configured to preprocess the current time period operation data of the elevator to obtain preprocessed current time period operation data;

[0215] Model building module, used to build a baseline model of personnel flow based on elevator operation data over historical time periods;

[0216] a comparison and identification module, configured to compare the pre-processed operation data of the current time period with the personnel flow baseline model to identify abnormal personnel flow events when the elevator is running in the current time period;

[0217] The classification and alarm module is used to classify abnormal crowd flow events when the elevator is running in the current time period and to issue an alarm based on the classification results of the abnormal crowd flow events.

[0218] Preferably, if Figure 4 As shown, the abnormal crowd flow identification system based on elevator sensor data is deployed in the elevator controller. The elevator controller outputs the alarm information of abnormal crowd flow events to the smart community cloud platform through the edge computing gateway, and receives feedback information from the smart community cloud platform through the edge computing gateway.

[0219] Typical Case 1: Deployment in a Single Office Building with Multiple Elevators

[0220] In a large 20-story office building, there are 30 to 60 employees on each floor, and a large number of people use the elevators during peak hours in the morning and evening. The deployment is as follows:

[0221] (1) Collect the following information in all elevator controllers: load value (kg) at each door opening, door opening / closing event timestamp (seconds), elevator destination floor (integer), and single up / down trip duration (seconds);

[0222] (2) Deploy using edge computing boxes: TensorFlow Lite model + Python data service + SQLite lightweight database;

[0223] (3) Identification process: Pre-build a baseline model of peak personnel flow during the working hours of 7:30-9:30 and 17:00-19:00 in the past 30 working days; perform a sliding window comparison between the peak real-time data stream and the baseline model of peak personnel flow. If the following behaviors occur, a "high risk alarm" is triggered:

[0224] A single elevator runs more than 10 times per minute (up / down trips). This is a hard-coded threshold, but combined with the baseline model, it can be adjusted to "the number of runs per minute is significantly higher than the baseline P95 of the elevator during the period."

[0225] Average load > 90% for 3 consecutive minutes;

[0226] The frequency of multiple elevators stopping at the same floor has increased significantly.

[0227] Typical case 2: abnormal gathering at night in a residential community

[0228] Deployed in a certain high-end residential community building to prevent abnormal visits, gatherings, or illegal entry at night.

[0229] (1) Data source: only use the opening and closing door frequency and load change of the building elevator, do not collect images;

[0230] (2) Pre-construct night personnel flow baseline model parameters: normal opening frequency mean <1 times / hour at night (23:00-05:00);

[0231] (3) Identification process: compare the real-time data flow at night with the night personnel flow baseline model in a sliding window, and if the following behaviors occur, issue a "building gathering at night" event alarm and push it to the property / neighborhood management platform:

[0232] The number of up and down runs of a certain elevator in 1 hour is ≥8 and the average load is >2 people;

[0233] The CO2 sensor concentration in the community changes suddenly by >100ppm.

[0234] The abnormal people flow identification method based on elevator sensor data provided by the embodiment of the present application, (1) privacy friendly: does not rely on images, voices or facial information, and fully utilizes existing sensor data for behavior modeling; (2) high frequency and real-time: based on sliding window and stream processing algorithm, realizes instant identification of behaviors such as personnel peak, gathering, abnormal use, etc.; (3) clear system architecture: decouples and combines data acquisition, preprocessing, behavior modeling, anomaly identification, alarm pushing, which is convenient for edge / cloud deployment; (4) strong scalability: suitable for multiple building scenarios such as residential buildings, office buildings, hospitals and shopping malls; (5) the present application also supports: multi-elevator collaborative analysis, different floor gathering behavior identification, event level division and intelligent response mechanism.

[0235] The above is only a preferred embodiment of the present application, and is not intended to limit the present application in any form. Although the present application has been disclosed as above with a preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content without departing from the scope of the technical solution of the present application, and any simple modification, equivalent change and modification of the above embodiments based on the technical essence of the present application are still within the scope of the technical solution of the present application.

[0236] It should be noted that the skilled in the art can understand that the method and system of the present application are not limited to the sensor data mentioned above. With the development of technology, future elevators may integrate more types of sensors (such as audio sensors, vibration sensors, infrared sensors, etc.), and the core idea of the present application is also applicable to the fusion of these new sensor data to establish a more rich-dimensional personnel flow baseline model and identify more complex abnormal behavior patterns, and these extensions shall fall within the protection scope of the present application.

Claims

1. A method for identifying abnormal crowd flow based on elevator sensor data, characterized in that: The abnormal crowd flow identification method based on elevator sensor data comprises the following steps: Step S1: Acquire the elevator's operating data for the current time period, wherein the elevator's operating data for the current time period at least includes load, door opening and closing status, operating floors, operating time, operating acceleration, and operating vibration data; Step S2: preprocessing the elevator's operating data for the current time period to obtain preprocessed operating data for the current time period; Step S3: Establishing a personnel flow baseline model based on the elevator operation data of the historical time period; Step S4: comparing the pre-processed operation data of the current time period with the personnel flow baseline model to identify abnormal personnel flow events when the elevator is running in the current time period; Step S5: classifying abnormal crowd flow events when the elevator is running in the current time period into different levels, and issuing an alarm based on the classification results of the abnormal crowd flow events.

2. The abnormal crowd flow identification method based on elevator sensor data according to claim 1 is characterized in that: The step S2 further includes: The current time period operation data of the elevator is standardized, denoised and time series integrated to obtain pre-processed current time period operation data.

3. The abnormal crowd flow identification method based on elevator sensor data according to claim 1 is characterized in that: The step S3 further includes: Obtain the historical time period operation data of the elevator, and perform standardization, denoising and time series integration on the historical time period operation data of the elevator to obtain the pre-processed historical time period operation data; wherein the pre-processed historical time period operation data includes date type, time window, elevator ID, operation floor ID and aggregation index, the aggregation index includes number index, load index, time index and spatial behavior index, the number index includes the average total number of elevator up / down times, the total number of elevator door opening and closing times, the number standard deviation / variance and the number quantile in the historical time period, the load index includes the average elevator load, the maximum elevator load, the load standard deviation and the load quantile in the historical time period, the time index includes the total elevator operation time, the single elevator operation time, the average elevator up time, the average elevator down time and the average elevator waiting time in the historical time period, the spatial behavior index includes the frequency of multiple elevators stopping at the same floor and the main service floors of the elevator in the historical time period; Group the pre-processed historical time period operation data according to "date type", "time window", "elevator ID" and "operating floor ID" to obtain the historical time period aggregation index under each group; Based on the historical time period aggregation indicators under each group, the personnel flow baseline model is established.

4. The abnormal crowd flow identification method based on elevator sensor data according to claim 1 is characterized in that: After step S3, the method further includes: The personnel flow baseline model is regularly updated using the latest elevator operation data of a historical time period; wherein, when updating the personnel flow baseline model, known abnormal personnel flow events in the latest elevator operation data of a historical time period are eliminated.

5. The abnormal crowd flow identification method based on elevator sensor data according to claim 1 is characterized in that: The step S4 further includes: Comparing the pre-processed current time period operation data with the personnel flow baseline model, and identifying abnormal personnel flow events when the elevator is operating in the current time period based on the comparison result; wherein the identification method includes frequency threshold judgment, spatial clustering identification, and load change trend analysis; (1) Frequency threshold judgment: monitor the number of times the elevator is used in the current time period and compare it with the normal number of times used in the same historical time period in the personnel flow baseline model; if the current time period is a peak period and the number of times the elevator is used in the current time period is greater than the normal number of times used in the personnel flow baseline model, then it is identified that the elevator has a peak congestion event in the current time period; if the current time period is a non-peak period and the number of times the elevator is used in the current time period is greater than the normal number of times used in the personnel flow baseline model, then it is identified that the elevator has an abnormal personnel flow event in the current time period; (2) Spatial gathering identification: monitor the behavior patterns of multiple elevators on the same floor or adjacent floors to determine whether there is an abnormal gathering of people; Continuously track the real-time stop locations and door opening and closing events of multiple elevators; within the current time period, count how many elevators stop at a certain floor at the same time, or how many "multiple elevators stop at the same floor" events occur; Compare the number of elevators that stop at the current floor simultaneously during the current time period with the normal number of elevators that stop at the same floor simultaneously during the same historical time period in the personnel flow baseline model. If the number of elevators that stop at the current floor simultaneously during the current time period is greater than the normal number of elevators in the personnel flow baseline model, it is identified that an abnormal crowd gathering event has occurred in the elevator during the current time period. (3) Load change trend analysis: Analyze the instantaneous load value of the elevator and the load value change trend over a period of time to reflect the actual crowding level and flow dynamics of personnel; If the elevator load value exceeds 80% of the rated load for multiple consecutive operating cycles in the current time period, it is recognized that the elevator has experienced a continuous high load event in the current time period; In the current time period, the instantaneous load value of the elevator changes suddenly, and an abnormal load change event of the elevator in the current time period is identified; During the peak period of the current time period, the number of elevator no-load runs is higher than the set value, identifying high-frequency no-load run events in the current time period; During the off-peak period of the current time period, the elevator load data is higher than the set value, and an abnormal personnel flow event is identified in the elevator during the current time period.

6. The abnormal crowd flow identification method based on elevator sensor data according to claim 5 is characterized in that: The spatial cluster identification further includes: The total load, total mileage, and total operating time of the entire elevator system of the building in the current time period are calculated. When the total load, total mileage, or total operating time far exceeds the normal threshold value in the historical period in the personnel flow baseline model, a building-level peak anomaly is identified.

7. The abnormal crowd flow identification method based on elevator sensor data according to claim 1 is characterized in that: The step of classifying the abnormal passenger flow events when the elevator is running in the current time period further includes: The abnormal passenger flow events when the elevator is running in the current time period are divided into low-risk abnormality level, medium-risk abnormality level and high-risk abnormality level, wherein, The types of abnormal crowd flow events are divided into excessive congestion / high load events during peak hours, abnormal activity events during non-peak hours, abnormal regional gathering of people events and abnormal loading behavior events. The detailed information of the abnormal crowd flow event includes: abnormal level, event type, occurrence time, elevator ID involved, floor involved and specific values ​​of abnormal indicators.

8. The abnormal crowd flow identification method based on elevator sensor data according to claim 7 is characterized in that: The step of issuing an alarm based on the classification result of the abnormal crowd flow event further includes: For abnormal crowd flow events with low risk abnormality level, output first response suggestions; wherein the first response suggestions include logging, dashboard display, regular reporting and not triggering immediate notification; For abnormal crowd flow events with a medium-risk abnormality level, output a second response suggestion; wherein the second response suggestion includes real-time notification, highlighting, manual confirmation suggestion, optimized scheduling, and emergency preparation; For abnormal crowd flow events with high-risk abnormality levels, a third response suggestion is output; wherein, the third response suggestion includes emergency notification, on-site verification, emergency dispatch, broadcast notification, external support, and event recording and analysis.

9. A system for identifying abnormal crowd flow based on elevator sensor data, used to implement the method for identifying abnormal crowd flow based on elevator sensor data according to any one of claims 1 to 8, characterized in that: The abnormal crowd flow recognition system based on elevator sensor data includes: An acquisition module is used to acquire the elevator's operating data for the current time period, wherein the elevator's operating data for the current time period at least includes load, door opening and closing status, operating floors, operating time, operating acceleration, and operating vibration data; A preprocessing module, configured to preprocess the elevator's operating data for the current time period to obtain preprocessed operating data for the current time period; Model building module, used to build a baseline model of personnel flow based on elevator operation data over historical time periods; a comparison and identification module, configured to compare the pre-processed operation data of the current time period with the personnel flow baseline model to identify abnormal personnel flow events when the elevator is running in the current time period; The classification and alarm module is used to classify abnormal crowd flow events when the elevator is running in the current time period and to issue an alarm based on the classification results of the abnormal crowd flow events.

10. The abnormal crowd flow recognition system based on elevator sensor data according to claim 9 is characterized in that: The abnormal crowd flow identification system based on elevator sensor data is deployed in the elevator controller. The elevator controller outputs the alarm information of abnormal crowd flow events to the smart community cloud platform through the edge computing gateway, and receives feedback information from the smart community cloud platform through the edge computing gateway.

Images