An elevator stopping discrimination method and device for heterogeneous multi-modal data

Abstract

The application provides an elevator stopping discrimination method and device for heterogeneous multi-modal data, and belongs to the field of elevator intelligent monitoring. The method comprises the following steps: obtaining initial operation data of all elevators in the same place; judging the relationship between the stopping position of the elevator to be discriminated and the non-level position based on the position data in the initial operation data; if not in the non-level position, calculating the ratio of the operation frequency of the elevator to be discriminated to the average operation frequency of other elevators in the same place; performing first abnormality judgment based on the ratio; calculating the basic probability of stopping caused by each fault, identifying the change trend of the fault mode; adjusting the basic probability according to the change trend to obtain the stopping probability; based on the elevator operation mode and fault characteristics in the initial operation data, using a second machine learning model to perform second abnormality judgment; combining the first abnormality judgment, the stopping probability and the second abnormality judgment, comprehensively judging whether the elevator to be discriminated is in the stopping state. The application can improve the safety and reliability of the elevator.

Description

Technical Field

[0001] This application relates to the field of elevator intelligent monitoring technology, and in particular to an elevator stop determination method and device based on heterogeneous multimodal data. Background Technology

[0002] With the acceleration of urbanization, elevators, as key equipment in vertical transportation, are experiencing a continuous increase in usage frequency and quantity. As people's reliance on elevators grows, elevator safety has also become a major concern. Currently, elevator operation safety monitoring and control technologies play a crucial role in ensuring reliable elevator operation, but existing technologies still have many shortcomings.

[0003] Regarding the current status of elevator operation monitoring, elevators are widely used in various types of buildings, from residential buildings to commercial buildings, covering a very wide range. However, the operating environments of elevators in different buildings are complex and diverse, and their operating conditions vary greatly, posing challenges to the unified management and safety monitoring of elevators. At the same time, with the development of Internet of Things (IoT) technology, more and more elevators are being equipped with various sensors and data acquisition devices, generating a large amount of operational data, but the efficiency of data utilization still needs to be improved.

[0004] Currently, traditional elevator intelligent monitoring systems primarily rely on the elevator's operational status after a malfunction to determine whether the elevator has stopped. The monitoring system collects real-time elevator operating parameters, such as speed, acceleration, and position, as well as relevant information about the malfunction, such as malfunction type and time. The elevator's operational status is determined by analyzing this data. On the hardware side, the elevator is equipped with various sensors, such as car position sensors and floor / door zone sensors, to detect the elevator's position and operational status. On the software side, simple algorithms are used to process and analyze the collected data.

[0005] However, current technology has significant problems. First, there's the issue of missed detections. Due to the limitations of monitoring equipment, some minor but potential faults are difficult to detect in a timely manner. When the elevator is unoccupied, the system may fail to capture minute fault signals, causing the elevator to continue operating despite potential malfunctions, increasing safety risks. Second, there's the problem of false alarms. Minor faults may be mistaken for serious ones in certain situations, especially when the elevator is idle or without passengers. For example, elevator sensors may generate erroneous signals due to small vibrations or power fluctuations, causing the system to incorrectly determine that the elevator has stopped. This not only causes unnecessary panic but also undermines the trust of maintenance companies and passengers in elevator safety. Furthermore, traditional technologies are insufficient in handling complex faults and comprehensively analyzing multi-source data, making it difficult to achieve a comprehensive and accurate assessment of elevator faults and failing to provide effective support for elevator maintenance and management. Summary of the Invention

[0006] In view of this, this application provides an elevator stop detection method and device based on heterogeneous multimodal data, which can identify potential faults in a timely manner, reduce the occurrence of missed and false alarms, and improve the safety and reliability of elevators.

[0007] Specifically, this application is implemented through the following technical solution:

[0008] The first aspect of this application provides a method for elevator stop determination based on heterogeneous multimodal data, the method comprising:

[0009] Obtain initial operating data for all elevators within the same location;

[0010] Based on the location data in the initial operation data, determine the relationship between the stopping position of the elevator to be determined and the non-level position. If it is not in a non-level position, calculate the ratio of the number of times the elevator to be determined runs to the average number of times other elevators in the same location run.

[0011] A first anomaly judgment is made based on the relationship between the ratio and the threshold number of runs.

[0012] Historical fault data is extracted from the elevator control system, maintenance records, and fault logs. Based on the historical fault data, the basic probability of each fault causing elevator stoppage is calculated, and the trend of fault mode changes over time is identified based on time series analysis.

[0013] The basic probability is adjusted based on the changing trend and the first machine learning model to obtain the elevator stop probability. The first machine learning model is used to analyze historical fault data extracted from the elevator control system, maintenance records and fault logs to identify factors with a greater impact on elevator stop than a threshold. An adjustment margin is determined based on the changing trend. An adjustment coefficient is calculated based on the difference between the factors output by the first machine learning model and the real-time fault information in the initial operating parameters. The basic probability is adjusted based on the adjustment margin and the adjustment coefficient.

[0014] Based on the elevator operation mode and fault characteristics in the initial operation data, a second machine learning model is used to make a second anomaly judgment.

[0015] Combining the first anomaly judgment, the elevator stop probability, and the second anomaly judgment, a comprehensive judgment is made as to whether the elevator to be judged is in a stopped state; wherein, the comprehensive anomaly judgment result is calculated based on the weighted sum of the first anomaly judgment, the elevator stop probability, and the second anomaly judgment.

[0016] The second aspect of this application provides an elevator stop determination device based on heterogeneous multimodal data, the device comprising an acquisition module, a judgment module, a calculation module, and an adjustment module;

[0017] The acquisition module is used to acquire the initial operating data of all elevators in the same location;

[0018] The judgment module is used to determine the relationship between the stopping position of the elevator to be judged and the non-leveling position based on the position data in the initial running data. If it is not in a non-leveling position, it calculates the ratio of the number of runs of the elevator to be judged to the average number of runs of other elevators in the same place.

[0019] The judgment module is also used to make a first anomaly judgment based on the relationship between the ratio and the running number threshold;

[0020] The calculation module is used to extract historical fault data from the elevator control system, maintenance records and fault logs, calculate the basic probability of each fault causing the elevator to stop based on the historical fault data, and identify the trend of fault mode changes over time based on time series analysis methods.

[0021] The adjustment module is used to adjust the basic probability according to the changing trend and the first machine learning model to obtain the elevator stop probability. The first machine learning model is used to analyze historical fault data extracted from the elevator control system, maintenance records and fault logs, identify factors with a greater impact on elevator stop than a threshold, determine the adjustment margin based on the changing trend, calculate the adjustment coefficient based on the difference between the factors output by the first machine learning model and the real-time fault information in the initial operating parameters, and adjust the basic probability based on the adjustment margin and the adjustment coefficient.

[0022] The judgment module is also used to make a second anomaly judgment based on the elevator operation mode and fault characteristics in the initial operation data using a second machine learning model.

[0023] The judgment module is further configured to combine the first anomaly judgment, the elevator stop probability, and the second anomaly judgment to comprehensively judge whether the elevator to be judged is in a stopped state; wherein, the comprehensive anomaly judgment result is calculated based on the weighted sum of the first anomaly judgment, the elevator stop probability, and the second anomaly judgment.

[0024] The elevator stop determination method and apparatus provided in this application, based on heterogeneous multimodal data, firstly, acquires the initial operating data of all elevators in the same location, and comprehensively determines whether an elevator has stopped by integrating three different modalities and configurations of data: location data, frequency data, and status data. This provides a comprehensive understanding of the elevator's operating status and offers a more solid data foundation for accurately determining the elevator's stop status compared to traditional methods that rely solely on a single type of data. Furthermore, by determining whether the elevator is not level with a floor, this method can promptly detect dangerous situations and issue warnings, helping relevant personnel to take swift action to ensure passenger safety and reduce the occurrence of accidents.

[0025] Secondly, corresponding to modal data, this invention improves the accuracy of elevator stop detection through a three-level comprehensive judgment based on different modal data. Specifically, the first level of judgment directly uses location data for rapid determination. When the location judgment is unclear, the second level is entered, which uses the number of runs data to obtain ratio information. The third level uses state data, employing both trend learning and direct machine learning judgment to obtain two fault judgment results. That is, the first, second, and third levels of fault judgment use different types of fault data, improving the utilization of different types of data and increasing the accuracy of elevator stop identification. At the same time, for data with complex identification methods, fault identification is performed simultaneously through two methods: direct machine learning judgment and machine learning that only learns trend changes, further improving the accuracy of fault identification.

[0026] Finally, by employing multi-dimensional judgment and comprehensive analysis, the limitations of a single judgment method are avoided, enabling a more comprehensive and accurate assessment of the elevator's actual operating status and effectively reducing the probability of misjudgments and omissions. Especially when facing complex operating environments and potential malfunctions, this multi-dimensional judgment method improves the reliability of the judgment results, providing stronger protection for the safe operation of elevators. Attached Figure Description

[0027] Figure 1 A flowchart of an embodiment of the elevator stop determination method based on heterogeneous multimodal data provided in this application;

[0028] Figure 2 A schematic diagram of the second embodiment of the elevator stop detection device for heterogeneous multimodal data provided in this application. Detailed Implementation

[0029] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0030] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used herein are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any and all possible combinations of one or more of the associated listed items.

[0031] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0032] The following specific embodiments are given to illustrate the technical solution of this application in detail.

[0033] Example 1:

[0034] Figure 1 This is a flowchart of an embodiment of the elevator stop determination method based on heterogeneous multimodal data provided in this application. Please refer to... Figure 1 The method provided in this embodiment may include:

[0035] S101. Obtain the initial operating data of all elevators in the same location.

[0036] It should be noted that "the same location" refers to a specific area with defined boundaries where multiple elevators operate. For example, in a large shopping mall, multiple elevators installed in different locations within the mall constitute the same location. Similarly, in a high-rise office building, multiple elevators used for vertical transportation share the same location within the building. Defining the location as "the same location" helps in collecting elevator data with similar operating environments and needs, facilitating subsequent comparison and comprehensive analysis.

[0037] Initial operational data encompasses a wide range of information reflecting the real-time operating status of the elevator. Specifically, initial operational data may include the elevator's location data, operational data, fault logs, sensor data, and the number of times other elevators in the same location have been used.

[0038] The operational data includes collected parameters across multiple dimensions, such as elevator speed (up and down), acceleration, deceleration, travel time, and dwell time. These parameters are key indicators for measuring elevator operating efficiency. For example, analyzing the speed can determine if the elevator is operating at normal speeds; changes in acceleration and deceleration reflect the smoothness of the acceleration and deceleration process; and travel time and dwell time help understand the elevator's workload and its patterns of stopping at different floors. Comprehensive analysis of these parameters can provide a preliminary assessment of any abnormal elevator behavior, such as abnormal speed or uneven acceleration and deceleration, which may indicate problems with the mechanical or control systems.

[0039] Inter-floor operation information is also a crucial component of operational data, recording the starting and ending floors for each trip. This is essential for determining whether the elevator has made any abnormal stops (such as stopping without leveling). Stopping without leveling can pose safety hazards for passengers entering and exiting the elevator. By monitoring and analyzing inter-floor operation information, such problems can be detected promptly.

[0040] Fault logs are extracted from the elevator control system and may include fault type (such as mechanical or electrical faults), fault occurrence time, fault duration, and related error codes. This information is crucial for understanding the elevator's historical fault patterns and their frequency. For example, if a certain type of fault occurs frequently, it is necessary to focus on that fault type, analyze its causes, and take targeted preventative and corrective measures. Furthermore, recording the fault occurrence time and duration helps assess the impact of the fault on the elevator's normal operation.

[0041] Sensor data is used to monitor the elevator's operating status in real time using various sensors (such as position sensors, load sensors, and door switch sensors). Position sensors accurately report the elevator's current position, load sensors monitor the load capacity, and door switch sensors monitor the door's open / closed status. This real-time data provides accurate feedback on the elevator's operating status. For example, when a load sensor detects an overload or a door switch sensor reports an abnormality such as a door not closing properly, appropriate measures can be taken promptly to ensure safe elevator operation. Furthermore, if the elevator is also equipped with environmental sensors (such as temperature, humidity, and vibration sensors), collecting this data is also essential. Environmental factors can affect elevator operation; for example, high temperatures may affect the performance of electronic components, high humidity may cause electrical equipment to become damp, and excessive vibration may indicate problems such as loosening of the elevator's mechanical structure. By monitoring and analyzing environmental data, potential faults can be detected early, allowing for appropriate protective measures to be taken.

[0042] In addition, elevator maintenance and repair records can be obtained to analyze past maintenance problems and the measures taken. By analyzing past maintenance records, it is possible to understand which elevator components are prone to failure and whether the maintenance measures taken were effective. If certain components are found to require frequent repairs, it is necessary to consider whether upgrades or replacements are needed to improve the elevator's reliability.

[0043] It should be noted that data acquisition for other elevators within the same location includes collecting the number of runs and operating parameters of other elevators in the same building to establish baseline data. By comparing the operating behavior of the elevator to be identified with other elevators, abnormal patterns can be identified. For example, if the number of runs of the elevator to be identified is significantly lower than that of other elevators, or if its operating parameters (such as speed, acceleration, etc.) differ significantly from those of other elevators, it may indicate a potential malfunction or low operating efficiency in the elevator to be identified. Analyzing the operating conditions of other elevators within the same location provides a more comprehensive understanding of the overall elevator operation in that location, offering a more accurate reference for fault analysis and stoppage identification of the elevator to be identified.

[0044] Finally, by integrating the various types of data obtained above and establishing a comprehensive dataset through a database management system or big data processing platform, scattered and different types of data can be centrally managed, facilitating unified analysis and processing in the future. This provides an efficient and reliable data foundation for elevator stop detection and fault diagnosis based on these data.

[0045] S102. Based on the location data in the initial operation data, determine the relationship between the stopping position of the elevator to be judged and the non-level position. If it is not in a non-level position, calculate the ratio of the number of times the elevator to be judged runs to the average number of times other elevators in the same place run.

[0046] It's important to note that before determining the relationship between the elevator's stopping position and its non-leveling position, preprocessing of the initial operating data is required. Specifically, preprocessing may include data cleaning, denoising, normalization, and feature selection. Data cleaning aims to ensure high-quality and consistent data. Methods include removing duplicate values, handling missing values ​​(deletion and imputation), formatting data, and outlier detection. Denoising primarily aims to reduce noise in the data, thereby improving accuracy. Common denoising methods include smoothing techniques, filters, and outlier correction. Normalization aims to scale the data to a standard range, preventing different dimensions or orders of magnitude from adversely affecting subsequent analysis results. Common normalization methods include min-max scaling and Z-score normalization. Feature selection aims to select the most influential features for fault identification from the preprocessed data, thereby reducing data dimensionality and improving model performance. Common feature selection methods include filtering, wrapping, embedding, and principal component analysis.

[0047] It should be noted that the specific implementation principles and processes of the above preprocessing methods can be found in the descriptions of the relevant technologies, and will not be repeated here.

[0048] Through the comprehensive and meticulous data cleaning, denoising, normalization, and feature selection steps described above, a deeply processed dataset is finally obtained. This dataset is characterized by high quality, low noise, standardization, and prominent key features, providing a solid and reliable data foundation for subsequent analysis, model training, and fault identification, and helping to significantly improve the accuracy and robustness of the model.

[0049] After preprocessing, the relationship between the elevator's stopping position and its non-leveling position is determined based on the position data in the preprocessed initial operating data. Specifically, this determination mainly relies on the elevator's inter-floor operating information and sensor data. The inter-floor operating information records the starting and ending floors of each elevator trip, while sensor data, such as that from position sensors, provides real-time feedback on the elevator's accurate position. This data allows for precise determination of the elevator's stopping position. The sensor data includes data detected by position sensors located in the elevator car and / or by floor door sensors. In practice, the actual stopping position of the elevator is compared with its leveling position. The leveling position refers to the position where the elevator car floor is flush with the floor floor when the elevator is normally stopped; this is the standard position for safe elevator operation and normal passenger entry and exit. If the elevator stops at a position other than level, i.e., a non-leveling position, it indicates an abnormal stopping situation.

[0050] Furthermore, determining the relationship between the elevator's stopping position and its non-leveling position also involves using at least one of the following data points: the elevator's motor operating data, its self-test data, its emergency device data, floor display data, and safety circuit monitoring data. The motor operating data and self-test data are considered part of the elevator's operating data. Specifically, to determine whether the elevator is stopped at a non-leveling position, one or more of the following methods can be combined:

[0051] The first method involves installing position sensors, such as encoders or limit switches, inside the elevator car to monitor the elevator's position in real time. The sensors periodically send position data to the control system. The control system then compares the received position data with the preset leveling position.

[0052] Position sensors provide high-precision position feedback, promptly detecting whether the elevator has stopped at a non-level position. In the event of a malfunction, abnormal signals emitted by the sensors can quickly trigger emergency measures to ensure passenger safety.

[0053] The second method involves installing door zone sensors at the entrance of each floor of the elevator to detect whether the elevator car has entered the door zone. When the elevator arrives at a floor, the sensors will identify whether the elevator is in the correct door zone. If the elevator stops outside the door zone, the sensors cannot detect a signal, and the system will determine that it is not level with the floor.

[0054] This method effectively prevents elevator doors from opening when the elevator is not level with the floor, reducing the risk of passenger injury. Door zone sensors also provide additional data to help maintenance personnel analyze the elevator's operating status and malfunctions.

[0055] The third method: The elevator's motor and braking system have a self-checking mechanism that continuously monitors motor operating parameters (such as speed and acceleration). The system analyzes the feedback from the motor and brake to identify any abnormalities, such as failure to decelerate or stop as expected. If an abnormality is detected, the system will record a fault log and issue an alarm.

[0056] Real-time monitoring allows for the timely detection of problems with the motor or braking system, preventing serious malfunctions. Fault logs facilitate subsequent repairs and maintenance, making problem-solving more efficient.

[0057] The fourth method: The control system performs periodic self-checks, monitoring parameters such as elevator acceleration, deceleration, running time, and running distance. If the actual operating parameters are found to be inconsistent with expectations, the control system will trigger a fault handling procedure. The system will stop the elevator, issue an alarm, and record fault information.

[0058] Self-inspection mechanisms can improve elevator safety and reduce the failure rate. By monitoring operational data, maintenance personnel can obtain the elevator's operating history, which helps in fault analysis.

[0059] The fifth method involves installing emergency operation devices (such as emergency buttons) and floor displays both inside and outside the elevator. When the elevator stops at a non-level floor, the floor display will show abnormal information or error codes. Passengers can then use the emergency devices to notify maintenance personnel for assistance.

[0060] Emergency operation devices provide passengers with a safety guarantee, enabling them to seek help promptly in the event of elevator malfunction. Abnormal status displayed on the monitor can help maintenance personnel quickly diagnose elevator problems.

[0061] The sixth method: The elevator is equipped with a safety circuit (such as a leveling sensor switch) to ensure that the elevator is level before the door opens. If the elevator attempts to open the door before leveling, the safety circuit will be cut off, triggering emergency braking. The system records such events and issues a fault alarm.

[0062] The safety circuit provides redundant protection, further reducing the risk of passenger injury. Fault alarm functionality enhances elevator safety, allowing maintenance personnel to address problems promptly.

[0063] By combining the above methods, a more in-depth analysis can be conducted based on factors such as the elevator's operating data, fault logs, sensor data, and the number of times other elevators in the same location have been used. For example, by comparing the elevator's operating data and fault logs, elevators with high failure rates can be identified and prioritized for maintenance. Simultaneously, analyzing the number of times other elevators have been used can help identify common faults or design flaws, thereby improving the overall safety and reliability of the elevator system.

[0064] When the elevator is determined to be stopped at a non-level floor, an early warning action is immediately generated. This is because a non-level stop can pose a significant safety hazard to passengers, such as the risk of falls or other accidents when entering or exiting the elevator. The early warning action can send an alarm message to elevator management personnel, such as via SMS, email, or system notification, while simultaneously activating audible and visual alarms inside the elevator car and on relevant floors to remind passengers to be aware of safety and allow for timely rescue and maintenance measures.

[0065] If the elevator stops at a level landing, it indicates that the elevator's current stopping status is normal. At this point, the ratio of the number of trips of the elevator to be assessed to the average number of trips of other elevators in the same location is calculated. Specifically, the number of trips of the elevator to be assessed can be directly obtained from its operational data, while the number of trips of other elevators is also part of the initial operational data collected in step S101. By statistically analyzing the number of trips of all elevators in the same location (excluding the elevator to be assessed) and calculating their average, the average number of trips of other elevators in the same location can be obtained.

[0066] It should be noted that the purpose of calculating this ratio is to determine whether the elevator under investigation is operating efficiently by comparing its operating frequency with that of other elevators in the same location. If the ratio deviates significantly from 1, it may indicate a potential problem with the elevator under investigation. For example, a ratio much less than 1 may indicate low demand for the elevator or a malfunction causing it to operate poorly; a ratio much greater than 1 may mean that the elevator is overloaded and operating at high load for extended periods, which could easily lead to malfunctions.

[0067] S103. Make a first anomaly judgment based on the relationship between the ratio and the running number threshold.

[0068] Before making the first anomaly judgment, the following points should be noted: Within the same location, the number of operations of multiple elevators is interconnected, especially evident in the design, management, and control of their operation. This relationship can be understood from the following perspectives:

[0069] Scheduling Algorithm: Associated elevators are typically scheduled by a central control system. This system selects which elevator to provide service based on parameters such as the current status of each elevator, the requested floor by passengers, and the number of times each elevator has been used. In this way, the system can achieve more efficient operation, avoid service delays caused by overloading a single elevator, and also prevent other elevators from being idle.

[0070] Load balancing: To prevent any one elevator from being overused, the control system attempts to balance the number of operations across all elevators. By balancing the load, the lifespan of the elevator equipment can be extended, and the maintenance requirements of a single elevator due to frequent operation can be reduced.

[0071] Service efficiency: If the elevator system is poorly designed, some elevators may be frequently used while others remain relatively idle. This not only affects passenger waiting time but may also reduce overall service efficiency. Therefore, it is necessary to rationally allocate the number of trips between related elevators to reduce congestion and passenger waiting time during peak hours.

[0072] Energy-saving control: Some elevator systems dynamically adjust the number of available elevators based on floor demand. During off-peak hours, the system may reduce the number of elevators in use, concentrating energy consumption by using only a few elevators. This dynamic adjustment not only improves energy efficiency but also optimizes elevator usage.

[0073] In summary, the number of trips between related elevators is interconnected, and scheduling, load balancing, and efficiency optimization all depend on the coordinated operation of elevators. Through scientific management and control, the overall performance of the elevator system and the user experience can be significantly improved.

[0074] Therefore, after calculating the ratio of the number of runs of the elevator to be judged to the average number of runs of other elevators in the same location, a first anomaly judgment is made on the elevator's operating status based on this ratio. Specifically, the first anomaly judgment is made based on the relationship between the ratio and a run count threshold, including: comparing the ratio with a preset run count threshold range, which includes an upper limit and a lower limit; if the ratio is less than the lower limit or greater than the upper limit, it is determined to be a first anomaly; if the ratio is between the lower limit and the upper limit, the elevator is judged to be operating normally.

[0075] As an optional embodiment, the preset threshold range of the number of runs is calculated by the scheduling process of the running scheduling algorithm. Calculating the preset threshold range of the number of runs includes: calculating the average number of runs for an elevator based on the elevator operation load balancing requirements of the identified area; calculating the highest and lowest running probabilities of the elevator based on the elevator scheduling algorithm; and determining the upper and lower limits of the preset threshold range of the number of runs based on the results of the highest and lowest running probabilities and the average number of runs, respectively.

[0076] It should be noted that the preset operation frequency threshold range includes both an upper and a lower limit. These two values ​​are determined comprehensively based on factors such as statistical analysis of a large amount of elevator operation data, actual operating experience, and the elevator's design performance. For example, by conducting long-term monitoring of multiple elevators of the same type and under the same usage scenario, analyzing the distribution of their operation frequency ratios, and identifying the range in which the ratio falls when most elevators are operating normally, a reasonable upper and lower limit can be determined. The preset operation frequency threshold range provides a standard reference for judging whether elevator operation is abnormal; it defines the range in which the operation frequency ratio of the elevator to be judged should fall under normal circumstances.

[0077] The calculated ratio is compared to a preset threshold range for the number of trips. If the ratio is less than the lower limit, it may indicate that the number of trips of the elevator under investigation is significantly lower than the average level of other elevators in the same location, potentially indicating problems such as unrepaired malfunctions, incorrect control system settings, or low usage demand. For example, the elevator may have minor malfunctions causing occasional stops, reducing the number of trips; or some functions of the elevator may be improperly configured, affecting its normal use. If the ratio is greater than the upper limit, it means that the number of trips of the elevator under investigation is significantly higher than the average level, potentially facing excessive load pressure. Long-term operation under such conditions can lead to accelerated equipment wear and tear and faster aging of components, thus increasing the probability of malfunctions. For example, a sudden increase in pedestrian traffic in the area where the elevator is located, or a malfunction in other elevators causing passengers to flock to this elevator, significantly increases its number of trips. When the ratio is either less than the lower limit or greater than the upper limit, it is identified as the first anomaly.

[0078] If the ratio is between the lower and upper limits, it means that the number of times the elevator to be judged runs is within a reasonable range relative to the average number of times other elevators in the same place. It can be judged that the elevator is currently in normal operation. This indicates that the elevator's operating frequency is adapted to the overall environment and there are no obvious abnormal fluctuations.

[0079] By comparing the ratio of the number of trips to a preset threshold range, it is possible to quickly and effectively determine whether the elevator in question has a primary anomaly. This method is simple and intuitive, and the threshold range, determined based on extensive data and experience, has high reliability. Timely detection of the primary anomaly allows elevator management personnel to take preventative measures. If the ratio is below the lower limit, the elevator can be inspected for malfunctions and repaired, or its operating settings can be adjusted. If the ratio is above the upper limit, the frequency of elevator maintenance can be increased, or passenger flow management strategies can be adjusted to ensure the safe and stable operation of the elevator.

[0080] S104. Extract historical fault data from the elevator control system, maintenance records, and fault logs. Calculate the basic probability of each fault causing elevator stoppage based on the historical fault data, and identify the trend of fault mode changes over time based on time series analysis.

[0081] It should be noted that the elevator control system, maintenance records, and fault logs are the primary sources of historical fault data. The elevator control system records various data during elevator operation, including the operating status and parameters at the time of the fault; maintenance records contain information such as problems found during each maintenance, repair measures, and repair time; and fault logs specifically record detailed information about the fault, such as the fault type, occurrence time, and duration. Extracting data from these sources allows for a comprehensive understanding of information related to historical elevator faults.

[0082] Specifically, based on the historical fault data, the basic probability of each fault causing elevator shutdown is calculated, and the trend of fault mode changes over time is identified using time series analysis methods, including:

[0083] (1) Arrange the historical data in chronological order.

[0084] It should be noted that historical data was extensively collected from the elevator control system, maintenance records, and fault logs. This data covers: elevator runtime, which directly reflects the elevator's usage intensity at different stages; elevator stop events and corresponding fault events, detailing each stop and the faults that caused it; fault event timestamps, accurately marking the time of fault occurrence and providing crucial evidence for subsequent time series analysis; fault types, such as power failures, mechanical failures, and control system failures, with different types exhibiting different mechanisms and impacts; maintenance records containing detailed information on each maintenance session, including time and frequency, reflecting the progress of elevator maintenance; and external factor data, such as usage frequency reflecting elevator activity levels, weather changes (e.g., high temperatures, humidity) potentially affecting elevator component performance, and building conditions (e.g., increased dust during renovations) potentially impacting elevator operation.

[0085] The collected historical data is arranged strictly in chronological order. For example, arranging the fault logs from previous years in chronological order of fault occurrence allows for a clear view of the occurrence of faults in different years and seasons, laying the foundation for subsequent analysis of the evolution of faults over time.

[0086] In addition, it should be noted that after acquiring historical data, the process also includes removing erroneous records, duplicate data, and incomplete data to ensure data quality; converting all data into a unified format for easy subsequent analysis; and identifying key features, such as the operating mode before the elevator stops and warning signals before a malfunction.

[0087] (2) Based on historical data, for each type of fault, the basic probability of the fault causing the elevator to stop is calculated using the conditional probability formula.

[0088] In real-world elevator operation, elevators may experience various types of malfunctions, each with a different probability of causing a stop. To accurately assess elevator operational risks, we quantify this probability by calculating the basic probability of each malfunction type causing a stop. This basic probability is based on historical data and reflects the probability of a specific malfunction type causing the elevator to stop during past operations.

[0089] Furthermore, different fault types have varying basic probabilities of causing elevator stoppage due to their inherent characteristics, scope of impact, and degree of correlation with critical elevator components. In general, calculating the basic probabilities of various fault types provides a clearer understanding of elevator malfunction risks, allowing for focused attention on fault types with higher basic probabilities. This enables preventative measures to be taken in advance, such as increasing the maintenance frequency of relevant components and stockpiling spare parts, thereby reducing the probability of elevator stoppages and ensuring the safe and stable operation of the elevator. For each fault type, the conditional probability formula P(stoppage|fault) = P(stoppage and fault) / P(fault) is used to calculate its basic probability of causing an elevator stoppage. In actual calculations, P(stoppage and fault) requires counting the number of times a specific fault type occurred simultaneously in historical data, resulting in an elevator stoppage, and then dividing by the total number of data records; P(fault) is calculated by counting the number of times that fault type occurred and dividing by the total number of data records. For example, in 1000 historical data points, if mechanical failures occur 100 times, and 30 of those failures cause the elevator to stop, then the basic probability of a mechanical failure causing an elevator stop is P(stop | mechanical failure) = (30 / 1000) / (100 / 1000) = 0.3. By performing this calculation for each type of failure, we can quantify the impact of each failure on the elevator stoppage and identify which failures are high-risk factors leading to elevator stoppages.

[0090] (3) Fit the change process of the basic probability over time to identify the trend of the failure mode.

[0091] It's important to note that time series analysis methods, such as the ARIMA model and moving averages, can be applied to deeply analyze elevator operation and failure data. The ARIMA model fully considers the autocorrelation, seasonality, and trend characteristics of the data, predicting future failure trends by fitting historical data. For example, by inputting the monthly failure frequency as time series data into the ARIMA model, and after model training and parameter adjustment, a predicted curve showing the frequency of failures over time can be obtained. The moving average method smooths data fluctuations and highlights long-term trends by calculating the average value of data over a certain period. For instance, using a 12-month moving average to process elevator running time for each month of the year allows for a clearer observation of the annual trend in running time and its correlation with failure frequency.

[0092] By carefully observing the relationship between the frequency of power outages and time, and using graphical analysis tools such as line charts and seasonality decomposition plots, seasonal or periodic variations in power outages can be identified. For example, by plotting a line chart of the frequency of power outages for each month of the year, if a significant increase in the frequency of power outages is observed in the summer (June-August), this indicates that power outages have a clear seasonal characteristic in summer. Using seasonality decomposition plots, which break down time series data into components such as trend, seasonality, and residuals, the influence of seasonal factors on the frequency of power outages can be quantified more accurately.

[0093] This step, by combining historical and initial operational data, accurately calculates the basic probability of each type of fault causing elevator stoppage and deeply identifies the changing trends of fault modes, providing crucial information for the safe operation and maintenance management of elevators. It helps elevator maintenance personnel anticipate potential fault risks and develop personalized maintenance plans for different fault types and their changing trends. For example, for fault types with a high probability of occurrence and an upward trend, maintenance frequency is increased and relevant spare parts are stockpiled in advance; for faults with seasonal or periodic characteristics, special inspections and maintenance are conducted before the corresponding season or period. In this way, the elevator failure rate is effectively reduced, ensuring stable elevator operation and improving passenger safety and comfort. Simultaneously, from an economic perspective, a reasonable maintenance plan can reduce unnecessary maintenance costs, extend the elevator's service life, and provide a scientific basis for the full life-cycle management of elevators.

[0094] S105. Adjust the basic probability according to the changing trend and the first machine learning model to obtain the elevator stop probability.

[0095] The first machine learning model is used to analyze historical fault data extracted from the elevator control system, maintenance records, and fault logs, identify factors that have an impact on elevator stoppage greater than a threshold, determine adjustment margin based on the changing trend, calculate adjustment coefficient based on the difference between the factors output by the first machine learning model and the real-time fault information in the initial operating parameters, and adjust the basic probability based on the adjustment margin and adjustment coefficient.

[0096] It should be noted that in this step, the previously calculated basic probability is adjusted by comprehensively considering the changing trends of failure modes and the analysis results of the first machine learning model, thus obtaining a more accurate probability of elevator stoppage that reflects the current situation. This process helps to more accurately assess the likelihood of elevator stoppage due to various failures, providing an important basis for subsequent anomaly detection and comprehensive elevator status assessment.

[0097] Specifically, the base probability is adjusted based on the changing trend and the first machine learning model to obtain the elevator stopping probability, including:

[0098] (1) Based on the changing trend of the fault mode identified during the change of the basic probability over time, determine the periodic changing trend of the fault occurrence.

[0099] Based on the changes in the basic probability over time obtained by using time series analysis methods (such as ARIMA model, moving average method, etc.) and chart analysis (such as line chart, seasonal decomposition chart, etc.) in step S104, the periodic patterns of fault occurrence are identified. For example, if the analysis shows that a certain type of fault occurs more frequently in the summer of each year, this indicates that the fault has a seasonal periodic change trend. (2) Determine the current running time of the elevator to be judged.

[0100] It should be noted that the current operating time of the elevator to be judged should be obtained accurately, down to a specific point in time or time period, so as to facilitate subsequent analysis in combination with the periodic change trend of the fault.

[0101] (3) Based on the current running time of the elevator to be judged, find the basic probability value corresponding to each time point in the determined fault periodic change trend.

[0102] It should be noted that, based on the established cyclical trend of the faults, the current operating time of the elevator to be assessed is used as an index to find the basic probability value corresponding to that time point in the relevant trend data. This basic probability value is calculated based on historical data and reflects the likelihood of a similar fault causing the elevator to stop at that time.

[0103] (4) Determine the cause of the fault and the adjustment coefficient based on the first machine learning model, and determine the adjustment margin based on the periodic change trend.

[0104] Furthermore, after determining the basic probability values, the causes of the fault and adjustment coefficients are determined based on the first machine learning model. This involves three main steps: selecting a machine learning algorithm, training the model and identifying key factors, and calculating the adjustment coefficients. In practice, machine learning algorithms such as random forests and decision trees are used to conduct in-depth analysis of historical fault data extracted from the elevator control system, maintenance records, and fault logs. Taking the random forest algorithm as an example, elevator operating parameters (such as speed, acceleration, and running time), fault types, and external factor data (such as usage frequency and weather changes) from historical data are used as input features, and the elevator stop event is used as the output label to construct a random forest model. Then, the constructed model is trained. After training, the importance of features in the model is analyzed to determine factors with an impact greater than a pre-set threshold on elevator stoppage. For example, after model training, it was found that prolonged high-load operation of the elevator (reflected by features such as running time and usage frequency) and unstable power supply voltage (reflected by external factor data) are the main causes of elevator stoppage due to electrical faults. Finally, the key factors affecting elevator stoppage output by the first machine learning model are compared with the real-time fault information in the initial operating parameters, and the difference between the two is calculated. Based on this difference, an adjustment coefficient is calculated using certain rules or algorithms. For example, if the model determines that high-load operation is an important factor, and the difference between the current real-time operating load of the elevator and historical high-load conditions is large, then the adjustment coefficient may be increased accordingly to more significantly adjust the basic probability.

[0105] Furthermore, the adjustment margin should be determined based on the cyclical trends of fault occurrence, such as long-term upward or downward trends in fault frequency, seasonal fluctuations, etc. If the fault frequency is on the rise, and the current phase is a significant increase in this trend, the adjustment margin can be appropriately increased; conversely, if the trend is stable or declining, the adjustment margin can be reduced accordingly. For example, if the frequency of a certain type of fault has been increasing every year for the past few years, and is currently in an accelerating phase, the adjustment margin can be set to a relatively large value.

[0106] (5) Determine the adjustment factor based on the basic probability value, fault cause, adjustment coefficient and adjustment margin, and calculate the elevator stop probability based on the adjustment factor.

[0107] It should be noted that the adjustment factor needs to be set by comprehensively considering information such as the basic probability and the cause of the failure. If the basic probability of a certain failure is high, and the main cause of the failure is more likely to occur under the current conditions (e.g., high temperatures in summer make the main cause of electrical failures more likely to occur), and the failure has obvious seasonal variations (e.g., electrical failures are more frequent in summer), then the adjustment factor needs to be increased. For example, for electrical failures in summer, assuming a basic probability of 0.3, since high temperatures in summer make the main cause of electrical failures (e.g., overheating of electrical components) more likely to occur, and the failure has a seasonal high incidence, the adjustment factor can be set to 1.5. Conversely, if the basic probability of a certain failure is low, and its main cause is effectively controlled, and there is no obvious seasonal or periodic variation, the adjustment factor can be set to a value less than 1, such as 0.8.

[0108] Specifically, the rules for setting the adjustment factor are determined by comprehensively considering the basic probability value, the cause of the failure, the adjustment coefficient, and the adjustment margin. When the cause of the failure is highly correlated with seasonal or periodic factors and has a high basic probability, the adjustment factor = 1 + k1 × basic probability + k2 × correlation coefficient (k1 and k2 are coefficients determined based on the actual situation; the correlation coefficient represents the degree of correlation between the cause of the failure and the seasonal or periodic factors, and its value ranges from 0 to 1). When the cause of the failure has a low correlation with seasonal or periodic factors and a low basic probability, the adjustment factor = 1 - k3 × basic probability - k4 × correlation coefficient (k3 and k4 are coefficients determined based on the actual situation). In addition, it is also necessary to consider the cases where the cause of the failure is highly correlated with seasonal or periodic factors but has a low basic probability, and the cases where the cause of the failure has a low correlation with seasonal or periodic factors but has a high basic probability. For example, when the basic probability is high but the correlation with seasonal or periodic factors is low, the adjustment factor can be set to a value between the two cases mentioned above, such as adjustment factor = 1 + k5 × basic probability - k6 × correlation coefficient (k5 and k6 are also determined based on the actual situation).

[0109] Finally, multiply the previously calculated basic probability by the set adjustment factor, i.e., elevator stop probability = basic probability × adjustment factor. For example, if the basic probability of a certain fault is 0.3 and the adjustment factor is 1.5, then the elevator stop probability caused by this fault is 0.3 × 1.5 = 0.45; if the basic probability is 0.2 and the adjustment factor is 0.8, then the elevator stop probability is 0.2 × 0.8 = 0.16.

[0110] It should be noted that the obtained elevator stop probability more accurately reflects the actual likelihood of the elevator stopping due to a certain malfunction under the current circumstances. This probability will serve as one of the important bases for subsequent judgments on whether the elevator should stop, helping maintenance personnel and related systems to more accurately assess the elevator's operational risks and make reasonable decisions, such as whether to carry out maintenance in advance or whether to strengthen monitoring.

[0111] Furthermore, after obtaining the elevator stop probability, the adjusted probability is validated using cross-validation or time series backtesting methods to ensure that the adjusted probability can accurately predict future elevator stop events. The model is also regularly evaluated and updated based on new failure event data to ensure its accuracy and effectiveness.

[0112] In this step, by comprehensively considering factors such as the changing trends of malfunctions, the causes of malfunctions, and seasonal or periodic changes, the basic probability is adjusted to obtain the elevator stop probability, making the calculation of elevator stop probability more scientific and accurate. It can more comprehensively reflect the impact of various factors on elevator stoppage during operation, providing a more reliable guarantee for the safe operation of elevators. Through accurate stop probability calculation, potential high-risk malfunctions can be detected in advance, allowing for timely preventative measures to reduce the occurrence of elevator malfunctions, improve elevator reliability and safety, and also help to rationally allocate maintenance resources and reduce maintenance costs.

[0113] S106. Based on the elevator operation mode and fault characteristics in the initial operation data, a second anomaly judgment is made using a second machine learning model.

[0114] It should be noted that, based on the elevator operating mode and fault characteristics in the initial operating data, a second machine learning model is used to perform a second anomaly judgment, including:

[0115] (1) Collect elevator operation data and fault event records, and preprocess the collected data.

[0116] It should be noted that the collection includes elevator operation data (such as operating time, load, number of frequent stops, etc.) and fault event records (fault type, fault time, fault duration, etc.). The data sources can include the elevator's control system, sensors (such as temperature and vibration sensors), maintenance records, etc.

[0117] Preprocessing can include handling missing values, outliers, and duplicate data to ensure data integrity and consistency. Specifically, raw data can be transformed into a format suitable for machine learning algorithms, such as numerical categorical variables (e.g., fault type) and normalized or standardized numerical features (e.g., running time, load capacity). Features related to elevator operation and faults are extracted. For example, average running time, load capacity, and running frequency before a fault occurs. In addition, time features (e.g., season, month, day of the week) can be considered, as these can all affect elevator usage patterns.

[0118] (2) Extract elevator operation mode and fault characteristics from the preprocessed data.

[0119] The preprocessed data contains a wealth of information from which elevator operating modes and fault characteristics are extracted. Operating mode characteristics can include elevator start-stop patterns, speed variation patterns, and floor stopping sequence; analyzing these characteristics helps determine if the elevator is operating normally. Fault characteristics focus on fault-related data, such as changes in operating parameters accompanying specific faults, the frequency and intervals of fault occurrences, etc. These characteristics are key evidence for determining whether the elevator has potential faults.

[0120] (3) Based on the extracted elevator operation mode and fault characteristics, a classification algorithm is selected to train the second machine learning model.

[0121] It's important to note that common classification algorithms include neural networks (which possess powerful non-linear fitting capabilities and can learn complex feature relationships, such as multilayer perceptrons and convolutional neural networks), decision trees (easy to understand and interpret, making classification decisions based on features), random forests (composed of multiple decision trees, improving model accuracy and stability through ensemble learning, capable of handling large-scale data and evaluating feature importance), and support vector machines (performing well in small samples and high-dimensional data, classifying by finding the optimal classification hyperplane). Preprocessed and feature-extracted data is input into the selected algorithm for training. During training, model parameters (such as the weights and biases of the neural network) are continuously adjusted, and optimization algorithms (such as stochastic gradient descent and the Adam optimizer) are used to minimize the error between the predicted and actual results. To prevent overfitting, regularization methods (such as L1 and L2 regularization) are used to constrain model complexity, while cross-validation (such as K-fold cross-validation) is employed to evaluate model performance, ensuring the model has good generalization ability.

[0122] (4) Deploy the trained second machine learning model into the elevator monitoring system, analyze the elevator operation data in real time, output the fault prediction results, and determine whether there is a second anomaly based on the fault prediction results.

[0123] It should be noted that the trained second machine learning model, deployed in the elevator monitoring system, can be integrated into local monitoring software or a cloud server to ensure real-time reception of elevator operation data. During elevator operation, the monitoring system collects data in real time and inputs it into the model. The model analyzes the data based on the patterns learned during training and outputs a fault prediction result, that is, predicting the probability of the elevator malfunctioning and stopping within a certain period. A reasonable threshold is set; when the model's predicted probability of stopping exceeds the threshold, a second anomaly is identified; if it is below the threshold, the elevator is considered to be operating normally. This data-driven anomaly detection method can promptly identify potential fault risks, providing strong support for elevator maintenance and management, and improving the safety and reliability of elevator operation. In this step, by analyzing elevator operation data in real time, potential anomalies during elevator operation can be detected promptly, especially when the first anomaly detection fails to identify the problem, further improving the accuracy and reliability of elevator anomaly detection. Timely detection of the second anomaly allows elevator maintenance personnel to take rapid measures to prevent elevator malfunctions, ensuring passenger safety and the normal operation of the elevator. Simultaneously, by continuously optimizing and improving the second machine learning model, its performance can be improved, better adapting to the operating characteristics of different elevators and environmental changes, providing strong support for intelligent elevator management.

[0124] S107. Combining the first anomaly judgment, the elevator stop probability, and the second anomaly judgment, comprehensively determine whether the elevator to be judged is in a stopped state; wherein, the comprehensive anomaly judgment result is calculated based on the weighted sum of the first anomaly judgment, the elevator stop probability, and the second anomaly judgment.

[0125] It should be noted that, combining the first anomaly detection, the elevator stop probability, and the second anomaly detection, the comprehensive judgment of whether the elevator to be judged is in a stopped state includes:

[0126] Under different operating scenarios, based on the importance of the first anomaly judgment, the elevator stop probability, and the second anomaly judgment in the elevator stop determination process, corresponding first weights, second weights, and third weights are calculated respectively; the sum of the first weight, second weight, and third weight is 1; the first anomaly judgment score is calculated based on the ratio and the first weight, the elevator stop probability score is calculated based on the product of the elevator stop probability and the second weight, and the second anomaly judgment score is calculated based on the second anomaly occurrence probability and the third weight; the total score is calculated based on the first anomaly judgment score, the elevator stop probability score, and the second anomaly judgment score, and the elevator status is determined based on the total score.

[0127] The calculation of the first anomaly detection score includes:

[0128] Determine the relationship between the ratio and 1. If the ratio is greater than or equal to 1, the first anomaly judgment score is equal to 1. If the ratio is less than 1, the first anomaly judgment score is equal to the difference between 1 and the product of the ratio and the first weight.

[0129] Calculate the elevator stop probability score, including:

[0130] The elevator stop probability is obtained from the elevator stop probability model constructed using a machine learning algorithm; an elevator stop probability score is calculated based on the elevator stop probability and the second weight; the elevator stop probability score is equal to the difference between 1 and the product of the elevator stop probability and the second weight.

[0131] The second anomaly detection score is calculated, including:

[0132] Based on the analysis results of the second machine learning model, the probability of the second anomaly occurring is predicted; based on the predicted probability of the second anomaly occurring and the third weight, the second anomaly judgment score is calculated; the second anomaly judgment score is equal to the product of the difference between 1 and the probability of the second anomaly occurring and the third weight.

[0133] It should be noted that, based on the relative importance of the first anomaly detection, the probability of elevator stoppage, and the second anomaly detection in the elevator stoppage determination process, corresponding first, second, and third weights are assigned, respectively. The sum of these three weights is 1. For example, if the first anomaly detection is considered to play a crucial role in the determination, the first weight can be set to 0.5; the probability of elevator stoppage is of secondary importance, so the second weight can be set to 0.3; and the second anomaly detection is of relatively lower importance, so the third weight can be set to 0.2. The specific weight settings need to be determined based on actual business needs, data characteristics, and past experience.

[0134] Referring to the above explanation, when calculating the first anomaly judgment score, if the ratio of the number of runs calculated in step S103 is greater than or equal to 1, it indicates that the number of runs of the elevator to be judged is not significantly abnormal compared with the average number of runs of other elevators in the same location, and the first anomaly judgment score is equal to 1.

[0135] If the ratio of the number of runs is less than 1, it indicates that the elevator to be judged has significantly fewer runs than other elevators, suggesting a certain possibility of an anomaly. The first anomaly judgment score is equal to the difference between 1 and the product of this ratio and the first weight. This can be expressed as: If the ratio < 1, the first anomaly judgment score = 1 - ratio × first weight. For example, if the first weight is 0.5 and the ratio is 0.8, then the first anomaly judgment score = 1 - 0.8 × 0.4 = 0.60.

[0136] When calculating the elevator stop probability score, the elevator stop probability is obtained from an elevator stop probability model constructed using a machine learning algorithm. This model is obtained in steps S104 and S105 through the analysis, processing, and adjustment of historical data and initial operating data.

[0137] The elevator stop probability score is calculated based on the elevator stop probability and a second weight. The elevator stop probability score is equal to the difference between 1 and the product of the elevator stop probability and the second weight. The formula is: Elevator stop probability score = 1 - Elevator stop probability × Second weight. For example, if the elevator stop probability is 0.2 and the second weight is 0.3, then the elevator stop probability score = 1 - 0.2 × 0.3 = 0.94.

[0138] When calculating the second anomaly detection score, the predicted probability of the second anomaly is obtained from the second anomaly detection model constructed using a machine learning algorithm. This model was built in step S106 using initial running data and elevator stop probability.

[0139] The second anomaly detection score is calculated based on the second anomaly prediction probability and the third weight. The second anomaly detection score is equal to the product of the difference between 1 and the second anomaly prediction probability and the third weight. The formula is: Second anomaly detection score = (1 - Second anomaly prediction probability) × Third weight. For example, if the second anomaly prediction probability is 0.1 and the third weight is 0.3, then the second anomaly detection score = (1 - 0.1) × 0.2 = 0.18.

[0140] Finally, the scores for the first anomaly detection, the elevator stop probability, and the second anomaly detection are added together to obtain a total score. This total score is used to determine the elevator's status. A threshold can be set: if the total score is greater than the threshold, the elevator is considered to be in normal operation; if the total score is less than or equal to the threshold, the elevator is considered to be stopped. For example, setting the threshold to 0.7: if the total score is greater than 0.7, the elevator is considered normal. If the total score is between 0.4 and 0.7, the elevator is considered to be in an unstable state and requires further monitoring. If the total score is less than 0.4, the elevator is considered to be stopped.

[0141] In this step, the comprehensive judgment method makes full use of the information obtained in the previous steps, reduces the limitations and errors of single judgment, improves the accuracy and reliability of elevator stop detection, helps to detect elevator malfunctions in a timely manner, and ensures the safe operation of elevators.

[0142] The method provided in this embodiment comprehensively improves the safety and reliability of elevator operation and reduces operating costs through multiple stages, including data acquisition and analysis, fault early warning, probability calculation, anomaly judgment, and integrated decision-making. Specifically, by acquiring multimodal initial operating data, including operating data, fault logs, sensor data, and other elevator operation counts, it provides rich information for accurate judgment. Through integrated analysis, the elevator operating status can be understood from multiple dimensions, making it easier to discover potential faults compared to traditional single data sources. Based on the initial operating data, the stopping position is determined, and an early warning is issued directly when the elevator is not level with the floor. This step can promptly detect and notify personnel to handle the danger of non-level stopping, such as issuing alarms to management personnel and providing audible and visual reminders in the car and on the floors, which can prevent passengers from falling or getting out of the elevator and effectively protect passenger safety. By performing the first anomaly judgment, potential elevator problems can be detected in advance when the ratio is abnormal. For example, too few operation counts may indicate an unrepaired fault, while too many may indicate high-load operation that is prone to causing a fault. After timely detection of anomalies, measures such as inspection and maintenance, adjustment of operating settings, or diversion of passengers can be taken to prevent faults from occurring.

[0143] Furthermore, by combining historical and initial operational data, the basic probability is calculated using conditional probability formulas, and the trend of failure mode changes is analyzed. Adjustments are made to the elevator shutdown probability, taking into account seasonality and periodicity. This allows maintenance personnel to rationally schedule maintenance for high-probability faults, such as increasing maintenance frequency, stockpiling spare parts, reducing unnecessary maintenance, lowering costs, extending elevator lifespan, and achieving scientific management throughout the entire lifecycle. A second anomaly assessment can promptly identify issues missed in the first anomaly assessment. Finally, by combining the results of the first anomaly assessment, the shutdown probability, and the second anomaly assessment, a score is calculated using assigned weights to determine the elevator's status. This multi-dimensional comprehensive assessment overcomes the limitations of single-dimensional assessments, reduces misjudgments and omissions, and assigns weights according to different situations, making the results more realistic.

[0144] Example 2:

[0145] Corresponding to the aforementioned embodiment of the elevator stop determination method based on heterogeneous multimodal data, this application also provides an embodiment of an elevator stop determination device based on heterogeneous multimodal data.

[0146] Figure 2 This is a schematic diagram of the second embodiment of the elevator stop detection device for heterogeneous multimodal data provided in this application. Please refer to... Figure 2 The device provided in this embodiment includes an acquisition module 210, a judgment module 220, a calculation module 230, and an adjustment module 240;

[0147] The acquisition module 210 is used to acquire the initial operating data of all elevators in the same location;

[0148] The judgment module 220 is used to determine the relationship between the stopping position of the elevator to be judged and the non-leveling position based on the position data in the initial running data. If it is not in a non-leveling position, it calculates the ratio of the number of runs of the elevator to be judged to the average number of runs of other elevators in the same place.

[0149] The judgment module 220 is also used to make a first anomaly judgment based on the relationship between the ratio and the running number threshold;

[0150] The calculation module 230 is used to extract historical fault data from the elevator control system, maintenance records and fault logs, calculate the basic probability of each fault causing the elevator to stop based on the historical fault data, and identify the trend of fault mode changes over time based on time series analysis method.

[0151] The adjustment module 240 is used to adjust the basic probability according to the changing trend and the first machine learning model to obtain the elevator stop probability. The first machine learning model is used to analyze historical fault data extracted from the elevator control system, maintenance records and fault logs, identify factors with a greater impact on elevator stop than a threshold, determine the adjustment margin based on the changing trend, calculate the adjustment coefficient based on the difference between the factors output by the first machine learning model and the real-time fault information in the initial operating parameters, and adjust the basic probability based on the adjustment margin and the adjustment coefficient.

[0152] The judgment module 220 is also used to perform a second anomaly judgment based on the elevator operation mode and fault characteristics in the initial operation data using a second machine learning model.

[0153] The judgment module 220 is further configured to combine the first anomaly judgment, the elevator stop probability, and the second anomaly judgment to comprehensively judge whether the elevator to be judged is in a stopped state; wherein, the comprehensive anomaly judgment result is calculated based on the weighted sum of the first anomaly judgment, the elevator stop probability, and the second anomaly judgment.

[0154] The apparatus of this embodiment can be used to perform... Figure 1 The steps of the method embodiment shown are similar in principle and process, and will not be repeated here.

[0155] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0156] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0157] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for elevator stop decision of heterogeneous multi-modal data, characterized in that, The method includes: Obtain the initial operating data of all elevators in the same location; Based on the location data in the initial operating data, determine the relationship between the stopping position of the elevator to be judged and the non-level position. If it is not in a non-level position, calculate the ratio of the number of times the elevator to be judged runs to the average number of times other elevators in the same place run. A first anomaly judgment is made based on the relationship between the ratio and the threshold number of runs. Historical fault data is extracted from the elevator control system, maintenance records, and fault logs. Based on the historical fault data, the basic probability of each fault causing elevator stoppage is calculated, and the trend of fault mode changes over time is identified based on time series analysis. The basic probability is adjusted based on the changing trend and the first machine learning model to obtain the elevator stop probability. The first machine learning model is used to analyze historical fault data extracted from the elevator control system, maintenance records and fault logs to identify factors with a greater impact on elevator stop than a threshold. An adjustment margin is determined based on the changing trend. An adjustment coefficient is calculated based on the difference between the factors output by the first machine learning model and the real-time fault information in the initial operating data. The basic probability is adjusted based on the adjustment margin and the adjustment coefficient. Based on the elevator operation mode and fault characteristics in the initial operation data, a second machine learning model is used to make a second anomaly judgment. Combining the first anomaly judgment, the elevator stop probability, and the second anomaly judgment, a comprehensive judgment is made as to whether the elevator to be judged is in a stopped state; wherein, the comprehensive anomaly judgment result is calculated based on the weighted sum of the first anomaly judgment, the elevator stop probability, and the second anomaly judgment.

2. The method of claim 1, wherein, The first anomaly judgment based on the relationship between the ratio and the running count threshold includes: The ratio is compared with a preset range of running times thresholds, which includes an upper limit and a lower limit. If the ratio is less than the lower limit or greater than the upper limit, it is determined to be the first anomaly; If the ratio is between the lower and upper limits, the elevator is considered to be operating normally.

3. The method according to claim 1, characterized in that, The calculation of the basic probability of each fault causing elevator shutdown based on the historical fault data, and the identification of the changing trend of fault modes over time based on time series analysis methods, includes: The historical fault data is arranged in chronological order; Based on historical fault data, for each fault type, the basic probability of the fault type causing the elevator to stop is calculated using the conditional probability formula; By fitting the change process of the basic probability over time, the changing trend of the fault mode can be identified.

4. The method according to claim 1, characterized in that, The step of adjusting the base probability based on the changing trend and the first machine learning model to obtain the elevator stopping probability includes: Based on the changing trend of fault modes identified during the change of basic probability over time, the periodic trend of fault occurrence is determined. Determine the current running time of the elevator to be judged; Based on the current running time of the elevator to be determined, find the basic probability value corresponding to each time point in the established fault periodic change trend; The cause of the fault and the adjustment coefficient are determined based on the first machine learning model, and the adjustment margin is determined based on the periodic change trend. The adjustment factor is determined based on the basic probability value, the cause of the failure, the adjustment coefficient, and the adjustment margin, and the probability of elevator stoppage is calculated based on the adjustment factor.

5. The method according to claim 1, characterized in that, The second anomaly detection, based on the elevator operating mode and fault characteristics in the initial operating data and utilizing a second machine learning model, includes: Collect elevator operation data and fault event records, and preprocess the collected data; Extract elevator operation modes and fault characteristics from the preprocessed data; Based on the extracted elevator operation mode and fault characteristics, a classification algorithm is selected to train the second machine learning model; The trained second machine learning model is deployed into the elevator monitoring system to analyze elevator operation data in real time, output fault prediction results, and determine whether a second anomaly exists based on the fault prediction results.

6. The method according to claim 1, characterized in that, The step of combining the first anomaly detection, the elevator stop probability, and the second anomaly detection to comprehensively determine whether the elevator to be judged is in a stopped state includes: Under different operating scenarios, based on the importance of the first anomaly judgment, the elevator stop probability, and the second anomaly judgment in the elevator stop judgment process, the corresponding first weight, second weight, and third weight are calculated respectively; the sum of the first weight, second weight, and third weight is 1; The first anomaly judgment score is calculated based on the ratio and the first weight; the elevator stop probability score is calculated based on the product of the elevator stop probability and the second weight; and the second anomaly judgment score is calculated based on the second anomaly occurrence probability and the third weight. The total score is calculated based on the first anomaly judgment score, the elevator stop probability score, and the second anomaly judgment score, and the elevator status is determined based on the total score.

7. The method according to claim 6, characterized in that, The first anomaly detection score is calculated, including: Determine the relationship between the ratio and 1. If the ratio is greater than or equal to 1, the first anomaly judgment score is equal to 1. If the ratio is less than 1, the first anomaly judgment score is equal to the difference between 1 and the product of the ratio and the first weight.

8. The method according to claim 6, characterized in that, Calculate the elevator stop probability score, including: The elevator stop probability is obtained from the elevator stop probability model constructed using machine learning algorithms; The elevator stop probability score is calculated based on the elevator stop probability and the second weight; the elevator stop probability score is equal to the difference between 1 and the product of the elevator stop probability and the second weight.

9. The method according to claim 6, characterized in that, The calculation of the second anomaly judgment score includes: Based on the analysis results of the second machine learning model, the probability of the second anomaly occurring is predicted. The second anomaly judgment score is calculated based on the predicted probability of the second anomaly and the third weight; the second anomaly judgment score is equal to the product of the difference between 1 and the probability of the second anomaly and the third weight.

10. An elevator stop determination device based on heterogeneous multimodal data, characterized in that, The device includes an acquisition module, a judgment module, a calculation module, and an adjustment module; The acquisition module is used to acquire the initial operating data of all elevators in the same location; The judgment module is used to determine the relationship between the stopping position of the elevator to be judged and the non-leveling position based on the position data in the initial running data. If it is not in a non-leveling position, it calculates the ratio of the number of runs of the elevator to be judged to the average number of runs of other elevators in the same place. The judgment module is also used to make a first anomaly judgment based on the relationship between the ratio and the running number threshold; The calculation module is used to extract historical fault data from the elevator control system, maintenance records and fault logs, calculate the basic probability of each fault causing the elevator to stop based on the historical fault data, and identify the trend of fault mode changes over time based on time series analysis. The adjustment module is used to adjust the basic probability according to the changing trend and the first machine learning model to obtain the elevator stop probability. The first machine learning model is used to analyze historical fault data extracted from the elevator control system, maintenance records and fault logs, identify factors with a greater impact on elevator stop than a threshold, determine the adjustment margin based on the changing trend, calculate the adjustment coefficient based on the difference between the factors output by the first machine learning model and the real-time fault information in the initial operating data, and adjust the basic probability based on the adjustment margin and the adjustment coefficient. The judgment module is also used to make a second anomaly judgment based on the elevator operation mode and fault characteristics in the initial operation data using a second machine learning model. The judgment module is further configured to combine the first anomaly judgment, the elevator stop probability, and the second anomaly judgment to comprehensively judge whether the elevator to be judged is in a stopped state; wherein, the comprehensive anomaly judgment result is calculated based on the weighted sum of the first anomaly judgment, the elevator stop probability, and the second anomaly judgment.

Images