Intelligent elevator automatic control method combining building user big data and AI perception

Abstract

The invention relates to the technical field of data perception control, and provides a building user big data and AI perception combined intelligent elevator automatic control method, which comprises the following steps: deploying a distributed network at a building, obtaining elevator perception data, encrypting, fusing and partitioning the elevator perception data, and obtaining a building user big data and AI perception combined intelligent elevator automatic control system. Feature extraction is carried out based on federated learning of secure multi-party calculation, and fused perception features are generated; a space-time diagram is constructed according to fused sensing features, elevator taking demand modes of different building types in different scenes are learned and predicted through a space-time diagram convolutional neural network and a generative adversarial network, operation areas are flexibly divided based on the building types and the scenes, and operation routes are dynamically planned through a partition scheduling strategy. The cross-regional collaborative scheduling of the elevator is realized; and elevator abnormal states including passenger overload, abnormal behaviors, elevator car environment abnormity and the like are detected in real time according to the fused sensing data, multi-stage abnormity classification is carried out, and a corresponding emergency control strategy is triggered based on a classification result.

Description

Technical Field

[0001] The present application relates to the field of data perception control technology, and in particular to an intelligent elevator automatic control method that combines building user big data with AI perception. Background Art

[0002] In modern building management, elevators are important vertical transportation tools. The operating efficiency and safety of elevators directly affect the user experience and the overall operation of the building. However, traditional elevator control methods mainly rely on simple logical control, such as responding to floor calls in order. This method lacks a comprehensive perception of the elevator's operating status and user needs. At the same time, for different types of buildings, there are significant differences in elevator demand patterns, but traditional elevator control methods lack adaptability to such differences and cannot be flexibly scheduled according to different building types and scenarios. Summary of the invention

[0003] In view of the shortcomings of the prior art, the present application provides an intelligent elevator automatic control method combining building user big data with AI perception, the method comprising: deploying a distributed network at the building to obtain elevator perception data, the elevator perception data including elevator operation data, car environment data and user elevator riding data, encrypting and fusing the elevator perception data and dividing it into perception blocks according to time periods, extracting features from the perception blocks based on federated learning of secure multi-party computing, and generating fused perception features; A spatiotemporal graph is constructed based on the fused perception features, and the spatiotemporal graph is analyzed through a spatiotemporal graph convolutional neural network to learn the elevator demand patterns of different building types, and the elevator demand patterns in different scenarios are predicted through a generative adversarial network. Based on building types and scenarios, the operation areas are divided and the operation routes are dynamically planned through zoning scheduling strategies; Detect abnormal elevator status based on fused perception data and trigger emergency control strategies.

[0004] As an optional implementation, the deployment logic of the distributed network includes: Install sensors in each elevator in the building to obtain elevator sensing data; Deploy edge nodes in each elevator to receive and process elevator sensing data; Determine whether to dynamically adjust the frequency of acquiring elevator perception data based on changes in the cabin environment data and user elevator data.

[0005] As an optional implementation, the generation logic of the fused perception feature includes: Encrypt and interpolate elevator sensing data in edge nodes; Assign weights to the elevator perception data, fuse the elevator perception data through the attention mechanism, and obtain fused perception data; Divide the fused perception data into perception blocks according to time periods; Federated learning based on secure multi-party computing is used to extract features from each perception block; The features extracted from each perception block are sent to the cloud for aggregation to obtain fused perception features.

[0006] As an optional implementation, the construction logic of the space-time graph includes: Determine and fill the nodes of the space-time graph according to the floor number, building type, number of passengers, average waiting time and environmental indicators to obtain a node attribute matrix; The calculated frequency of users taking the elevator between floors is used as the edge weight to connect the nodes on different floors. The edge weight is determined by the length of the time interval to connect the nodes in different time periods on the same floor to obtain the adjacency matrix. The node attribute matrix and adjacency matrix of the spatiotemporal graph are updated in real time based on the fused perception features.

[0007] As an optional implementation, the analysis logic of the space-time graph includes: The spatial dependency between floors is extracted through the graph convolution layer based on the node attribute matrix and adjacency matrix of the spatiotemporal graph; The changes in elevator demand data are analyzed based on the fused perception features of the time step through the time convolution layer; After forward propagation of multiple layers of graph convolutional layers and temporal convolutional layers, the elevator demand patterns of different building types in different time and space are learned.

[0008] As an optional implementation, the prediction logic of the generative adversarial network includes: Initialize the generator and discriminator in the generative adversarial network, and use the generator to generate elevator demand data in different scenarios; Obtain real elevator demand data under different historical scenarios, alternately train the generator and discriminator for adversarial training, use the discriminator to distinguish the difference between the generated elevator demand data and the real elevator demand data, and obtain the difference feedback of the discriminator; The parameters of the generator are adjusted according to the difference feedback of the discriminator. After multiple rounds of adversarial training, the trained generator is used to predict the elevator demand patterns of different building types in different scenarios.

[0009] As an optional implementation manner, the operation area division logic includes: Generate dynamic heat maps based on elevator demand data in different building types and scenarios; Determine whether to merge adjacent floors into the same operating area based on the changes in the dynamic thermal map; Set static elevator priority based on floor attributes, and dynamically adjust elevator priority based on elevator abnormal status; Each elevator determines the service operation area based on the status information and broadcasts it to other elevators.

[0010] As an optional implementation, the partition scheduling strategy includes: Obtain status information of each elevator in real time and assign elevator identification; Calculate the operating cost of each elevator based on the status information and operating area, and determine the operating route of each elevator; Based on the broadcast between elevators, a status consensus is reached, and each elevator is collaboratively scheduled according to the elevator priority.

[0011] As an optional implementation, the triggering logic of the emergency control strategy includes: Continuously obtain fusion perception data during elevator operation; Determine the abnormal state of the elevator based on the changing trend of the fused sensing data, including whether the elevator is overloaded, whether the passengers in the elevator have abnormal behavior, and whether the car environment is abnormal; Determine multi-level abnormal classification based on abnormal status of elevator; Trigger emergency control strategies based on multi-level anomaly classification.

[0012] As an optional implementation, the generation logic of the dynamic heat map includes: Divide the floor plan of the building into a plurality of grid cells; Calculate the elevator demand density of each grid cell in different time windows; Color and transparency are assigned to each grid cell according to the density of elevator demand to generate a dynamic heat map.

[0013] Compared with the prior art, the beneficial effects of the present application are: by deploying a distributed network at the building and acquiring elevator perception data, a comprehensive perception of the elevator operating status and user needs is achieved, the elevator perception data is encrypted, fused and processed in blocks, and feature extraction is performed based on federated learning of secure multi-party computing to generate fused perception features, which not only ensures the security of the elevator perception data, but also improves the quality and availability of the fused perception data, providing a reliable data foundation for subsequent analysis and control.

[0014] A space-time graph is constructed based on the fused perception features. The space-time graph convolutional neural network and generative adversarial network are used to learn and predict the elevator demand patterns of different building types in different scenarios. This can accurately predict the elevator demand in advance, providing a scientific basis for elevator scheduling. The operating areas are flexibly divided based on building types and scenarios, and the operating routes are dynamically planned through the zoning scheduling strategy, which realizes cross-regional collaborative scheduling of elevators, improves the operating efficiency of elevators, reduces passengers' waiting time, and enhances user experience.

[0015] According to the fused perception data, the abnormal status of the elevator is detected in real time, including passenger overload, abnormal behavior and abnormal cabin environment, and multi-level abnormal classification is performed. The corresponding emergency control strategy is triggered based on the classification results. This method can timely discover and handle various abnormal situations during the operation of the elevator, thereby ensuring the safety of passengers' lives and the safe operation of the elevator to the greatest extent. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them: Figure 1 A method flow chart of an intelligent elevator automatic control method combining building user big data and AI perception provided in an embodiment of the present application; Figure 2 A time-space diagram for constructing a logic flow chart of the intelligent elevator automatic control method combining building user big data and AI perception provided in the embodiment of the present application; Figure 3 A generative adversarial network prediction logic flow chart of an intelligent elevator automatic control method combining building user big data with AI perception provided in an embodiment of the present application; Figure 4 A logical flow chart of the operation area division of the intelligent elevator automatic control method combining building user big data and AI perception provided in the embodiment of the present application; Figure 5 A zoning scheduling strategy diagram for an intelligent elevator automatic control method that combines building user big data with AI perception provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] In order to make the objectives, technical solutions and advantages of the embodiments of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application are clearly and completely described below in conjunction with the drawings in the specification. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments.

[0018] Example

[0019] like Figure 1 As shown, it is a method flow chart of the intelligent elevator automatic control method combining building user big data and AI perception provided in an embodiment of the present application, and the method includes: Deploy a distributed network in the building to obtain elevator perception data, which includes elevator operation data, car environment data, and user elevator data. Encrypt and fuse the elevator perception data and divide it into several perception blocks according to time periods. Based on federated learning of secure multi-party computing, extract features from the perception blocks and generate fused perception features. The deployment logic of the distributed network includes: Install sensors in each elevator in the building to obtain elevator sensing data; Deploy edge nodes in each elevator to receive and process elevator sensing data; Determine whether to dynamically adjust the frequency of acquiring elevator perception data based on changes in the cabin environment data and user elevator data.

[0020] In order to obtain comprehensive and accurate elevator perception data, it is necessary to install various sensors at key locations of each elevator in the building. Elevator perception data includes elevator operation data, car environment data, and user elevator data. Elevator operation data includes the elevator's operating speed, acceleration, location floor, and door status. Car environment data includes the temperature, humidity, and air quality inside and outside the elevator. User elevator data includes passenger identity, number of passengers, and weight distribution. Passenger identity is an identifier used to distinguish different passenger groups and ensure the anonymity of the data. For example, for office buildings, it is necessary to distinguish whether the passenger is an employee, visitor, or a person with special privileges. Sensitive personal information is not involved to ensure the legality and compliance of the passenger identity. Install an acceleration sensor to measure the acceleration of the elevator during operation, install a temperature and humidity sensor on the wall of the elevator car to detect the temperature and humidity in the car, install a camera and a pressure sensor at the elevator door. The camera is used to identify the identity and number of passengers, and the pressure sensor is used to calculate the weight distribution of passengers. Install a speed sensor and a position sensor on the elevator track to obtain information on the elevator's operating speed and location floor. Install a door movement sensor at the elevator door to monitor the opening and closing status of the elevator door, and install air quality sensors inside and outside the elevator car to detect the air quality inside and outside the elevator; thereby obtaining rich elevator perception data to provide a more comprehensive understanding of the elevator's operating status and passenger needs, which helps to improve elevator operating efficiency and safety.

[0021] In order to perform preliminary processing on the elevator perception data obtained by the sensors locally and reduce the computing pressure on the cloud, edge computing devices are deployed inside the machine room of each elevator as edge nodes, and each sensor is connected through a wired network to receive the elevator perception data obtained by the sensors in real time, reducing data transmission delays, reducing the occupation of cloud computing resources, and improving the overall response speed.

[0022] In order to find a balance between data volume and computing resources, to avoid overloading edge nodes when the data volume is too large, and to capture key information in time when the data volume is too small, the frequency of obtaining elevator perception data is determined based on changes in cabin environment data and user elevator data. The judgment criteria include when the number of passengers in the cabin is greater than the number threshold (for example, 80% of the elevator's rated capacity) and the duration is greater than the time threshold (for example, 5 minutes), or when the number of people waiting outside the elevator is greater than the number threshold (for example, 10 people), it is determined to be the elevator peak period; and when the temperature and humidity in the cabin are not within the normal range (for example, the temperature is between 22 and 2 6℃, humidity between 40% and 60%) and the duration is greater than the time threshold (for example, 10 minutes), or the environmental indicators (such as formaldehyde content and PM2.5 concentration and other air quality) are not within the safety standard, it is determined that the cabin environment is abnormal; at this time, the way to dynamically adjust the acquisition frequency of the elevator perception data is to increase the acquisition frequency of the elevator perception data when it is in the peak period of elevator riding or the cabin environment is abnormal, such as from once per minute to once every 10 seconds, otherwise under other normal circumstances, maintain the original acquisition frequency, thereby achieving the goal of effectively saving edge node resources and improving data processing efficiency while ensuring the acquisition of key data.

[0023] The generation logic of fusion perception features includes: Encrypt and interpolate elevator sensing data in edge nodes; Assign weights to the elevator perception data, fuse the elevator perception data through the attention mechanism, and obtain fused perception data; Divide the fused perception data into several perception blocks according to time periods; Federated learning based on secure multi-party computing is used to extract features from each perception block; The features extracted from each perception block are sent to the cloud for aggregation to obtain fused perception features.

[0024] In practical applications, sensors may have missing data or abnormal values ​​during the data acquisition process. In order to ensure the integrity and accuracy of elevator perception data, data interpolation compensation is required. At the same time, considering the security of elevator perception data, it is necessary to operate in a homomorphic encryption environment. The encrypted elevator perception data is processed at the edge node through the Lagrange interpolation algorithm based on homomorphic encryption. For missing or abnormal elevator perception data, the interpolation result is calculated through the Lagrange interpolation formula based on the normal elevator perception data before and after the missing or abnormal elevator perception data, thereby replacing the missing or abnormal elevator perception data, providing a reliable data foundation for the subsequent fusion and feature extraction of elevator perception data, while ensuring data privacy.

[0025] The function expression of data interpolation compensation is as follows: ; In the formula, Indicates the time point after compensation by Lagrange interpolation formula The corresponding missing or abnormal elevator sensing data, Indicates at a point in time The corresponding elevator perception data, Indicates the number of elevator sensing data, Indicates a point in time.

[0026] At the same time, different types of elevator perception data have different degrees of importance to elevator control. Assigning different weights through the attention mechanism for fusion can more accurately reflect the actual operating status of the elevator and passenger needs. According to the actual situation, initial weights are assigned to the elevator's operating speed, acceleration, location floor and door status, the temperature and humidity and air quality inside and outside the elevator, as well as the passenger identity, number of passengers and weight distribution of the passengers. Then, through the attention mechanism, according to the correlation between the elevator operation scenario and the elevator perception data, for example, the correlation score of each elevator perception data with the elevator operation scenario is obtained by dot product calculation, and the weight value is obtained after normalization of the correlation score, so as to dynamically adjust the weight, and the weighted sum of the elevator perception data after the weight adjustment is performed to obtain the fused perception data. In this way, the fused elevator perception data can more comprehensively and accurately reflect the elevator operation status and passenger needs, and provide data support for subsequent analysis.

[0027] In order to facilitate the processing and analysis of large-scale fused perception data, the fused perception data is divided into blocks according to time periods. The fused perception data is divided into time periods of 15 minutes to form several perception blocks, thereby reducing the complexity of processing the fused perception data and improving the efficiency of processing the fused perception data, which is convenient for subsequent feature extraction and analysis.

[0028] After the elevator perception data is homomorphically encrypted, the perception blocks are also in an encrypted state, and the features of the perception blocks cannot be effectively extracted. Federated learning based on secure multi-party computing can realize joint feature extraction of perception blocks under the premise of protecting data privacy. Each edge node uses the fused perception data corresponding to the perception block, including the calculation of average speed, acceleration change rate, peak number of elevator passengers, average weight and other fused perception features. Through the secure multi-party computing protocol, the fused perception features calculated by each edge node are summarized and aggregated in an encrypted state, thereby realizing the effective integration and feature extraction of the fused perception data corresponding to the perception block without leaking the original elevator perception data, providing key features for subsequent elevator control and optimization.

[0029] In order to ensure the security of fused perception features during transmission and storage, and to realize centralized analysis of fused perception features, the features extracted from each perception block are transmitted to the cloud through a secure network. The cloud aggregates the received features to obtain fused perception features, providing safe and reliable feature data for subsequent analysis and prediction of elevator demand patterns and coordinated elevator dispatching and control.

[0030] A spatiotemporal graph is constructed based on the fused perception features, and the spatiotemporal graph is analyzed through a spatiotemporal graph convolutional neural network to learn the elevator demand patterns of different building types, and the elevator demand patterns in different scenarios are predicted through a generative adversarial network. like Figure 2 As shown in the figure, the construction logic of the space-time graph includes: Determine and fill the nodes of the space-time graph according to the floor number, building type, number of passengers, average waiting time and environmental indicators to obtain a node attribute matrix; The calculated frequency of users taking the elevator between floors is used as the edge weight to connect the nodes on different floors. The edge weight is determined by the length of the time interval to connect the nodes in different time periods on the same floor to obtain the adjacency matrix. The node attribute matrix and adjacency matrix of the spatiotemporal graph are updated in real time based on the fused perception features.

[0031] In order to effectively analyze and learn the elevator demand patterns of different building types at different times and spaces, it is necessary to convert the fused perception features into a structured space-time graph form for subsequent processing using the space-time graph convolutional neural network. First, the nodes of the space-time graph are determined. The nodes of the space-time graph include floor numbers, building types, number of passengers, average waiting time, and environmental indicators. The corresponding values ​​of these data are added to the nodes of the space-time graph to obtain a node attribute matrix. For example, the node attribute matrix for the 10th floor in an office building is [10, office building, 20, 3min, 24℃, 50%, 26PM2.5], where 50% refers to the humidity in the car, and 26PM2.5 refers to the PM2.5 concentration of 26 micrograms / cubic meter.

[0032] The frequency of users taking the elevator between floors is calculated using the frequency formula, and the calculated frequency of users taking the elevator between floors is used as the weight of the edge to connect the nodes on different floors. The function expression of the frequency formula is as follows: ; In the formula, Indicates floor To floor The frequency of users taking the elevator between Indicates from the floor To floor The historical number of elevator rides, Represents the floor index, Indicates the total number of floors, represents the frequency weight, Indicates from the floor To floor The number of elevator requests in the current time period, Indicates the total number of elevator requests for all floors in the current time period. Indicates the request weight.

[0033] It should be noted that: Floor To floor Frequency of users taking the elevator The weight of the edge in the space-time graph is determined by dynamically adjusting the historical number of elevator rides and the number of elevator rides requested. The value of is between 0 and 1; From the floor The total number of historical elevator rides to all floors, used to normalize the number of elevator rides from the floor To floor Historical number of elevator rides , eliminate the difference in elevator demand between floors; frequency weight Refers to the weight of the historical number of elevator rides, which is used to reflect the impact of the historical number of elevator rides on the frequency of users taking the elevator. The value range is between 0 and 1; the total number of elevator ride requests for all floors in the current time period Used to normalize from floor To floor The number of elevator rides requested in the current time period , eliminating the difference in the number of requests in different time periods, and ; Request weight Refers to the weight of the number of elevator requests in the current time period, which is used to reflect the impact of the number of elevator requests in the current time period on the frequency of elevator rides by users. The value range is between 0 and 1, and .

[0034] The unit of the time period is set to 15 minutes, and the weight of the edge is determined according to the length of the interval between different time periods. When the number of passengers changes greatly between two adjacent time periods, for example, from 10 people in one time period to 30 people in the next time period, then according to the calculation result of the elevator frequency formula, the weight of the edge is adjusted to a larger value, and the weight of the edge will increase accordingly, such as 0.8; conversely, when the number of passengers changes little between two adjacent time periods, then the weight of the edge is adjusted to a smaller value, and the weight of the edge will decrease accordingly, such as 0.2. In this way, the nodes in different time periods on the same floor can be connected to obtain the adjacency matrix.

[0035] The node attribute matrix and adjacency matrix of the space-time graph are updated in real time based on the fused perception features. The update frequency of the space-time graph is set to be updated every 15 minutes. When new fused perception feature data arrives, such as when the number of passengers on a certain floor is detected to have changed, or when the environmental indicators are updated, the corresponding node attribute matrix in the space-time graph is updated. At the same time, the adjacency matrix is ​​updated according to the new information such as the frequency of users taking the elevator between floors and the change in the number of passengers on the same floor in different time periods. Thus, a space-time graph containing the node attribute matrix and the adjacency matrix is ​​obtained, which can intuitively reflect the elevator demand data of different floors at different times and the relationship between floors and different time periods on the same floor. It provides a structured data basis for the subsequent analysis using the space-time graph convolutional neural network, enabling the elevator demand pattern to be learned and analyzed from the two dimensions of time and space, thereby improving the accuracy and comprehensiveness of the analysis.

[0036] The analysis logic of the space-time graph includes: The spatial dependency between floors is extracted through the graph convolution layer based on the node attribute matrix and adjacency matrix of the spatiotemporal graph; The changes in elevator demand data are analyzed based on the fused perception features of the time step through the time convolution layer; After forward propagation of multiple layers of graph convolutional layers and temporal convolutional layers, the elevator demand patterns of different building types in different time and space are learned.

[0037] After constructing the space-time graph, it is necessary to use the space-time graph convolutional neural network to extract the spatial dependencies and time demand changes contained in the space-time graph, so as to learn the elevator demand patterns of different building types at different times and spaces. The node attribute matrix and adjacency matrix of the space-time graph are input into the graph convolution layer of the space-time graph convolutional neural network. The graph convolution layer calculates the node attribute matrix and the adjacency matrix to extract the spatial dependencies between floors. For example, the correlation between the 10th floor and the surrounding floors (such as the 9th or 11th floor) in elevator demand can be obtained through calculation, and it can be judged which floors have more frequent passenger flow and which floors have relatively less connection.

[0038] The fused perception features of different time periods are input into the temporal convolution layer. The temporal convolution layer captures the changes in elevator demand data over time based on these fused perception features. For example, by analyzing data such as the number of passengers and the average waiting time in different time periods, the peak and trough of elevator demand in different time periods of the day, as well as the trend of changes in elevator demand, can be determined.

[0039] After forward propagation of multiple layers of graph convolutional layers and temporal convolutional layers, the spatiotemporal graph convolutional neural network continuously learns and optimizes, and finally learns the elevator demand patterns of different building types at different times and spaces. For example, for office buildings, it learns that 8:00 to 10:00 a.m. on weekdays is the rush hour, when there are a large number of passengers on each floor and the elevators run frequently, while 5:00 to 7:00 p.m. is the rush hour, when passengers mainly flow from high floors to low floors. In this way, relevant information on the elevator demand patterns of different building types at different times and spaces is obtained, including the spatial dependency between floors and the changing patterns of elevator demand over time. This can deeply understand the elevator demand characteristics of different building types, provide a strong basis for subsequent elevator coordinated scheduling and optimization, and help improve elevator operation efficiency and passenger satisfaction.

[0040] like Figure 3 As shown in Figure 1, the prediction logic of the generative adversarial network includes: Initialize the generator and discriminator in the generative adversarial network, and use the generator to generate elevator demand data in different scenarios; Obtain real elevator demand data under different historical scenarios, alternately train the generator and discriminator for adversarial training, use the discriminator to distinguish the difference between the generated elevator demand data and the real elevator demand data, and obtain the difference feedback of the discriminator; The parameters of the generator are adjusted according to the difference feedback of the discriminator. After multiple rounds of adversarial training, the trained generator is used to predict the elevator demand patterns of different building types in different scenarios.

[0041] In order to predict the elevator demand patterns of different building types in different scenarios, a generative adversarial network is used. The generator generates elevator demand data, and the discriminator judges the authenticity of the elevator demand data. The generator is optimized through adversarial training to achieve accurate prediction. First, it is necessary to initialize the generator and discriminator and generate elevator demand data in different scenarios. The scene classification labels (such as weekdays, weekends, working hours and off-get off work hours, etc.) are embedded into the input of the generator so that the generator can generate elevator demand data in different scenarios, such as the number of passengers and elevator frequency on each floor of an office building during working hours on weekdays.

[0042] The real elevator demand data under different historical scenarios are obtained, such as the actual number of passengers and elevator records of the office building in different time periods in the past month, and the generator and discriminator are alternately trained for adversarial training. The discriminator receives the elevator demand data generated by the generator and the real elevator demand data, and distinguishes the difference between the two types of data by calculating the authenticity probability of the two types of data (for example, using the cross entropy loss function to calculate the difference between the generated elevator demand data and the real elevator demand data), and obtains the difference feedback of the discriminator. At the same time, the scene classification label is also input into the discriminator, so that the discriminator can judge the authenticity of the generated elevator demand data according to different scenarios.

[0043] The parameters of the generator are adjusted according to the difference feedback of the discriminator, so that the elevator demand data generated by the generator is closer to the real elevator demand data. After multiple rounds of adversarial training, when the elevator demand data generated by the generator can better deceive the discriminator, the trained generator is used to predict the elevator demand patterns of different building types in different scenarios. For example, the elevator demand patterns of an office building in different time periods on different weekdays and weekends in the next week are predicted, which provides a reference for elevator collaborative scheduling, so that the elevator demand patterns can be predicted in advance, helping the elevator collaborative scheduling control to prepare in advance, optimize the elevator operation route and zoning scheduling strategy, improve the elevator operation efficiency and service quality, and reduce passenger waiting time.

[0044] Based on building types and scenarios, the operation areas are flexibly divided, and the operation routes are dynamically planned through the zoning scheduling strategy to achieve cross-regional coordinated scheduling of elevators; like Figure 4 As shown in the figure, the logic of dividing the operation area includes: Generate dynamic heat maps based on elevator demand data in different building types and scenarios; Determine whether to merge adjacent floors into the same operating area based on the changes in the dynamic thermal map; Set static elevator priority based on floor attributes, and dynamically adjust elevator priority based on elevator abnormal status; Each elevator determines the service operation area based on the status information and broadcasts it to other elevators.

[0045] There are differences in elevator demand data under different building types and scenarios. In order to dispatch elevators more efficiently, it is necessary to flexibly divide the operating areas according to the elevator demand data so that the elevators can serve the passengers in the corresponding areas more targetedly. First, the elevator demand data under different building types and different scenarios are collected, including the number of passengers on each floor, the direction of the elevator (up or down) and the waiting time, etc. Then, the floor plan of the building is divided into several small grid units through the density-based dynamic heat map generation algorithm. For each grid unit, the elevator demand density within a certain time window (such as 15 minutes) is calculated. The elevator demand density is calculated as the number of passengers in the grid unit divided by the area of ​​the grid unit. According to the calculated elevator demand density, different colors and transparency are assigned to each grid unit. The higher the elevator demand density, the darker the color and the lower the transparency, thereby generating a dynamic heat map. For example, during the rush hour from 8 to 10 a.m. on weekdays, the dynamic heat map of some floors of the office building shows darker colors, indicating that the elevator demand on these floors is large.

[0046] Observe the changes in elevator demand on adjacent floors in the dynamic heat map. If the difference in elevator demand density on adjacent floors is small and the elevator directions are the same, it is determined that these adjacent floors can be merged into the same operating area. The small difference in elevator demand density on adjacent floors is obtained by comparing with the difference threshold. When the difference in elevator demand density on adjacent floors is less than the difference threshold, it means that the difference in elevator demand density on adjacent floors is small. The same elevator direction means that 80% of the passengers are going up or down at the same time. For example, in an office building from the 10th floor to the 12th floor in a certain period of time, the elevator demand density is high and mainly for passengers in the upward direction, then these three floors can be merged into one operating area.

[0047] Static elevator priorities are set in combination with floor attributes. For example, the elevator priorities on each floor of a shopping mall, office building, and hospital are different. The specific elevator priorities can be represented by numbers. For example, the elevator priority for the fire escape on the ground floor of a shopping mall is 3, the elevator priority for the high-rise office building is 4, the elevator priority for the emergency floor of a hospital is 5, and the elevator priority for other floors is 2 or 1.

[0048] The elevator priority is dynamically adjusted in combination with the abnormal status of the elevator. When an elevator fails, for example, the sensor detects that the elevator is running abnormally or the door status is faulty, the elevator priority of the elevator is reduced, for example, from the original 3 to 1. At the same time, the priority of other elevators that are operating normally and can cover the service area of ​​the faulty elevator is increased to ensure the overall service quality of the elevators in the building.

[0049] Each elevator determines the operating area it serves based on its own status information (including the current floor position, operating direction, and whether there is a fault, etc.) and the divided operating area. For example, if an elevator is currently located on the 15th floor of an office building, and this floor belongs to the extended range of the 10-12 floor operating area previously divided, then this elevator determines to serve this operating area. This elevator broadcasts its served operating area to other elevators through the wireless network so that other elevators can understand the overall operating area allocation, thereby making the subsequent zoning scheduling of the elevator more in line with the elevator demand patterns under different building types and scenarios, improving the operating efficiency of the elevator, reducing the waiting time of passengers, and being able to adjust the zoning scheduling strategy in time when an elevator abnormality occurs to ensure the stability of the overall service.

[0050] like Figure 5 As shown, the partition scheduling strategies include: Obtain status information of each elevator in real time and assign elevator identification; Calculate the operating cost of each elevator based on the status information and operating area, and determine the operating route of each elevator; Based on the broadcast between elevators, a status consensus is reached, and each elevator is collaboratively scheduled according to the elevator priority.

[0051] After the operating areas are divided, in order to achieve cross-regional coordinated scheduling of elevators, it is necessary to obtain the elevator status information in real time, calculate the operating cost of each elevator, determine the operating route of each elevator, reach a status consensus through communication between elevators, and perform coordinated scheduling based on the elevator priority.

[0052] Various sensors installed on the elevators are used to obtain real-time status information of each elevator, including the elevator's current position, running speed, running direction, and whether there are any faults. At the same time, each elevator is assigned a unique elevator ID, such as numbers 1, 2, 3, etc., to facilitate distinction and management in the subsequent scheduling process.

[0053] The operating cost of each elevator is calculated based on the elevator's status information and operating area. The calculation of the operating cost needs to take into account multiple factors, such as the elevator's operating distance, waiting time and energy consumption. The waiting time is calculated based on the time the passenger requests the elevator and the elevator arrives. The energy consumption is measured by current and voltage sensors and calculated in combination with the operating time. A corresponding weight is assigned to each factor, such as a weight of 0.4 for the operating distance, a weight of 0.3 for the waiting time, and a weight of 0.3 for the energy consumption. The operating cost of each elevator is calculated by weighted summation, and then the operating route of each elevator is determined based on the operating cost of each elevator. For example, for a passenger requesting an elevator on a certain floor, the operating costs of all elevators that can serve the passenger are compared, and the elevator with the lowest operating cost is selected to go to that floor to pick up the passenger, and the corresponding operating route is planned.

[0054] Based on the broadcast between elevators, each elevator can receive the status information and service area information of other elevators, and coordinately dispatch each elevator according to the elevator priority. For example, when there are multiple passenger requests in a certain operating area, the elevator with a higher priority will respond first, and other elevators will reasonably avoid or assist according to the elevator priority and their own status information to achieve efficient dispatch of the entire elevator in the building, improve the dispatch efficiency and coordination of the elevator, reduce the overall operating cost, enhance the passengers' elevator experience, and enable elevators to operate more intelligently and efficiently to meet the elevator needs in different building types and scenarios.

[0055] Detect abnormal elevator status based on fused sensing data and trigger emergency control strategies; The trigger logic of the emergency control strategy includes: Continuously obtain fusion perception data during elevator operation; Determine the abnormal state of the elevator based on the changing trend of the fused sensing data, including whether the elevator is overloaded, whether the passengers in the elevator have abnormal behavior, and whether the car environment is abnormal; Determine multi-level abnormal classification based on abnormal status of elevator; Trigger emergency control strategies based on multi-level anomaly classification.

[0056] In order to detect the abnormal state of the elevator in time, it is necessary to obtain various data that can reflect the elevator's operating status, car environment and passenger conditions in real time and continuously. The fused perception data integrates various information and is the basis for judging the abnormal state of the elevator. Therefore, it is necessary to obtain it continuously. The edge node performs preliminary processing and fusion of these elevator perception data to form fused perception data and continuously updates it. For example, various sensor data are acquired and fused once a second to obtain the latest fused perception data, which provides real-time and comprehensive data support for subsequent accurate judgment of the abnormal state of the elevator, so that potential abnormal states of the elevator can be discovered in time.

[0057] Based on the latest acquired fusion perception data, by analyzing the changing trend of the fusion perception data, it is determined whether the elevator is in an abnormal state. This is a key step to achieve timely emergency response. It can detect different types of abnormal situations in a targeted manner. The edge node determines whether it is overloaded based on the weight distribution of the elevator passengers and the number of passengers in the fusion perception data. For example, it is known that the rated load capacity of the elevator is 1000 kg and the rated passenger capacity is 14 people. When the total weight of the elevator detected by the pressure sensor is greater than 1000 kg, or the number of passengers recognized by the camera is greater than 14 people, it is determined that the elevator is overloaded and is judged to be in an overload abnormal state.

[0058] The images inside the elevator car taken by the camera are analyzed through the image recognition algorithm, and some abnormal behavior patterns are pre-set, such as passengers falling, fighting, and leaning against the elevator door for a long time. When it is detected that the passenger's behavior meets these preset abnormal behavior patterns, it is determined that the passenger has exhibited abnormal behavior. For example, if the image recognition algorithm detects that a passenger falls in the car, it is determined that the passenger has exhibited abnormal behavior.

[0059] Whether the cabin environment is abnormal is determined based on the temperature, humidity and air quality indicators in the fused perception data. For example, the normal temperature range is set to 22℃-26℃, the humidity range is 40%-60%, the formaldehyde content in the air quality indicators is not more than 0.08mg / m³, and the PM2.5 concentration is not more than 35μg / m³. When the temperature detected by the temperature and humidity sensor is lower than 22℃ or higher than 26℃, the humidity is lower than 40% or higher than 60%, or the air quality sensor detects that the formaldehyde content is greater than 0.08mg / m³, and the PM2.5 concentration is greater than 35μg / m³, it is determined that the cabin environment is abnormal, so as to promptly discover various abnormal situations during the operation of the elevator, provide a clear basis for the subsequent adoption of corresponding emergency control measures, and ensure the safety and comfort of passengers.

[0060] Different degrees of elevator abnormal conditions require different levels of emergency measures. Therefore, the elevator abnormal conditions are classified into multiple levels of abnormalities in order to allocate resources more reasonably and take corresponding treatment methods. When there is an abnormal overload state or the car environment is abnormal and the degree of the elevator abnormality is low, it is judged as a level one abnormality. The low degree of the elevator abnormality can be judged by the passenger weight distribution and the degree of environmental indicators. For example, for the overload abnormal state, when the total weight of the elevator is greater than the rated load by less than 5%, or the number of passengers is greater than the rated number by less than 1 person; for the abnormal car environment, when the temperature deviates from the normal range by less than 1°C, the humidity deviates from the normal range by less than 5%, and only one of the air quality indicators is slightly exceeded, it is judged as the degree of the elevator abnormality is low.

[0061] When there is an abnormal overload state or the cabin environment is abnormal and the degree of elevator abnormality is medium, it is judged as a secondary abnormality. For example, for an abnormal overload state, when the total weight of the elevator is greater than the rated load by 5%-10%, or the number of passengers is greater than the rated number of 2-3 people; for an abnormal cabin environment, when the temperature deviates from the normal range by 1℃-3℃, the humidity deviates from the normal range by 5%-10%, or two of the air quality indicators exceed the standard, it is judged as a medium degree of elevator abnormality.

[0062] When there is an abnormal overload state or the passengers have abnormal behavior or the cabin environment is abnormal and the degree of elevator abnormality is high, it is judged as a third-level abnormality. For example, for an abnormal overload state, when the total weight of the elevator is more than 10% greater than the rated load, or the number of passengers is more than 3 people greater than the rated number; for an abnormal cabin environment, when the temperature deviates from the normal range by more than 3°C, the humidity deviates from the normal range by more than 10%, or multiple air quality indicators seriously exceed the standard, and the passengers have abnormal behavior, it is judged that the degree of elevator abnormality is high, so as to provide a clear standard for the subsequent triggering of the corresponding level of emergency control strategy, making emergency handling more scientific and reasonable, and being able to take appropriate measures according to the severity of the abnormality to improve the efficiency and effectiveness of emergency handling.

[0063] Different levels of abnormalities require emergency handling of different intensities and methods. The corresponding emergency control strategy is triggered according to the multi-level abnormality classification results to ensure the safe operation of the elevator and the safety of passengers. When it is determined to be a level one abnormality, a warm reminder is sent to the passengers through the voice broadcasting system in the car, informing them that the elevator is currently slightly overloaded or the car environment is slightly abnormal. Passengers are asked to pay attention to safety. At the same time, the abnormal information is sent to the monitoring system of the property management center for the attention of property personnel.

[0064] For level 2 abnormalities, in addition to voice prompts, the elevator's operating parameters are automatically adjusted. For example, in the case of overload, the elevator's operating speed is reduced to ensure safe operation. For abnormal cabin environment, the ventilation equipment in the cabin is started or the air-conditioning system is adjusted to try to improve the environment. At the same time, an alarm message is sent to the property management center, and staff are arranged to go to the floor where the elevator is located for inspection and processing.

[0065] Once it is determined to be a level 3 abnormality, emergency braking measures will be taken immediately, the elevator will be stopped at the nearest safe floor, the elevator door will be opened, and the passengers will be evacuated. At the same time, an emergency alarm will be sent to the property management center to notify maintenance personnel to rush to the scene for emergency repairs, and detailed abnormality information and handling status will be sent to all relevant personnel via text messages or mobile phone applications. For example, when passengers are detected fighting in the car and it is determined to be a level 3 abnormality, the elevator will immediately stop at the nearest floor, open the door to evacuate the passengers, and notify relevant personnel to handle it, thereby maximizing the protection of passengers' life safety and the safe operation of the elevator, reducing the hazards that may be caused by abnormal situations, and improving the emergency handling capability and reliability of the elevator.

Claims

1. An intelligent elevator automatic control method combining building user big data with AI perception, characterized in that: include: Deploy a distributed network in the building to obtain elevator perception data, which includes elevator operation data, car environment data, and user elevator data. Encrypt and fuse the elevator perception data and divide it into perception blocks by time period. Based on federated learning of secure multi-party computing, extract features from the perception blocks and generate fused perception features. A spatiotemporal graph is constructed based on the fused perception features, and the spatiotemporal graph is analyzed through a spatiotemporal graph convolutional neural network to learn the elevator demand patterns of different building types, and the elevator demand patterns in different scenarios are predicted through a generative adversarial network. Based on building types and scenarios, the operation areas are divided and the operation routes are dynamically planned through zoning scheduling strategies; Detect abnormal elevator status based on fused perception data and trigger emergency control strategies.

2. The intelligent elevator automatic control method combining building user big data and AI perception as claimed in claim 1 is characterized in that: The deployment logic of the distributed network includes: Install sensors in each elevator in the building to obtain elevator sensing data; Deploy edge nodes in each elevator to receive and process elevator sensing data; Determine whether to dynamically adjust the frequency of acquiring elevator perception data based on changes in the cabin environment data and user elevator data.

3. The intelligent elevator automatic control method combining building user big data and AI perception as claimed in claim 2 is characterized in that: The generation logic of the fusion perception feature includes: Encrypt and interpolate elevator sensing data in edge nodes; Assign weights to the elevator perception data, fuse the elevator perception data through the attention mechanism, and obtain fused perception data; Divide the fused perception data into perception blocks according to time periods; Federated learning based on secure multi-party computing is used to extract features from each perception block; The features extracted from each perception block are sent to the cloud for aggregation to obtain fused perception features.

4. The intelligent elevator automatic control method combining building user big data and AI perception as claimed in claim 3 is characterized in that: The construction logic of the space-time graph includes: Determine and fill the nodes of the space-time graph according to the floor number, building type, number of passengers, average waiting time and environmental indicators to obtain a node attribute matrix; The calculated frequency of users taking the elevator between floors is used as the edge weight to connect the nodes on different floors. The edge weight is determined by the length of the time interval to connect the nodes in different time periods on the same floor to obtain the adjacency matrix. The node attribute matrix and adjacency matrix of the spatiotemporal graph are updated in real time based on the fused perception features.

5. The intelligent elevator automatic control method combining building user big data and AI perception as claimed in claim 4 is characterized in that: The analysis logic of the space-time diagram includes: The spatial dependency between floors is extracted through the graph convolution layer based on the node attribute matrix and adjacency matrix of the spatiotemporal graph; The changes in elevator demand data are analyzed based on the fused perception features of the time step through the time convolution layer; After forward propagation of multiple layers of graph convolutional layers and temporal convolutional layers, the elevator demand patterns of different building types in different time and space are learned.

6. The intelligent elevator automatic control method combining building user big data and AI perception as claimed in claim 5 is characterized in that: The prediction logic of the generative adversarial network includes: Initialize the generator and discriminator in the generative adversarial network, and use the generator to generate elevator demand data in different scenarios; Obtain real elevator demand data under different historical scenarios, alternately train the generator and discriminator for adversarial training, use the discriminator to distinguish the difference between the generated elevator demand data and the real elevator demand data, and obtain the difference feedback of the discriminator; The parameters of the generator are adjusted according to the difference feedback of the discriminator. After multiple rounds of adversarial training, the trained generator is used to predict the elevator demand patterns of different building types in different scenarios.

7. The intelligent elevator automatic control method combining building user big data and AI perception as claimed in claim 6 is characterized in that: The division logic of the operation area includes: Generate dynamic heat maps based on elevator demand data in different building types and scenarios; Determine whether to merge adjacent floors into the same operating area based on the changes in the dynamic thermal map; Set static elevator priority based on floor attributes, and dynamically adjust elevator priority based on elevator abnormal status; Each elevator determines the service operation area based on the status information and broadcasts it to other elevators.

8. The intelligent elevator automatic control method combining building user big data and AI perception as claimed in claim 7 is characterized in that: The partition scheduling strategy includes: Obtain status information of each elevator in real time and assign elevator identification; Calculate the operating cost of each elevator based on the status information and operating area, and determine the operating route of each elevator; Based on the broadcast between elevators, a status consensus is reached, and each elevator is collaboratively scheduled according to the elevator priority.

9. The intelligent elevator automatic control method combining building user big data and AI perception as claimed in claim 8, characterized in that: The triggering logic of the emergency control strategy includes: Continuously obtain fusion perception data during elevator operation; Determine the abnormal state of the elevator based on the changing trend of the fused sensing data, including whether the elevator is overloaded, whether the passengers in the elevator have abnormal behavior, and whether the car environment is abnormal; Determine multi-level abnormal classification based on abnormal status of elevator; Trigger emergency control strategies based on multi-level anomaly classification.

10. The intelligent elevator automatic control method combining building user big data and AI perception as claimed in claim 9, characterized in that: The generation logic of the dynamic heat map includes: Divide the floor plan of the building into a plurality of grid cells; Calculate the elevator demand density of each grid cell in different time windows; Color and transparency are assigned to each grid cell according to the density of elevator demand to generate a dynamic heat map.

Images