Self-learning elevator floor selection method and system based on passenger behavior habits

Abstract

The application discloses a self-learning elevator floor selection method and system based on passenger elevator riding behavior habits, and the method comprises the following steps: recognizing a passenger, and acquiring the elevator riding behavior habit characteristics of the passenger; constructing a feature matrix through the elevator riding behavior habit characteristics of the passenger; predicting the destination floor of the passenger by using the feature matrix, and acquiring a prediction result; training a floor prediction model based on the feature matrix and the prediction result, and acquiring an updated floor prediction model; and predicting the destination floor of the passenger based on the updated floor prediction model. The application can automatically predict the destination floor of the passenger according to the elevator riding behavior habits of the passenger, avoid direct contact in the elevator, improve the elevator riding efficiency, and has the characteristics of non-contact, high efficiency and no feeling.

Description

Technical Field

[0001] This invention belongs to the field of elevator intelligent scheduling technology, and particularly relates to a self-learning elevator floor selection method and system based on passenger elevator riding behavior habits. Background Technology

[0002] Elevators, as an important form of vertical transportation, are widely used in various aspects of life. Currently, people are paying more attention to personal hygiene and avoiding direct contact with potential bacteria and viruses, such as avoiding touching elevator buttons. Based on each passenger's elevator usage habits—for example, if residents in a community live on a fixed floor and generally travel between the first and their home floors daily—their destination floor can be predicted. This avoids passengers touching floor selection buttons and selecting floors each time they enter the elevator, thus preventing indirect contact between passengers, reducing the risk of infection. It also improves the elevator experience when passengers are carrying heavy objects or holding children, making it difficult to use their hands, achieving a contactless, efficient, and seamless elevator experience.

[0003] Existing technologies primarily consider the behavioral habit of elevator travel direction and rely on the frequency with which a passenger selects a specific destination floor in a recent period to perform self-learning. For example, if a passenger has traveled from the 1st floor to the 3rd floor twice, to the 5th floor five times, and to the 8th floor three times recently, the method predicts that the passenger's next destination floor when taking the elevator from the 1st floor will be the 5th floor. This self-learning method is relatively simple in its approach, relies on a single feature dimension, and is only suitable for relatively simple destination floor prediction tasks. For instance, if a passenger takes the elevator to the -1st floor at 7:00 AM to drive their child to school, to the 2nd floor at 10:00 AM to buy groceries, and to the 1st floor at 7:00 PM for a walk, the number of times they travel from their residential floor to the -1st, 2nd, and 1st floors is almost the same, making it impossible for this method to accurately predict the passenger's destination floor. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention aims to propose a self-learning elevator floor selection method and system based on passenger elevator riding habits, thereby improving the user's elevator riding experience. The system can learn passenger elevator riding habits through model training and predict the passenger's destination floor accordingly.

[0005] To achieve the above objectives, this invention provides a self-learning elevator floor selection method based on passenger elevator riding habits, including:

[0006] Identify passengers and obtain their elevator riding behavior characteristics;

[0007] A feature matrix is ​​constructed based on the elevator riding behavior habits of the passengers.

[0008] The target floor of the occupant is predicted using the feature matrix, and the prediction result is obtained.

[0009] Based on the feature matrix and the prediction results, a floor prediction model is trained to obtain an updated floor prediction model.

[0010] The destination floor of the occupant is predicted based on the updated floor prediction model.

[0011] Optionally, identifying passengers and obtaining their elevator-riding behavior characteristics includes:

[0012] The face of the passenger is recognized and matched with the system database to determine the passenger's identity and obtain the passenger's elevator riding behavior characteristics.

[0013] Optionally, occupant identification also includes: acquiring the face detection box and body detection box of the occupant, matching the face detection box and the body detection box, determining the location of each occupant, and tracking them.

[0014] Optionally, the passenger's elevator riding behavior characteristics include: the passenger's identification number, residential floor, current elevator riding time, elevator riding direction, current floor, and elevator riding information.

[0015] Optionally, after inputting the feature matrix into the floor prediction model and obtaining the prediction result, the method further includes:

[0016] The prediction result is judged. If the prediction result is incorrect, the destination floor reached by the passenger is added to the passenger's elevator riding behavior characteristics. If the prediction result is correct, the predicted destination floor is added to the passenger's elevator riding behavior characteristics.

[0017] Optionally, based on the feature matrix and the prediction results, training the floor prediction model to obtain the updated floor prediction model includes:

[0018] Obtain a new feature matrix, filter the new feature matrix, and obtain the updated feature matrix;

[0019] The updated feature matrix is ​​input into the floor prediction model to obtain the target floor for prediction.

[0020] The floor prediction model is updated based on the predicted target floor and the correct target floor.

[0021] Optionally, obtaining a new feature matrix and filtering the new feature matrix includes:

[0022] The accuracy of the floor prediction model is obtained. If the accuracy of the floor prediction model is less than a set threshold, the new feature matrix is ​​added to the feature matrix; otherwise, the new feature matrix is ​​input into the floor prediction module.

[0023] If the prediction result of the floor prediction model is incorrect, the new feature matrix is ​​added to the feature matrix; otherwise, the overlap between the new feature matrix and the feature matrix is ​​calculated.

[0024] Based on the overlap, it is determined whether to add the new feature matrix to the feature matrix, thereby completing the filtering of the new feature matrix.

[0025] Optionally, updating the floor prediction model based on the predicted target floor and the correct target floor includes:

[0026] Based on the predicted target floor and the correct target floor, calculate the loss function and obtain the loss value of the loss function;

[0027] If the loss value of the loss function is greater than the set loss value, then the floor prediction model continues to be trained and updated; otherwise, the update of the prediction model is completed.

[0028] On the other hand, to achieve the above objectives, the present invention also provides a self-learning elevator floor selection system based on passenger elevator behavior habits, including: a camera module, a feature recognition module, a wireless transmission module, a model inference module, a cloud training module, a transfer learning module, and a data filtering module.

[0029] The camera module is used to identify occupants;

[0030] The feature recognition module is used to identify the passenger's elevator riding behavior habits.

[0031] The model inference module is used to predict the passenger's destination floor based on the passenger's elevator riding behavior habits.

[0032] The cloud training module is used to train the floor prediction model based on the passenger's elevator riding behavior characteristics and the passenger's destination floor, and to update the trained floor prediction model to the model inference module.

[0033] The wireless transmission module is used to transmit the passenger's elevator riding behavior characteristics to the model inference module, and to transmit the passenger's elevator riding behavior characteristics and the passenger's destination floor to the cloud training module.

[0034] The transfer learning module is used to transfer parameters from the floor prediction model to another model;

[0035] The data filtering module is used to filter the elevator riding behavior characteristics of the passengers.

[0036] The technical advantages of this invention are as follows: By filtering data, this invention reduces repetitive and useless passenger elevator behavior characteristic data, lowers storage requirements, and avoids model overfitting. This invention can transfer trained model parameters to another model and utilize small samples for training and fine-tuning, thereby establishing a high-performing model in a short time. This invention can automatically predict the passenger's destination floor based on their elevator riding habits, avoiding direct contact inside the elevator, improving elevator efficiency, and featuring contactless, high-efficiency, and seamless operation. Attached Figure Description

[0037] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0038] Figure 1 This is a flowchart illustrating the self-learning elevator floor selection method based on passenger elevator riding behavior habits, as described in an embodiment of the present invention.

[0039] Figure 2 This is a schematic diagram of the passenger behavior habit feature matrix according to an embodiment of the present invention;

[0040] Figure 3 This is a schematic diagram of the cloud training module of the self-learning elevator floor selection system based on passenger elevator behavior habits in an embodiment of the present invention;

[0041] Figure 4 This is a schematic diagram of the model reasoning module of the self-learning elevator floor selection system based on passenger elevator behavior habits in an embodiment of the present invention;

[0042] Figure 5 This is a schematic diagram of the transfer learning module of the self-learning elevator floor selection system based on passenger elevator riding behavior habits in an embodiment of the present invention;

[0043] Figure 6 This is a schematic diagram of the data filtering module of the self-learning elevator floor selection system based on passenger elevator behavior habits in an embodiment of the present invention;

[0044] Figure 7 This is a schematic diagram of the camera module of the self-learning elevator floor selection system based on passenger elevator behavior habits, according to an embodiment of the present invention. Detailed Implementation

[0045] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0046] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0047] This embodiment provides a self-learning elevator floor selection method based on passenger elevator riding behavior habits, including:

[0048] Identify passengers and obtain their elevator usage habits;

[0049] A feature matrix is ​​constructed based on the elevator riding behavior habits of passengers;

[0050] The target floor of the occupant is predicted using the feature matrix, and the prediction result is obtained;

[0051] Based on the feature matrix and the prediction results, a floor prediction model is trained to obtain an updated floor prediction model.

[0052] The destination floor of the occupant is predicted based on the updated floor prediction model.

[0053] like Figure 1 As shown, this embodiment also provides a self-learning elevator floor selection system based on passenger elevator riding behavior habits. The system includes a camera module, a feature recognition module, a wireless transmission module, a model inference module, a cloud training module, a transfer learning module, and a data filtering module. The camera module completes passenger identification; the feature recognition module completes passenger elevator riding behavior habit feature recognition, which is used as model input data and training data; the wireless transmission module completes data transmission, transmitting passenger feature data to the model inference module and transmitting passenger feature data and passenger destination floor to the cloud training module respectively; after receiving passenger features, the model inference module performs model inference and outputs the passenger destination floor; the cloud training module trains the floor prediction model based on the received data, and updates it to the model inference module after training is completed; the transfer learning module can complete the learning of passenger behavior habits with a small amount of data; the filtering module can reduce repetitive and useless passenger elevator riding behavior habit feature data, reduce storage occupation, and avoid model overfitting.

[0054] When a passenger rides the elevator for the first time, the elevator will output a default floor. If it is not the passenger's destination floor, the passenger can select the correct destination floor through the floor selection button. The system will input the corresponding behavioral characteristics and the correct destination floor into the cloud training module for model training. After the passenger rides the elevator multiple times, the floor prediction model can learn the passenger's elevator riding behavior habits and accurately predict the destination floor based on this.

[0055] Each time a passenger uses the elevator, their elevator-riding habits are identified and uploaded for training. The difference is that if the passenger does not change their destination floor, the model is considered to have output the correct destination floor; if the passenger changes their destination floor, the model is considered to have output the incorrect destination floor. The former strengthens the floor prediction model, while the latter corrects it.

[0056] Specifically, passenger A arrives at the elevator entrance and presses the elevator call button.

[0057] When the elevator call button is pressed, the facial recognition module is triggered to identify the passenger ID. At the same time, the feature recognition module begins to identify the passenger's elevator riding behavior characteristics, which are obtained from the system as follows: ① the passenger's residential floor; ② the current time, whether it is a weekday / weekend / holiday; ③ the elevator riding direction, determined by whether the passenger pressed the up or down call button, and the current floor. At this time, the destination floor is defaulted to 0 (at this time, the elevator riding behavior characteristics have not yet been input into the model inference module, and there is no output of the destination floor); ④ the order of elevator riding, which is retrieved from the system from the passenger's previous three elevator riding records.

[0058] The elevator riding behavior habits identified by the feature recognition module are constructed into a feature matrix and input into the model inference module. The model inference module outputs the predicted target floor, and the corresponding floor button on the floor selection button lights up.

[0059] After the elevator arrives at the floor, the passenger enters. If the passenger presses a button for another floor, the model is considered to have made a prediction error, and the floor corresponding to the pressed button is added to the destination floor in the elevator direction section of the passenger's elevator behavior habit features. If the passenger does not press a button for another floor before the end of the elevator ride, the model is considered to have made a prediction correct, and the predicted floor is added to the destination floor in the elevator direction section of the passenger's elevator behavior habit features. At this point, the passenger's elevator behavior habit features are complete, and a complete feature matrix is ​​constructed. The correct destination floor is also used as a label in the cloud-based model training.

[0060] The elevator then travels to the correct destination floor, and passenger A completes the elevator ride after reaching the destination floor.

[0061] Passenger behavior characteristics are mainly divided into passenger ID, residential floor, time of elevator ride, direction of elevator ride, and order of elevator ride.

[0062] Passenger ID: Passenger identification number;

[0063] Residential floors: Passenger living floors / Passenger office floors;

[0064] Elevator travel time: Year-Month-Day-Hour-Minute-Second, Weekday / Weekend / Holiday;

[0065] Elevator direction: Up / Down, Current floor - Destination floor;

[0066] Elevator ride sequence: Information on the first three elevator rides, for example: 5th floor to 1st floor, 1st floor to 6th floor, 6th floor to 5th floor.

[0067] Methods for obtaining passenger behavior and habit characteristics

[0068] Passenger ID: Facial recognition and query system;

[0069] Residential Floor: Inquiry System;

[0070] Elevator travel time: Get the current system time and query the system;

[0071] Elevator direction: Determine the up or down direction and obtain the current floor based on the button pressed when calling the elevator; determine the destination floor based on the selected floor.

[0072] Elevator boarding order: Information saved in the query system.

[0073] like Figure 2 The diagram shown illustrates the occupant behavior habit feature matrix according to an embodiment of the present invention. After acquiring occupant behavior habit features, the system constructs a feature matrix. The data stored in the feature matrix from left to right are: occupant ID, residential floor, elevator time, elevator direction, and elevator order. After the feature matrix is ​​constructed, it is transmitted to the cloud training module via a wireless transmission module.

[0074] like Figure 3 The diagram shown illustrates the cloud-based training module of this invention. Besides the occupant behavior habit feature matrix, the correct target floor also needs to be input for calculating the loss function. The training process is as follows: Figure 3 .

[0075] The specific steps are as follows:

[0076] First, the feature dataset is input into the model, which performs forward propagation and outputs a predicted target floor. This predicted floor and the correct target floor are then input into the loss function to calculate the loss value. If the loss value is greater than or equal to a set threshold, the model has not converged. The model continues with backpropagation, updating the model weights, and repeating the forward propagation process. When the loss value is less than the set threshold, the model is considered to have converged, training stops, and the model is saved and output.

[0077] After the model training is completed, it is deployed to the model inference module.

[0078] After training, the model can learn the behavioral habits of passengers. It does not rely on a single feature, but rather integrates various features to make judgments.

[0079] For example, passenger A lives on the 6th floor of a residential building. They leave home at 7 AM and return at 8 PM every day. The model learns their characteristics, and their elevator-riding habits are defined as leaving in the morning and returning home in the evening. One day, they arrive at the elevator at 6:45 AM. Their elevator-riding habits at this time are: elevator arrival time - 6:45 AM, elevator direction - downhill, starting from the 6th floor. These characteristics are input into the model. Based on its learned elevator-riding habits—morning, downhill, starting from the 6th floor—the model predicts they need to leave, with their destination being the 1st floor. In this case, elevator arrival time and direction contribute significantly to the decision. When they arrive at the elevator at 8:20 PM, their main habits are: elevator arrival time - 8:20 PM, elevator direction - uphill, starting from the 1st floor. Based on their habits—evening, uphill, starting from the 1st floor—the model predicts they are returning home, outputting their residential floor as the 6th floor.

[0080] For example, B is a company secretary who frequently needs to go down from the 8th floor to the 1st floor to pick up a package on weekdays, then go to the 9th floor to give it to her supervisor, and finally return to the 8th floor. Her behavioral habits follow a small cycle: 8th floor → 1st floor → 9th floor → 8th floor. When she needs to go downstairs to pick up the package at 9:00 AM, her behavioral characteristics are: elevator time - 9:00 AM, elevator direction - downhill, starting from the 8th floor, elevator sequence - start of the small cycle of picking up the package (i.e., 8th floor → 1st floor). Because the arrival time of the package is uncertain, the elevator time feature contributes little to the prediction. The features of going downhill and starting from the 8th floor contribute significantly, but these features are too singular to guarantee prediction accuracy. Therefore, the contribution of the elevator sequence becomes crucial. Based on her habits—working hours, downhill, starting from the 8th floor, small cycle of picking up the package—the model predicts that the person will go downstairs to pick up the package and outputs the destination floor as 1st floor. The same applies to the next two elevator cycles.

[0081] like Figure 4 The diagram shows a schematic of the model inference module in an embodiment of the present invention. The main function of the model inference module is to output the predicted elevator floor after receiving the input passenger behavior habit features. Since the model has already been trained and has learned the behavioral features, there is no need to perform backpropagation here; the floor output by the model is directly used as the target floor.

[0082] like Figure 5 The diagram shows a transfer learning module according to an embodiment of the present invention. In daily life, different people often share some common habits. For example, in one residential community, most people leave early and return late, and many residents are accustomed to taking a walk downstairs after dinner. Similarly, similar habits exist in other communities. In one company, employees typically start work at 9:00 AM and finish at 6:00 PM. Even though in another company, the start time might be 8:30 AM and the finish time 5:30 PM, aside from the different times, their other behavioral habits are highly similar.

[0083] When a new residential community is built and residents move in, the number of times users use the elevators is limited, making it difficult to build a good dataset of elevator usage habits in a short period, thus hindering the model's predictive performance. Therefore, based on transfer learning, the parameters of a model trained in another community can be transferred to the model in this community, and the model can be trained and fine-tuned using a small sample size, allowing for the establishment of a high-performing model in a short time.

[0084] like Figure 6 The diagram shows the data filtering module of this invention. A large amount of repetitive and similar data can easily lead to model overfitting, reducing model robustness and overall performance. Such data is abundant in passenger elevator usage habits. A person's habits are long-term; for example, a resident of a community might live there for over a decade. If data from over a decade were used as the dataset for model training, it would not only result in a large data volume and wasted storage space, but also lead to long training times, untimely updates, and most importantly, model overfitting. When a passenger's elevator usage habit changes, the model would struggle to correct it. Therefore, filtering this data is necessary.

[0085] The specific steps are as follows: First, obtain the model's accuracy. When the accuracy is less than or equal to a set threshold, add the new data directly to the feature dataset. When the accuracy is greater than the set threshold, further filter the new data. In the next filtering step, if the new data outputs an error when input to the model inference module, add it to the feature dataset; if the output is correct, proceed to the next filtering step. In the next filtering step, calculate the overlap between the new data and the feature dataset. If the overlap is less than or equal to a set threshold, add it to the feature dataset; if it is greater than the set threshold, discard it.

[0086] When only one passenger is using the elevator, if the destination floor button is changed, it will be assumed by default that the passenger changed the destination floor.

[0087] When multiple passengers are using the elevator, if a new floor button is pressed, meaning one of the destination floor buttons has been changed, the system cannot determine which passenger changed the destination floor. Therefore, the following method is proposed to solve this problem.

[0088] like Figure 7 The camera module shown is installed on the wall opposite the elevator car door.

[0089] The camera performs facial recognition on passengers entering the elevator and matches the results with the system database to determine the passengers' identities.

[0090] The camera performs both facial recognition and human body detection. The obtained face detection bounding boxes (composed of x, y, w, h – the x-coordinate of the box center point, the y-coordinate of the box center point, the width of the box, and the height of the box) are matched with the human body detection bounding boxes. The matching principle is that the face detection box is within the human body detection box, and the face detection box is located within the upper quarter of the human body detection box. Finally, the human body detection boxes are mapped one-to-one with the occupants, and their positions are determined in real time through target tracking.

[0091] When the elevator door is detected to be opening in the video, and the occupant's human detection frame gradually moves towards the edge of the video (near the door) and eventually disappears from the video, it is determined that the occupant has exited the stairwell and reached the destination floor. In other words, when the elevator reaches the newly selected destination floor, the system determines who changed the destination floor based on which occupant exits the elevator on that floor.

[0092] Based on the above, it can be determined which passengers changed their destination floor when one or more passengers changed their destination floor during an elevator ride.

[0093] Further optimization involves not uploading the feature matrix data immediately after each passenger completes their elevator ride, but saving it and uploading it all at once when there are enough of the data or at regular intervals. This achieves high efficiency and reduces power consumption. Moreover, cloud model training also requires a certain amount of time, which is in line with reality.

[0094] Further optimization involves adding L1 regularization to the loss function, which uses the sum of the absolute values ​​of the weights to penalize the model. This limits the sum of the absolute values ​​of the weights to a certain value, preventing the weights from becoming too large and ensuring that the model does not fit too closely to the training set, thus preventing overfitting.

[0095] To further optimize the model, a Dropout mechanism is added to the network model. During each training iteration, some parameters are randomly masked, i.e., set to 0. This ensures that the masked parameters are different in each training round, treating the network model as a combination of multiple networks. This captures more random factors, enhances model performance, and prevents overfitting.

[0096] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A self-learning elevator floor selection method based on passenger elevator riding behavior habits, characterized in that, include: Identify passengers and obtain their elevator riding behavior characteristics; A feature matrix is ​​constructed based on the elevator riding behavior habits of the passengers. The target floor of the occupant is predicted using the feature matrix, and the prediction result is obtained. Based on the feature matrix and the prediction results, a floor prediction model is trained to obtain an updated floor prediction model. Training the floor prediction model based on the feature matrix and the prediction results, and obtaining the updated floor prediction model includes: Obtain a new feature matrix, filter the new feature matrix, and obtain the updated feature matrix; The updated feature matrix is ​​input into the floor prediction model to obtain the target floor for prediction. The floor prediction model is updated based on the predicted target floor and the correct target floor; Obtaining a new feature matrix and filtering the new feature matrix includes: Obtain the accuracy of the floor prediction model. If the accuracy of the floor prediction model is less than a set threshold, add the new feature matrix to the feature matrix; otherwise, input the new feature matrix into the floor prediction model. If the prediction result of the floor prediction model is incorrect, the new feature matrix is ​​added to the feature matrix; otherwise, the overlap between the new feature matrix and the feature matrix is ​​calculated. Based on the overlap, it is determined whether to add the new feature matrix to the feature matrix, thereby completing the filtering of the new feature matrix; The destination floor of the occupant is predicted based on the updated floor prediction model.

2. The self-learning elevator floor selection method based on passenger elevator riding behavior habits as described in claim 1, characterized in that, Identifying passengers and obtaining their elevator-riding behavior characteristics includes: The face of the passenger is recognized and matched with the system database to determine the passenger's identity and obtain the passenger's elevator riding behavior characteristics.

3. The self-learning elevator floor selection method based on passenger elevator riding behavior habits as described in claim 1, characterized in that, The occupant identification process also includes: acquiring the face detection box and the body detection box of the occupant, matching the face detection box and the body detection box, determining the location of each occupant, and tracking them.

4. The self-learning elevator floor selection method based on passenger elevator riding behavior habits as described in claim 1 or 2, characterized in that, The elevator riding behavior characteristics of the passenger include: the passenger's identification number, residential floor, current elevator riding time, elevator riding direction, current floor, and elevator riding information.

5. The self-learning elevator floor selection method based on passenger elevator riding behavior habits as described in claim 1, characterized in that, After obtaining the prediction result, the process also includes: The prediction result is judged. If the prediction result is incorrect, the destination floor reached by the passenger is added to the passenger's elevator riding behavior characteristics. If the prediction result is correct, the predicted destination floor is added to the passenger's elevator riding behavior characteristics.

6. The self-learning elevator floor selection method based on passenger elevator riding behavior habits as described in claim 1, characterized in that, Updating the floor prediction model based on the predicted target floor and the correct target floor includes: Based on the predicted target floor and the correct target floor, calculate the loss function and obtain the loss value of the loss function; If the loss value of the loss function is greater than the set loss value, then the floor prediction model continues to be trained and updated; otherwise, the update of the prediction model is completed.

7. A system for implementing the self-learning elevator floor selection method based on passenger elevator riding behavior habits as described in claim 1, the system comprising: Camera module, feature recognition module, wireless transmission module, model inference module, cloud training module, transfer learning module, data filtering module; The camera module is used to identify occupants; The feature recognition module is used to identify the passenger's elevator riding behavior habits. The model inference module is used to predict the passenger's destination floor based on the passenger's elevator riding behavior characteristics; The cloud training module is used to train the floor prediction model based on the passenger's elevator riding behavior characteristics and the passenger's destination floor, and to update the trained floor prediction model to the model inference module. Training the floor prediction model to obtain the updated floor prediction model includes: Obtain a new feature matrix, filter the new feature matrix, and obtain the updated feature matrix; The updated feature matrix is ​​input into the floor prediction model to obtain the target floor for prediction. The floor prediction model is updated based on the predicted target floor and the correct target floor; Obtaining a new feature matrix and filtering the new feature matrix includes: Obtain the accuracy of the floor prediction model. If the accuracy of the floor prediction model is less than a set threshold, add the new feature matrix to the feature matrix; otherwise, input the new feature matrix into the floor prediction model. If the prediction result of the floor prediction model is incorrect, the new feature matrix is ​​added to the feature matrix; otherwise, the overlap between the new feature matrix and the feature matrix is ​​calculated. Based on the overlap, it is determined whether to add the new feature matrix to the feature matrix, thereby completing the filtering of the new feature matrix; The wireless transmission module is used to transmit the passenger's elevator riding behavior characteristics to the model inference module, and to transmit the passenger's elevator riding behavior characteristics and the passenger's destination floor to the cloud training module. The transfer learning module is used to transfer parameters from the floor prediction model to another model; The data filtering module is used to filter the elevator riding behavior characteristics of the passengers.

Images