A method for recognizing a group control elevator traffic pattern

Abstract

The application discloses a traffic mode recognition method for group control of elevators, relates to an elevator mode recognition method, and discloses a passenger flow condition detection model constructed by using a yolov5 target detection technology, which is used for detecting passenger flow, extracting passenger flow data, and completing recognition of a traffic mode and optimization of dispatching of elevators. The traffic mode recognition can recognize real-time, effective and accurate passenger flow data, analyzes the actual passenger flow, and allocates a traffic mode in line with the current passenger flow condition. The application adds an improved image processing technology to a traditional elevator group control system, designs a traffic mode recognition module based on actual and effective passenger flow conditions, so that the passenger flow condition recognized by the module is time-effective and real, and is significant for the elevator group control system.

Description

TECHNICAL FIELD

[0001] The application relates to an elevator mode recognition method, in particular to a traffic mode model construction method for group control elevator operation. BACKGROUND

[0002] Elevators have developed for more than 100 years from mechanical control to motor drag control and now to intelligent control, and especially after the appearance of PLC, elevator control has gradually tended to maturity. Modern building construction and planning are no longer single such as housing or office, and more tend to commercialization and diversification, the floors are higher and higher, the business modes are more and more, and the people flow in the building is larger and larger, so the traditional elevator is difficult to meet the demand of the modern building.

[0003] As the main traffic tool in the building, the elevator undertakes the responsibility of conveying personnel to move up and down in the building. The elevator first appeared in people's vision in a mechanical way, and then with the development of electrical control, the single-step elevator dominated by electrical control became the mainstream. However, the single-step elevator is difficult to meet the requirements of conveying passengers when facing a large number of people and a large number of floors, so the group control elevator gradually replaces the single-step elevator. The performance indicators of the group control elevator are very important to passengers, and the performance of the group control system will determine the waiting time and the time of taking the elevator of the passengers.

[0004] The elevator has nonlinearity, discreteness and uncertainty, so it is difficult to establish an accurate mathematical model. Therefore, how to meet the scheduling of the group control elevator system will be the main problem to improve the performance of the group control system. SUMMARY

[0005] The purpose of the application is to provide a group control elevator traffic mode recognition method. The application takes target detection as the starting point, detects the people in the elevator car and the waiting people in real time, and then obtains the actual passenger flow condition. The control module in the system integrates the passenger flow data to obtain the traffic mode conforming to the current passenger flow, and improves the operation efficiency of the group control system.

[0006] The purpose of the application is achieved by the following technical scheme:

[0007] A group control elevator traffic mode recognition method, the method comprises constructing a traffic mode recognition image detection network model, a traffic mode recognition module extracts data from the image model and distributes traffic modes according to real-time conditions, and specifically comprises the following steps:

[0008] 1) prepare passenger data set for target detection of elevator group control system;

[0009] 2) Configure the network environment using the deep learning pytorch framework, and complete the model construction of yolov5 in this environment;

[0010] 3) Add CA attention mechanism to the original model framework of yolov5 to optimize the model;

[0011] 4) The preprocessed data set is used as the input of the network and is trained, the yolov5s pre-training weight is loaded, and CIOU is used as the loss function;

[0012] 5) Use the video of the elevator passengers to detect the network model;

[0013] 6) The traffic mode recognition module sorts the passenger flow data detected by the system, and decides the suitable elevator traffic mode according to the passenger flow data.

[0014] The traffic mode recognition method of the elevator group control, the system includes preprocessing the picture data set input to the model; for elevator group control traffic mode recognition, the data set uses the picture data of the passengers when they get on or wait for the elevator; collect the video of the monitoring equipment and process it frame by frame to obtain the images of the passengers in various situations, collect and arrange the elevator passenger pictures and convert them to JPG format, then use the labelimg tool to label the picture data, and then output the label format in xml format, since the label format of yolo is in txt, the xml format label needs to be converted to txt format through code; when training the network, the data needs to be divided into training set and test set, here 80% of the data set is used as the training set of the network, and the remaining 20% of the data set is used as the test set.

[0015] The traffic mode recognition method of the elevator group control, the preprocessing of the picture data set is completed, and the image processing target detection model is built; the model architecture is divided into input, backbone network, neck network layer and head output four parts, the input end processes the input picture, uses Mosaic data enhancement, adaptive anchor frame calculation, adaptive picture scaling and other forms to improve the speed of model training and the accuracy of network, the backbone network uses Focus structure and CSP structure to build image feature neural network, the neck network uses FPN+PAN structure to strengthen the network feature fusion ability, the head uses CIOU as the loss function to predict the image feature, generates the boundary box and class confidence; the model loads the yolov5s pre-training weight, the initial learning rate is 0.01, the momentum is set to 0.957, the loss gain is set to 0.53, the loss function uses CIOU, and the rest of the parameters uses the default value.

[0016] The target detection model is built, in order to further improve the precision of the model, the CA attention mechanism is used to optimize the network model, the attention mechanism is added in the built yolov5 backbone network, the position information is embedded into the channel attention, the attention weight of the feature map in height and width is obtained, finally the attention weight is fused on the original feature map, and the network detection precision is strengthened; the CA attention mechanism formula is as follows:

[0017] (1)

[0018] Compared with the initial model, the map of the model added with the CA attention mechanism increases by about 0.5%, which is significant for the model to detect passenger flow.

[0019] The optimization of the model is completed, the number of data input into the network, the number of training rounds, the working thread and some parameters are set, and the model is trained; after the model training is completed, whether the performance index of the model is reasonable is checked, the camera is started to detect the input data, and the confidence of the prediction frame is checked.

[0020] The model training is completed and tested, the target detection network is used for passenger data extraction in a three-part ten-layer elevator group control system, the system records the number of waiting passengers on each floor and the number of passengers in the car, records the average boarding time and average waiting time of passengers and the target floor stay time and other data in a certain interval, and transmits the data to the information management module for optimizing elevator dispatching; the traffic mode recognition module obtains the target detection passenger flow data, and selects the traffic mode suitable for the current situation according to the real-time detection passenger flow data.

[0021] The advantages and effects of the present application are:

[0022] 1. The present application proposes an application of a target detection algorithm based on yolov5 in group control elevator traffic mode recognition. The yolov5 target detection algorithm is used to detect the passengers in the elevator car and the waiting passengers, the real-time passenger flow condition is obtained, the data is transmitted to the corresponding control module, and finally the system selects the traffic mode suitable for the current situation according to the passenger flow data. Compared with the traditional time sequence model, the target detection technology has high efficiency, accuracy and real-time performance, can obtain the current actual effective passenger flow data, and can improve the accuracy of traffic mode recognition, and finally realizes the improvement of the dispatching strategy.

[0023] 2. The application adopts a target detection-based method to add image recognition to the elevator group control system. By detecting passengers inside and outside the real car, the passenger flow condition can be obtained in real time. Compared with traditional time series prediction algorithms, it has strong authenticity and timeliness. By transmitting the actual passenger flow data to the traffic pattern recognition module, the traffic pattern that best matches the current passenger flow can be determined. In addition, data such as average boarding time and average waiting time of passengers can be detected to optimize the scheduling algorithm. In the yolov5 target detection algorithm, the CA attention mechanism is added to optimize the original algorithm and improve the accuracy of the network, which helps the system traffic pattern recognition. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is the function structure diagram of the three-part ten-layer elevator group control system of the application;

[0025] Figure 2 is the yolov5 model architecture diagram of the application;

[0026] Figure 3 is the CA attention mechanism network structure diagram of the application;

[0027] Figure 4 is the model training flowchart of the application;

[0028] Figure 5 is the map precision diagram of the application;

[0029] Figure 6 is the loss function diagram of the application;

[0030] Figure 7 is the flowchart of the application. DETAILED DESCRIPTION

[0031] The application of a yolov5-based target detection algorithm in the group control elevator traffic pattern recognition is as follows:

[0032] 1. Prepare passenger data set for elevator group control system target detection;

[0033] 2. Configure the network environment using the deep learning pytorch framework. The model construction of yolov5 is completed in this environment;

[0034] 3. Add CA attention mechanism to the original model framework of yolov5 to optimize the model;

[0035] 4. The preprocessed data set is used as the input of the network and trained, yolov5s pre-training weight is loaded, and CIOU is used as the loss function;

[0036] 5. Using the video of the elevator passengers to detect the network model;

[0037] 6. The traffic pattern recognition module sorts the passenger flow data detected by the system and decides the appropriate elevator traffic pattern according to the passenger flow data.

[0038] The specific implementation steps of step 1 are:

[0039] Collect pictures of elevator passengers and label the pictures using labelimg;

[0040] Convert the labeled data into yolo txt format;

[0041] Divide the data set into training set and test set.

[0042] The specific implementation steps of step 2 are:

[0043] Import the configuration environment required by the project;

[0044] Build the yolo model, the main body is Input input end, Backbone main network, Neck network layer, Head output end.

[0045] The specific implementation steps of step 3 are:

[0046] Introduce CA (Coordinate Attention) coordinate attention mechanism to strengthen the optimization of the network and improve the accuracy of the network; CA includes two parts, coordinate information embedding and coordinate attention generation, CA module will divide the input feature map into width and height two directions respectively to get two direction feature output, the output formula is as follows: (1)

[0047] (2)

[0048] Where, c is the channel, h is the height, w is the width, X is the input feature map; The two direction feature maps are spliced, and after completion, they are transformed into 1*1 shared convolution to get new feature output f, the formula of f is as follows:

[0049] (3)

[0050] Where, is the activation function, is the feature map after normalization processing; After obtaining f, it needs to be sliced into two separate tensors along the spatial dimension and , using two 1*1 convolutions and Two tensors are transformed to the same number of channels as the input feature map X, and finally pass through the sigmoid activation function to get the attention weight of the feature map in width and height respectively. The attention weight formula is as follows:

[0051] (4)

[0052] (5)

[0053] After getting the attention weight in width and height direction, weighted calculation is performed on the original feature map, and finally the feature map with attention weight in width and height direction is obtained, and the final output formula is as follows:

[0054] (6)

[0055] The feature map gets the attention weight in width and height direction, which improves the accuracy of the model.

[0056] The specific implementation steps of step 4 are:

[0057] Load the pre-trained weights of yolov5s to improve the speed and accuracy of network training;

[0058] The preprocessed data set is divided into training set and test set and sent into the network;

[0059] Set the bath size to 16, the epochs to 100, the input size to 640, and the hyperparameters to the default settings;

[0060] Select precision and recall as the indicators to measure the model, and the formulas are as follows:

[0061] (7)

[0062] (8)

[0063] Where TP is the correct identification of positive samples, FP is the identification of negative samples as positive samples, and FN is the correct identification of negative samples; In order to solve the divergence problem of IOU in the training process, CIOU is used as the loss function of Bounding box, which considers the distance between the target and the anchor, the overlap rate, the scale and the penalty factor, strengthens the stability of target frame regression, and speeds up the convergence speed of predicted frame and real frame regression, the formula is as follows:

[0064] (9)

[0065] (10)

[0066] (11)

[0067] wherein, is a weight function, is the width-height aspect ratio of the predicted box and the real box, is the Euclidean distance between the center points of the predicted box and the real box, C is the diagonal distance of the minimum closed region that can contain the predicted box and the real box; and finally substituted into to obtain the loss function, and the loss function formula is as follows:

[0068] (12)

[0069] The specific implementation steps of step 5 are:

[0070] The performance indicators of the detection training model, such as Precision, Recall, F1 score, map, loss function, etc. are detected; the passenger video in the real scene is input into the network, and the confidence of the network recognition is detected.

[0071] The specific implementation steps of step 6 are:

[0072] The system extracts the passenger data in front of each layer of the car door and the passenger data in the car, obtains the real-time specific passenger flow condition, calculates the waiting time of the passenger, the time of taking the elevator and the time of the elevator staying at a floor under different passenger flow; these data will be transmitted to the data management module, and the data management module will arrange the data and transmit them to the corresponding data module; the traffic mode recognition module will obtain the passenger flow condition of each floor, and select the appropriate traffic mode according to the current passenger proportion; when the low floor passenger is more and the uplink call signal is preferentially responded, the system will select the uplink peak traffic mode; when the high floor passenger is more and the downlink call signal is preferentially responded, the system will select the downlink peak traffic mode; when the passengers are scattered in each floor, the system will select the interlayer traffic mode; when the number of passengers is small, the system will select the idle traffic mode. Embodiment

[0073] The application is applied to the traffic mode recognition of the elevator group control system by the target detection method, taking a three-part ten-layer elevator group control system as an example, the system is divided into two parts, namely the target detection algorithm network and the traffic mode recognition, and the three-part ten-layer elevator group control system structure diagram is as shown in Figure 1 .

[0074] Before building the network, the input picture data set needs to be preprocessed. First, the photographed elevator passenger pictures are collected and sorted, and the pictures are converted to JPG format. Second, the picture data is labeled using the labelimg tool, and then the label format is output in xml format. Since the label format of yolo is in txt, the xml format label needs to be converted to txt format through code. When training the network, the data needs to be divided into training set and test set. Here, 80% of the data set is used as the training set of the network, and the remaining 20% of the data set is used as the test set.

[0075] After completing the preprocessing of the data set, the network model of yolov5 algorithm needs to be built and the running environment needs to be configured. Pycharm is used as the IDE of the project, and the downloaded pytorch environment is imported into the project using conda. After the environment configuration is completed, the parameters of yolov5 need to be set and the model needs to be built. The yolov5 model is shown in Figure 2 The input end, backbone network, neck network layer and head output end are mainly included. The input end processes the input picture, adopts Mosaic data enhancement, adaptive anchor frame calculation and adaptive picture scaling to improve the speed and accuracy of model training; the backbone network adopts Focus structure and CSP structure to build image feature neural network; the neck network adopts FPN+PAN structure to strengthen network feature fusion capability; the head adopts CIOU as loss function to predict image features, generate boundary box and class confidence.

[0076] After the initial model of yolov5 is built, CA attention mechanism is added to the backbone network to optimize the network model. CA is a novel mobile network attention mechanism that can improve the accuracy of the network. The CA attention mechanism module diagram is shown in Figure 3 The CA is added to the yolov5 model in the original model framework. The first step is to write the CA function in common to complete the model structure of CA; the second step is to modify the CA model in yolo; the third step is to rewrite the Backbone network of yolov5s model, and add CA to the backbone network model.

[0077] After the model is improved, the preprocessed picture dataset needs to be input into the network for training. Before training the model, the pre-trained weights of yolov5s are added to optimize the training of the model, shorten the network training time and improve the prediction accuracy of the model. When changing the parameters of the trained model, set the path, category and number of the training dataset; load the pre-trained weight type and path of the model; select the hyperparameters and activation functions that meet the training type of the dataset; set the number of working threads, the number of pictures input into the network and the number of training rounds. After the model parameters are set, the model is trained to obtain the model weight that meets the system. The training process of the model is as shown in Figure 4 .

[0078] After the model is trained, the performance indicators such as Precision, Recall, F1 score and map are used to check the quality of the model training. The map accuracy of the model is as shown in Figure 5 , and the loss function is as shown in Figure 6 . The video of the elevator passengers is used as the input of the model to detect the accuracy of the passenger prediction box identified by the model.

[0079] The three ten-story elevator group control system uses the target detection network to extract the passenger data. The system records the number of passengers waiting on each floor and the number of passengers in the car. In a certain interval, the average boarding time and the average waiting time of the passengers and the target floor stay time are recorded, and the data are transmitted to the information management module for optimizing the elevator dispatch. The traffic mode recognition module obtains the target detection passenger flow data, and selects the traffic mode suitable for the current situation according to the real-time detection passenger flow data. The traffic mode recognition of the elevator is realized by the target detection method, which can effectively detect the real passenger flow data in real time.

Claims

1. A method for traffic pattern recognition in a group-controlled elevator, characterized in that, The method includes constructing an image detection network model for traffic pattern recognition. The traffic pattern recognition module extracts data from the image model and assigns traffic patterns based on real-time conditions. Specifically, it includes the following steps: 1) Create a passenger dataset for target detection in elevator group control systems; 2) Configure the network environment using the deep learning PyTorch framework, and complete the YOLOv5 model construction within this environment; construct the YOLO model, which mainly consists of the Input layer, the Backbone network, the Neck network layer, and the Head layer. 3) The CA attention mechanism was added to the original YOLOv5 model framework to optimize the model. CA consists of two parts: coordinate information embedding and coordinate attention generation. The CA module divides the input feature map into two directions, width and height, and performs global average pooling on each direction to obtain feature outputs in two directions. The output formula is as follows: ； Where c is the channel, h is the height, w is the width, and X is the input feature map; the feature maps in the two directions are concatenated, and then transformed into a 1*1 shared convolution to obtain a new feature output f, the formula for f is as follows: ； in, For activation function, The feature map is normalized; after obtaining f, it needs to be sliced ​​into two separate tensors along the spatial dimension. and Using two 1x1 convolutions and The two tensors are transformed to have the same number of channels as the input feature map X, and finally, the attention weights for the width and height of the feature map are obtained by passing them through the sigmoid activation function. The attention weight formula is as follows: （4）； （5）； After obtaining the attention weights in the width and height directions, a weighted calculation is performed on the original feature map to finally obtain a feature map with attention weights in the width and height directions. The final output formula is as follows: （6）； The feature map provides attention weights in the width and height directions, which improves the model's accuracy. 4) Use the preprocessed dataset as input to the network for training, load the YOLOv5s pre-trained weights, and use CIOU as the loss function; 5) Use videos of elevator passengers to test the network model; test the performance metrics of the trained model, including Precision, Recall, F1 score, map, and loss function; input real-world passenger videos into the network and test the confidence of the network's recognition. 6) The system extracts passenger data in front of the elevator doors and passenger data inside the elevator car on each floor to obtain real-time passenger flow information, and calculates passenger waiting time, elevator riding time, and the time the elevator stays on a certain floor under different passenger flow levels. All this data is transmitted to the data management module, which organizes the data and transmits it to the corresponding data modules. The traffic pattern recognition module obtains the passenger flow information on each floor and selects an appropriate traffic mode based on the current passenger proportion. When there are more passengers on lower floors and the upward elevator call signal is responded to first, the system will select the upward peak traffic mode; when there are more passengers on higher floors and the downward elevator call signal is responded to first, the system will select the downward peak traffic mode; when passengers are relatively dispersed across different floors, the system will select the inter-floor traffic mode; when there are fewer passengers, the system will select the idle traffic mode.

Images