Canteen people flow intelligent scheduling and dish recommendation system

Abstract

The invention relates to the technical field of intelligent logistics, in particular to a canteen people flow intelligent scheduling and dish recommendation system, which comprises a people flow monitoring module used for collecting people flow data of each region of a canteen in real time; the dish database is used for storing dish components, tastes, nutrition data and historical sales volume; the central processing unit is used for dynamically analyzing the data and generating a scheduling strategy; the user terminal is used for receiving the recommendation information and displaying a navigation path; according to the invention, the multi-mode sensing network is used for monitoring the distribution of people flow in the canteen in real time, the dynamic queuing model and the collaborative filtering algorithm are combined, the optimal distribution of window resources and personalized dish recommendation are realized, the average waiting time is reduced, the dining order in the peak period is ensured through the digital twin technology and the emergency response mechanism, and the dining efficiency is improved. The operation efficiency and the service quality of the canteen are integrally improved, and meanwhile, the dining path selection of the user is optimized through AR navigation, and the problem of local congestion is relieved.

Description

Technical Field

[0001] The present invention relates to the field of smart logistics technology, and in particular to a canteen passenger flow intelligent scheduling and dish recommendation system. Background Art

[0002] As a centralized dining place, canteens usually adopt a fixed-window meal supply model. Diners need to choose a window to queue up. During peak hours, some windows are prone to congestion and long waiting times, while other windows are idle. In addition, the selection of dishes mostly relies on manual experience or random recommendations, and lacks personalized guidance. As a result, some dishes are left over and wasted due to insufficient exposure, while popular dishes are in short supply, affecting the dining experience.

[0003] Although some existing canteen management systems have adopted electronic menus or simple queuing reminders, crowd scheduling still relies on static rules and cannot be dynamically adjusted to adapt to real-time changes in crowd flow, resulting in uncontrollable queuing time during peak hours. In addition, the match between dish recommendations and actual user needs is poor and cannot be combined with current inventory, dining time and personal health data. At the same time, there is a lack of emergency relief mechanisms, which makes it easy to cause chaos when sudden crowds gather.

[0004] Therefore, in order to solve the above problems, the present invention proposes an intelligent canteen traffic scheduling and dish recommendation system. Summary of the Invention

[0005] In order to overcome the problem that the existing cafeteria management system cannot be dynamically adjusted to adapt to real-time changes in passenger flow, the present invention proposes an intelligent cafeteria passenger flow scheduling and dish recommendation system.

[0006] The technical solution of the present invention is: a canteen traffic intelligent scheduling and dish recommendation system, including:

[0007] The crowd monitoring module collects real-time crowd flow data in various areas of the cafeteria through an infrared sensor array, a video analysis unit, and a mobile device signal capture device. The infrared sensor array is deployed at the entrance and exit, the video analysis unit uses a CNN algorithm to identify crowd density, and the mobile device signal capture device uses WiFi probes to count the number of people in real time. The output data includes the number of people in each area and the concentration information, providing dynamic crowd distribution input for the central processing unit.

[0008] The dish database stores dish ingredients, flavors, nutritional data, and historical sales information. The dish ingredient and nutritional data are synchronized and updated with the supply chain system. Flavor parameters are labeled according to grading standards, and historical sales records are associated with dates, time periods, and weather factors. It supports multi-conditional searches based on dish IDs, provides data support for the dietary profile generation module, and participates in the construction of the user-dish rating matrix for the collaborative filtering algorithm.

[0009] The central processing unit dynamically analyzes input data from the crowd monitoring module and the dish database, calculates the expected waiting time for each window based on a queuing theory model, generates diversion paths based on real-time crowd heat maps, updates recommendation weights based on remaining dish predictions, invokes user preference data from the meal profile generation module to execute the dish recommendation algorithm, and outputs the scheduling strategy to user terminals and on-site guidance equipment.

[0010] The user terminal receives the recommendation information generated by the central processing unit and displays the navigation path. The integrated positioning function enables the AR navigation interface to display the best meal collection path, provides nutritional balance index visualization and allergen warning functions, triggers a pop-up warning when it detects user-registered taboo ingredients, and supports user feedback on recommendation results to optimize the preference model of the dietary portrait generation module.

[0011] Preferably, the hardware composition of the crowd monitoring module includes: a 16-line infrared laser radar deployed at the entrance and exit of the cafeteria, which uses a point cloud clustering algorithm to count the number of people entering and leaving and distinguish the direction; a 4K resolution wide-angle camera group installed above the meal pick-up line, which uses an improved YOLOv7-tiny model to detect the status of plate holding in real time, combined with the DeepSORT algorithm to track the movement trajectory of people; distributed WiFi6 probe devices, which calculate the regional density by capturing the MAC address hash value of the smart terminal. All sensor data are time-synchronized and input into the Kalman filter for multi-source data fusion, and output a crowd distribution matrix with confidence score. When abnormal aggregation is detected, a graded warning signal is triggered.

[0012] Preferably, the core algorithm of the central processing unit includes three-level processing logic: the first layer adopts the M / M / c queuing model, where c is the number of service windows, calculates the expected waiting time of each window based on the real-time arrival rate and service rate, and dynamically displays the countdown prediction on the electronic menu screen; the second layer generates dynamic path planning based on the Dijkstra algorithm, converts the crowd heat map into a weighted directed graph, and outputs the optimal diversion route with the goal of minimizing the average waiting time; the third layer integrates the XGBoost regressor to analyze the correlation between the historical sales of dishes and weather / time period / price, and automatically reduces the weight of the dish in the recommendation list when the inventory is lower than the safety threshold.

[0013] Preferably, the dietary portrait generation module constructs a user's dietary preference model through multi-dimensional data fusion: importing 90 consecutive days of historical consumption records from the campus card system, extracting the frequency and combination pattern of dish selection, connecting to the electronic health record API to obtain the user's registered allergen list and BMI index, receiving the dietary goals actively submitted by the mobile APP, and using the hierarchical analysis method to quantify the above data into taste preference vectors, nutritional requirement matrices and taboo lists. User clustering groups are updated through federated learning every 24 hours, and cross-user recommendations are triggered when new popular dishes are detected in similar groups.

[0014] Preferably, the implementation method of the collaborative filtering recommendation algorithm includes: establishing a user-dish rating matrix based on a time decay factor, calculating the k-nearest neighbor user group by cosine similarity, where k=15; introducing a context-aware mechanism, strengthening the feature weight of porridge / noodles during the breakfast period, increasing the recommendation intensity of high-protein dishes during the lunch period, and giving priority to low-calorie combinations during the dinner period; setting up an emergency response module, when a window causes a surge in queues due to equipment failure, immediately push the navigation path of the alternative window and the AR real-scene preview of the window's special dishes to the affected user's terminal to guide the rapid evacuation of people.

[0015] Preferably, the user terminal implements centimeter-level positioning AR navigation through the ARKit / ARCore engine, superimposes a colored light band with distance indication on the camera image, and the width of the light band is dynamically adjusted according to the congestion of the path.

[0016] As a preference, nutritional data visualization uses the D3.js engine to render a five-dimensional radar chart, with the coordinate axes representing calorie / protein / fat / carbohydrate / sodium content, and coloring them in comparison with the recommended values ​​in the Chinese Dietary Guidelines; the allergen detection process is initiated when the user scans the plate chip, and the dish ingredient list is matched with the user's pre-stored taboo list in AC automatic mode. When high-risk ingredients in taboo foods are detected, a full-screen flashing warning is triggered and the risk details are announced by voice.

[0017] As a preferred approach, a BIM modeling tool was used to construct a high-precision three-dimensional model of the cafeteria. Real-time IoT data streams were imported through the Unity3D engine, and the pedestrian flow simulation in each area was updated every second. A sandbox test mode was set up to preview the evacuation efficiency of three diversion schemes, and the NSGA-II multi-objective optimization algorithm was used to select the Pareto optimal solution.

[0018] Preferably, when the infrared sensor detects that the population density in a certain area exceeds 3 people / ㎡ and lasts for 120 seconds, the three-level response plan is activated. The first-level response controls the electronic menu screen of the area to switch to the "peak shifting recommended" prompt. The second-level response activates the ceiling directional audio to play guidance voice. The third-level response links the access control system to temporarily close the entrance and trigger the release of the electromagnetic lock of the fire passage; at the same time, the optimal detour route is pushed to the user terminals within the affected radius of 10 meters.

[0019] As a preferred method, a comparison matrix of predicted and actual crowd flow distribution is established, and the difference in error distribution is calculated through KL divergence. At the same time, the actual adoption behavior of users on recommended dishes is recorded, and the confusion matrix is ​​used to calculate the precision and recall rate.

[0020] Beneficial effects of the present invention:

[0021] 1. This invention uses a multimodal sensor network to monitor the distribution of cafeteria traffic in real time, combines a dynamic queuing model with a collaborative filtering algorithm, achieves optimal allocation of window resources and personalized dish recommendations, reduces average waiting time, and ensures dining order during peak hours through digital twin technology and emergency response mechanisms, thereby improving the overall cafeteria's operational efficiency and service quality.

[0022] 2. AR navigation optimizes users’ dining route choices and alleviates local congestion. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 Shown is a schematic diagram of the system framework of the present invention;

[0024] Figure 2 What is shown is a schematic diagram of the workflow of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0026] See also Figure 1 The present invention provides an embodiment: a cafeteria traffic intelligent scheduling and dish recommendation system, including:

[0027] The crowd monitoring module collects real-time crowd flow data in various areas of the cafeteria through an infrared sensor array, a video analysis unit, and a mobile device signal capture device. The infrared sensor array is deployed at the entrance and exit, the video analysis unit uses a CNN algorithm to identify crowd density, and the mobile device signal capture device uses WiFi probes to count the number of people in real time. The output data includes the number of people in each area and the concentration information, providing dynamic crowd distribution input for the central processing unit.

[0028] The dish database stores dish ingredients, flavors, nutritional data, and historical sales information. The dish ingredient and nutritional data are synchronized and updated with the supply chain system. Flavor parameters are labeled according to grading standards, and historical sales records are associated with dates, time periods, and weather factors. It supports multi-conditional searches based on dish IDs, provides data support for the dietary profile generation module, and participates in the construction of the user-dish rating matrix for the collaborative filtering algorithm.

[0029] The central processing unit dynamically analyzes input data from the crowd monitoring module and the dish database, calculates the expected waiting time for each window based on a queuing theory model, generates diversion paths based on real-time crowd heat maps, updates recommendation weights based on remaining dish predictions, invokes user preference data from the meal profile generation module to execute the dish recommendation algorithm, and outputs the scheduling strategy to user terminals and on-site guidance equipment.

[0030] The user terminal receives the recommendation information generated by the central processing unit and displays the navigation path. The integrated positioning function enables the AR navigation interface to display the best meal collection path, provides nutritional balance index visualization and allergen warning functions, triggers a pop-up warning when it detects user-registered taboo ingredients, and supports user feedback on recommendation results to optimize the preference model of the dietary portrait generation module.

[0031] Furthermore, the hardware components of the crowd monitoring module include: a 16-line infrared laser radar (scanning frequency 10Hz, detection accuracy ±3cm) deployed at the entrance and exit of the cafeteria, which uses a point cloud clustering algorithm to count the number of people entering and exiting and distinguish the direction; a 4K resolution wide-angle camera group installed above the food pickup line, which uses an improved YOLOv7-tiny model to detect the status of plate holding in real time, combined with the DeepSORT algorithm to track the movement trajectory of people; distributed deployment of WiFi6 probe devices hidden in the ceiling, which calculates the regional density by capturing the MAC address hash value of the smart terminal. All sensor data are time-synchronized and input into the Kalman filter for multi-source data fusion, and output a crowd distribution matrix with confidence score. When abnormal gathering is detected (density >2 people / ㎡ for 30 seconds), a graded warning signal is triggered.

[0032] The core algorithm of the central processing unit includes three levels of processing logic: the first level adopts the M / M / c queuing model, where c is the number of service windows, calculates the expected waiting time of each window based on the real-time arrival rate and service rate, and dynamically displays the countdown prediction on the electronic menu screen; the second level generates dynamic path planning based on the Dijkstra algorithm, converts the crowd heat map into a weighted directed graph, and outputs the optimal diversion route with the goal of minimizing the average waiting time; the third level integrates the XGBoost regressor to analyze the correlation between the historical sales volume of dishes and weather / time period / price, and automatically reduces the weight of the dish in the recommendation list when the inventory level is lower than 20% of the set daily average sales volume.

[0033] The dietary profile generation module constructs a user dietary preference model through multi-dimensional data fusion: it imports 90 consecutive days of historical consumption records from the campus card system to extract the frequency and combination patterns of dish selection; it connects to the electronic health record API to obtain the user's registered allergen list and BMI index; it receives the dietary goals actively submitted by the mobile app (such as fat loss / muscle gain / sugar control), and uses the hierarchical analysis method (AHP) to quantify the above data into a taste preference vector (sweet / salty / spicy weights), a nutritional requirement matrix (daily protein / dietary fiber intake gap) and a taboo list (prohibited ingredient coding table). The user clustering grouping is updated every 24 hours through federated learning, and cross-user recommendations are triggered when new popular dishes are detected in similar groups.

[0034] The system adopts a collaborative filtering recommendation algorithm, and the implementation of the collaborative filtering recommendation algorithm includes: establishing a user-dish rating matrix based on a time decay factor (the weight of recent consumption records decays exponentially according to e^(-0.1t)), and calculating the k-nearest neighbor user group (k=15) through cosine similarity; introducing a context-aware mechanism to strengthen the feature weight of porridge / noodles during breakfast time, increase the recommendation intensity of high-protein dishes during lunch time, and give priority to low-calorie combinations during dinner time; setting up an emergency response module. When a window causes a surge in queues due to equipment failure, the navigation path of an alternative window and an AR real-life preview of the window's featured dishes are immediately pushed to the affected user's terminal to guide the rapid evacuation of the flow of people.

[0035] The details of the interactive function implementation of the user terminal are as follows: AR navigation with centimeter-level positioning is realized through the ARKit / ARCore engine, and a colored light band with distance indication is superimposed on the camera image. The width of the light band is dynamically adjusted according to the congestion of the path, where the green colored light band indicates unobstructed traffic and the red colored light band indicates congestion.

[0036] Nutritional data visualization uses the D3.js engine to render a five-dimensional radar chart, with the coordinate axes representing calorie / protein / fat / carbohydrate / sodium content, and coloring them in comparison with the recommended values ​​in the Chinese Dietary Guidelines. The allergen detection process is initiated when the user scans the plate chip, matching the dish ingredient list with the user's pre-stored taboo list in AC automatic mode. When high-risk ingredients in taboo foods are detected, a full-screen flashing warning is triggered and the risk details are announced by voice.

[0037] The implementation process of the digital twin technology is as follows: use BIM modeling tools to build a three-dimensional high-precision model of the cafeteria (including CAD data of tables, chairs, windows, and passages), import real-time IoT data streams through the Unity3D engine, and update the pedestrian flow simulation in each area every second; set up a sandbox test mode to rehearse the evacuation efficiency of three diversion schemes: opening a spare window / adjusting the one-way flow of the meal pickup line / activating the emergency exit.

[0038] The triggering logic of the emergency response mechanism is as follows: when the infrared sensor detects that the population density in a certain area exceeds 3 people / m2 and lasts for 120 seconds, a three-level response plan is activated. The first-level response (density 3-4 people / m2) controls the electronic menu screen in the area to switch to the "peak shifting" prompt; the second-level response (4-5 people / m2) activates the ceiling directional audio to play guidance voice; the third-level response (>5 people / m2) links the access control system to temporarily close the entrance and trigger the release of the electromagnetic lock of the fire escape; at the same time, the optimal detour route (avoiding areas with a density >2 people / m2) is pushed to user terminals within a radius of 10 meters affected.

[0039] The data processing method of the post-effect evaluation module includes: establishing a predicted-actual crowd flow distribution comparison matrix, calculating the error distribution difference through KL divergence, recording the user's actual adoption behavior of recommended dishes (click / ignore / bad review), and using the confusion matrix to calculate the precision and recall rate.

[0040] See also Figure 2 , further, the workflow of the present invention is described in detail:

[0041] After the system is started, the infrared sensor arrays deployed in various areas of the cafeteria scan the movement trajectories of people in real time. The video analysis unit uses the CNN algorithm to calculate the number of people queuing at each window and their movement speed. At the same time, the WiFi probe device starts capturing smart terminal signals to estimate the regional density. At the same time, the food database synchronously updates the daily menu, inventory and nutritional data. The user terminal receives personal health records and dietary preferences. All data is transmitted to the central processing unit after time stamp alignment, forming a complete data set including crowd flow heat map, food status and user portraits.

[0042] The central processing unit predicts the changes in pedestrian flow in each area in the next 15 minutes based on the LSTM model, and uses operations research algorithms to calculate the optimal window resource allocation plan. At the same time, the collaborative filtering engine analyzes the user's historical data and current inventory to generate a personalized recommendation list. After using digital twin technology to preview the effects of different diversion plans, it outputs a comprehensive strategy that includes recommended dishes, navigation paths and dynamic pricing. All decision data is transmitted to the terminal device and on-site display screen through an encrypted channel.

[0043] After the user's mobile APP receives the recommendation information, the optimal meal pickup route navigation arrow is superimposed through the AR interface, and the window waiting time and discounted dishes are displayed in real time. When the user approaches the recommendation window, the NFC automatic ordering function is triggered. If an allergy is detected, a full-screen warning will pop up, and nutritional data will be displayed in the form of a three-dimensional chart to display the composition of the dish. After the user confirms the selection, the system automatically updates the inventory and adjusts the subsequent recommendation strategy.

[0044] The system reassesses the flow of people every two minutes. When the queue at a window exceeds the threshold, the emergency mechanism is automatically triggered, thereby guiding the diversion through the electronic screen and adjusting the food supply of adjacent windows. All adjustment data is synchronized to the digital twin model for effect verification, forming a closed-loop optimization process of "monitoring-decision-making-execution-feedback".

[0045] After the dining peak ends, the system automatically generates an operation report, compares the error rate between the predicted and actual flow of people, analyzes the adoption rate of recommended dishes and user ratings, optimizes model parameters through reinforcement learning algorithms, updates the dietary portrait database, and accumulates experience data for the next day's operations, thus completing a complete work cycle.

[0046] Through the above steps, the cafeteria traffic distribution is monitored in real time through a multimodal sensor network. Combined with a dynamic queuing model and collaborative filtering algorithm, the optimal allocation of window resources and personalized dish recommendations are achieved, reducing the average waiting time. At the same time, AR navigation is used to optimize the user's dining path selection and alleviate local congestion problems. This solves the problem that the existing cafeteria management system cannot dynamically adjust to adapt to real-time changes in traffic flow, resulting in uncontrollable queuing time during peak hours, and the dish recommendations have a low match with users' actual needs and cannot be combined with current inventory, dining time and personal health data.

Claims

1. The canteen traffic intelligent scheduling and dish recommendation system is characterized by: Includes: The crowd monitoring module collects real-time crowd flow data in various areas of the cafeteria through an infrared sensor array, a video analysis unit, and a mobile device signal capture device. The infrared sensor array is deployed at the entrance and exit, the video analysis unit uses a CNN algorithm to identify crowd density, and the mobile device signal capture device uses WiFi probes to count the number of people in real time. The output data includes the number of people in each area and the concentration information, providing dynamic crowd distribution input for the central processing unit. The dish database stores dish ingredients, flavors, nutritional data, and historical sales information. The dish ingredient and nutritional data are synchronized and updated with the supply chain system. Flavor parameters are labeled according to grading standards, and historical sales records are associated with dates, time periods, and weather factors. It supports multi-conditional searches based on dish IDs, provides data support for the dietary profile generation module, and participates in the construction of the user-dish rating matrix for the collaborative filtering algorithm. The central processing unit dynamically analyzes input data from the crowd monitoring module and the dish database, calculates the expected waiting time for each window based on a queuing theory model, generates diversion paths based on real-time crowd heat maps, updates recommendation weights based on remaining dish predictions, invokes user preference data from the meal profile generation module to execute the dish recommendation algorithm, and outputs the scheduling strategy to user terminals and on-site guidance equipment. The user terminal receives the recommendation information generated by the central processing unit and displays the navigation path. The integrated positioning function enables the AR navigation interface to display the best meal collection path, provides nutritional balance index visualization and allergen warning functions, triggers a pop-up warning when it detects user-registered taboo ingredients, and supports user feedback on recommendation results to optimize the preference model of the dietary portrait generation module.

2. The canteen flow intelligent scheduling and dish recommendation system according to claim 1 is characterized in that: The hardware components of the crowd monitoring module include: a 16-line infrared laser radar deployed at the entrance and exit of the cafeteria, which uses a point cloud clustering algorithm to count the number of people entering and exiting and distinguish the direction; a 4K resolution wide-angle camera group installed above the food pickup line, which uses an improved YOLOv7-tiny model to detect the status of plate holding in real time and combines the DeepSORT algorithm to track the movement trajectory of people; distributed WiFi6 probe devices calculate the regional density by capturing the MAC address hash value of the smart terminal. All sensor data are time-synchronized and input into the Kalman filter for multi-source data fusion, and output a crowd distribution matrix with confidence score. When abnormal aggregation is detected, a graded warning signal is triggered.

3. The cafeteria crowd intelligent scheduling and dish recommendation system according to claim 1 is characterized by: The core algorithm of the central processing unit includes three levels of processing logic: the first level uses the M / M / c queuing model, where c is the number of service windows. The expected waiting time for each window is calculated based on the real-time arrival rate and service rate, and the countdown forecast is dynamically displayed on the electronic menu screen. The second level generates dynamic path planning based on the Dijkstra algorithm, converting the passenger flow heat map into a weighted directed graph to output the optimal diversion route with the goal of minimizing the average waiting time. The third layer integrates the XGBoost regressor to analyze the correlation between the historical sales volume of dishes and weather / time period / price. When the inventory level falls below the safety threshold, the weight of the dish in the recommendation list is automatically reduced.

4. The cafeteria crowd intelligent scheduling and dish recommendation system according to claim 1 is characterized by: The dietary profile generation module constructs a user's dietary preference model through multi-dimensional data fusion: it imports 90 consecutive days of historical consumption records from the campus card system, extracts the frequency and combination patterns of dish selection, connects to the electronic health record API to obtain the user's registered allergen list and BMI index, receives the dietary goals actively submitted by the mobile app, and uses the hierarchical analysis method to quantify the above data into taste preference vectors, nutritional requirements matrices and taboo lists. User clustering groups are updated every 24 hours through federated learning, and cross-user recommendations are triggered when new popular dishes are detected in similar groups.

5. The cafeteria crowd intelligent scheduling and dish recommendation system according to claim 1 is characterized by: The system adopts a collaborative filtering recommendation algorithm, and the implementation of the collaborative filtering recommendation algorithm includes: establishing a user-dish rating matrix based on a time decay factor, calculating the k-nearest neighbor user group through cosine similarity, where k=15; introducing a context-aware mechanism to strengthen the feature weight of porridge / noodles during breakfast time, increase the recommendation intensity of high-protein dishes during lunch time, and prioritize low-calorie combinations during dinner time; and setting up an emergency response module. When a window causes a surge in queues due to equipment failure, the navigation path of an alternative window and an AR real-life preview of the window's featured dishes are immediately pushed to the affected user's terminal to guide the rapid evacuation of the flow of people.

6. The cafeteria crowd intelligent scheduling and dish recommendation system according to claim 1 is characterized by: The user terminal implements centimeter-level AR navigation through the ARKit / ARCore engine, superimposing a colored light band with distance indication on the camera image, and the width of the light band dynamically adjusts according to the congestion of the path.

7. The cafeteria crowd intelligent scheduling and dish recommendation system according to claim 1 is characterized by: Nutrition data visualization uses the D3.js engine to render a five-dimensional radar chart, with the coordinate axes representing calorie / protein / fat / carbohydrate / sodium content, and coloring them in comparison with the recommended values ​​in the Chinese Dietary Guidelines. The allergen detection process starts when the user scans the chip on the plate. The AC automatic machine mode is used to match the ingredient list of the dish with the user's pre-stored taboo list. When high-risk ingredients of taboo foods are detected, a full-screen flashing warning is triggered and the risk details are announced by voice.

8. The cafeteria crowd intelligent scheduling and dish recommendation system according to claim 1 is characterized by: The system adopts digital twin technology, which uses BIM modeling tools to build a three-dimensional high-precision model of the cafeteria, imports real-time Internet of Things data streams through the Unity3D engine, and updates the pedestrian flow simulation in each area every second. A sandbox test mode is set up to preview the evacuation efficiency of three diversion schemes, and the NSGA-II multi-objective optimization algorithm is used to select the Pareto optimal solution.

9. The cafeteria crowd intelligent scheduling and dish recommendation system according to claim 1 is characterized by: The system has an emergency response mechanism. When the infrared sensor detects that the population density in a certain area exceeds 3 people / m2 and lasts for 120 seconds, the three-level response plan is activated. The first-level response controls the electronic menu screen in the area to switch to the "peak shifting" prompt. The second-level response activates the ceiling directional speakers to play guidance voice. The third-level response links the access control system to temporarily close the entrance and trigger the release of the electromagnetic lock of the fire escape. At the same time, the optimal detour route is pushed to user terminals within a radius of 10 meters affected.

10. The cafeteria crowd intelligent scheduling and dish recommendation system according to claim 1 is characterized by: The data processing method of the post-effect evaluation module includes: establishing a predicted-actual crowd flow distribution comparison matrix, calculating the error distribution difference through KL divergence, recording the user's actual adoption behavior of recommended dishes, and using a confusion matrix to calculate the precision and recall rate.

Images