Multi-source data driven cooking machine cluster intelligent scheduling system and method

Abstract

The invention provides a multi-source data-driven intelligent scheduling system and method for a cooker cluster, and the system comprises the steps: receiving take-out / ordering orders and historical data through a cloud end, carrying out the real-time fusion of the orders, inventory, equipment states and other multi-source information through an edge end, carrying out the dynamic combination of the orders through a multi-factor weight algorithm, calculating the order insertion priority, and carrying out the balanced distribution of the order insertion priority to the cooker cluster, thereby achieving the intelligent scheduling of the cooker cluster. Meal delivery time estimation and closed-loop fault-tolerant scheduling with the precision of + / -1 minute are realized, so that the catering efficiency and the user experience are improved. According to the method, intelligent dynamic scheduling of the cooking machine cluster is realized through side cloud cooperation and multi-source data fusion, the meal delivery time error can be controlled within + / -1 minute, the equipment utilization rate and the order fulfillment rate are improved, the manual intervention and the order returning risk are reduced, and finally the catering operation efficiency and the customer satisfaction are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of catering order scheduling, in particular to a multi-source data driven intelligent scheduling system and method for a cluster of cooking machines. BACKGROUND

[0002] With the popularization of intelligent cooking equipment, it has functions such as automatic feeding, stirring, and washing pots. With the development of seasoning and food supply chain, the stability of the taste is ensured, but the intelligent level of task scheduling (order scheduling) is still a bottleneck. Although the MES (Manufacturing Execution System) of the manufacturing industry provides a production scheduling model, there are significant differences when directly transplanted to the catering industry (see Table 1).

[0003] Table 1 Comparison of differences between MES production scheduling and catering cooking machine order scheduling

[0004] Compared with the prior art, such as patents “CN119850299A Intelligent matching cooking method based on big data”, “CN114115026A Method for controlling unmanned cooking machine based on order payment system”, “CN113971619A Intelligent restaurant management method, system, restaurant management device, and readable storage medium”, only the received orders are arranged in order of ordering time. Lack of multi-device collaborative scheduling capability, order scheduling rigidity, lack of emergency demand response mechanism, and weak risk control capability.

[0005] Disadvantages of the prior art: 1. Order scheduling rigidity: cannot combine orders for similar dishes, etc., resulting in low delivery efficiency. When there are multiple orders at the same time, it is not possible to have each order served every certain period of time, resulting in long customer waiting times. Stir-frying spicy and non-spicy dishes together results in frequent pot washing.

[0006] 2. Lack of emergency demand response mechanism: for urgent orders, customer order acceleration, and unprepared food materials, it is not possible to insert or adjust orders, and simply cooking in order may ignore the time pressure of high-priority orders, leading to customer complaints during peak periods.

[0007] 3. Insufficient multi-device collaborative scheduling capability: unable to respond to device failures, material waiting, etc.; unable to dynamically allocate tasks and balance loads based on real-time device status (such as cleaning, material shortage downtime, and faults), even if there are multiple cooking machines, resulting in overloading of some devices and idling of others.

[0008] 4. Weak risk control capability: not effectively linked with inventory systems and order platforms, material shortages cannot be fed back to the order platform, resulting in customer order cancellation after ordering due to material shortages. Unable to dynamically calculate the preparation time of different food materials, lack of risk prediction ability for “material waiting”, resulting in order scheduling interruption and device idling.

[0009] Therefore, the prior art has defects, and further improvement is needed. SUMMARY

[0010] In view of the problems existing in the prior art, the application provides a multi-source data driven cluster intelligent scheduling system and method for a stir-fry machine.

[0011] To achieve the above-mentioned purpose, the specific scheme of the application is as follows: The application provides a multi-source data driven cluster intelligent scheduling system for a stir-fry machine, which comprises: A cloud computing layer module is deployed on a remote server and comprises: An API interface docking unit is used to access the order data interface of a takeout platform and an ordering software, receive and store order information; A user interaction unit provides a customer end and a store manager end operation interface, is used to push an estimated delivery time and receive a scheduling parameter configuration instruction, and the configuration instruction has a higher priority than an automatic scheduling rule; A historical data analysis unit is based on an AI model to analyze order history records, construct a user portrait library and update dish process parameters; A system configuration management unit provides a set of configurable scheduling parameters, including a single combination time window, a protection window number, a maximum single combination number and a device load threshold; An edge computing layer module is deployed on a local node of a store and comprises: A core scheduling engine executes a multi-factor weight algorithm, dynamically combines orders, calculates order insertion priority and allocates device load balancing; A device state monitoring unit collects real-time stir-fry machine device state data, constructs an available device pool and triggers a fault tolerance mechanism; A real-time task allocation unit allocates tasks to a stir-fry machine cluster based on a device load rate, and the load rate = total remaining task frying time / unit time; A multi-source data fusion module comprises: An order processing unit filters invalid orders, extracts dish type, quantity, estimated delivery time attributes, and generates taste preference markers through AI analysis of user notes; An inventory linkage unit communicates with a food material inventory system, checks food material readiness status and feeds back material shortage warning; A recipe process database stores dish frying parameters for task decomposition and time estimation; A closed-loop control module responds to order state changes, device failures or order cancellation events, triggers queue recalculation and real-time issuance of updated task sequences to the stir-fry machine cluster, and simultaneously pushes high-precision estimated delivery time with an error of ±1 minute to the user interaction unit.

[0012] Further, the historical data analysis unit is specifically used for: Analyzing user notes in historical orders through a natural language processing model to generate structured taste preference labels, including quantitative ratings of spiciness, saltiness, and sweetness; Based on user order cancellation records and order frequency, dynamically updating the customer order cancellation rate portrait library; According to the cooking process parameter adjustment record, the frying temperature and time parameters in the menu database are optimized.

[0013] Further, when the core scheduling engine performs dynamic order merging: The time window is set to be configurable 5-10 minutes; The number of protection windows is fixed to the first 8 tasks in the queue; The maximum number of merged order thresholds is ≤4 portions per single order.

[0014] Further, in the real-time task allocation unit: The device load rate preset threshold is 80%; When the load rate exceeds 80%, the task is temporarily stored in the order queue pool; The order queue pool backlog threshold is set to 10 tasks, and after triggering an alarm, the store manager end is notified for manual intervention.

[0015] Further, the fault tolerance mechanism triggered by the device state monitoring unit includes: When a device failure is detected, all uncompleted tasks in the device queue are automatically returned to the order queue pool; Based on the urgency of the remaining tasks, they are redistributed to high-availability devices, and the expected maintenance duration of the faulty device is marked; Push an alarm notification to the store manager end, including the fault code and the number of affected orders or the expected delay duration.

[0016] The present application also provides a multi-source data driven intelligent scheduling method for a stir-fry machine cluster, applied to the field of food ordering, based on the above system, the method comprises the following steps: S1, order receiving: through the API interface of the cloud computing layer, the order data interface of the take-out platform and the ordering software is connected, and multi-source order data is received; S2, order processing: Filter invalid orders and extract order key attributes, including dish type, portion, order time, and expected delivery time; Associate the user portrait library through the user unique identifier, and generate structured taste preference labels based on AI model analysis of user notes and portrait data; Synchronize the food material inventory status and feedback material shortage warning; S3, task decomposition and resource evaluation: Decompose orders into independent cooking task units; Check the ability of the cooking machine to support the recipe process and the readiness of the ingredients; Estimate the cooking time of each task, and set a delay scheduling time according to the risk probability of order cancellation; S4, dynamically merge orders: Apply merging rules: only merge the same type of dishes within a certain time window, set a protection window for the first N tasks to prevent merging, limit the maximum number of merged orders threshold, and ensure that merging does not cause the original order to exceed the time limit; Optimize the task queue to achieve balanced serving of multi-table orders and overall efficient cooking; S5, priority calculation of inserted orders: Use a multi-factor weighted scoring model, including operator intervention forced top task and urgent order weight promotion coefficient; Evaluate the global impact of inserted orders: calculate the time saved by inserted orders and the total overtime increment caused by other orders, to maximize the benefit-cost ratio to determine the optimal insertion position; S6, device scheduling strategy: Build a pool of available devices, filter out idle, sufficient ingredients, and normal operation of the cooking machine; Calculate the device load rate, where load rate = total cooking time of remaining tasks / unit time; Assign new tasks to the device with the lowest load rate, or temporarily store tasks in the order pool when the load rate exceeds the preset threshold, and trigger an alarm when the order pool backlog reaches the threshold; In response to device failure or ingredient waiting events, automatically return unfinished tasks to the order pool and reschedule; S7, real-time feedback and adjustment: Respond to order status changes, device failures, or order cancellation events to trigger task queue recalculation; Issue updated task sequences to the cooking machine cluster and push the estimated delivery time to the user end, with an accuracy of ±1 minute.

[0017] Further, in the S2 order processing step, generating structured taste preference labels includes: Use the BERT model to analyze keywords in user comments, including "less spicy" and "no onions"; Convert the analysis results into device executable seasoning instruction codes; Correlate historical preference data in the user profile library to generate a final taste score.

[0018] Further, the multi-factor weighted scoring model in the S5 inserted order priority calculation step is: Inserted order priority score = (order base weight x urgent coefficient) + operator intervention weight; Wherein: The order basis weight is the inherent priority score of the order itself; The urgent coefficient value range is 1.0-2.0, and the default value is 1.5; The operator intervention weight is a Boolean value of 0 or 1.

[0019] Further, in the S6 device scheduling strategy step: The unit time is set to 60 minutes; The load rate = the total remaining task frying time / 60; When the lowest load rate in the available device pool is > 80%, the task enters the order pool for waiting.

[0020] Further, in response to device failure in the S6 device scheduling strategy step: Immediately interrupt the current frying task of the device; Sort the unfinished tasks in ascending order of remaining frying time and return to the order pool; When reassigning, preferentially schedule tasks with a remaining time of less than 3 minutes.

[0021] The technical scheme of the present application has the following beneficial effects: 1. Efficient and stable automatic scheduling: By integrating various scheduling rules (dynamic order combining, intelligent order inserting, load balancing) through the core order scheduling engine, and supporting rich configurable parameters, the system can efficiently and stably handle various complex order scenarios and constraint conditions, with processing efficiency and consistency far exceeding manual scheduling (tested, device switching frequency reduced by about 30%, customer average waiting time reduced by about 20%).

[0022] 2. Standardization and adaptability: By flexibly configuring scheduling parameters (such as order combining time window, protection window number, maximum order combining number, load threshold, etc.), the system can accurately adapt to different store operation styles and meal requirements (such as fast food pursuing speed, and fine dining emphasizing serving rhythm), helping to establish standardized operation procedures (SOP), and facilitating subsequent continuous optimization.

[0023] 3. Improve service transparency and customer satisfaction: Algorithm-based automatic scheduling ensures the stability and predictability of the serving process. Combined with real-time order status feedback (such as high-precision estimated serving time), it significantly improves service transparency and enhances customers' sense of control over waiting time, thereby effectively improving customer satisfaction.

[0024] 4. Reduce operational dependence and cost: Significantly reduces the high dependence on specific personnel (such as experienced store managers), reduces the communication and coordination costs between the front and back kitchens, and provides strong and low-cost operational support for the expansion of chain restaurant stores.

[0025] 5. Improve equipment resource utilization: through real-time monitoring of equipment status (idle / busy / failure / waiting for material) and dynamic task allocation based on load rate, efficient cooperation of multiple frying machines is realized, equipment utilization is maximized, and idle is reduced.

[0026] 6. Enhance system robustness: built-in fault-tolerant mechanism (such as automatic task transfer of device failure, order pool backlog alarm) effectively deals with abnormal situations such as device failure and resource shortage, and guarantees the continuous and stable operation of the system. BRIEF DESCRIPTION OF DRAWINGS

[0027] Fig. 1 is the system architecture diagram of the present application; Fig. 2 is the logic block diagram of the present application; Fig. 3 is the overall flowchart of the present application. DETAILED DESCRIPTION

[0028] The present application will be further described in detail below in conjunction with the drawings and examples; it should be understood that the specific examples described herein are only used to explain the present application, but not to limit the present application; in addition, it should be noted that only the parts related to the present application are shown in the drawings for the convenience of description, but not all.

[0029] In conjunction with Figs. 1-3 shown, the present application provides a multi-source data driven frying machine cluster intelligent scheduling system, which comprises: a cloud computing layer module deployed on a remote server, comprising: an API interface docking unit: for accessing the order data interface of the take-out platform and the ordering software, receiving and storing order information; a user interaction unit: providing customer and store owner operation interfaces, for pushing the estimated delivery time and receiving scheduling parameter configuration instructions, and the configuration instructions have higher priority than the automatic scheduling rules; a historical data analysis unit: based on AI model analysis of order history records, building user portrait library and updating dish process parameters; a system configuration management unit: providing a set of configurable scheduling parameters, including order combining time window, protection window number, maximum order combining number and equipment load threshold; an edge computing layer module deployed on a local node of the store, comprising: a core scheduling engine: executing a multi-factor weight algorithm, dynamically combining orders, calculating order insertion priority and distributing equipment load balance; a device state monitoring unit: real-time collection of frying machine device state data, construction of available device pool and triggering of fault-tolerant mechanism; Real-time task allocation unit: allocate tasks to the wok cluster based on device load rate, load rate = total remaining task frying time / unit time; Multi-source data fusion module, comprising: Order processing unit: filter invalid orders, extract dish type, portion, and estimated delivery time attributes, and generate taste preference markers through AI analysis of user notes; Inventory linkage unit: communicate with the food material inventory system, check the food material readiness status and feedback material shortage warning; Recipe process database: store dish frying parameters for task decomposition and time estimation; Closed-loop control module: in response to order state changes, device failures or order cancellation events, trigger queue recalculation and real-time issuance of updated task sequences to the wok cluster, while pushing high-precision estimated delivery time, error ±1 minute, to the user interaction unit.

[0030] The historical data analysis unit is specifically used for: Analyzing user notes in historical orders through a natural language processing model to generate structured taste preference labels, including quantitative ratings of spiciness, saltiness, and sweetness; Based on user order cancellation records and order frequency, dynamically update the customer order cancellation rate portrait library; According to dish process parameter adjustment records, optimize the frying temperature and time parameters in the recipe database.

[0031] When the core scheduling engine performs dynamic order merging: The time window is set to be configurable 5-10 minutes; The protection window number is fixed at the first 8 tasks in the queue; The maximum number of merged portions threshold is total portions per order ≤4 portions.

[0032] In the real-time task allocation unit: The device load rate threshold is preset to 80%; When the load rate exceeds 80%, tasks are temporarily stored in the order queue pool; The order queue pool backlog threshold is set to 10 tasks, triggering an alarm to notify the store manager for manual intervention.

[0033] The fault tolerance mechanism triggered by the device state monitoring unit includes: When a device failure is detected, all uncompleted tasks in the device queue are automatically rolled back to the order queue pool; Based on the remaining task urgency, reallocate to high-availability devices and mark the estimated repair time of the faulty device; Push an alarm notification to the store manager including the fault code and the number of affected orders or the estimated delay time.

[0034] The application also provides a multi-source data-driven cluster intelligent scheduling method for a stir-fry machine, applied to the field of catering order placement, based on the above system, the method comprises the following steps: S1, order receiving: receiving multi-source order data through the API interface of the cloud computing layer to connect the order data interface of the takeout platform and the ordering software; S2, order processing: Filter invalid orders and extract order key attributes, including dish type, portion, order time, and estimated delivery time; Associate the user portrait library through the user's unique identifier, and generate structured taste preference markers based on AI model analysis of user notes and portrait data; Synchronize the inventory status of food materials and provide a lack of material warning; S3, task decomposition and resource assessment: Decompose the order into independent stir-frying task units; Verify the ability of the stir-fry machine to support the recipe process and the readiness of the food materials; Estimate the stir-frying time of each task and set a delay scheduling time according to the order cancellation risk probability; S4, dynamic order merging: Apply merging rules: only merge the same type of dishes within a set time window, set a protection window for the first N tasks to prevent merging, limit the maximum single portion threshold, and ensure that merging does not cause the original order to exceed the time limit; Optimize the task queue to achieve balanced serving of multi-table orders and overall efficient meal preparation; S5, priority calculation of inserted orders: Use a multi-factor weighted scoring model, including operator intervention forced task and urgent order weight promotion coefficient; Evaluate the global impact of inserted orders: calculate the time saved by inserted orders and the total overtime increment caused by other orders, and determine the optimal inserted order placement by maximizing (benefit-cost); S6, device scheduling strategy: Build a pool of available devices, filter stir-fry machines that are idle, have sufficient food materials, and are functioning properly; Calculate the device load rate, where load rate = total stir-frying time of remaining tasks / unit time; Assign new tasks to the device with the lowest load rate, or temporarily store tasks in the order placement pool when the load rate exceeds the preset threshold, and trigger an alarm when the order placement pool backlog reaches the threshold; In response to device failure or material waiting events, automatically return incomplete tasks to the order placement pool and reschedule them; S7, real-time feedback and adjustment: In response to order status changes, device failures, or order cancellation events, trigger task queue recalculation; The updated task sequence is issued to the cluster of wok machines, and the estimated delivery time is pushed to the user end, with a precision control within ±1 minute.

[0035] Generating structured taste preference labels in the S2 order processing step includes: Using the BERT model to analyze keywords in user notes, including "less spicy" and "no onions"; Convert the analysis results into device executable seasoning instruction code; Correlate historical preference data in the user profile library to generate a final taste score.

[0036] The multi-factor weighted scoring model in the S5 order insertion priority calculation step is: Order insertion priority score = (order base weight x urgent coefficient) + operator intervention weight; Where: The order base weight is the inherent priority score of the order itself; The urgent coefficient has a value range of 1.0-2.0, with a default value of 1.5; The operator intervention weight is a Boolean value of 0 or 1.

[0037] In the S6 device scheduling strategy step: The unit time is set to 60 minutes; Load rate = total remaining task frying time / 60; When the lowest load rate in the available device pool is > 80%, the task enters the order pool and waits.

[0038] When responding to device failures in the S6 device scheduling strategy step: Immediately interrupt the current frying task of the device; Sort the unfinished tasks in ascending order of remaining frying time and return them to the order pool; When reassigning, prioritize tasks with a remaining time of less than 3 minutes.

[0039] Embodiment 1: (1) Order placement algorithm flow Step 1: Order reception The system cloud layer uniformly interfaces with the API interfaces of major food delivery platforms and ordering software to receive order data. This design reduces the complexity of upgrading the store order placement system when interfacing with new platforms.

[0040] Step 2: Order processing Filter invalid orders (such as canceled orders).

[0041] Extract key attributes: dish type, portion size, order time, estimated delivery time, store information, table number information, etc.

[0042] When the system acquires an order, the user's portrait is marked in the order through the unique identification of the ordering user, which is used for subsequent calculation of order emergency weight and customer order cancellation rate.

[0043] Then the user's order notes and user portrait are analyzed and recognized by AI to convert into individual dish taste preferences, which can be directly marked on the equipment for subsequent frying to prevent chefs from missing.

[0044] The system data platform regularly performs AI analysis on order history records to update user portraits for adjusting dish technology and taste.

[0045] Synchronize and check the status of food material inventory (link with inventory system), and feedback material shortage information to staff.

[0046] Step 3: Task decomposition and resource assessment Decompose the order into frying task units that can be independently executed by the frying machine.

[0047] Check whether the equipment has the ability to execute the dish (such as recipe support) and whether the required food materials are ready.

[0048] Estimate the frying time of each task. Check the order cancellation risk and set a delayed scheduling time according to the probability.

[0049] Step 4: Dynamic order merging Merging rules: Only the same type of dish tasks within a set time window (such as within N minutes) are allowed to be merged.

[0050] Set a "protection window" for the first N (such as 8) tasks in the current queue to prevent being merged by subsequent orders, ensuring the continuity of meal serving.

[0051] The number of dishes merged at a time does not exceed the pre-set maximum threshold of merged dishes.

[0052] The extended time after merging does not cause the original order to exceed the time limit.

[0053] Scheduling goal: dynamically calculate the waiting time of each task and the customer's estimated waiting time, optimize the queue to achieve balanced serving (especially for multi-table orders) and overall efficient dish preparation.

[0054] Step 5: Insertion priority calculation Use a multi-factor weighted scoring model to determine the feasibility and optimal position of insertion: Operator intervention authority: kitchen operators (such as chefs) have the authority to forcibly top certain tasks (for example, when a dish needs to be delayed due to unready food materials, the tasks following it in the queue can be prioritized).

[0055] Urgent order processing: the front desk can mark orders as urgent (weighting factor, e.g. 1.5). The system evaluates the impact of the insertion operation on the overall queue: Calculate the order's potential reduced waiting time (multiplied by the urgency factor).

[0056] Calculate the total overtime increment of other orders potentially caused by the insertion.

[0057] Determine the optimal insertion position based on the evaluation results (benefit-cost) (i.e. the position where (insertion time saved * urgency factor) - total overtime increment of other orders is maximized or exceeds a threshold).

[0058] Step 6: Equipment scheduling strategy Build "available equipment pool": filter out the woks with status "idle", sufficient ingredients, and normal running status.

[0059] Calculate equipment load rate: load rate = (total remaining task frying time) / (unit time). This indicator reflects the relative busyness of the equipment.

[0060] Task allocation: Allocate new tasks to the equipment with the lowest load rate in the available equipment pool.

[0061] If the load rate of all available equipment exceeds the preset threshold, the task is temporarily stored in the order queue pool.

[0062] When the backlog of tasks in the order queue pool reaches a configurable threshold, trigger an alarm notification to the store manager.

[0063] Fault tolerance mechanism: when a device failure or long "waiting for ingredients" is detected, the system automatically returns the unfinished tasks in its queue to the order queue pool, re-computes the scheduling, and sends an alarm notification to the store manager.

[0064] Step 7: Real-time feedback and adjustment The system responds to key events (such as the completion of a single dish frying, the arrival of a new order, the occurrence of a cancellation, and the change of equipment status) to trigger the re-computation and scheduling of the queue.

[0065] Push the updated estimated delivery time (with an accuracy of ±1 minute) to the customer side (e.g. mini-program).

[0066] Compared with existing technologies (including traditional manual order queuing, simple sequential execution of wok systems, and industrial MES scheduling not suitable for catering scenarios), the core invention of the present application lies in: Firstly apply the concept of industrial scheduling to fast cooking: In view of the current situation that intelligent cooking machines on the market generally lack mature order scheduling systems, the concept and method system of planning and scheduling in the field of industrial manufacturing are innovatively introduced and adapted to the fast cooking industry for the first time, solving the high-frequency, short-cycle and strong timeliness order scheduling problems specific to the industry.

[0067] Using accumulated historical order data, building a user portrait library, and processing natural language data into structured data through AI assisted analysis to assist order scheduling.

[0068] Highly configurable dynamic scheduling engine: A highly configurable scheduling rule engine and parameter system is designed, which is optimized for fast cooking scenarios such as high-frequency combined orders, urgent orders, multi-table uniform serving, and device state sensitivity. The core is: Support for dynamic time window combined order strategy for similar dishes and protection window mechanism.

[0069] Intelligent order insertion algorithm based on multi-factor weighted scoring model (operator intervention, urgent weight, global influence assessment).

[0070] Consider multi-table order uniform serving strategy.

[0071] Load balancing algorithm based on real-time load rate of equipment (load rate = total remaining task frying time / unit time capacity).

[0072] Edge-cloud collaborative system architecture: The architecture design adopts edge computing and cloud computing collaboration: The core scheduling algorithm engine, device state monitoring, real-time task allocation and other modules with high requirements for low latency and stability are deployed on edge computing nodes close to the cooking machine, ensuring that the core scheduling function can still run reliably when the network fluctuates or the cloud is interrupted.

[0073] The third-party takeout platform API interface, user interaction interface (customer end / store owner end), historical data analysis, system configuration management and other modules are deployed in the cloud, which is convenient for continuous iteration and upgrade, cross-store management and optimization of user experience. This architecture effectively balances real-time performance, reliability and maintainability / expandability.

[0074] The above only describes the preferred embodiments of the present application, and does not limit the patent scope of the present application. Any equivalent structural transformation made under the inventive concept of the present application, or direct / indirect application in other related technical fields is included in the protection scope of the present application.

Claims

1. A multi-source data driven cooking machine cluster intelligent scheduling system, characterized by: The system includes: The cloud computing layer module is deployed on a remote server and includes: API interface docking unit: used to access the order data interface of the food delivery platform and ordering software, receive and store order information; User Interaction Unit: Provides an operation interface for customers and store managers to push estimated meal delivery times and receive scheduling parameter configuration instructions. Configuration instructions have higher priority than automatic scheduling rules. Historical data analysis unit: Analyzes order history records based on AI models, builds user profiles, and updates dish processing parameters; System Configuration Management Unit: Provides a set of configurable scheduling parameters, including the order consolidation time window, number of protection windows, maximum number of orders consolidating, and device load threshold; The edge computing layer module is deployed on the store's local node and includes: Core scheduling engine: Executes a multi-factor weighting algorithm to dynamically merge orders, calculate order priority, and balance device load. Equipment status monitoring unit: collects cooking machine equipment status data in real time, builds an available equipment pool and triggers the fault tolerance mechanism; Real-time task allocation unit: allocates tasks to the cooking machine cluster based on the device load rate, where load rate = total cooking time of remaining tasks / unit time; Multi-source data fusion module, including: Order processing unit: Filters invalid orders, extracts dish type, quantity, and estimated delivery time attributes, and generates taste preference tags through AI analysis of user notes; Inventory linkage unit: communicates with the food inventory system to verify the food readiness status and provide early warning of food shortages; Recipe process database: stores cooking parameters for dishes, used for task decomposition and time estimation; Closed-loop control module: Responds to order status changes, equipment failures, or order returns, triggers queue recalculation, and sends the updated task sequence to the cooking machine cluster in real time. It also pushes a high-precision estimated delivery time, with an error of ±1 minute, to the user interaction unit.

2. The multi-source data driven cooking machine cluster intelligent scheduling system according to claim 1, characterized in that: The historical data analysis unit is specifically used for: A natural language processing model is used to parse user notes in historical orders and generate structured taste preference labels, including quantitative scores for spiciness, saltiness, and sweetness. Dynamically update the customer return rate profile database based on user return records and order frequency; Adjust the records according to the dish process parameters and optimize the stir-frying temperature and time parameters in the recipe database.

3. The multi-source data driven cooking machine cluster intelligent scheduling system according to claim 1, characterized in that: When the core scheduling engine performs dynamic order merging: The time window is set to a configurable 5-10 minutes; The number of protection windows is fixed to the first 8 tasks in the queue; The maximum number of copies in a single order is ≤ 4 copies.

4. The multi-source data driven cooking machine cluster intelligent scheduling system according to claim 1, characterized in that: In the real-time task allocation unit: The preset threshold value of the equipment load rate is 80%; When the load rate exceeds 80%, the task is temporarily stored in the order pool; The backlog threshold of the order pool is set to 10 tasks, and after the alarm is triggered, the store manager is notified for manual intervention.

5. The multi-source data driven cooking machine cluster intelligent scheduling system according to claim 1, characterized in that: The fault tolerance mechanism triggered by the device status monitoring unit includes: When a device failure is detected, all unfinished tasks in the device queue are automatically rolled back to the order pool; Redistribute remaining tasks to high-availability equipment based on their urgency, and mark the estimated repair time for faulty equipment; Push an alarm notification containing the fault code and the affected order quantity or expected delay time to the store manager.

6. A multi-source data-driven cooking machine cluster intelligent scheduling method, applied to the field of catering order scheduling, based on the system according to any one of claims 1-5, characterized in that: The method The following steps are involved: S1, order reception: Connect to the order data interface of the food delivery platform and the ordering software through the API interface of the cloud computing layer to receive multi-source order data; S2, Order Processing: Filter out invalid orders and extract key order attributes, including dish type, number of portions, order time, and estimated delivery time; The user profile database is linked to the user's unique identifier, and user notes and profile data are analyzed based on the AI ​​model to generate structured taste preference tags; Synchronously verify the inventory status of ingredients and provide early warning of shortages; S3, task decomposition and resource assessment: Decompose the order into independent cooking task units; Verify the cooking machine's ability to support recipe processing and the ingredients' readiness; Estimate the processing time for each task and set a delayed scheduling time based on the probability of chargeback risk; S4, Dynamic Merge Orders: Apply merging rules: Only merge tasks of the same type within a set time window, set a protection window for the first N tasks in the queue to prohibit merging, set a maximum threshold for the number of combined orders, and ensure that merging orders does not cause the original order to time out; Optimize the task queue to achieve balanced serving of orders from multiple tables and overall efficient meal preparation; S5, order priority calculation: A multi-factor weighted scoring model is used, including operator intervention to force top tasks and expedited order weighting factors; Evaluate the global impact of order insertion: Calculate the time saved by inserting an order and the total timeout increase that may result for other orders, and determine the optimal insertion location by maximizing the (benefit-cost) ratio. S6, device scheduling strategy: Build a pool of available devices and filter out cooking machines that are idle, have sufficient ingredients, and are operating normally; Calculate the device load rate, where load rate = total remaining task processing time / unit time; Assign new tasks to the device with the lowest load rate, or temporarily store tasks in the order pool when the load rate exceeds the preset threshold, and trigger an alarm when the backlog in the order pool reaches the threshold; In response to equipment failure or material waiting events, unfinished tasks are automatically rolled back to the order pool and rescheduled; S7, real-time feedback and adjustment: Respond to order status changes, equipment failures, or order returns, triggering task queue recalculation; The updated task sequence is sent to the cooking machine cluster, and the estimated meal delivery time is pushed to the user end, with an accuracy of ±1 minute.

7. The multi-source data driven cooking machine cluster intelligent scheduling method according to claim 6, characterized in that: Generating structured taste preference markup in the S2 order processing step includes: Use the BERT model to parse keywords in user comments, including "less spicy" and "no onion"; Convert the parsing results into seasoning instruction codes that can be executed by the device; The historical preference data in the user portrait library is associated and weighted to generate the final taste score.

8. The multi-source data driven cooking machine cluster intelligent scheduling method according to claim 6, characterized in that: The multi-factor weighted scoring model in the S5 order priority calculation step is: Insertion order priority score = (order base weight × expedited coefficient) + operator intervention weight; in: The order base weight is the inherent priority score of the order itself; The expedited coefficient ranges from 1.0 to 2.0, with a default value of 1.5; Operator intervention weight is a Boolean value 0 or 1.

9. The multi-source data driven cooking machine cluster intelligent scheduling method according to claim 6, characterized in that: In the S6 device scheduling strategy step: The unit time is set to 60 minutes; Load rate = total cooking time of remaining tasks / 60; When the lowest load rate in the available device pool is greater than 80%, the task enters the queue pool and waits.

10. The multi-source data driven cooking machine cluster intelligent scheduling method according to claim 6, characterized in that: When responding to a device failure in the S6 device scheduling strategy step: Immediately interrupt the current cooking task of the device; Sort unfinished tasks in ascending order by remaining processing time and return them to the order pool; When reallocating tasks, priority will be given to tasks with a remaining duration of less than 3 minutes.

Images