Multi-station microwave oven intelligent cooking combined control method and system

Abstract

The application relates to the technical field of data processing, and discloses a multi-station microwave oven intelligent cooking combined control method and system. The method comprises the following steps: collecting temperature, water content and weight data of each station to establish a multi-station comprehensive state vector; inputting the state vector into a graph neural network to establish a station interaction model; determining a parallel execution scheme based on the interaction model by using an out-of-order execution strategy; returning to a stable state when the pre-computed execution state is inconsistent; and dynamically adjusting power and time slice allocation to generate a cooperative control instruction. The application solves the problem of lack of cooperative control and intelligent scheduling of each station in a multi-station microwave oven, and improves the overall efficiency of multi-station cooking and the consistency of food cooking quality.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method and system for intelligent cooking combination control of a multi-station microwave oven. Background Technology

[0002] Current microwave oven control technology primarily employs a single-station independent control method, heating a single food item by pre-setting time and power parameters. Traditional control methods are based on fixed cooking programs; users select the appropriate heating mode according to the type of food, and the system executes the heating process according to preset parameters. Some advanced microwave ovens have incorporated sensor technology, adjusting heating parameters by detecting changes in food temperature and humidity, but these technologies remain limited to a single-station independent operation mode.

[0003] However, existing technologies have significant shortcomings in the application of multi-station microwave ovens. First, there is a lack of effective coordination mechanisms between the stations, which can easily lead to power competition and heat interference when multiple stations are working simultaneously, resulting in inconsistent cooking outcomes. Second, traditional control methods employ linear execution strategies, failing to optimize the task execution sequence based on the cooking requirements of different foods and the interrelationships between stations. Third, existing systems lack the ability to predict and handle abnormal states; when a deviation occurs at a station, it cannot be adjusted in a timely manner, affecting the overall cooking quality.

[0004] Because each workstation operates independently and lacks mutual awareness, resource allocation becomes unreasonable and execution efficiency is low. Further analysis reveals that the core technical challenges in a multi-workstation environment include: how to establish an accurate model of the interrelationships between workstations; how to achieve intelligent scheduling based on task characteristics and resource constraints; how to predict and handle abnormal states during execution; and how to dynamically adjust resource allocation strategies based on real-time conditions. These problems are interconnected and progressively deepen, requiring comprehensive technological innovation across the entire process, from data acquisition and relationship modeling to task scheduling, anomaly handling, and resource optimization, to achieve effective solutions. Summary of the Invention

[0005] This application provides a method and system for intelligent cooking combination control of multi-station microwave ovens, which solves the problem of lack of coordinated control and intelligent scheduling of each station in a multi-station microwave oven, thereby improving the overall efficiency of multi-station cooking and the consistency of food cooking quality.

[0006] In a first aspect, this application provides a multi-station microwave oven intelligent cooking combination control method, the multi-station microwave oven intelligent cooking combination control method comprising:

[0007] Step S1: Collect temperature distribution data, moisture content data and weight change data of each workstation, and establish a multi-workstation comprehensive state vector after data fusion processing;

[0008] Step S2: Input the multi-station integrated state vector into the graph neural network to identify the heat transfer relationship and power interaction relationship between each station, and establish a station interaction model;

[0009] Step S3: Analyze the execution constraint relationship between cooking tasks based on the workstation interaction model, rearrange the start timing of each workstation task using a random execution strategy, and determine the parallel execution scheme;

[0010] Step S4: Pre-calculate the execution state of the next stage for each task node in the parallel execution scheme. When the actual execution result does not match the pre-calculated result, roll back to the previous stable state.

[0011] Step S5: Based on the actual execution progress and resource requirements of each workstation, dynamically adjust the power allocation ratio and time slice allocation strategy to generate multi-workstation collaborative control instructions.

[0012] Optionally, step S1 includes:

[0013] A two-dimensional temperature distribution matrix is ​​obtained by scanning and detecting the food surface in each workstation using an infrared temperature sensor array.

[0014] The frequency of food in each workstation is scanned using a dielectric property detection device to obtain a loss factor sequence that reflects the change in the internal water content of the food.

[0015] High-precision weight sensors are installed at the bottom of each workstation for continuous weighing and monitoring to obtain data on the time-series changes in food weight.

[0016] The two-dimensional temperature distribution matrix, the loss factor sequence, and the weight time-series change data are subjected to Kalman filtering for noise reduction to obtain a multi-station integrated state vector.

[0017] Optionally, step S2 includes:

[0018] The multi-station integrated state vector is constructed as a dynamic weighted graph structure, where each station is a graph node, and the physical distance and power level difference between stations are edge weights.

[0019] The dynamic weighted graph structure is subjected to multi-layer propagation computation of graph convolutional neural network. By convolving the adjacency matrix and the node feature matrix, a node embedding vector containing information about the mutual influence between workstations is obtained.

[0020] The heat transfer coefficient and electromagnetic interference coefficient between each workstation are calculated based on the node embedding vector, and the influence intensity weight distribution between workstations is generated by softmax normalization.

[0021] The heat transfer coefficient, the electromagnetic interference coefficient, and the influence intensity weight distribution are combined to construct the workstation interaction model matrix.

[0022] Optionally, step S3 includes:

[0023] Based on the workstation interaction model matrix analysis, the resource dependencies and temporal constraints among each cooking task are analyzed, and a cooking task dependency graph is constructed using a topological sorting algorithm.

[0024] Critical path analysis was performed on the cooking task dependency graph to identify the set of independent tasks without pre-dependent conditions and the chain of related tasks with execution order constraints.

[0025] The set of independent tasks and the chain of related tasks are input into the out-of-order execution scheduling algorithm, and the genetic algorithm is used to optimize and solve the task rearrangement sequence that breaks the traditional linear execution order.

[0026] Based on the task rearrangement sequence and the power carrying capacity constraints of each workstation, a parallel execution scheme containing task start time, execution duration, and workstation allocation information is generated.

[0027] Optionally, the step of analyzing the resource dependencies and temporal constraints among cooking tasks based on the workstation interaction model matrix, and constructing a cooking task dependency graph using a topological sorting algorithm, includes:

[0028] The heat transfer coefficient and electromagnetic interference coefficient between each workstation are extracted from the workstation interaction model matrix, and the workstation pairs that have mutual influence are identified by the threshold judgment method.

[0029] Based on the workstation pair combination, a resource competition relationship matrix is ​​established between cooking tasks, wherein tasks within the same workstation pair are marked as resource conflicts, and tasks within different workstation pairs are marked as resource independence.

[0030] A depth-first traversal is performed on the resource competition relationship matrix to calculate the in-degree and out-degree values ​​of each cooking task, and a set of directed edges reflecting the order of task execution is generated.

[0031] Each cooking task is treated as a graph node, and the set of directed edges is used as the node connection relationship. The Kahn algorithm is used to perform topological sorting to obtain a cooking task dependency graph that satisfies resource dependency constraints.

[0032] Optionally, step S4 includes:

[0033] Based on the current execution status and historical execution trajectory of each task node in the parallel execution scheme, the temperature change trend and power demand of each workstation in the next time window are predicted by a long short-term memory neural network.

[0034] The temperature change trend and the power demand are compared with the preset execution state threshold range. When the predicted value exceeds the threshold range, the execution deviation detection mechanism is triggered.

[0035] A status snapshot is saved for the workstation that triggers the execution deviation detection mechanism, recording the power configuration parameters, temperature distribution data and task execution progress at the current moment, and constructing checkpoint status data;

[0036] Based on the checkpoint status data, the execution status of the deviation station is rolled back to the most recent stable historical time node, and the execution timing arrangement of the relevant task nodes in the parallel execution scheme is updated.

[0037] Optionally, step S5 includes:

[0038] Real-time monitoring of the completion rate and remaining execution time of cooking tasks at each workstation, combined with the current power consumption, calculates the resource utilization efficiency index of each workstation.

[0039] Based on the resource utilization efficiency index, a power demand priority queue is established, and high-efficiency, low-power workstations are set as priority allocation objects, while low-efficiency, high-power workstations are set as restricted allocation objects.

[0040] Based on the power demand priority queue and the total power capacity limit of the system, the power allocation quota and time slice length of each workstation are recalculated using the shortest job first scheduling algorithm;

[0041] The recalculated power allocation quota and time slice length are encapsulated into a multi-station collaborative control instruction containing station number, execution instructions, and timing control parameters.

[0042] Secondly, this application provides a multi-station microwave oven intelligent cooking combination control system, the multi-station microwave oven intelligent cooking combination control system comprising:

[0043] The fusion module is used to collect temperature distribution data, moisture content data and weight change data of each workstation, and to establish a multi-workstation comprehensive state vector after data fusion processing.

[0044] The input module is used to input the multi-station integrated state vector into the graph neural network to identify the heat transfer relationship and power interaction relationship between each station and establish a station interaction model.

[0045] The startup module is used to analyze the execution constraint relationship between cooking tasks based on the workstation interaction model, rearrange the startup timing of each workstation task using a randomized execution strategy, and determine the parallel execution scheme.

[0046] The calculation module is used to pre-calculate the execution state of the next stage for each task node in the parallel execution scheme, and to fall back to the previous stable state when the actual execution result does not match the pre-calculated result.

[0047] The generation module is used to dynamically adjust the power allocation ratio and time slice allocation strategy according to the actual execution progress and resource requirements of each workstation, and generate multi-workstation collaborative control instructions.

[0048] Thirdly, a multi-station microwave oven intelligent cooking combination control device is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the multi-station microwave oven intelligent cooking combination control device to execute the above-described multi-station microwave oven intelligent cooking combination control method.

[0049] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to perform the above-described multi-station microwave oven intelligent cooking combination control method.

[0050] The technical solution provided in this application effectively solves the technical problems of lack of spatiotemporal correlation and data isolation between devices in traditional battery test data acquisition by collecting test parameters such as battery voltage, temperature, and current of each downhole loader and assigning equipment number, spatial coordinates, and timestamp identifiers to each data point to establish a multi-device hierarchical data structure. The establishment of the multi-device hierarchical data structure enables the correlation analysis of battery test data from different devices under a unified spatiotemporal coordinate system, providing a data foundation for subsequent anomaly tracing and fault location. By establishing a three-dimensional data storage model of the multi-device hierarchical data structure according to time, space, and device dimensions, and constructing a spatiotemporal data cube that supports rapid retrieval, the storage efficiency and query speed of massive battery test data are significantly improved. Compared with traditional relational database storage methods, the spatiotemporal data cube can complete complex multi-dimensional data retrieval within milliseconds. When an abnormal battery status is detected, the data change trajectory before the anomaly occurs is traced back in the spatiotemporal data cube, and the propagation path of the abnormal signal among multiple devices is identified. This achieves a technological leap from passive anomaly response to proactive anomaly prediction and tracing, enabling the system to quickly locate the source of the anomaly and analyze its impact range. This approach analyzes the temporal correlation of battery state changes among devices based on propagation paths and quantifies the intensity of direct and indirect causal influences between devices. It overcomes the limitations of traditional methods that can only identify surface-level relationships, achieving accurate modeling of deep-seated causal relationships in complex multi-device systems. By adjusting the acquisition frequency and range of battery test data for each device based on the intensity of direct and indirect causal influences and optimizing the multi-device collaborative test data acquisition strategy, an adaptive data acquisition mechanism is formed. This allows the system to dynamically adjust the data acquisition strategy according to the actual influence relationships between devices, ensuring the accuracy of key data acquisition while avoiding resource waste.

[0051] The spatiotemporal data cube construction algorithm, through the design of the MapReduce parallel computing framework and multi-level index structure, enables the system to efficiently process large-scale time-series data generated in downhole multi-device environments. Especially when dealing with surges in abnormal data during equipment failures, the algorithm's parallel processing capability ensures the real-time performance and accuracy of data processing. The anomaly propagation chain tracing algorithm employs a breadth-first search strategy combined with spatiotemporal backtracking computation, enabling rapid identification of the propagation path of abnormal signals in complex multi-device networks. Compared to traditional single-point anomaly detection methods, this algorithm reveals the inherent laws governing anomaly propagation between devices, providing a scientific basis for preventative maintenance. The causality quantification algorithm, through the combined application of Granger causality testing and transit entropy theory, achieves mathematical modeling of complex inter-device interactions. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a schematic diagram of one embodiment of the intelligent cooking combination control method for multi-station microwave ovens in this application.

[0054] Figure 2 This is a schematic diagram of the interaction coefficient matrix between workstations in an embodiment of this application;

[0055] Figure 3 This is a schematic diagram illustrating the dynamic power allocation process of a multi-station microwave oven in an embodiment of this application.

[0056] Figure 4 This is a schematic diagram of one embodiment of the multi-station microwave oven intelligent cooking combination control system in this application.

[0057] Figure 5 This is a schematic block diagram of the structure of the multi-station microwave oven intelligent cooking combination control device in this embodiment of the invention. Detailed Implementation

[0058] This application provides a multi-station microwave oven intelligent cooking combination control method and system. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0059] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the intelligent cooking combination control method for multi-station microwave ovens in this application includes:

[0060] Step S1: Collect temperature distribution data, moisture content data and weight change data of each workstation, and establish a multi-workstation comprehensive state vector after data fusion processing;

[0061] Step S2: Input the multi-station integrated state vector into the graph neural network to identify the heat transfer relationship and power interaction relationship between each station and establish a station interaction model;

[0062] Step S3: Analyze the execution constraint relationship between cooking tasks based on the workstation interaction model, rearrange the start timing of each workstation task using a random execution strategy, and determine the parallel execution scheme;

[0063] Step S4: Pre-calculate the execution state of the next stage for each task node in the parallel execution scheme. When the actual execution result does not match the pre-calculated result, roll back to the previous stable state.

[0064] Step S5: Based on the actual execution progress and resource requirements of each workstation, dynamically adjust the power allocation ratio and time slice allocation strategy to generate multi-workstation collaborative control instructions.

[0065] It is understood that the executing entity of this application can be a multi-station microwave oven intelligent cooking combination control system, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiments use a server as an example for illustration.

[0066] Specifically, an infrared temperature sensor array scans the food surface at each workstation 10 times per second to obtain a two-dimensional temperature distribution matrix, where each element represents the temperature value at a specific location. A dielectric property detection device scans the food by emitting a 2.45 GHz frequency signal, detecting signal attenuation and converting the attenuation value into a loss factor, which directly reflects changes in the food's internal moisture content. High-precision weight sensors are installed at the bottom of each workstation to continuously monitor weight changes at a sampling interval of 0.1 seconds, recording the temporal weight changes of the food during the heating process. Kalman filtering preprocesses the raw sensor data, eliminating noise caused by downhole vibration and electromagnetic interference through prediction and update steps. The filtered data is then fused according to a preset vector format to obtain a multi-device state vector.

[0067] A dynamic weighted graph structure is constructed using multi-device state vectors as input data, with each workstation acting as a graph node. The connection weights between nodes are calculated based on physical distance and power level differences. A graph convolutional neural network performs multi-layer propagation calculations on this graph structure. Each convolutional operation aggregates and propagates the feature information of neighboring nodes through mathematical operations on the adjacency matrix and the node feature matrix. During the forward propagation of the network, each node collects information from its neighbors and updates its own feature vector after processing by a nonlinear activation function. After multi-layer propagation, node embedding vectors containing information about the mutual influence between workstations are obtained. Based on the node embedding vectors, the heat transfer coefficient and electromagnetic interference coefficient between each workstation are calculated. The heat transfer coefficient reflects the intensity of heat flow between workstations, and the electromagnetic interference coefficient describes the degree of mutual influence of microwave power. Softmax normalization is used to generate the influence intensity weight distribution between workstations. Normalization ensures that the sum of all weight values ​​is 1, facilitating subsequent analysis. The heat transfer coefficient, electromagnetic interference coefficient, and influence intensity weight distribution are combined to construct the workstation interaction model matrix.

[0068] Based on a workstation interaction model matrix analysis, resource dependencies and temporal constraints among cooking tasks are analyzed. A topology sorting algorithm analyzes task dependencies, identifying independent task sets without pre-existing dependencies and related task chains with execution order constraints. Critical path analysis calculates the earliest and latest start times of task execution to determine the sequence of critical tasks affecting the overall completion time. An out-of-order execution scheduling algorithm breaks the traditional limitation of execution according to user input order, rearranging the execution order based on workstation resource conditions and task characteristics. A genetic algorithm, as an optimization tool, encodes task execution schemes as chromosomes, sets a fitness function to evaluate the merits of the schemes, and gradually optimizes the task rearrangement sequence through evolutionary operations such as selection, crossover, and mutation. After convergence, the algorithm obtains the optimal task allocation scheme, which considers the power carrying capacity constraints of each workstation and generates a parallel execution scheme containing task start time, execution duration, and workstation allocation information.

[0069] A prediction mechanism is established for each task node in the parallel execution scheme. A Long Short-Term Memory (LSTM) neural network receives the current execution state and historical execution trajectory of each task node as input. Through the coordinated operation of forget gates, input gates, and output gates, the network selectively retains and updates historical information to predict the execution state of each workstation in the next time window. The prediction result is numerically compared with a preset execution state threshold range. When the predicted value exceeds the threshold range, an execution deviation detection mechanism is triggered. A state snapshot saving mechanism records the power configuration parameters, temperature distribution data, and task execution progress at the time deviation detection is triggered; this data constitutes checkpoint state data. A rollback mechanism restores the execution state of the deviating workstation to the most recent stable historical time point based on the checkpoint state data, while simultaneously updating the execution sequence of relevant task nodes in the parallel execution scheme.

[0070] A real-time monitoring mechanism is established to track the completion rate and remaining execution time of cooking tasks at each workstation. The resource utilization efficiency index is calculated as the ratio of current power consumption to task completion rate, reflecting the energy efficiency level of each workstation. A power demand priority queue is established based on the resource utilization efficiency index, prioritizing high-efficiency, low-power workstations and restricting the allocation of resources to low-efficiency, high-power workstations. The shortest job first scheduling algorithm recalculates the power allocation quota for each workstation based on the power demand priority queue and the system's total power capacity limit, prioritizing resource allocation for tasks with short execution times and high efficiency. The time slice length is dynamically adjusted according to task priority, with higher-priority tasks receiving longer time slices. During the dynamic adjustment process, the algorithm monitors the status changes of each workstation in real time, immediately recalculating the allocation scheme when a change in resource demand is detected. The recalculated power allocation quota and time slice length are encapsulated into control instructions, which include the workstation number, execution command, and timing control parameters.

[0071] In one specific embodiment, step S1 includes:

[0072] A two-dimensional temperature distribution matrix is ​​obtained by scanning and detecting the food surface in each workstation using an infrared temperature sensor array.

[0073] Frequency scanning of food at each workstation is performed using a dielectric property detection device to obtain a loss factor sequence that reflects changes in the internal moisture content of the food.

[0074] High-precision weight sensors are installed at the bottom of each workstation for continuous weighing and monitoring to obtain data on the time-series changes in food weight.

[0075] Kalman filtering was applied to the two-dimensional temperature distribution matrix, loss factor sequence, and weight time-series variation data to denoise, resulting in a multi-station integrated state vector.

[0076] Specifically, during the infrared temperature sensor array scanning detection process, the sensors are arranged in a matrix above each workstation, with each sensor corresponding to a specific area of ​​the food surface. Scanning sampling is performed according to a preset time interval and spatial resolution. The sensors receive infrared radiation signals emitted from the food surface, converting the radiation intensity into temperature values. The scanning process covers the entire food surface, resulting in a two-dimensional data structure containing multiple temperature measurement points. Each element in the two-dimensional temperature distribution matrix corresponds to the temperature value at a specific coordinate location; the rows and columns of the matrix represent the horizontal and vertical coordinates of the food surface, respectively. Temperature data acquisition relies on the Stefan-Boltzmann law. The sensors calculate the corresponding surface temperature based on the received radiation power density, and the voltage signal output by the sensors is converted into a digital temperature value by an analog-to-digital converter.

[0077] The frequency scanning process of the dielectric property testing device is based on the principle of microwave dielectric heating. The device emits electromagnetic wave signals of a specific frequency into the food and analyzes the food's dielectric properties by measuring the propagation and attenuation of the signal within the food. Frequency scanning refers to continuously changing the transmitted signal frequency within a preset frequency range and recording the signal attenuation level at each frequency point. The loss factor sequence reflects the food's absorption capacity for microwave signals of different frequencies; a higher loss factor value indicates stronger microwave absorption at that frequency. The device calculates the signal attenuation value by comparing the transmitted and received power. The attenuation value is directly related to the food's water content; foods with higher water content absorb microwave signals more strongly, resulting in a higher loss factor value. During the scanning process, the device records the loss factor value at each frequency point, arranging them in frequency order to obtain a loss factor sequence. Each element in the sequence corresponds to the dielectric loss characteristics at a specific frequency.

[0078] Continuous weighing monitoring using high-precision weight sensors is achieved through strain gauge technology. The sensors are installed at the bottom of each workstation, directly bearing the total weight of the food and containers. During weighing monitoring, the sensors continuously output weight signals, with a fixed sampling interval. Each sample captures the weight value at the current moment. The time-series data recording of food weight variations document the pattern of weight change over time during cooking, primarily due to the evaporation and volatilization of moisture. The sensor converts mechanical pressure into resistance changes, which are then converted into a voltage signal via a Wheatstone bridge circuit. This voltage signal is amplified and input to an analog-to-digital converter, where it is converted into a digital weight value. Continuous monitoring means the sensor operates continuously throughout the entire cooking process, recording weight data at preset time intervals, resulting in a data sequence containing timestamps and weight values.

[0079] Kalman filtering, as a data preprocessing step, removes random noise from the three types of raw data. The Kalman filter algorithm, based on a state-space model, estimates the true signal value through iterative processes of prediction and update steps. The prediction step predicts the current state value based on the previous time-stamped state estimate and the system dynamic model. The update step combines the predicted value with the current observation value to calculate the optimal state estimate. During the filtering process, the algorithm maintains the covariance matrix of the state estimate, which reflects the degree of uncertainty of the estimate. The algorithm achieves optimal filtering by minimizing the variance of the estimation error, preserving the true signal characteristics while eliminating random noise interference. The processing includes independent filtering of each element in the two-dimensional temperature distribution matrix, time-series filtering of the values ​​at each frequency point in the loss factor sequence, and continuous filtering of the weight time-series variation data. The three types of filtered data are then fused according to a preset vector format. This fusion process organizes different types of data into a unified data structure, resulting in a multi-station integrated state vector containing temperature, dielectric, and weight information.

[0080] In one specific embodiment, step S2 includes:

[0081] The multi-station integrated state vector is constructed as a dynamic weighted graph structure, where each station is a graph node and the physical distance and power level difference between stations are edge weights.

[0082] A multi-layer propagation computation of a graph convolutional neural network is performed on the dynamic weighted graph structure. By convolving the adjacency matrix and the node feature matrix, a node embedding vector containing information about the mutual influence between workstations is obtained.

[0083] The heat transfer coefficient and electromagnetic interference coefficient between each workstation are calculated based on the node embedding vector, and the influence intensity weight distribution between workstations is generated by softmax normalization.

[0084] The workstation interaction model matrix is ​​constructed by combining the heat transfer coefficient, electromagnetic interference coefficient, and influence intensity weight distribution.

[0085] Specifically, the dynamic weighted graph structure construction process maps the data of each workstation in the multi-workstation integrated state vector to graph nodes. Each node contains information on temperature distribution, moisture content, and weight change for that workstation. The connections between graph nodes are represented by edge weights, which are calculated based on two key factors: physical distance between workstations and power level differences. Physical distance measures the Euclidean distance between the center points of each workstation; closer workstations have stronger heat transfer and electromagnetic influence, resulting in larger edge weights. Power level differences reflect the differences in microwave power output between different workstations; workstations with greater power differences experience higher levels of mutual interference, leading to increased edge weights. The dynamic characteristic is reflected in the fact that edge weights adjust over time. When the workstation state changes, the corresponding edge weight values ​​are updated synchronously, maintaining consistency between the graph structure and the actual physical conditions.

[0086] Graph convolutional neural networks (CNNs) employ multi-layer propagation computation based on a dynamically weighted graph structure. Information propagation is achieved through mathematical operations on the adjacency matrix and node feature matrix. The adjacency matrix records the connections and weights between all nodes in the graph, with each element corresponding to the edge weight between two nodes. The node feature matrix contains the state feature information of each workstation node; rows represent different workstations, and columns represent different feature dimensions. During convolution, each node collects feature information from its neighboring nodes. The collected information is weighted and summed according to the edge weights, and the sum is processed by a non-linear activation function to update the node features. Multi-layer propagation refers to the neural network containing multiple convolutional layers, with information propagating layer by layer, and the output of each layer serving as the input to the next. During propagation, node features gradually fuse with information from neighboring nodes, resulting in a node embedding vector containing global information. The node embedding vector has a fixed dimension, with each element representing the node's value in a specific feature dimension. The embedding vector comprehensively reflects the interaction between the workstation and other workstations.

[0087] The coefficient calculation process based on node embedding vectors transforms abstract vector information into specific physical parameters. The heat transfer coefficient is calculated based on the temperature-related feature dimension of the node embedding vector, mapping vector elements to heat transfer intensity values ​​through a linear transformation. The heat transfer coefficient reflects the strength of heat flow between workstations; a larger coefficient indicates more active heat exchange between the two workstations. The electromagnetic interference (EMI) coefficient is calculated based on the power-related feature dimension of the node embedding vector, obtaining EMI intensity values ​​through a similar linear transformation. The EMI coefficient describes the degree of mutual influence of microwave power between workstations; a larger coefficient indicates a more significant impact of power changes at one workstation on another. Softmax normalization converts all coefficient values ​​into a probability distribution, ensuring that the sum of the influence intensities of all workstations on a specific workstation equals 1. The influence intensity weight distribution expresses the relative influence between workstations in probabilistic form, with each element in the weight distribution corresponding to the proportion of influence intensity between specific workstation pairs.

[0088] The workstation interaction model matrix construction process combines three types of data—heat transfer coefficient, electromagnetic interference coefficient, and influence intensity weight distribution—into a unified mathematical model. The matrix construction adopts a block matrix format, with different regions of the matrix storing different types of coefficient information. The heat transfer coefficient is filled in the upper triangular region of the matrix, the electromagnetic interference coefficient in the lower triangular region, and the influence intensity weight distribution, as a diagonal element, reflects the influence intensity of each workstation. The matrix combination process includes three steps: data format unification, numerical range standardization, and matrix dimension alignment. Data format unification converts coefficient data from different sources into the same numerical type; numerical range standardization ensures that different types of coefficients are within the same numerical range; and matrix dimension alignment ensures the correct placement of each part of the data in the matrix. The constructed workstation interaction model matrix comprehensively describes the complex interrelationships between multiple workstations.

[0089] Figure 2 This is a schematic diagram of the interaction coefficient matrix between workstations in an embodiment of this application. Figure 2 As shown, this 4×4 matrix illustrates the mutual influence strength among four workstations (Workstation A, Workstation B, Workstation C, and Workstation D). The diagonal elements are all 1.000, representing the self-influence strength of each workstation; the off-diagonal elements reflect the degree of interaction between different workstations, with larger values ​​indicating stronger mutual influence. The matrix shows a strong mutual influence between workstation A and workstation B (coefficients of 0.856 and 0.734 respectively), relatively low influence coefficients between workstation C and other workstations (range 0.312-0.612), and a moderate level of mutual influence between workstations and all other workstations. The matrix visually displays the weighted distribution of influence strength between workstations after softmax normalization through color variations.

[0090] In one specific embodiment, step S3 includes:

[0091] Based on the workstation interaction model matrix analysis, the resource dependencies and temporal constraints among various cooking tasks are analyzed, and a cooking task dependency graph is constructed using a topological sorting algorithm.

[0092] Critical path analysis was performed on the cooking task dependency graph to identify the set of independent tasks without pre-dependent conditions and the chain of related tasks with execution order constraints.

[0093] The set of independent tasks and the chain of related tasks are input into the out-of-order execution scheduling algorithm, and the genetic algorithm is used to optimize and solve the task rearrangement sequence that breaks the traditional linear execution order.

[0094] Based on the task rearrangement sequence and the power carrying capacity constraints of each workstation, a parallel execution scheme containing task start time, execution duration, and workstation allocation information is generated.

[0095] Specifically, the resource dependency and temporal constraint analysis process is based on data information in the workstation interaction model matrix. The heat transfer coefficient in the matrix reflects the degree of heat interaction between different cooking tasks, and the electromagnetic interference coefficient describes the power constraints between tasks. Resource dependencies refer to the fact that certain cooking tasks must wait for other tasks to complete or reach a specific state before they can begin execution. The determination of dependencies is based on the magnitude of the coefficient values ​​in the workstation interaction model matrix; workstation pairs with coefficient values ​​exceeding a preset threshold have strong dependencies. Temporal constraints include the earliest start time, latest finish time, and time interval requirements between tasks. These constraints are determined based on the physical characteristics of the cooking process; for example, defrosting tasks must be completed before heating tasks, and baking tasks cannot be performed simultaneously with high-humidity tasks. The topology sorting algorithm transforms these dependencies and constraints into a directed acyclic graph structure. Each node in the graph represents a cooking task, and directed edges represent the execution order requirements between tasks. The cooking task dependency graph construction process includes three stages: node creation, edge connection, and cycle detection. Node creation assigns a unique identifier to each cooking task; edge connection establishes directed connections between relevant nodes based on dependencies; and cycle detection ensures that there are no cyclic dependencies in the graph structure.

[0096] Critical path analysis performs a depth-first traversal and computation on the cooking task dependency graph, calculating the earliest and latest start times for each task node. The earliest start time is obtained through forward computation, starting from nodes without preceding tasks and progressively calculating the earliest possible start times for subsequent tasks. The latest start time is obtained through backward computation, starting from nodes without subsequent tasks and calculating the latest allowed start times for preceding tasks. The critical path is the longest path from the start node to the end node, and the task nodes on the critical path are called critical tasks; their execution time directly affects the overall cooking completion time. The independent task set contains task nodes without preceding dependencies and not on the critical path. These tasks have greater flexibility in execution time scheduling and can be adjusted based on resource availability. A related task chain refers to a sequence of tasks with direct or indirect dependencies. The execution order of tasks in the chain is fixed, but the start time of the entire chain can be adjusted. The identification process is implemented using a graph traversal algorithm. The algorithm starts from each node without preceding dependencies and performs a depth-first search, recording all nodes on the search path to obtain the related task chains.

[0097] The out-of-order execution scheduling algorithm receives a set of independent tasks and a chain of related tasks as input data. The core idea of ​​the algorithm is to break the traditional constraint of executing tasks according to the user-input order and rearrange the execution order based on workstation resource conditions and task characteristics. A genetic algorithm, as an optimization tool, encodes the task scheduling problem as a chromosome structure, with each chromosome representing a task execution scheme. Chromosome encoding uses an integer sequence, where each position corresponds to a task, and the position value represents the workstation number to which the task is assigned. The fitness function of the genetic algorithm comprehensively considers task completion time, resource utilization, and constraint violations; chromosomes with higher fitness values ​​represent better scheduling schemes. The algorithm generates new chromosomes through three genetic operations: selection, crossover, and mutation. Selection selects superior individuals based on fitness values ​​to enter the next generation; crossover exchanges gene segments between two parent chromosomes to produce offspring; and mutation randomly changes the values ​​of certain genes in the chromosome to introduce new solutions. After multiple generations of evolution, the algorithm converges to the optimal solution. The chromosome corresponding to the optimal solution is the task rearrangement sequence. The arrangement of tasks in the sequence breaks the original linear constraints, achieving parallelization and optimization of task execution.

[0098] The parallel execution plan generation process combines the task rearrangement sequence with the power capacity constraints of each workstation to generate a specific execution plan. Power capacity constraints include the maximum power output, power adjustment range, and power stability requirements of each workstation; these constraints stem from the physical characteristics and safety specifications of the microwave oven hardware. The plan generation algorithm calculates the power allocation scheme for each workstation at different time periods based on the task order in the task rearrangement sequence, combined with the power requirements and expected execution time of each task. The task start time calculation considers the completion time of preceding tasks and the availability of workstations, ensuring that the corresponding workstation is idle and meets power supply requirements when the task starts. The execution duration is calculated based on the task type, food characteristics, and allocated power. The calculation process considers the mutual influence between workstations; when adjacent workstations are working simultaneously, the execution time is appropriately extended to compensate for heat loss. Workstation allocation information records the specific workstation number assigned to each task, the power setpoint for that workstation during task execution, and the temperature control parameters.

[0099] In one specific embodiment, the process of performing the step of analyzing the resource dependencies and temporal constraints between cooking tasks based on the workstation interaction model matrix can specifically include the following steps:

[0100] The heat transfer coefficient and electromagnetic interference coefficient between each workstation are extracted from the workstation interaction model matrix, and the workstation pairs that have mutual influence are identified by the threshold judgment method.

[0101] A resource competition matrix is ​​established based on workstation pair combinations, where tasks within the same workstation pair are marked as resource conflicts, and tasks within different workstation pairs are marked as resource independence.

[0102] Perform a depth-first traversal of the resource competition relationship matrix, calculate the in-degree and out-degree values ​​of each cooking task, and generate a set of directed edges that reflect the order in which the tasks are executed.

[0103] Each cooking task is treated as a graph node, and the set of directed edges is used as the node connection relationship. The Kahn algorithm is used to perform topological sorting to obtain a cooking task dependency graph that satisfies resource dependency constraints.

[0104] Specifically, the data extraction process of the workstation interaction model matrix locates specific regions within the matrix that store the heat transfer coefficient and electromagnetic interference coefficient. The heat transfer coefficient is located in the upper triangular region of the matrix, reflecting the intensity of heat flow between workstations, while the electromagnetic interference coefficient is located in the lower triangular region, describing the degree of mutual influence of microwave power between workstations. A threshold judgment method filters workstation pairs whose coefficient values ​​exceed a preset standard by setting numerical limits. The threshold judgment calculation process includes a direct numerical comparison between the coefficient value and the threshold. When the heat transfer coefficient is greater than the heat threshold or the electromagnetic interference coefficient is greater than the interference threshold, the corresponding workstation pair is marked as having a mutual influence relationship. Workstation pair combinations refer to workstation pairings with mutual influence relationships. The combination data is stored in binary form, containing the numbers of the two workstations and an influence type identifier between them. The influence type identifier distinguishes between two basic types: heat influence and electromagnetic influence. The extraction process is implemented through a matrix traversal algorithm. The algorithm scans the matrix elements row by row and column by column, recording the positions of elements that meet the threshold conditions, converting the row and column indices into workstation numbers, and obtaining a list of workstation pair combinations.

[0105] The resource competition matrix construction process is based on workstation pair combination data, transforming the physical influence relationships at the workstation level into resource competition relationships at the task level. The core logic of matrix construction is that tasks within the same workstation pair experience resource conflicts due to sharing the same physical influence environment, while tasks within different workstation pairs maintain resource independence due to independent influence environments. Resource conflict marking indicates a resource contention relationship between cooking tasks executed within the same workstation pair. The root cause of the conflict is the constraint of physical space and energy transfer; when two tasks are executed simultaneously at mutually influencing workstations, their cooking effects interfere with each other. Resource independence marking indicates that there is no direct resource competition between tasks executed within different workstation pairs; these tasks can be performed simultaneously without negative impacts. The matrix construction process includes three stages: matrix initialization, conflict relationship filling, and independence relationship confirmation. Matrix initialization creates a two-dimensional array with cooking tasks as row and column indices. Conflict relationship filling marks conflict states at corresponding positions based on workstation pair combination data. Independence relationship confirmation sets unmarked conflict positions as independent. The value of each element in the matrix represents the competition intensity between the corresponding task pairs; conflicting task pairs have positive values, while independent task pairs have zero values.

[0106] The Depth-First Search (DFS) algorithm systematically analyzes the resource contention matrix, calculating the in-degree and out-degree values ​​of each cooking task within its dependency relationships. DFS starts from any task node and searches as deeply as possible along the dependency relationships until it can no longer proceed, at which point it backtracks to the previous node and continues searching other branches. The in-degree calculation counts the number of dependencies pointing to a specific task, reflecting the number of preceding tasks that the task needs to wait for. A task with an in-degree of zero indicates no preceding dependencies and can begin execution immediately. The out-degree calculation counts the number of dependencies from a specific task to other tasks, reflecting the number of subsequent tasks that can be triggered after the task is completed. A task with an out-degree of zero indicates no subsequent dependencies and is the end of the entire task sequence. During the traversal, the algorithm maintains an access marker array to record processed task nodes, preventing duplicate calculations and infinite loops. The process of generating a set of directed edges transforms the dependencies between tasks into an edge structure in graph theory. Each directed edge contains three attributes: the starting task node, the target task node, and the dependency strength. The dependency strength is determined by the value in the resource competition matrix; the larger the value, the stronger the dependency.

[0107] Kahn's algorithm, through topological sorting, organizes cooking tasks and the set of directed edges into a directed acyclic graph (DAG). The algorithm iteratively eliminates all edges in the graph, generating a task execution order that satisfies dependency constraints. Topological sorting refers to linearly sorting all nodes in the DAG, ensuring that the starting node of any directed edge precedes the target node in the sort. The Kahn algorithm's execution process includes three stages: initialization, iterative processing, and result output. The initialization stage calculates the initial in-degree of all task nodes and creates a queue of nodes with zero in-degree. The iterative processing stage repeatedly performs node dequeueing, updates the in-degree of adjacent nodes, and adds new nodes with zero in-degree. The result output stage generates the topologically sorted sequence. During execution, each time a node is taken from the queue of nodes with zero in-degree and added to the sorting result, all outgoing edges of that node are traversed, and the in-degree of the target node is decremented by one. When the in-degree of the target node becomes zero, it is added to the queue of nodes with zero in-degree. After the cooking task dependency graph is constructed, it contains all cooking tasks as graph nodes and directed edges reflecting the execution order. The graph structure clearly expresses the complex dependencies between tasks and satisfies resource dependency constraints.

[0108] In one specific embodiment, step S4 includes:

[0109] Based on the current execution status and historical execution trajectory of each task node in the parallel execution scheme, the temperature change trend and power demand of each workstation in the next time window are predicted by a long short-term memory neural network.

[0110] The temperature change trend and power demand are compared with the preset execution state threshold range. When the predicted value exceeds the threshold range, the execution deviation detection mechanism is triggered.

[0111] Save a status snapshot of the workstation that triggers the deviation detection mechanism, record the power configuration parameters, temperature distribution data and task execution progress at the current moment, and construct checkpoint status data;

[0112] Based on the checkpoint status data, the execution status of the deviation workstation is rolled back to the most recent stable operation historical time node, and the execution sequence arrangement of relevant task nodes in the parallel execution plan is updated.

[0113] Specifically, the speculative execution and state rollback mechanisms in the multi-station microwave oven intelligent cooking combination control method achieve predictive control and anomaly handling of task execution status through four core data processing stages. The Long Short-Term Memory (LSTM) neural network prediction process is based on real-time status data and historical execution trajectory data of each task node in the parallel execution scheme. The neural network receives the current execution status as input data, including the current temperature value, current power output value, and task completion percentage of each station. The historical execution trajectory data records the sequence of changes in the status parameters of each station over a past period. The trajectory data is arranged chronologically to form a time series reflecting the station's operating mode. The LTM neural network processes the time series data through its unique forget gate, input gate, and output gate mechanisms. The forget gate determines which information to discard from long-term memory, the input gate determines which new information to store in long-term memory, and the output gate determines which information to output based on the current input and long-term memory. The network's computation process includes three stages: hidden state update, memory unit update, and output computation. Hidden state update combines the current input and the hidden state from the previous time step to calculate the new hidden state. Memory unit update updates the long-term memory content through the coordinated action of the forget gate and the input gate. Output computation generates prediction results based on the updated hidden state and memory units. Temperature trend prediction outputs the expected temperature value for each station in the next time window, and power demand prediction outputs the power value required for each station to maintain the target cooking state. The prediction results are represented in the form of a numerical vector, where each element corresponds to the prediction parameter for a specific station.

[0114] The execution state threshold comparison process compares the predicted values ​​output by the neural network with the preset safe operating range. The execution state threshold range comprises two core components: a temperature threshold range and a power threshold range. The temperature threshold range is set according to the process requirements of different cooking tasks, while the power threshold range is determined based on the safe operating limits of the microwave oven hardware. The threshold range is defined as an upper and lower limit, with the upper limit representing the maximum permissible value and the lower limit representing the minimum permissible value. The numerical comparison process is implemented by determining the magnitude of these values. When the predicted temperature change trend exceeds the temperature threshold range or the predicted power demand exceeds the power threshold range, the comparison result triggers the execution deviation detection mechanism. The execution deviation detection mechanism is an abnormal state identification and response mechanism. Once activated, the corresponding workstation is immediately marked as having a deviation state, and subsequent state saving and rollback processes are initiated. The deviation detection judgment logic includes two cases: single-parameter deviation and multi-parameter deviation. Single-parameter deviation refers to only the temperature or power parameter exceeding the threshold range, while multi-parameter deviation refers to both the temperature and power parameters exceeding their respective threshold ranges simultaneously. The detection mechanism also considers the duration of the deviation. Short-term deviations and long-term deviations are handled with different strategies. Short-term deviations are corrected by adjusting parameters, while long-term deviations trigger a state rollback operation.

[0115] The status snapshot saving process comprehensively records and stores the status information of the workstation that triggers deviation detection. A status snapshot is a complete record of the workstation's operating status at a specific moment, and the snapshot content includes three core information categories: power configuration parameters, temperature distribution data, and task execution progress. Power configuration parameters record the workstation's current power output setting, power adjustment mode, and power control strategy. Parameter data is stored in key-value pairs, where the key represents the parameter name and the value represents the parameter's specific value or status. Temperature distribution data records the temperature values ​​at each measurement point within the workstation. The data is organized in a two-dimensional array, with the array's rows and columns corresponding to the spatial coordinates of the temperature measurement points. Task execution progress records the current cooking task's completion percentage, remaining execution time, and elapsed time. Progress data is represented as a combination of numerical values ​​and timestamps. The checkpoint status data construction process integrates these three types of information according to a unified data format. The integrated data structure includes four fields: timestamp, workstation number, status type, and specific numerical value. Status data saving adopts a first-in, first-out (FIFO) caching strategy. New checkpoint data is added to the end of the cache queue, and old data at the front of the queue is deleted when cache space is insufficient. The cache capacity is dynamically adjusted based on system memory resources and the importance of historical data. Important stable state data is retained for a longer time, while abnormal state data is retained for a relatively shorter time.

[0116] The status rollback operation restores the workstations that deviated from their previous state to their most recent stable operating state based on checkpoint status data. Status rollback refers to resetting the current operating parameters of the workstation to the values ​​of its historical stable state. The rollback process includes three stages: parameter restoration, status verification, and execution plan update. The parameter restoration process extracts the specific parameter values ​​of the target historical state from the checkpoint status data and resets these values ​​to the corresponding workstation control unit. Restoration operations include power output adjustment, temperature control target correction, and execution timing reset. The status verification process checks whether the workstation state after rollback meets expectations. Verification methods include parameter value verification, status consistency checks, and functional normality tests. The execution plan update process adjusts the task node information related to the rolled-back workstation in the parallel execution plan. Updates include task restart times, corrected execution durations, and adjusted workstation allocation relationships. The execution timing update considers the impact of the rollback operation on other workstations and tasks. When a workstation experiences status rollback, subsequent tasks dependent on that workstation need to have their start times delayed accordingly, and parallel tasks competing for resources with that workstation need to have their execution priorities reassessed. The update algorithm uses dependency graph analysis to determine the scope of affected tasks, recalculates the optimal execution sequence of these tasks, and ensures the coordination and effectiveness of the overall cooking plan.

[0117] Figure 3 This is a schematic diagram illustrating the dynamic power allocation process of a multi-station microwave oven in an embodiment of this application. Figure 3As shown in the figure, the horizontal axis represents cooking time, and the vertical axis represents the power allocation value of each workstation. The solid line represents the power allocation curve of workstation A, the dashed line represents the power allocation curve of workstation B, the dotted-dashed line represents the power allocation curve of workstation C, and the dotted line represents the power allocation curve of workstation D. The figure shows that workstations A and B continuously allocate power throughout the cooking process, reflecting the characteristics of continuous cooking tasks; workstation C's power drops to zero after performing the defrosting task in the first 15 minutes, reflecting resource release after task completion; workstation D only begins allocating power after the 10th minute, reflecting the delayed start mechanism in the out-of-order execution strategy. The power allocation curves of each workstation exhibit a dynamic trend, reflecting the real-time adjustment process based on resource utilization efficiency indicators and power demand priority queues, verifying that the multi-workstation collaborative control command can dynamically optimize power allocation according to the actual execution progress and resource demand of each workstation.

[0118] In one specific embodiment, step S5 includes:

[0119] Real-time monitoring of the completion rate and remaining execution time of cooking tasks at each workstation, combined with the current power consumption, calculates the resource utilization efficiency index of each workstation.

[0120] A power demand priority queue is established based on resource utilization efficiency indicators, and high-efficiency, low-power workstations are set as priority allocation objects, while low-efficiency, high-power workstations are set as restricted allocation objects.

[0121] Based on the power demand priority queue and the total power capacity limit of the system, the power allocation quota and time slice length of each workstation are recalculated using the shortest job first scheduling algorithm;

[0122] The recalculated power allocation quota and time slice length are encapsulated into a multi-station collaborative control instruction that includes station number, execution instructions, and timing control parameters.

[0123] Specifically, the real-time monitoring process continuously tracks the cooking task execution status of each workstation. Monitoring data includes three key parameters: cooking task completion rate, remaining execution time, and current power consumption. The cooking task completion rate is calculated by comparing the completed cooking steps to the total cooking steps, expressed as a percentage, reflecting the actual progress of the task. The remaining execution time is calculated by subtracting the time already consumed from the estimated total execution time, recorded in minutes to dynamically reflect changes in the task's time requirements. Current power consumption is monitored by real-time collection of instantaneous power consumption values ​​at each workstation using power meters, expressed in watts, recording the actual energy usage at each workstation. The resource utilization efficiency index calculation process correlates task completion rate with power consumption. The calculation method divides the task completion rate by the corresponding cumulative power consumption value to obtain the task completion level per unit of power consumption. A higher efficiency index value indicates that the workstation completes a greater amount of work per unit of power consumption, demonstrating more efficient resource utilization. The calculation process also takes into account the impact of time factors. By dividing the efficiency index by the time consumed, the time-corrected efficiency value is obtained. The corrected index more accurately reflects the comprehensive resource utilization level of each workstation.

[0124] The power demand priority queue establishment process sorts and categorizes elements based on their resource utilization efficiency indicators for each workstation. A priority queue is a special data structure where elements are sorted according to their priority, with higher-priority elements at the front and lower-priority elements at the back. The queue establishment process compares the efficiency indicators of all workstations, identifies the workstations with the highest and lowest efficiency indicators, and then sorts all workstations in descending order of efficiency indicator values. High-efficiency, low-power workstations, characterized by high efficiency indicators and relatively low current power consumption, are prioritized for resource allocation in the queue. Low-efficiency, high-power workstations, characterized by low efficiency indicators and relatively high current power consumption, are subject to limited allocation, with their resource acquisition restricted. The dynamic nature of the priority queue is reflected in the real-time adjustment of workstation priorities as efficiency indicators change. When the efficiency indicator of a workstation changes, the queue automatically reorders to reflect the latest priority relationships. The queue management algorithm is implemented using the heap sort data structure, which ensures that the insertion, deletion and search operations of the queue have high execution efficiency.

[0125] The Shortest Job First (SJF) scheduling algorithm recalculates by comprehensively considering two constraints: the power demand priority queue and the system's total power capacity limit. SJF is a classic resource scheduling strategy that prioritizes allocating resources to tasks with the shortest execution time, reducing average waiting time and overall completion time. The core idea is to use remaining execution time as a measure of job length, with jobs having shorter remaining time receiving higher resource allocation priority. The power allocation quota calculation process determines the system's total power capacity limit and then allocates corresponding power shares based on each job's position in the priority queue and its remaining execution time. Quota allocation uses a proportional allocation strategy, with higher-priority jobs receiving larger power quotas and lower-priority jobs receiving smaller power quotas. The time slice length calculation process determines the execution time of each job within a scheduling cycle based on its remaining execution time and task complexity. The time slice length is directly proportional to the job's priority; higher-priority jobs receive longer time slices. During the algorithm execution process, a feasibility check is also required to ensure that the total power quota of all workstations does not exceed the total power capacity of the system. When the total quota exceeds the limit, the algorithm will reduce the quota of low-priority workstations according to priority order until the constraint conditions are met.

[0126] The multi-station collaborative control instruction encapsulation process integrates the recalculated power allocation quotas and time slice lengths into a standardized control instruction format. Control instruction encapsulation is the process of converting abstract calculation results into concrete executable instructions. The encapsulated instruction contains three core components: station number, execution instruction, and timing control parameters. The station number identifies the target execution unit of the instruction, using a unique identifier to ensure accurate transmission and execution. The execution instruction contains specific power setpoints, operating mode selections, and control strategy parameters. The instruction content is organized in a structured data format for easy parsing and execution by the control unit. The timing control parameters specify the instruction's execution time, duration, and synchronization requirements, ensuring that each station operates according to a coordinated timing sequence. The instruction encapsulation process also includes instruction verification and error checking. Verification checks the correctness of the instruction format and the rationality of the parameters, while error checking identifies and handles abnormal data in the instruction. The encapsulated control instructions are sent to the control units of each station via a communication interface. After receiving the instructions, the control units parse the instruction content and execute the corresponding control operations. The multi-station collaborative feature is reflected in the fact that the control commands of all stations are coordinated in timing, ensuring that the operation of each station will not conflict or interfere with each other. The collaborative control mechanism achieves precise coordination between stations through unified clock synchronization and status monitoring.

[0127] The above describes the intelligent cooking combination control method for multi-station microwave ovens in the embodiments of this application. The following describes the intelligent cooking combination control system for multi-station microwave ovens in the embodiments of this application. Please refer to [link / reference]. Figure 4One embodiment of the multi-station microwave oven intelligent cooking combination control system in this application includes:

[0128] The fusion module is used to collect temperature distribution data, moisture content data and weight change data of each workstation, and to establish a multi-workstation comprehensive state vector after data fusion processing.

[0129] The input module is used to input the multi-station integrated state vector into the graph neural network to identify the heat transfer relationship and power interaction relationship between each station and establish a station interaction model.

[0130] The startup module is used to analyze the execution constraint relationship between cooking tasks based on the workstation interaction model, rearrange the startup timing of each workstation task using a randomized execution strategy, and determine the parallel execution scheme.

[0131] The calculation module is used to pre-calculate the execution state of the next stage for each task node in the parallel execution scheme, and to fall back to the previous stable state when the actual execution result does not match the pre-calculated result.

[0132] The generation module is used to dynamically adjust the power allocation ratio and time slice allocation strategy according to the actual execution progress and resource requirements of each workstation, and generate multi-workstation collaborative control instructions.

[0133] above Figure 4 The multi-station microwave oven intelligent cooking combination control system in this embodiment of the invention is described in detail from the perspective of modular functional entities. The multi-station microwave oven intelligent cooking combination control device in this embodiment of the invention is described in detail from the perspective of hardware processing.

[0134] Reference Figure 5 This invention also provides a multi-station microwave oven intelligent cooking combination control device, which can be a server, and its internal structure can be as follows: Figure 5 As shown, this multi-station microwave oven intelligent cooking combination control device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor, designed as a computer, provides computing and control capabilities. The memory of the multi-station microwave oven intelligent cooking combination control device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the multi-station microwave oven intelligent cooking combination control device stores the data corresponding to this embodiment. The network interface of the multi-station microwave oven intelligent cooking combination control device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0135] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the multi-station microwave oven intelligent cooking combination control device to which the present invention is applied.

[0136] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the multi-station microwave oven intelligent cooking combination control method.

[0137] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0138] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a multi-station microwave oven intelligent cooking combination control device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0139] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent cooking combination control of a multi-station microwave oven, characterized in that, The method includes: Step S1: Collect temperature distribution data, moisture content data and weight change data of each workstation, and establish a multi-workstation comprehensive state vector after data fusion processing; Step S2: Input the multi-station integrated state vector into the graph neural network to identify the heat transfer relationship and power interaction relationship between each station, and establish a station interaction model; Step S3: Analyze the execution constraint relationship between cooking tasks based on the workstation interaction model, rearrange the start timing of each workstation task using a random execution strategy, and determine the parallel execution scheme; Step S4: Pre-calculate the execution state of the next stage for each task node in the parallel execution scheme. When the actual execution result does not match the pre-calculated result, roll back to the previous stable state. Step S5: Based on the actual execution progress and resource requirements of each workstation, dynamically adjust the power allocation ratio and time slice allocation strategy to generate multi-workstation collaborative control instructions.

2. The intelligent cooking combination control method for multi-station microwave ovens according to claim 1, characterized in that, Step S1 includes: A two-dimensional temperature distribution matrix is ​​obtained by scanning and detecting the food surface in each workstation using an infrared temperature sensor array. The frequency of food in each workstation is scanned using a dielectric property detection device to obtain a loss factor sequence that reflects the change in the internal water content of the food. High-precision weight sensors are installed at the bottom of each workstation for continuous weighing and monitoring to obtain data on the time-series changes in food weight. The two-dimensional temperature distribution matrix, the loss factor sequence, and the weight time-series change data are subjected to Kalman filtering for noise reduction to obtain a multi-station integrated state vector.

3. The intelligent cooking combination control method for multi-station microwave ovens according to claim 1, characterized in that, Step S2 includes: The multi-station integrated state vector is constructed as a dynamic weighted graph structure, where each station is a graph node and the physical distance and power level difference between stations are edge weights. The dynamic weighted graph structure is subjected to multi-layer propagation computation of graph convolutional neural network. By convolving the adjacency matrix and the node feature matrix, a node embedding vector containing information about the mutual influence between workstations is obtained. The heat transfer coefficient and electromagnetic interference coefficient between each workstation are calculated based on the node embedding vector, and the influence intensity weight distribution between workstations is generated by softmax normalization. The heat transfer coefficient, the electromagnetic interference coefficient, and the influence intensity weight distribution are combined to construct the workstation interaction model matrix.

4. The intelligent cooking combination control method for multi-station microwave ovens according to claim 1, characterized in that, Step S3 includes: Based on the workstation interaction model matrix analysis, the resource dependencies and temporal constraints among each cooking task are analyzed, and a cooking task dependency graph is constructed using a topological sorting algorithm. Critical path analysis was performed on the cooking task dependency graph to identify the set of independent tasks without pre-dependent conditions and the chain of related tasks with execution order constraints. The set of independent tasks and the chain of related tasks are input into the out-of-order execution scheduling algorithm, and the task rearrangement sequence that breaks the traditional linear execution order is optimized and solved by the genetic algorithm. Based on the task rearrangement sequence and the power carrying capacity constraints of each workstation, a parallel execution scheme containing task start time, execution duration, and workstation allocation information is generated.

5. The intelligent cooking combination control method for multi-station microwave ovens according to claim 4, characterized in that, The process involves analyzing the resource dependencies and temporal constraints among cooking tasks based on the workstation interaction model matrix, and constructing a cooking task dependency graph using a topological sorting algorithm, including: The heat transfer coefficient and electromagnetic interference coefficient between each workstation are extracted from the workstation interaction model matrix, and the workstation pairs that have mutual influence are identified by the threshold judgment method. Based on the workstation pair combination, a resource competition relationship matrix is ​​established between cooking tasks, wherein tasks within the same workstation pair are marked as resource conflicts, and tasks within different workstation pairs are marked as resource independence. A depth-first traversal is performed on the resource competition relationship matrix to calculate the in-degree and out-degree values ​​of each cooking task, and a set of directed edges reflecting the order of task execution is generated. Each cooking task is treated as a graph node, and the set of directed edges is used as the node connection relationship. The Kahn algorithm is used to perform topological sorting to obtain a cooking task dependency graph that satisfies resource dependency constraints.

6. The intelligent cooking combination control method for multi-station microwave ovens according to claim 1, characterized in that, Step S4 includes: Based on the current execution status and historical execution trajectory of each task node in the parallel execution scheme, the temperature change trend and power demand of each workstation in the next time window are predicted by a long short-term memory neural network. The temperature change trend and the power demand are compared with the preset execution state threshold range. When the predicted value exceeds the threshold range, the execution deviation detection mechanism is triggered. A status snapshot is saved for the workstation that triggers the execution deviation detection mechanism, recording the power configuration parameters, temperature distribution data and task execution progress at the current moment, and constructing checkpoint status data; Based on the checkpoint status data, the execution status of the deviation station is rolled back to the most recent stable operation historical time node, and the execution sequence arrangement of the relevant task nodes in the parallel execution scheme is updated.

7. The intelligent cooking combination control method for multi-station microwave ovens according to claim 1, characterized in that, Step S5 includes: Real-time monitoring of the completion rate and remaining execution time of cooking tasks at each workstation, combined with the current power consumption, calculates the resource utilization efficiency index of each workstation. Based on the resource utilization efficiency index, a power demand priority queue is established, and high-efficiency, low-power workstations are set as priority allocation objects, while low-efficiency, high-power workstations are set as restricted allocation objects. Based on the power demand priority queue and the total power capacity limit of the system, the power allocation quota and time slice length of each workstation are recalculated using the shortest job first scheduling algorithm; The recalculated power allocation quota and time slice length are encapsulated into a multi-station collaborative control instruction containing station number, execution instructions, and timing control parameters.

8. A multi-station microwave oven intelligent cooking combination control system, characterized in that, For implementing the multi-station microwave oven intelligent cooking combination control method as described in any one of claims 1-7, the multi-station microwave oven intelligent cooking combination control system comprises: The fusion module is used to collect temperature distribution data, moisture content data and weight change data of each workstation, and to establish a multi-workstation comprehensive state vector after data fusion processing. The input module is used to input the multi-station integrated state vector into the graph neural network to identify the heat transfer relationship and power interaction relationship between each station and establish a station interaction model. The startup module is used to analyze the execution constraint relationship between cooking tasks based on the workstation interaction model, rearrange the startup timing of each workstation task using a randomized execution strategy, and determine the parallel execution scheme. The calculation module is used to pre-calculate the execution state of the next stage for each task node in the parallel execution scheme, and to fall back to the previous stable state when the actual execution result does not match the pre-calculated result. The generation module is used to dynamically adjust the power allocation ratio and time slice allocation strategy according to the actual execution progress and resource requirements of each workstation, and generate multi-workstation collaborative control instructions.

9. A multi-station microwave oven intelligent cooking combination control device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the multi-station microwave oven intelligent cooking combination control method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the multi-station microwave oven intelligent cooking combination control method as described in any one of claims 1 to 7.

Images