Cooking control method

Abstract

The invention relates to a cooking control method, and the method comprises the steps: obtaining and quantifying the cooking intention of a user, and setting a clear control target; and a'prediction-compensation-correction 'closed-loop control framework is constructed by generating a reference curve, introducing feed-forward compensation based on historical efficiency and combining feedback adjustment of real-time monitoring. According to the architecture, the control system not only depends on a static model, but also can actively adapt to dynamic attenuation of equipment performance, so that the core defect of inconsistent cooking effects caused by equipment aging in the background technology is directly overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cooking equipment, in particular to a cooking control method. BACKGROUND

[0002] Modern household cooking equipment, especially micro-steam ovens, is committed to reducing user operation complexity and improving the consistency of cooking results through intelligentization. The core goal is that the user only needs to specify the desired cooking result (such as "steak - three minutes"), and the equipment can automatically complete the whole process from recognition to cooking, and finally stably output the expected dish. This requires the system to have the ability to convert vague user intentions into precise control instructions and overcome various uncertainties in the cooking process.

[0003] Currently, the common technical solutions to achieve the above goal mainly rely on two methods. The first is fixed program control based on a preset database. A large number of standard temperature-time curves for specific food types and weights are built into the device. After the user selects, the device executes the heating according to the established program. The second is to introduce limited feedback on the basis of the fixed program, for example, through temperature probes or cameras to monitor the food state at the end of cooking, and stop heating when a certain threshold is reached. However, both of these two schemes have significant shortcomings. Fixed program control is completely based on the ideal performance model of the device when it leaves the factory, and its fundamental defect is that it ignores the inevitable decay of the core performance of the cooking equipment (such as heating pipe efficiency, cavity heat exchange efficiency) over time. This directly leads to a serious consequence: executing the same program, a new device may reach perfect three minutes in 10 minutes, while an old device after one year needs 15 minutes, and the internal doneness may have been overcooked, that is, "device aging leads to cooking effect drift". Simple end feedback control can correct the end point of a single cooking, but it cannot systematically compensate for the entire heating process lasting for several minutes to several tens of minutes, and cannot correct the accumulated deviation of heat due to performance decay in the early and middle stages of cooking. Therefore, how to build an intelligent cooking control system that can perceive and compensate for the performance decay of the device itself, so that it can stably output consistent cooking results throughout its life cycle, has become a key and unsolved technical problem in the field. SUMMARY

[0004] Therefore, it is necessary to provide a cooking control method in view of the above-mentioned problems of the traditional cooking control method.

[0005] The present application provides a cooking control method, comprising: obtaining a cooking intention input by a user; converting the cooking intention into a target physical parameter; collecting food material information to identify the food material category and obtain the physical characteristics of the food material; generating a reference cooking curve based on the food category, the physical characteristics of the food, and the target physical parameter; performing feedforward compensation on the reference cooking curve based on historical cooking data of the cooking device to obtain a feedforward control curve; during the cooking process, performing feedback adjustment on the feedforward control curve based on real-time monitoring data of the multi-source sensor to generate a final control instruction for controlling the cooking device; terminating the cooking when the target physical parameter is reached.

[0006] The present application relates to a cooking control method, which sets a clear control target by obtaining and quantifying the user's cooking intention, and further constructs a "prediction-compensation-correction" closed-loop control architecture by generating a reference curve, introducing feedforward compensation based on historical performance, and combining real-time monitoring feedback adjustment. This architecture makes the control system not only rely on static models, but also actively adapt to the dynamic decay of device performance, thereby directly solving the core defect of inconsistent cooking results caused by device aging in the background technology. BRIEF DESCRIPTION OF DRAWINGS

[0007] Figure 1 A method flowchart of a cooking control method provided by an embodiment of the present application. DETAILED DESCRIPTION

[0008] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0009] As shown in Figure 1 The present application provides a cooking control method. It should be noted that the cooking control method provided by the present application is applied to any type of cooking device.

[0010] In addition, the cooking control method provided by the present application does not limit its execution subject. Alternatively, the execution subject of the cooking control method provided by the present application can be a cooking device. The cooking device is built-in with a cooking system (hereinafter referred to as "system"). Specifically, the execution subject of the cooking control method provided by the present application can be one or more processors in the cooking device.

[0011] In an embodiment of the present application, the cooking control method comprises: S100, obtaining the user's input cooking intention; S200, converting the cooking intention into a target physical parameter; S300, collecting food information to identify the food category and obtain the physical characteristics of the food; S400, Generate a benchmark cooking curve based on the food category, the physical characteristics of the food, and the target physical parameters; S500, based on the historical cooking data of the cooking equipment, the reference cooking curve is fed forward compensation to obtain the feedforward control curve; S600, during the cooking process, the feedforward control curve is adjusted based on real-time monitoring data from multiple sources of sensors to generate the final control command for controlling the cooking equipment. S700, cooking is terminated when the target physical parameters are reached.

[0012] Specifically, this embodiment defines the core closed-loop process of this method. Take, for example, a user wanting to cook a rare steak. When the user selects the "Steak - Rare" button, it's equivalent to the user inputting a cooking intention. Of course, pressing a function button on the cooking device is only one possible way for the user to input a cooking intention. Users can also input cooking intentions in other ways, such as inputting remote computer commands through a mobile app.

[0013] The system first parses the data into target physical parameters, such as a core temperature target value T_target = 55℃. Next, it acquires images and weight information of the steak using a camera and weighing sensors, identifies it as a "sirloin steak," and calculates its critical thermal thickness D_eff (e.g., 2.5 cm). Then, based on the typical physical properties of the "sirloin steak," D_eff, and T_target, the system uses a simplified heat transfer model for forward prediction and solves for a series of future cavity temperature setpoints using a model predictive control algorithm, forming an initial baseline cooking curve. Then, the system retrieves historical data on similar steaks cooked in the past, analyzes the average attenuation of the cooking equipment's heating efficiency during specific cooking stages (e.g., the first 3 minutes), calculates the attenuation coefficient (e.g., 0.2), and superimposes the corresponding compensation temperature offset (e.g., +10℃) onto the baseline cooking curve to form a feedforward control curve. After cooking begins, the system monitors the cavity temperature and the surface condition of the steak in real time using temperature sensors and a camera. It then uses a state estimation algorithm (there are many methods, which are not the focus of this application; for example, an extended Kalman filter could be used) to infer the core temperature of the steak in real time. Alternatively, a thermometer can be inserted into the core of the steak to obtain the core temperature. This core temperature is compared with the target temperature to identify the deviation, and the current instantaneous heating efficiency is evaluated. A comprehensive calculation of the feedback adjustment amount is then performed, dynamically fine-tuning the feedforward control curve to generate the final control command to drive the heating elements, such as the heating tube. When the system's estimated core temperature stabilizes at 55±1℃ and the surface color meets the "rare" characteristic, cooking automatically terminates.

[0014] In this embodiment, a clear control objective is set by acquiring and quantifying the user's cooking intention. Then, a closed-loop control architecture of "prediction-compensation-correction" is constructed by generating a baseline curve, introducing feedforward compensation based on historical performance, and combining real-time monitoring feedback adjustment. This architecture enables the control system to not only rely on a static model but also actively adapt to the dynamic degradation of equipment performance, thereby directly solving the core defect in the prior art of inconsistent cooking results caused by equipment aging.

[0015] In one embodiment of this application, S300 includes, namely, collecting food information to identify food categories and obtain the physical characteristics of food, including: S310 simultaneously acquires visual images of food from multiple spatial perspectives to obtain multi-view visual images; S320, Identify the specific category of the food based on the multi-view visual image; S330, Reconstruct a three-dimensional model of the food based on the multi-view visual images, and determine the key thermal thickness of the food based on the three-dimensional model.

[0016] Specifically, this embodiment refines the information acquisition and feature extraction steps. For example, after the user places the steak in and closes the door, multiple cameras located on the top and side walls of the oven cavity are simultaneously triggered, acquiring RGB images of the steak from top and side angles. This collection of RGB images constitutes a multi-view visual image. After preprocessing, these RGB images are fed into a trained convolutional neural network for multi-view feature fusion and classification, initially identifying it as belonging to the "steak" category. To further determine the specific part, the system calculates the Mahalanobis distance between its fused feature vector and template feature vectors such as "sirloin" and "tenderloin" in the database, ultimately classifying it as "sirloin steak." Simultaneously, using these multi-view visual images, a dense 3D point cloud of the steak is reconstructed through a stereo vision algorithm. The system calculates the minimum bounding cuboid of this point cloud and defines the minimum size of this cuboid (in the case of non-irregularly shaped ingredients, this is the thickness of the steak; this application does not consider the case of irregularly shaped ingredients) as the critical thermal thickness D_eff. This parameter is crucial because it dominates the time required for heat to transfer from the surface to the center.

[0017] In this embodiment, multi-view synchronous acquisition is adopted to overcome the blind spots in recognition caused by the posture or occlusion of food in a single view. The key thermal thickness is determined by three-dimensional reconstruction, which provides the core geometric parameters that determine the heating time for the subsequent thermodynamic control model. This allows the control benchmark to be established on actual physical properties rather than estimated weight, thus improving the accuracy of the model.

[0018] In one embodiment of this application, S320 includes, namely, identifying the specific category of food based on the multi-view visual image, including: S321, Extract the fusion features of the multi-view visual images; S322, sequentially calculate the Mahalanobis distance between the fused feature and the image features of each pre-stored food category template; S323, the food category of the pre-stored food category template corresponding to the minimum Mahalanobis distance is determined as the specific category of the food for multi-view visual image recognition.

[0019] Specifically, this embodiment defines a specific category recognition algorithm. The system inputs multi-view visual images into a CNN (Convolutional Neural Network) with shared weights, extracting high-dimensional feature vectors for each. Subsequently, an attention fusion layer assigns weights to the features of each viewpoint and performs a weighted summation to obtain a global fusion feature vector F. The fusion feature vector F is the fusion feature of the multi-view visual images in S321. When identifying the specific category of food, instead of simply calculating the Euclidean distance between the fusion feature vector F and the image features E of the pre-stored food category template, the Mahalanobis distance is calculated. The Mahalanobis distance formula considers the covariance matrix Σ of the features of samples of the same category in the database, and its calculation formula is as follows: .

[0020] Where μ is the average feature vector of image features for a pre-stored food category template in the database. The distance metric of μ considers the correlation and scale of change between the various dimensions of the features. For "sirloin steak" and "filet mignon," the distribution difference of their fat texture (one of the feature dimensions) is captured by the covariance matrix. Dm is the Mahalanobis distance. Σ is the covariance matrix of image features for this food category template in the database. -1 is a superscript, indicating that the matrix Σ is inverted. If Σ is not invertible, a small regularization term is added in practice to make it invertible. F is the fused feature vector extracted from the current multi-view visual image. T is a superscript, indicating that the vector (F-μ) is transposed, converting it from a column vector to a row vector for matrix multiplication.

[0021] Mahalanobis distance is essentially a custom-designed, scaled flexible ruler (i.e., Σ) for different food categories and different feature dimensions. -1 This measuring stick will flexibly measure the differences based on the natural distribution of the characteristics of this type of food (which dimensions fluctuate the most, which dimensions are related). For example, even if F and μ differ greatly in a certain dimension, if this dimension naturally fluctuates greatly (has a large variance) in this type of food, then Σ -1 The system will consider this difference to be relatively small, thus assigning it a smaller weight. This effectively reduces misjudgments and improves the robustness of identification.

[0022] The system calculates the Mahalanobis distance between F and all candidate pre-stored food category templates such as "sirloin" and "tenderloin," and uses the food category with the smallest Mahalanobis distance (e.g., "sirloin steak") as the final recognition result. This method is more robust to changes in lighting and local occlusion.

[0023] In this embodiment, the category determination is performed by calculating the Mahalanobis distance between the fused features and the template features. This distance metric considers the variance and covariance of each dimension of the features. Compared with Euclidean distance, it is not sensitive to changes in feature scale and correlation. Therefore, in real-world scenarios with complex lighting and morphological changes, it can more robustly distinguish similar food subcategories and improve recognition reliability.

[0024] In one embodiment of this application, S400 includes generating a baseline cooking curve based on the ingredient category, the physical characteristics of the ingredient, and the target physical parameters, comprising: S410, Construct a simplified heat transfer model based on the key thermal thickness of the food and the typical thermal properties of the food category; S420, using model predictive control, takes the target physical parameters as the control target, and calculates the future cavity temperature setpoint sequence through rolling time-domain optimization based on the simplified heat transfer model, and uses the future cavity temperature setpoint sequence as the reference cooking curve.

[0025] Specifically, this embodiment defines a method for generating a baseline cooking curve. Based on the identified "sirloin steak" tag, the system retrieves its typical specific heat capacity c_p, thermal conductivity k, and density ρ from a database. Combined with the measured critical thermal thickness D_eff, a simplified first-order lumped-parameter heat transfer model is constructed. The core of this model is a thermal time constant τ=(ρ×c_p×(D_eff / 2)^2) / (2×k), which describes the inertia of the steak's center temperature in response to changes in the cavity temperature.

[0026] τ is the thermal time constant, measured in seconds (s). Its physical meaning is an inertial measure of how quickly the core temperature of the food responds to changes in the cavity temperature. The larger the τ value, the slower the food heats up. ρ is the density of the food, measured in kilograms per cubic meter (kg / m³). c_p is the specific heat capacity of the food, measured in joules per kilogram per Kelvin (J / (kg·K)). D_eff is the critical thermal thickness of the food, measured in meters (m). D_eff is typically the minimum dimension of the 3D model. k is the thermal conductivity of the food, measured in watts per meter per Kelvin (W / (m·K)).

[0027] The model predicts the relationship between the next core temperature T_core(k+1) and the current core temperature T_core(k) and the current cavity set temperature T_cavity_set(k): T_core(k+1) = T_core(k) + (Ts / τ) × [T_cavity_set(k) - T_core(k)], where Ts is the control period. The model predictive control method operates on this model: In each control period k, the algorithm starts with the currently estimated core temperature and, over a future prediction time domain (e.g., the next 10 periods), continuously optimizes and solves for a series of T_cavity_sets, ensuring that the predicted core temperature is as close as possible to T_target = 55℃ at the end of the time domain, while preventing excessively drastic temperature changes. The optimization solution is typically transformed into a constrained quadratic programming problem. Finally, only the first value T_cavity_set(k) in the optimization sequence is output as the current baseline setpoint, and prediction and optimization are repeated in the next period. The resulting series of T_cavity_sets constitutes the dynamic baseline cooking curve.

[0028] In this embodiment, by constructing a simplified lumped parameter heat transfer model, the distributed parameter system is transformed into a computational model suitable for real-time control. Model predictive control is used for rolling time-domain optimization, which can proactively solve for the optimal temperature sequence that satisfies the final temperature target at each control moment. This achieves global dynamic planning of the cooking process and avoids the response lag and overshoot that may exist in local PID control.

[0029] In one embodiment of this application, S500 includes performing feedforward compensation on the reference cooking curve based on historical cooking data from the cooking equipment to obtain a feedforward control curve, including: S510 calculates the attenuation coefficient reflecting the heating efficiency of the cooking equipment within multiple preset cooking time periods based on historical cooking data. S520, Based on the attenuation coefficient, determine the temperature compensation amount for the cooking curve at each time period; S530, the reference cooking curve is corrected using the temperature compensation amount to obtain the feedforward control curve.

[0030] Specifically, this embodiment defines a feedforward compensation mechanism. The system divides the entire cooking process (e.g., estimated 10 minutes) into multiple consecutive time periods, such as stage A for the first 0-3 minutes, stage B for 3-6 minutes, and so on. For the "sirloin steak" to be cooked, the system retrieves the most recent N records of cooking the same type of steak from the historical database. For each time period j, the system calculates the ratio ξ of the actual average temperature rise rate of the cavity during that time period to the set temperature rise rate at that time. The actual average temperature rise rate of the cavity can be calculated based on pre-stored cavity temperature sensor data. ξ less than 1 indicates that the actual temperature rise is slower than expected, resulting in performance degradation. The average performance ratio ξ_avg_j is obtained by weighting all historical ξ values ​​within that time period (the more recent the record, the higher the weight), and the degradation coefficient K_j = 1 - ξ_avg_j. The larger K_j is, the more severe the performance loss during that time period. Then, the system uses a preset mapping relationship, such as the linear function ΔT_j=α×K_j, where α is the learned gain coefficient, to convert K_j into a specific temperature compensation amount ΔT_j. Before cooking begins, ΔT_j is directly added to the cavity temperature setpoint for the corresponding time period j of the baseline cooking curve, thereby generating a feedforward control curve with compensation preset at the starting point.

[0031] In this embodiment, by analyzing historical data to calculate the performance degradation coefficient of the equipment at each stage of cooking, and using this to perform feedforward compensation on the baseline curve, the correction amount for known equipment performance degradation can be preset before cooking begins. This is equivalent to establishing a "baseline" for the control system that self-adjusts according to the equipment state, which is a key step in combating performance degradation from the source.

[0032] In one embodiment of this application, S510 includes calculating the attenuation coefficient reflecting the heating efficiency of the cooking equipment over multiple preset cooking time periods based on historical cooking data, including: S511, for each preset cooking time period, calculates the ratio of the actual average temperature rise rate of the cavity to the set temperature rise rate during multiple historical cooking processes. S512, the attenuation coefficient corresponding to the preset cooking time period is calculated based on the ratio.

[0033] Specifically, this embodiment describes the specific calculation method of the attenuation coefficient K. Taking stage A (0-3 minutes) as an example, the system retrieves data from 5 historical cooking sessions of "sirloin steak". For the first historical record, the system reads the actual measured values ​​of cavity temperature T_act(0) and T_act(3) at 0 minutes and 3 minutes, and calculates the actual average temperature rise rate v_act1=(T_act(3)-T_act(0)) / 3. At the same time, it reads the set temperature rise rate v_set1 of this cooking session in stage A (determined by the baseline cooking curve at that time). The ratio ξ1=v_act1 / v_set1 is calculated. Similarly, ξ2 to ξ5 are obtained. The system performs statistical analysis on these ratios, for example, calculating their weighted average ξ_avg_A. Then the attenuation coefficient K_A of stage A is K_A=1-ξ_avg_A. If these 5 cooking cycles are relatively recent, and ξ_avg_A = 0.85, then K_A = 0.15, indicating that the heating efficiency of the equipment experiences an average decrease of 15% during the initial cooking phase. Note that the attenuation coefficient is expressed using K when describing the principle, but K_A is used in the example. A indicates which specific preset cooking time period, and K_A represents the attenuation coefficient for that preset cooking time period.

[0034] In this embodiment, by specifically quantifying the ratio of the "actual average temperature rise rate of the cavity to the set temperature rise rate" during historical cooking, a clear and repeatable definition of the attenuation coefficient is provided. This ratio directly reflects the efficiency of the heating system in converting the set power into effective heat, making the calculation of the attenuation coefficient have clear physical meaning and engineering measurability.

[0035] In one embodiment of this application, S520 includes determining the temperature compensation amount for the cooking curve at each time period based on the attenuation coefficient, which includes: S521, the attenuation coefficients corresponding to each time period are converted into compensation temperature offsets through a preset linear mapping relationship.

[0036] Specifically, this embodiment defines the conversion rules from the attenuation coefficient to the temperature compensation amount. The system maintains a linear mapping relationship, typically in the form ΔT_comp = β × K, where K is the attenuation coefficient obtained in S511 to S512. β is a positive coefficient obtained through regression analysis of a large amount of historical data, measured in degrees Celsius. For example, the system learns and determines β = 65°C. For the attenuation coefficient K_A = 0.15 calculated in stage A, the corresponding compensation temperature offset ΔT_comp_A = 65 × 0.15 ≈ 10 degrees Celsius. This means that when generating the feedforward control curve, the reference cavity temperature setpoint at each moment within stage A (0-3 minutes) will be temporarily increased by degrees Celsius, hoping to "recover" the 15% heat lost due to equipment aging through stronger heating. This linear mapping is simple and effective, and the coefficient β can be continuously optimized through background learning.

[0037] In this embodiment, the dimensionless attenuation coefficient is converted into a specific temperature compensation bias through a preset linear mapping relationship, establishing a quantitative compensation rule between the two. This step is simple in design, and the parameters (mapping coefficients) are easily determined through regression of historical data, ensuring the feasibility and maintainability of the compensation mechanism.

[0038] In one embodiment of this application, S600 includes, during the cooking process, adjusting the feedforward control curve based on real-time monitoring data from multiple sensors to generate a final control command for the cooking equipment, including: S610, construct a state-space model with the core temperature of the food and the equivalent thermal time constant as state variables; S620 uses the real-time collected cavity temperature, food surface temperature, and image color features as observations to jointly estimate the core temperature of the food and the equivalent thermal time constant online. S630, Based on the results of the joint online estimation, calculate the feedback adjustment amount for the feedforward control curve.

[0039] Specifically, S610 aims to establish a mathematical model that can describe the internal dynamics of the cooking process. This model uses two key internal states as state variables: the first is the core temperature of the food, referring to the temperature of the geometric center region of the food, which is a direct physical quantity determining the cooking endpoint; the second is the equivalent thermal time constant, a comprehensive parameter reflecting the overall rate at which heat is transferred from the surface of the food to its core under current cooking conditions. Its value is influenced by both the actual physical properties of the food (such as moisture content) and the heat transfer environment of the cavity (such as steam content). The vector composed of these two state variables is denoted as the state vector x, i.e., x = [T_core, τ]^T, where T_core represents the core temperature of the food, τ represents the equivalent thermal time constant, and the superscript T indicates transpose. The core of this state-space model is a discrete-time state transition equation: x(k)=f(x(k-1),u(k-1))+w(k-1), where k is the discrete-time index, f(•) is a nonlinear function describing the state evolution law, u(k-1) is the system input at time k-1 (i.e., the cavity temperature setpoint T_cavity_set_ff(k-1) provided by the feedforward control curve), and w(k-1) is the process noise, used to characterize the uncertainty of the model.

[0040] Step S620 uses real-time data from multiple sensors to infer the state variable x, which cannot be directly measured as defined in S610. The system uses three types of real-time acquireable physical quantities as observations, forming an observation vector z(k): The cavity temperature, denoted as T_cavity(k), is obtained by measuring and calculating the average value of multiple temperature sensors arranged in the cooking cavity, reflecting the thermodynamic state of the heating environment.

[0041] The surface temperature of the food, denoted as T_surf(k), is obtained through a non-contact infrared temperature sensor and reflects the thermal state of the food surface.

[0042] Image color features, denoted as H(k): obtained by capturing food images with a camera and extracting the average value of the hue component in its color space (such as HSV space). This feature is strongly correlated with the degree of protein denaturation and indirectly reflects the core temperature.

[0043] The observation vector z(k) and the state vector x(k) are related through an observation equation z(k) = h(x(k)) + v(k), where h(•) is the nonlinear observation function and v(k) is the observation noise. The system uses an extended Kalman filter, a recursive estimation algorithm, to perform joint online estimation. In each control cycle k, the algorithm performs two calculations: first, prediction is performed, that is, based on the state estimate x^(k-1|k-1) and the input u(k-1) at the previous time step, the prior estimate x^(k|k-1) of the current state is predicted through the state transition equation f; then, an update is performed, that is, using the actual observation z(k) at the current time step, the Kalman gain is calculated and the prior estimate is corrected, thus obtaining the posterior optimal estimate of the state vector x^(k|k) = [T_core_est(k), τ_est(k)]^T. Where T_core_est(k) is the optimal estimate of the core temperature of the food, and τ_est(k) is the optimal estimate of the equivalent thermal time constant. Through this process, the system achieves real-time, high-precision joint estimation of the food's core temperature T_core_est and the current equivalent thermal time constant τ_est.

[0044] Based on the joint estimation results T_core_est(k) and τ_est(k) output by S620, and combined with the preset target and control logic, S630 calculates the feedback quantity used to correct the feedforward control curve. The feedback adjustment quantity ΔT_fb(k) consists of two superimposed components: a first adjustment quantity ΔT_fb1(k) and a second adjustment quantity ΔT_fb2(k). The specific calculation process will be explained in subsequent embodiments.

[0045] In this embodiment, by constructing a state-space model and using the extended Kalman filter algorithm for multi-sensor data fusion, real-time and optimal estimation of the core temperature of the food and the thermal time constant reflecting the current comprehensive heat transfer conditions—which cannot be directly measured—is achieved. This provides crucial internal state information for high-precision feedback regulation.

[0046] In one embodiment of this application, S630 includes calculating the feedback adjustment amount of the feedforward control curve based on the result of the joint online estimation, comprising: S631, calculate the first adjustment amount based on the estimated deviation between the core temperature of the food and the target physical parameters; S632, calculates the second adjustment amount based on the real-time evaluation of the instantaneous heating efficiency of the cooking equipment; S633, the first adjustment amount and the second adjustment amount are superimposed to form the feedback adjustment amount.

[0047] Specifically, step S631 aims to eliminate the deviation between the core temperature of the food and the user-set target. First, the core temperature deviation e(k) is calculated: e(k)=T_target-T_core_est(k).

[0048] Where T_target is the target core temperature value converted according to the user's cooking intention (e.g., 55℃ for a rare steak), and T_core_est(k) is the optimal estimate of the core temperature of the food output by step S620. Subsequently, a proportional-integral-derivative controller algorithm is used to calculate the first adjustment ΔT_fb1(k) based on the deviation e(k) and its historical accumulation and trend: ΔT_fb1(k)=K_p×e(k)+K_i×Σ_{j=0}^{k}e(j)×Δt+K_d×[e(k)-e(k-1)] / Δt.

[0049] Wherein, K_p is the proportional gain coefficient, which generates a rapid response based on the current deviation magnitude; K_i is the integral gain coefficient, which eliminates steady-state error by accumulating historical deviations; K_d is the derivative gain coefficient, which provides anticipatory adjustment based on the deviation's changing trend to suppress overshoot; Δt is the control period. The first adjustment variable ΔT_fb1 forms the basis of feedback regulation, used to achieve precise tracking of the core temperature to the target value.

[0050] Step S632 aims to address the instantaneous decrease in heating efficiency caused by unforeseen circumstances during cooking, such as the generation of large amounts of steam or grease splattering. First, the instantaneous heating efficiency η(k) of the cooking equipment is evaluated: η(k) = v_act(k) / v_exp(k).

[0051] Where v_act(k) is the actual cavity temperature rise rate calculated in real time based on cavity temperature sensor data; v_exp(k) is the expected cavity temperature rise rate predicted based on the current heating tube input power, cavity thermodynamic model, and current state τ_est(k). The ratio η(k) intuitively reflects the instantaneous decay of the actual heating capacity relative to the theoretical model. A performance threshold η_th (e.g., 0.85) is set. When η(k) is continuously lower than η_th, it indicates a sudden performance drop requiring additional compensation. At this time, the second adjustment amount ΔT_fb2(k) is calculated: ΔT_fb2(k)=γ×[1-η(k)]×ΔT_max.

[0052] Here, γ is a gain coefficient between 0 and 1, used to adjust the compensation intensity; ΔT_max is the maximum allowable instantaneous compensation temperature value of the system, a safety limit parameter to prevent over-compensation. If η(k) is higher than or equal to the threshold η_th, then ΔT_fb2(k) = 0. The second adjustment amount ΔT_fb2 provides a feedforward compensation capability in the feedforward-feedback composite control structure, specifically for quickly suppressing sudden disturbances.

[0053] In S633, the first adjustment amount and the second adjustment amount calculated above are algebraically superimposed to form the total feedback adjustment amount ΔT_fb(k): ΔT_fb(k) = ΔT_fb1(k) + ΔT_fb2(k).

[0054] The total feedback adjustment ΔT_fb(k) represents the amount of immediate correction required to the feedforward control curve setpoint based on real-time status and performance assessment.

[0055] In this embodiment, by decomposing the feedback adjustment amount into a conventional adjustment amount (first adjustment amount) for target tracking deviation and an enhanced compensation amount (second adjustment amount) for sudden drop in equipment efficiency, the control system can simultaneously cope with two types of problems: setpoint tracking and sudden disturbance suppression, thereby enhancing the robustness of the system in the face of instantaneous disturbances such as juice evaporation and steam blockage during the cooking process.

[0056] In one embodiment of this application, after obtaining the user's input cooking intention, the cooking control method further includes: W100 performs a self-test step before starting the cooking appliance. The self-test step includes: W110, measuring the cold resistance value of the heating element and comparing the cold resistance value of the heating element with the factory nominal value to generate marking information to characterize the aging state of the element.

[0057] Specifically, this embodiment adds a startup self-test step. After the user closes the door and presses the start button, before the main cooking program begins, the system first performs a rapid electrical test on the heating element. The system obtains the cold-state DC resistance R_now of the main heating element at the current room temperature through a measurement circuit. The system reads the factory-rated cold-state resistance R0 of the heating element stored in non-volatile memory. Then, it calculates the resistance change rate: δR = |R_now - R0| / R0. If δR exceeds a set threshold (e.g., 10%), the system generates a component aging flag, for example, by setting the internal flag Heater_Aging_Flag to "TRUE" or recording an aging level. This flag information is not used to prevent cooking, but rather serves as an important prior factor passed to subsequent control algorithms. For example, when the feedforward compensation module detects that the flag is TRUE, it may increase the confidence weight of the attenuation coefficient K_j, or slightly increase the coefficient β in the linear mapping, making the compensation strategy more aggressive.

[0058] In this embodiment, by performing a self-test before cooking starts, measuring the cold-state resistance of the heating element and comparing it with the factory value, a preliminary hardware-level diagnosis of the aging degree of the heating element can be achieved. The generated element aging marker can serve as an important input or confidence factor for the software-layer adaptive compensation algorithm, realizing a closed-loop information flow from hardware state perception to software control strategy.

[0059] After cooking is completed, this method includes a crucial data archiving and model update step, which is central to achieving continuous adaptation. The system defines this cooking session as a valid learning sample and creates a new historical record. This record includes: ingredient category ("sirloin steak"), key heat thickness D_eff, set temperature rise sequence for each cooking stage, actual measured cavity temperature rise sequence, and calculated actual efficiency ratio ξ sequence for each stage. This data, after being encrypted, along with timestamps and device identifiers, is uploaded to a cloud server or stored in local extended storage.

[0060] In the cloud or on-device backend, an asynchronous model learning service is periodically triggered. This service aggregates anonymized historical data from a group of devices of the same model. For a specific food category and thickness range, the service recalculates the average performance ratio ξ_avg_j for each time period across all relevant historical records and updates the lookup table for the global attenuation coefficient K. Simultaneously, it utilizes a larger-scale dataset to re-optimize the global gain coefficient β in the feedforward compensation linear mapping through regression analysis, making it more consistent with the general aging patterns of this device model. The updated model parameters (K table, β values) are securely distributed to each device.

[0061] In this way, each device not only learns from its own cooking history but also benefits from the collective experience of the entire device group, thereby accelerating its adaptation process. This allows new devices to establish accurate compensation models more quickly, and the entire system achieves continuous, distributed performance optimization.

[0062] The technical features of the above embodiments can be combined arbitrarily, and the execution order of the method steps is not restricted. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0063] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A cooking control method, characterized in that, include: Obtain the user's cooking intent; Convert cooking intentions into target physical parameters; Collect ingredient information to identify ingredient categories and obtain the physical characteristics of ingredients; A baseline cooking curve is generated based on the food category, the physical characteristics of the food, and the target physical parameters; The reference cooking curve is fed forward compensation based on the historical cooking data of the cooking equipment to obtain the feedforward control curve. During the cooking process, the feedforward control curve is adjusted based on real-time monitoring data from multiple sensors to generate the final control command for the cooking equipment. Cooking will cease when the target physical parameters are reached.

2. The cooking control method according to claim 1, characterized in that, The process of collecting food information to identify food categories and obtain the physical characteristics of food includes: Visual images of ingredients are simultaneously acquired from multiple spatial perspectives to obtain multi-view visual images; Based on the multi-view visual images, the specific category of the food ingredients can be identified; A three-dimensional model of the food is reconstructed based on the multi-view visual images, and the key thermal thickness of the food is determined based on the three-dimensional model.

3. The cooking control method according to claim 2, characterized in that, The specific categories of food ingredients identified based on the multi-view visual images include: Extract the fusion features of the multi-view visual images; Calculate the Mahalanobis distance between the fused features and the image features of each pre-stored food category template in sequence; The food category of the pre-stored food category template corresponding to the minimum Mahalanobis distance is determined as the specific category of the food for multi-view visual image recognition.

4. The cooking control method according to claim 2, characterized in that, The process of generating a baseline cooking curve based on the ingredient category, the physical characteristics of the ingredient, and the target physical parameters includes: A simplified heat transfer model is constructed based on the key thermal thickness of the food and the typical thermal properties of the food category. Using a model predictive control method, with the target physical parameters as the control objective, and based on the simplified heat transfer model, the future cavity temperature setpoint sequence is solved through rolling time-domain optimization calculation, and the future cavity temperature setpoint sequence is used as the reference cooking curve.

5. The cooking control method according to claim 1, characterized in that, The reference cooking curve is fed forward to compensate based on historical cooking data from the cooking equipment to obtain a feedforward control curve, including: Based on historical cooking data, calculate the attenuation coefficient reflecting the heating efficiency of the cooking equipment within multiple preset cooking time periods; The temperature compensation amount for the cooking curve at each time period is determined based on the attenuation coefficient. The reference cooking curve is corrected using the temperature compensation amount to obtain the feedforward control curve.

6. The cooking control method according to claim 5, characterized in that, The calculation of the attenuation coefficient reflecting the heating efficiency of the cooking equipment over multiple preset cooking time periods based on historical cooking data includes: For each preset cooking time period, calculate the ratio of the actual average temperature rise rate of the cavity to the set temperature rise rate during multiple historical cooking processes. The attenuation coefficient for the corresponding preset cooking time period is calculated based on the ratio.

7. The cooking control method according to claim 5, characterized in that, The determination of the temperature compensation amount for the cooking curve at each time period based on the attenuation coefficient includes: The attenuation coefficients corresponding to each time period are converted into compensation temperature offsets through a preset linear mapping relationship.

8. The cooking control method according to claim 1, characterized in that, During the cooking process, the feedforward control curve is adjusted based on real-time monitoring data from multiple sensors to generate final control commands for the cooking equipment, including: Construct a state-space model with the core temperature of the food and the equivalent thermal time constant as state variables; The cavity temperature, food surface temperature, and image color features collected in real time are used as observations to jointly estimate the core temperature of the food and the equivalent thermal time constant online. Based on the results of the joint online estimation, the feedback adjustment amount for the feedforward control curve is calculated.

9. The cooking control method according to claim 8, characterized in that, The step of calculating the feedback adjustment amount for the feedforward control curve based on the results of the joint online estimation includes: The first adjustment amount is calculated based on the estimated deviation between the core temperature of the food and the target physical parameters; The second adjustment amount is calculated based on the instantaneous heating efficiency of the cooking equipment as assessed in real time; The first adjustment amount and the second adjustment amount are superimposed to form the feedback adjustment amount.

10. The cooking control method according to claim 1, characterized in that, After obtaining the user's cooking intent, it also includes: Before starting the cooking equipment, a self-test step is performed; the self-test step includes: measuring the cold resistance value of the heating element and comparing the cold resistance value of the heating element with the factory nominal value to generate marking information to characterize the aging state of the element.

Images