Frying and baking equipment energy-saving heat-preservation self-adaptive dynamic control method and system

Abstract

The invention provides an energy-saving thermal-insulation self-adaptive dynamic control method and system for frying and baking equipment, and relates to the technical field of data processing, and the method comprises the steps: 1, collecting the surface temperature distribution, thickness gradient and environment temperature and humidity data of food materials in real time, extracting three-dimensional thermal diffusion morphological features, and generating an initial morphological feature vector; and step 2, based on a real-time deviation value of a set target temperature and an actually measured food material core temperature, a pseudo partial derivative estimation value is calculated in combination with the morphological feature vector, and periodic intermittent power parameters including a duty ratio and a period are generated according to the pseudo partial derivative estimation value. Through multi-source data acquisition and analysis, parameter dynamic calculation and optimization and knowledge base intelligent matching, accurate temperature control, high efficiency and energy saving are realized, and the service life of equipment is prolonged while the food material cooking quality is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a method and system for adaptive dynamic control of energy saving and heat preservation of frying and grilling equipment. Background Art

[0002] In the existing technology, there is room for improvement in the equipment's dynamic perception of the thermal diffusion characteristics of food and its strategy adjustment capabilities. In actual applications, the heating quality and energy utilization problems caused by temperature control deviations need to be optimized urgently.

[0003] Traditional control methods mostly use fixed power or simple PID adjustment. The association mechanism between the three-dimensional thermal diffusion characteristics of food (such as thickness gradient and surface curvature) and heating parameters needs to be improved. When processing food with different shapes, local uneven heating is prone to occur, which needs to be improved through parameter adjustment, affecting processing efficiency. There is room for improvement in the thermal insulation technology in combining the food shape with the dynamic adjustment strategy of ambient temperature and humidity. The ability to compensate for heat loss when the environment changes is insufficient, which may lead to temperature fluctuations during the thermal insulation stage. The global uniform heating mode of traditional equipment has an imperfect mechanism for differentiated heat demand in different areas of food. Food with complex geometric shapes is prone to local over-cooking or under-cooking, requiring manual intervention and adjustment, which affects processing consistency.

[0004] Some studies have optimized heating control through sensor data or logical algorithms, but there are still limitations in establishing a direct mapping relationship between the thermal diffusion characteristics of ingredients and heating parameters. When processing ingredients of different thicknesses or curvatures, the optimization effects of temperature control accuracy and energy utilization need to be improved. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an adaptive dynamic control method and system for energy saving and heat preservation of frying and grilling equipment, so as to achieve precise temperature control and energy saving and heat preservation during the frying and grilling process.

[0006] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0007] In a first aspect, a method for adaptive dynamic control of energy conservation and heat preservation of frying and grilling equipment is provided, the method comprising:

[0008] Step 1: Real-time data collection of food surface temperature distribution, thickness gradient, and ambient temperature and humidity are collected to extract three-dimensional thermal diffusion morphological features and generate an initial morphological feature vector.

[0009] Step 2: Based on the real-time deviation between the set target temperature and the measured food core temperature, combined with the morphological feature vector, a pseudo partial derivative estimate is calculated, and based on the pseudo partial derivative estimate, periodic intermittent power parameters, including duty cycle and period, are generated;

[0010] Step 3: Input the morphological feature vector into the dynamic compensation function, output the morphological correction coefficient, and dynamically correct the pseudo partial derivative estimate based on the periodic intermittent power parameter to generate an adaptive partial derivative;

[0011] Step 4: Input the adaptive partial derivatives into the dynamic power allocator, combine the intermittent power constraint and the temperature error tolerance, optimize the step weight and correction strength through the directed parameter search algorithm, and generate the spatial power allocation parameter set;

[0012] Step 5: Associate the spatial power allocation parameter set with the corresponding morphological feature vector and periodic power parameter and store them in the heat diffusion knowledge base. When new food is detected, the spatial power allocation parameter set is automatically matched.

[0013] Furthermore, the surface temperature distribution, thickness gradient and ambient temperature and humidity data of the food are collected in real time to extract the three-dimensional thermal diffusion morphological features and generate the initial morphological feature vector, including:

[0014] The infrared thermal imaging array is used to collect the surface temperature distribution data of the food in real time, the laser thickness sensor is used to obtain the thickness gradient data of the food, and the temperature and humidity sensor is used to collect the ambient temperature and humidity data;

[0015] Extract three-dimensional thermal diffusion morphological features based on surface temperature distribution, thickness gradient, and ambient temperature and humidity data;

[0016] The three-dimensional thermal diffusion morphological features are fused to generate the initial morphological feature vector.

[0017] Furthermore, the three-dimensional heat diffusion morphological characteristics include:

[0018] Geometric shape factor, i.e. the thermal diffusion rate parameter of the food surface curvature distribution;

[0019] Thickness distribution factor, which is the heat conduction efficiency parameter of the thickness difference of different areas of food;

[0020] The surface curvature factor is the parameter of convective heat transfer efficiency on the concave and convex surface of food.

[0021] Furthermore, based on the real-time deviation between the set target temperature and the measured food core temperature, combined with the morphological feature vector, the pseudo partial derivative estimate is calculated, and based on the pseudo partial derivative estimate, the periodic intermittent power parameters, including duty cycle and period, are generated, including:

[0022] Obtain the initial morphological feature vector and calculate the deviation between the set target temperature and the measured food core temperature in real time;

[0023] The deviation value and the initial morphological feature vector are input into the pseudo partial derivative estimator, the thickness distribution factor in the initial morphological feature vector is processed to generate a thickness weight coefficient, and the thickness weight coefficient is fused with the historical deviation data. The dynamic weighted least squares algorithm is used for rolling optimization and the pseudo partial derivative estimate is output. The pseudo partial derivative estimate represents the intensity of the impact of unit power change on the core temperature.

[0024] The estimated value of the pseudo partial derivative is input into the intermittent power generator to calculate the basic duty cycle, wherein the basic duty cycle is inversely proportional to the estimated value of the pseudo partial derivative; the geometric shape factor in the initial morphological feature vector is obtained, the curvature compensation is performed on the period value according to the geometric shape factor, and the periodic intermittent power parameters are output, including the duty cycle optimized by the pseudo partial derivative and the period value compensated by the geometric shape factor.

[0025] Furthermore, the morphological feature vector is input into the dynamic compensation function, the morphological correction coefficient is output, and based on the periodic intermittent power parameter, the pseudo partial derivative estimate is dynamically corrected to generate an adaptive partial derivative, including:

[0026] The morphological feature vector is input into the dynamic compensation function, and the geometric shape factor and the surface curvature factor are nonlinearly mapped using Gaussian kernel to generate the curvature-diffusion coupling coefficient;

[0027] Based on the interaction between the coupling coefficient and the thickness distribution factor, the morphology correction coefficient is output to quantify the intensity of the compensation demand for thermal diffusion on the food surface;

[0028] The periodic intermittent power parameters are obtained, and the power density characteristics in the duty cycle and the switching frequency characteristics in the period value are analyzed. The morphology correction coefficient and the power switching frequency characteristics are integrated to construct a dynamic weight matrix. The morphology correction coefficient is used as the main correction factor to perform variable gain correction on the pseudo partial derivative estimate. The adaptive partial derivative is output to represent the actual impact intensity of the unit power on the core temperature after compensation.

[0029] Furthermore, the adaptive partial derivatives are input into the dynamic power allocator. Combined with the intermittent power constraint and temperature error tolerance, the step weight and correction strength are optimized through a directed parameter search algorithm to generate a spatial power allocation parameter set, including:

[0030] Input the adaptive partial derivatives into the dynamic power allocator, use the adaptive partial derivatives as the gradient descent direction reference, and initialize the parameter search path;

[0031] Combined with intermittent power constraints, including the maximum power threshold limit and the minimum intermittent duration constraint, and temperature error tolerance, including the core temperature fluctuation range and the surface temperature uniformity threshold, axial detection is performed along the adaptive partial derivative direction through a directional parameter search algorithm, where the step weight is mapped to the power adjustment step amount and the correction intensity is mapped to the thermal response inertia compensation coefficient.

[0032] During the axial detection process, the Pareto final solution set is generated within the constraint boundary, which includes the partition radiation power intensity value, partition heating duration sequence and cross-zone power switching timing scheme.

[0033] Furthermore, the spatial power allocation parameter set is associated with the corresponding morphological feature vector and periodic power parameter and stored in the heat diffusion knowledge base. When a new food is detected, the spatial power allocation parameter set is automatically matched, including:

[0034] Establishing a primary correlation mapping between the spatial power allocation parameter set and the morphological feature vector; establishing a secondary adjustment mapping between the spatial power allocation parameter set and the periodic power parameter;

[0035] The real-time geometric shape factor in the morphological feature vector is used as the thermal field distribution benchmark key, the real-time duty cycle dynamic range in the periodic power parameter is used as the power regulation benchmark key, and the spatial power allocation parameter set is associated and stored to construct a heat diffusion knowledge base storage structure;

[0036] When new food is detected, the real-time morphological feature vector and periodic power parameters of the new food are extracted, and the degree of fit of the thermal diffusion characteristics with the thermal field distribution benchmark key is calculated based on the real-time geometric shape factor, and the dynamic compatibility with the power regulation benchmark key is verified based on the real-time duty cycle dynamic range; the spatial power allocation parameter set that has passed the double verification is retrieved, and the partitioned radiation power intensity value, partitioned heating duration sequence and cross-zone power switching timing plan are loaded.

[0037] Secondly, the adaptive dynamic control system for energy saving and heat preservation of frying and grilling equipment includes:

[0038] The feature extraction module is used to collect food surface temperature distribution, thickness gradient and ambient temperature and humidity data in real time, extract three-dimensional thermal diffusion morphological features and generate initial morphological feature vectors;

[0039] A power parameter calculation module is used to calculate the pseudo partial derivative estimate based on the real-time deviation between the set target temperature and the measured food core temperature, combined with the morphological feature vector, and generate periodic intermittent power parameters;

[0040] A dynamic correction module is used to input the morphological feature vector into the dynamic compensation function to output the morphological correction coefficient, dynamically correct the pseudo partial derivative estimate based on the periodic intermittent power parameter, and generate an adaptive partial derivative;

[0041] The parameter optimization module is used to input the adaptive partial derivatives into the dynamic power allocator, optimize the step weight and correction strength through a directed parameter search algorithm based on the intermittent power constraint and temperature error tolerance, and generate a spatial power allocation parameter set;

[0042] The matching module is used to associate the spatial power allocation parameter set with the corresponding morphological feature vector and periodic power parameter and store them in the heat diffusion knowledge base, and automatically match the spatial power allocation parameter set when new food is detected.

[0043] According to a third aspect, a computing device includes:

[0044] one or more processors;

[0045] The storage device is used to store one or more programs, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method.

[0046] In a fourth aspect, a computer-readable storage medium stores a program, which implements the method when executed by a processor.

[0047] The above solution of the present invention includes at least the following beneficial effects:

[0048] By collecting data such as the surface temperature and thickness gradient of ingredients in real time, combined with dynamic compensation and correction mechanisms, the changes in the thermal conductivity characteristics of ingredients can be accurately captured, and the core temperature deviation can be controlled within a very small range to avoid local overcooking or undercooking, ensuring that the ingredients are heated evenly, and improving the cooking quality and taste. Pseudo-partial derivative estimates are used to generate periodic intermittent power parameters, and combined with a directed search algorithm to optimize power distribution, avoiding energy waste caused by continuous high-power heating, and reducing operating costs compared to traditional frying and grilling equipment. Regardless of how the shape and thickness of the ingredients change, or whether the ambient temperature and humidity fluctuate, this method can dynamically adjust the power allocation strategy based on real-time data, quickly adapting from thin slices of fish to thick pieces of meat, from humid and rainy weather to dry environments, to ensure stable cooking results.

[0049] During power allocation, intermittent power constraints, such as limiting the maximum power threshold and setting a minimum intermittent duration, are strictly adhered to. This prevents equipment from operating at high loads for extended periods or from frequent starts and stops. This reduces equipment wear and tear, lowers failure rates, extends overall equipment life, and reduces maintenance costs. A heat diffusion knowledge base stores the correspondence between different food forms and power parameters. As this data accumulates, it automatically matches the optimal power allocation plan for similar ingredients. When new ingredients are added to the equipment, parameters can be quickly retrieved, reducing preheating and commissioning time, improving cooking efficiency, and gradually achieving intelligent equipment upgrades. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 It is a flow chart of the adaptive dynamic control method for energy saving and heat preservation of frying and grilling equipment provided by an embodiment of the present invention.

[0051] Figure 2It is a schematic diagram of an energy-saving and heat-insulating adaptive dynamic control system for frying and grilling equipment provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0052] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0053] like Figure 1 As shown, an embodiment of the present invention provides an adaptive dynamic control method for energy saving and heat preservation of grilling equipment, which includes the following steps:

[0054] Step 1: Real-time data collection of food surface temperature distribution, thickness gradient, and ambient temperature and humidity are collected to extract three-dimensional thermal diffusion morphological features and generate an initial morphological feature vector.

[0055] Step 2: Based on the real-time deviation between the set target temperature and the measured food core temperature, combined with the morphological feature vector, a pseudo partial derivative estimate is calculated, and based on the pseudo partial derivative estimate, periodic intermittent power parameters, including duty cycle and period, are generated;

[0056] Step 3: Input the morphological feature vector into the dynamic compensation function, output the morphological correction coefficient, and dynamically correct the pseudo partial derivative estimate based on the periodic intermittent power parameter to generate an adaptive partial derivative;

[0057] Step 4: Input the adaptive partial derivatives into the dynamic power allocator, combine the intermittent power constraint and the temperature error tolerance, optimize the step weight and correction strength through the directed parameter search algorithm, and generate the spatial power allocation parameter set;

[0058] Step 5: Associate the spatial power allocation parameter set with the corresponding morphological feature vector and periodic power parameter and store them in the heat diffusion knowledge base. When new food is detected, the spatial power allocation parameter set is automatically matched.

[0059] In an embodiment of the present invention, by real-time collection of temperature distribution, thickness gradient and environmental parameters, a three-dimensional heat diffusion feature vector is constructed to achieve comprehensive modeling of food morphology and heat conduction characteristics, solve the problem of one-sided thermal field perception in traditional control, and provide data support for dynamic adjustment. Based on the temperature deviation and morphological characteristics, pseudo-partial derivatives are calculated to generate periodic intermittent power parameters (duty cycle, period), which can dynamically adjust the heating power according to the real-time state of the food, reduce energy waste in constant power mode, and improve energy saving efficiency. The dynamic compensation function is used to generate the morphological correction coefficient, and the intermittent power parameter is combined to correct the partial derivative, effectively compensating for the influence of morphological differences such as food thickness and curvature on heat conduction, reducing the core temperature control deviation, and improving heating uniformity.

[0060] A targeted parameter search algorithm optimizes spatial power distribution, enabling differentiated heating based on the heat requirements of different food areas. This avoids the overcooking or undercooking of food caused by traditional uniform heating, improving the consistency and pass rate of food processing. Power parameters are associated with morphological characteristics and stored in a knowledge base, enabling automatic parameter matching for new ingredients, reducing manual debugging costs, enhancing the equipment's adaptability to different ingredients, and promoting intelligent control upgrades.

[0061] In a preferred embodiment of the present invention, the above step 1, collecting food surface temperature distribution, thickness gradient, and ambient temperature and humidity data in real time, extracting three-dimensional thermal diffusion morphological features, and generating an initial morphological feature vector, may include:

[0062] Step 100: collecting food surface temperature distribution data in real time through an infrared thermal imaging array, obtaining food thickness gradient data through a laser thickness sensor, and collecting ambient temperature and humidity data through a temperature and humidity sensor;

[0063] Step 101, based on the surface temperature distribution, thickness gradient and ambient temperature and humidity data, extracting three-dimensional heat diffusion morphological features, specifically includes:

[0064] Geometric shape factor, i.e. the thermal diffusion rate parameter of the food surface curvature distribution;

[0065] Thickness distribution factor, which is the heat conduction efficiency parameter of the thickness difference of different areas of food;

[0066] Surface curvature factor, i.e., the parameter of convective heat transfer efficiency on the concave and convex surfaces of food;

[0067] Step 102 : Fusing the three-dimensional heat diffusion morphological features to generate an initial morphological feature vector.

[0068] In this embodiment of the present invention, a 16×16 infrared thermal imaging sensor array is evenly distributed across the top of the grilling chamber, with adjacent sensors spaced 5mm apart. This array covers a 20cm diameter food area, ensuring at least four temperature measurement points per square centimeter of surface area. The sensor utilizes an uncooled microbolometer with a response band of 8-14μm and a temperature measurement range of -20°C to 300°C. A built-in thermoelectric cooler maintains a constant detector temperature, preventing ambient temperature from affecting measurement accuracy. The sampling frequency is set to 10Hz, and a two-point calibration is performed before each acquisition. The first step is to sample blackbody radiation at 0°C and 100°C using an electronically controlled shutter. A table of offset and gain factors for each pixel is established (e.g., a pixel offset of -0.3°C and a gain factor of 1.02) to compensate for temperature deviations in real time. The raw temperature matrix is ​​then median filtered (3×3 window) to eliminate salt and pepper noise, and the temperature field is then smoothed using a Gaussian filter (σ=1.0) to reduce measurement fluctuations caused by surface texture. The imaging area was calibrated weekly at nine points (four corners and the center) using a high-precision platinum resistance thermometer (accuracy ±0.1°C). A polynomial correction model (quadratic polynomial) for temperature-pixel value was established. A laser thickness measurement module (semiconductor laser wavelength 650 nm, CMOS image sensor resolution 1280 × 720) was used using triangulation. This module was mounted on the side rails of the instrument and could move two-dimensionally along the X and Y axes, with a scanning range of 30 cm × 30 cm and a vertical measurement accuracy of ±0.05 mm.

[0069] Static food: A zigzag scanning path is used with a scanning pitch of 0.5mm. It takes 15 seconds to complete a 20cm×20cm area scan.

[0070] Dynamic food (such as food on a conveyor belt): Sampling is triggered by an encoder, and a thickness point is collected every 0.5mm of movement. Dynamic thickness profile reconstruction is achieved in conjunction with the movement speed (maximum 10mm / s).

[0071] Abnormal points are removed from the original thickness data: if the thickness difference between a point and the adjacent 8 points exceeds 0.5mm, it is judged as splash or oil drop interference and repaired using bilinear interpolation; cubic spline interpolation is used to encrypt the sparse scanning points to generate a thickness matrix with a resolution of 1mm×1mm, and the edge area (within 5mm from the contour) is encrypted to 0.5mm×0.5mm to ensure that concave and convex details are not lost.

[0072] The temperature sensor uses a PT100 platinum resistor (Class A accuracy, ±0.1°C), and the humidity sensor uses a capacitive polymer thin film sensor (accuracy ±1.5% RH). These sensors are integrated into a stainless steel probe and placed within the heating chamber, 10 cm from the food surface, out of direct sunlight from the heating tube. The probe surface is coated with an oil-resistant coating to reduce contamination. Temperature sampling is at a 10Hz frequency, and humidity at a 5Hz frequency. The signals are converted to digital signals using a 24-bit ADC. The PT100 resistance is linearly corrected based on the sensor's internal temperature, as monitored by the probe's built-in NTC thermistor (e.g., a PT100 resistance of 109.7Ω at 25°C corresponds to the actual ambient temperature after compensation). Actual ambient humidity is calculated through linear interpolation using a pre-calibrated humidity-temperature correction table (e.g., 40% RH at 25°C is corrected to 38.5% RH at 30°C).

[0073] Step 101: Perform Canny edge detection on the temperature distribution matrix. First, smooth the image with a Gaussian filter (σ = 1.5), then extract edges using non-maximum suppression and double threshold segmentation. Finally, use morphological closing (3×3 rectangular kernel) to connect the broken edges to obtain the complete outline of the food. For each point on the outline, a five-point quadratic polynomial is used to fit the local curve and calculate the curvature value of the point. The specific steps are as follows:

[0074] Take two adjacent points before and after the current point and construct a quadratic polynomial y=ax 2 +bx+c, solve the coefficients and curvature by the least squares method Among them, x is the horizontal coordinate (such as along the length of the food); y is the longitudinal coordinate (such as along the thickness or height of the food); a is the core parameter of the local surface concave and convex degree, which determines the opening direction and curvature of the curve (a>0 means the opening is upward, a<0 ​​means the opening is downward); b is the position of the symmetry axis that affects the curve; c is the intercept of the curve on the y-axis. For non-contour points (inside the food), the distance to the nearest contour is calculated through distance transformation as a virtual curvature reference value. A curvature-thermal diffusion rate lookup table is established, and the thermal diffusion coefficient under different curvatures (such as curvature 0.1mm) is measured. -1 Corresponding diffusion coefficient 1.2W / (m·K), curvature -0.1mm -1 Corresponding to 0.8 W / (m·K), the calculated curvature values ​​are linearly interpolated to generate a thermal diffusion rate parameter matrix, where each element in the matrix corresponds to the diffusion rate at a specific position on the food surface.

[0075] Thickness distribution factor calculation:

[0076] Based on the laser thickness measurement data, the K-means clustering algorithm is used to divide the food into several homogeneous areas. The clustering parameters are set as: distance threshold 0.5mm, minimum area 5mm 2For irregular ingredients (such as meat with bones), first identify key areas (bone blocks, lean meat, fat) through contour convex hull analysis, and then cluster them separately. Calculate the ratio of the average thickness of each area to the overall average thickness as the thickness difference coefficient (such as the average thickness of the area is 10mm, the overall average is 8mm, and the coefficient is 1.25). For areas with sudden thickness changes (such as the edge drops sharply from 10mm to 2mm), introduce a gradient factor. The larger the gradient, the higher the coefficient weight (the weight is multiplied by 1.5 when the gradient is greater than 1mm / mm). According to Fourier's law of heat conduction, a thickness-heat conduction efficiency model is established: the greater the thickness, the lower the efficiency. Where α is the material's thermal conductivity (beef α = 0.5 mm / W, vegetables α = 0.3 mm / W), and Δh is the thickness variation coefficient -1. This yields the thermal conductivity efficiency value for each region. A table lookup method is used to retrieve the corresponding α value for each food type, ultimately generating a thickness distribution factor (thermal conductivity parameter matrix) by spatial arrangement.

[0077] Surface curvature factor calculation:

[0078] Combining temperature distribution and geometric factors, it can identify convex, concave and flat areas on the surface of food. Convex areas are defined as curvature greater than 0.05mm. -1 , concave curvature <-0.05mm -1 , and the rest are planes. Using a simplified natural convection model, for the convex area, the heat transfer coefficient Where ΔT is the temperature difference between the surface and the environment (unit: °C); concave area For the plane area, h = 5. In actual calculations, the model is modified using shape correction factors (e.g., a 1.2 correction factor for hemispherical convex surfaces and a 0.8 correction factor for V-shaped concave surfaces). The calculated heat transfer coefficient is normalized to the interval [0, 1]. Taking the plane area as the reference (value of 1), the convex area value is greater than 1, and the concave area value is less than 1. A surface curvature factor matrix is ​​generated, and the matrix elements reflect the multiples of the convective heat transfer efficiency of each area relative to the plane.

[0079] Step 102: Find the maximum value Vmax and minimum value Vmin of the thermal diffusion rate parameter matrix. Apply minimum-maximum normalization to map the data to [0, 1]. For outliers (such as mutation points caused by noise), use the 3σ principle to identify and replace them with the mean of the adjacent points. Similarly, the thermal conductivity efficiency parameter matrix is ​​normalized. However, considering that thickness differences may contain negative values ​​(such as regional thickness less than the overall average), Z-score normalization is applied. Finally, the data is mapped to [0, 1] using a translation transformation. Since the heat transfer coefficient has been normalized, smoothing is performed directly. A 5×5 Gaussian kernel convolution is used to eliminate local fluctuations, with a kernel standard deviation of σ = 1.0.

[0080] Through orthogonal experimental design, the geometric shape factor weight w1 = 0.4, thickness distribution factor w2 = 0.35, and surface curvature factor w3 = 0.25 were determined. The experimental variables included the type of food (beef, chicken breast, vegetables), thickness range (2-10mm), and surface curvature (flat, convex, concave). The weight was optimized with heating uniformity as the evaluation index. When the standard deviation of food thickness is detected to be greater than 1.5mm, w2 is automatically increased to 0.45 and w1 is reduced to 0.35; when the standard deviation of surface curvature is greater than 0.1mm, the weight is automatically increased to 0.45 and w1 is reduced to 0.35. -1 When the thickness is greater than 1.5 mm, w1 is increased to 0.45 and w3 is reduced to 0.2. The dynamic correction threshold is obtained through historical data statistics (for example, correction is triggered when the thickness standard deviation is greater than 1.5 mm with a probability of 15%). Principal component analysis (PCA) is performed on the three standardized matrices. The first three principal components of the geometric shape factor (cumulative variance contribution rate greater than 90%), the first two principal components of the thickness distribution factor, and the first two principal components of the surface curvature factor are retained to reduce the vector dimension. The factor vectors after dimensionality reduction are weighted and summed according to their weights. For example, when the vector dimension is n, the i-th element of the initial morphological feature vector F(i) = w1×G(i) + w2×T(i) + w3×S(i), where G, T, and S are the reduced dimensionality vectors of the geometric shape, thickness, and curvature, respectively. Spatial coordinates (X, Y) and timestamps are attached to each initial morphological feature vector to form a four-dimensional feature vector (X, Y, T, F), where F is the initial morphological feature vector.

[0081] Through infrared thermal imaging, laser thickness measurement, and environmental sensing, real-time collection of food surface temperature, thickness gradient, and environmental parameters is achieved, resolving the problem of missing thermal field information caused by traditional single-point temperature measurement and providing comprehensive data support for dynamic control. Factors such as geometric shape, thickness distribution, and surface curvature are extracted to quantify the impact of food shape on heat conduction, construct a feature model that matches the actual heat diffusion process, and enhance the pertinence of the control strategy. Multidimensional features are integrated through standardization and weight distribution to form a comprehensive feature vector that can adapt to the thermal characteristics of food with different shapes (such as irregular shapes and uneven thicknesses).

[0082] In a preferred embodiment of the present invention, the above step 2, based on the real-time deviation between the set target temperature and the measured food core temperature, combined with the morphological feature vector, calculates the pseudo partial derivative estimate, and generates the periodic intermittent power parameters including the duty cycle and period according to the pseudo partial derivative estimate, which may include:

[0083] Step 200: Obtain an initial morphological feature vector and calculate in real time the deviation between the set target temperature and the measured food core temperature;

[0084] Step 201: Input the deviation value and the initial morphological feature vector into a pseudo partial derivative estimator, process the thickness distribution factor in the initial morphological feature vector, generate a thickness weight coefficient, fuse the thickness weight coefficient with the historical deviation data, perform rolling optimization using a dynamic weighted least squares algorithm, and output a pseudo partial derivative estimate. The pseudo partial derivative estimate represents the intensity of the impact of a unit power change on the core temperature.

[0085] In step 202, the pseudo partial derivative estimate is input into the intermittent power generator to calculate the basic duty cycle, wherein the basic duty cycle is inversely proportional to the pseudo partial derivative estimate; a geometric shape factor in the initial morphological feature vector is obtained, and curvature compensation is performed on the period value according to the geometric shape factor, and periodic intermittent power parameters are output, including the duty cycle optimized by the pseudo partial derivative and the period value compensated by the geometric shape factor.

[0086] In the embodiment of the present invention, the vector after the fusion of the three-dimensional thermal diffusion morphological features is read from the output result of step 1. The vector includes components such as geometric shape factor, thickness distribution factor, surface curvature factor, etc. (for example, the dimension is 1×n, where n is determined by the food scanning accuracy, such as per 1cm 2 Corresponding to 1 feature point). Cache the latest 5 sets of feature vectors and use the sliding average method for smoothing:

[0087] The current vector and the previous four vectors are weighted based on their temporal proximity (e.g., weight 0.4 for the most recent vector, 0.3 for the previous vector, and so on, decreasing in order) to eliminate feature jumps caused by slight food movement or sensor noise. The thickness distribution factor is mapped to [0, 1] (0 represents the thinnest area, 1 represents the thickest area). The geometric shape factor is mapped to [-1, 1] (negative values ​​indicate concave surfaces, positive values ​​indicate convex surfaces) based on curvature calculation. The surface curvature factor is mapped to [0, 2] based on convective heat transfer efficiency. A 0.5mm diameter K-type thermocouple is embedded in the geometric center of the food (to a depth of half the thickness), and the sampling frequency is set to 10Hz. Before each sampling, the ambient temperature is measured using a PT100 sensor, and cold-junction compensation is applied to the thermocouple (for example, if the ambient temperature is 25°C, the thermoelectric potential corresponding to 25°C is superimposed on the thermocouple output voltage). The target temperature is set by the user (e.g., 60°C for medium doneness of steak), and the real-time deviation is the difference between the target temperature and the measured temperature. To filter high-frequency noise, a first-order low-pass filter is used: current deviation = 0.7 × current measured deviation + 0.3 × previous deviation. The filter coefficient is dynamically adjusted based on the thermal inertia of the food (0.8 for thick food and 0.6 for thin food). A 50-second FIFO queue is established to store the filtered deviation values. Each data point is timestamped (accurate to 0.1 second). When new data is input, the oldest data point is removed from the end of the queue to ensure that the queue always contains the deviation data for the last 5 seconds (sampling interval is 0.1 second).

[0088] Step 201: Extract the thickness distribution factor component from the eigenvector. The original thickness data (e.g., 2-10 mm) is linearly mapped to [0, 1]. Calculate the standard deviation of the thickness distribution factor. If the standard deviation is less than 0.5 (corresponding to actual thickness fluctuation less than 1 mm), the weight coefficient is wth = 0.5 + 0.5 × thickness distribution factor mean, where mean ∈ [0, 1], so wth ∈ [0.5, 1]. If there is a region with thickness less than 1 / 2 of the average thickness (e.g., average thickness 8 mm, and a region less than 4 mm), the weight coefficient is wth = 0.8 × maximum thickness distribution factor + 0.2 × mean, where maximum ∈ [0.5, 1], mean ∈ [0.3, 0.7], so wth ∈ [0.62, 0.94]. Traverse each point in the thickness distribution factor and calculate the difference coefficient (regional thickness / average thickness); if more than 30% of the points have a difference coefficient greater than 1.5, the weight enhancement is triggered: wth = wth × 1.2, where wth = wth × 1.2 is essentially a linear amplification operation on the current weight coefficient. The "wth" on the left is the updated weight, and the "wth" on the right is the weight before the update. 1.2 is a dynamic enhancement factor, which is used to quantify the weight increment when the thickness suddenly changes, and is finally truncated to [0.5, 1.2] (to avoid weight overlimit).

[0089] For each deviation value efilter(i) in the history queue, calculate the time difference Δt = current time - ti, where ti is the sampling time of the i-th deviation value in the history queue; the weight function is exponential decay: The time constant τ = 10 seconds. Specifically, it is expressed as:

[0090] When Δt < 2 seconds, w(i) > 0.8 (e.g., weight ≈ 0.90 at 1 second);

[0091] When Δt = 5 seconds, w(i) ≈ 0.61;

[0092] When Δt>10 seconds, w(i)<0.37 (e.g., weight ≈0.22 at 15 seconds).

[0093] Multiply the thickness weight coefficient wth by the time-decay weight w(i) of each deviation value, then multiply by the deviation value itself, and accumulate the accumulated values ​​to obtain a fusion value = ∑(wth × w(i) × efilter(i)). This value reflects the coupling strength between the thickness difference and the temperature deviation. A larger wth and a larger absolute value of the recent deviation result in a larger absolute value of the fusion value. Assume that the core temperature change ΔT and the power change ΔP satisfy a linear relationship: ΔT = φ × ΔP + ε, where φ is the pseudo-partial derivative, representing the effect of unit power (W) on the core temperature (°C), and ε is the random error. A sliding window of length 20 is used to store the latest (ΔP, ΔT) data pairs, and the oldest data set is removed each time new data is input. The data within the window must satisfy |ΔP|>50W (to filter small power fluctuations) and |ΔT|>0.5°C (to ensure significant temperature changes). Each data set within the window is assigned a weight w = wth × w(i) (where w(i) is the time-decay weight of the data set) and is calculated separately:

[0094] Numerator: ∑(ΔP×ΔT×w), is the weighted coupling of power-temperature variation;

[0095] Denominator: ∑(ΔP 2 ×w), is the weighted sum of squares of power changes;

[0096] Pseudopartial derivatives The result is rounded to 3 decimal places and the unit is ℃ / W;

[0097] If φ<0.01℃ / W or φ>0.1℃ / W, the boundary value is taken and an early warning is triggered (indicating that there may be a sensor failure or food abnormality).

[0098] Step 202: The basic duty cycle D is inversely proportional to the pseudo partial derivative φ, and the calculation formula is: Where k1 and k2 are calibration coefficients (e.g., k1 = 0.05, k2 = 0.2), ensuring D∈[20%, 80%]; the larger φ is (faster heating per unit power), the smaller the required duty cycle is to avoid overheating; the smaller φ is (slower heating per unit power), the larger the duty cycle is, accelerating the heating speed; if D×rated power > 90%×maximum power, the duty cycle is proportionally reduced: D=D×(0.9×maximum power)÷(D×rated power) to ensure that the power output does not exceed the upper limit of the device. Extract the geometric shape factor from the eigenvector and calculate the average curvature Cavg of the food surface:

[0099] Convex area: Cavg>0.05mm -1 (e.g. hemispherical surface);

[0100] Concave area: Cavg<-0.05mm -1 (such as spoon-shaped grooves);

[0101] Plane area: |Cavg|≤0.05mm -1 .

[0102] Cycle compensation calculation:

[0103] Basic cycle T0 = 10 seconds (applicable to flat food);

[0104] When the convex surface is dominant, the compensation coefficient A = 1-0.5×(Cavg-0.05)÷0.15, so that A∈[0.7, 1], and the final period T = T0×A (speeding up power switching to avoid excessive heat dissipation of the convex surface);

[0105] When the concave surface is dominant, the compensation coefficient B = 1 + 0.3 × (-Cavg - 0.05) / 0.15, so that B∈[1, 1.2], and the final period T = T0 × B (extending the period to balance the heat accumulation on the concave surface);

[0106] Planar area: T = T0; if T < 5 seconds, set it to 5 seconds (to avoid frequent relay operation); if T > 20 seconds, set it to 20 seconds (to ensure temperature stability).

[0107] By integrating morphological features with temperature deviations in real time, the system can track changes in the thermal properties of food during heating (such as thickness decay and water evaporation), shortening temperature control response latency and improving real-time heating performance. Dynamic weights generated based on thickness distribution factors enable differentiated power control for different thickness regions of food, effectively resolving the issue of overcooked thin areas and undercooked thick areas caused by traditional uniform heating and improving heating uniformity. Intermittent power energy-saving optimization: An inverse adjustment mechanism based on the duty cycle and pseudo-partial derivative automatically adjusts average power output based on the food's thermal conductivity, avoiding energy waste in constant power mode. Periodic adjustment of curvature compensation further optimizes thermal convection efficiency, improving overall energy utilization. A rolling optimization algorithm continuously updates the pseudo-partial derivative estimate, adapting to sudden changes in food shape (such as flipping) or environmental changes (such as heat dissipation caused by door opening), quickly reconverging to a stable control state and reducing temperature fluctuations. Boundary constraints and feasibility verification mechanisms in parameter calculation ensure safe operation of devices of varying power levels, preventing power overloads and frequent hardware activation, and extending device life.

[0108] In a preferred embodiment of the present invention, the above step 3, inputting the morphological feature vector into the dynamic compensation function, outputting the morphological correction coefficient, and dynamically correcting the pseudo partial derivative estimate based on the periodic intermittent power parameter to generate the adaptive partial derivative, may include:

[0109] Step 300 , inputting the morphological feature vector into a dynamic compensation function, performing Gaussian kernel nonlinear mapping on the geometric shape factor and the surface curvature factor, and generating a curvature-diffusion coupling coefficient;

[0110] Step 301: Based on the interaction between the coupling coefficient and the thickness distribution factor, a morphology correction coefficient is output to quantify the intensity of the compensation requirement for heat diffusion on the food surface;

[0111] Step 302: Obtain periodic intermittent power parameters, analyze the power density characteristics in the duty cycle and the switching frequency characteristics in the period value; fuse the morphology correction coefficient with the power switching frequency characteristics to construct a dynamic weight matrix, and use the morphology correction coefficient as the main correction factor to perform variable gain correction on the pseudo partial derivative estimate, and output an adaptive partial derivative to represent the actual impact intensity of the unit power on the core temperature after compensation.

[0112] In the embodiment of the present invention, the geometric shape factor and the surface curvature factor are accurately separated from the initial morphological feature vector. The geometric shape factor is stored in the form of a matrix, and each element corresponds to 1cm of the food surface. 2 The curvature value of the area, the curvature of the convex area is positive (such as the curvature of the hemispherical convex area is 0.1mm -1 ), the concave surface is negative (such as the curvature of the groove is -0.08mm -1 ), and the flat area is close to 0. The surface curvature factor is also stored in a matrix. The larger the value, the higher the convective heat transfer efficiency (for example, a smooth surface has a value of 0.6, and a rough surface has a value of 0.9). Using the minimum-maximum normalization method, the geometric shape factor value is mapped to the range [-1, 1], and the surface curvature factor is mapped to the range [0, 1]. For example, the original value of 0.6 remains 0.6 after normalization (because the original range is 0-1). During the normalization process, the original maximum value is automatically recorded.

[0113] The Gaussian kernel function is applied to the normalized geometric shape factor and surface curvature factor for nonlinear transformation. The core logic of the Gaussian kernel function is to calculate the "similarity" of each factor value with the kernel center (set to 0), and output a weight value between 0 and 1.

[0114] For example, consider the geometric shape factor: for a region with a normalized curvature value of 0.75, the Gaussian kernel function calculates the distance from the kernel center (0.75 - 0 = 0.75) and combines this with the bandwidth parameter (default 0.3) to determine the weight of that point. The closer the distance (the closer the value is to 0), the closer the weight is to 1; the farther the distance, the closer the weight is to 0. The same principle applies to the surface curvature factor. For example, a value of 0.6, after being calculated using the Gaussian kernel, is weighted based on its "similarity" to the kernel center.

[0115] The elements are multiplied element-by-element according to the preset weights (60% for the geometric shape factor and 40% for the surface curvature factor) and then summed. For example, if the geometric shape factor has a weight of 0.8 at a point and the surface curvature factor has a weight of 0.7 at the corresponding point, the fused curvature-diffusion coupling coefficient is 0.8 × 0.6 + 0.7 × 0.4 = 0.76. This ultimately generates a coupling coefficient matrix with the same dimensions as the original factors, where each element reflects the comprehensive impact of the corresponding area's surface morphology on thermal diffusion.

[0116] Step 301, the thickness distribution factor extracted in step 101 is interactively calculated with the curvature-diffusion coupling coefficient generated in step 300. The thickness distribution factor is presented in matrix form, with each element corresponding to 1cm of food. 2 The normalized thickness value of the region (0 represents the thinnest part, 1 represents the thickest part). Multiply the corresponding elements of the thickness distribution factor matrix and the coupling coefficient matrix to obtain the intermediate variable matrix. For example, if the thickness distribution factor value of a certain region is 0.8 (thicker) and the coupling coefficient value is 0.76, then the intermediate variable is 0.8×0.76=0.608, indicating that the region has a higher thermal diffusion requirement due to the superposition of thickness and surface morphology. This process fully considers the synergistic effect of internal heat conduction (thickness) and surface diffusion (curvature) of the food.

[0117] The intermediate variable matrix is ​​normalized, using min-max normalization to map values ​​to the interval [0, 1] to generate a morphology correction coefficient matrix. To eliminate local fluctuations caused by sensor noise or measurement errors, the morphology correction coefficient matrix is ​​smoothed using a 5×5 sliding average filter (for larger ingredients, a 3×3 window is used for smaller ingredients). For a given point, the average of the 25 surrounding points (a 5×5 window) is calculated and used to replace the original value, ensuring continuous coefficient changes and conforming to physical laws. The final output morphology correction coefficient matrix accurately quantifies the thermal diffusion compensation requirements for each area of ​​the ingredient surface.

[0118] Step 302 extracts the duty cycle and period value from the periodic intermittent power parameters output in step 202. The duty cycle indicates the proportion of heating power on per unit time (e.g., a duty cycle of 60% indicates heating for 6 seconds and off for 4 seconds in every 10 seconds), reflecting the average power density; the period value determines the power switching frequency (e.g., a period of 8 seconds indicates a heating-off cycle is completed every 8 seconds).

[0119] Perform eigendecomposition on the duty cycle and period values:

[0120] Duty cycle classification: divided into three ranges: high (greater than 70%), medium (30%-70%), and low (less than 30%). For example, a duty cycle of 75% is classified as "high", indicating a high power density requirement;

[0121] Cycle value classification: divided into three intervals: fast (less than 5 seconds), medium (5-10 seconds), and slow (more than 10 seconds). For example, a cycle of 4 seconds is classified as "fast", indicating high-frequency thermal regulation requirements;

[0122] Each parameter value is quickly matched to the corresponding feature label through a table lookup method to form a discrete power regulation state description.

[0123] A two-dimensional dynamic weight matrix is ​​constructed, with the row dimension corresponding to the morphology correction coefficient (divided into 11 levels from 0 to 1 with intervals of 0.1, such as 0.0, 0.1, ..., 1.0), and the column dimension corresponding to the three levels of power switching frequency (fast, medium, and slow). When the morphology correction coefficient is 1.0 (strong compensation requirement) and the power switching frequency is slow, the weight is set to 0.8; when the morphology correction coefficient is 0.0 (no compensation requirement) and the power switching frequency is fast, the weight is set to 0.2. During the weight matrix establishment process, multiple groups of experiments were conducted for different food types (such as steak, chicken breast) and heating stages (heating, keeping warm). The weight distribution was optimized through cross-validation to ensure that the matrix covers the correction needs of all scenarios.

[0124] The morphology correction coefficient is used as the main correction factor and combined with the dynamic weight matrix to perform variable gain correction on the pseudo partial derivative estimate in step 201. The specific steps are as follows:

[0125] According to the current shape correction coefficient (such as 0.6) and the power switching frequency characteristics (such as "medium speed"), the weight value of the corresponding element is found from the weight matrix (assuming it is 0.5);

[0126] Multiplying the estimated pseudo-partial derivative by the weight value yields the adaptive partial derivative. For example, if the estimated pseudo-partial derivative is 0.06°C / W, the adaptive partial derivative is 0.06 × 0.5 = 0.03°C / W, which reflects the actual temperature response after factoring in the food shape and power adjustment.

[0127] To prevent over-correction, boundary constraints are set for the adaptive partial derivatives: the lower limit is 80% of the pseudo-partial derivative estimate (i.e., 0.06×0.8=0.048℃ / W), and the upper limit is 120% (i.e., 0.06×1.2=0.072℃ / W). If the calculated result exceeds the range (e.g., 0.03℃ / W is lower than the lower limit), the boundary value (0.048℃ / W) is directly taken to ensure that the control parameters are stable and reliable and to avoid abnormal operation of the equipment.

[0128] Through Gaussian kernel mapping and multi-factor fusion, the dynamic compensation function can accurately capture the synergistic effects of the surface curvature and thickness distribution of ingredients on thermal diffusion. Compared with traditional methods, it effectively solves the problem of uneven heating caused by morphological differences. The linked optimization of the morphological correction coefficient and the power parameter can adjust the correction intensity of the pseudo-partial derivative and the temperature control response speed in real time according to the changes in the shape of the ingredients (such as changes in thickness due to water evaporation during heating) and power adjustment requirements (such as rapid heating or insulation stages), thereby reducing temperature overshoot or lag. The adaptive partial derivative generation mechanism avoids power waste and temperature runaway, such as reducing overheating in concave areas and preventing convex areas from dissipating heat too quickly, thereby improving the quality of food processing. The dynamic weight matrix and boundary constraint mechanism ensure that the system can operate stably under different food types (such as thin slices, thick blocks), heating stages and environmental conditions, improve anti-interference capabilities, and extend the service life of the equipment.

[0129] In a preferred embodiment of the present invention, the above step 4, inputting the adaptive partial derivatives into the dynamic power allocator, combining the intermittent power constraint and the temperature error tolerance, optimizing the step weight and the correction strength through a directed parameter search algorithm, and generating a spatial power allocation parameter set, may include:

[0130] Step 400: input the adaptive partial derivatives into the dynamic power allocator, use the adaptive partial derivatives as the gradient descent direction reference, and initialize the parameter search path;

[0131] Step 401: In combination with intermittent power constraints, including a maximum power threshold limit and a minimum intermittent duration constraint, and temperature error tolerances, including an allowable core temperature fluctuation range and a surface temperature uniformity threshold, axial detection is performed along the adaptive partial derivative direction using a directional parameter search algorithm, where the step weight is mapped to the power adjustment step amount, and the correction strength is mapped to the thermal response inertia compensation coefficient.

[0132] Step 402 : During the axial detection process, a Pareto final solution set is generated within the constraint boundary, including partition radiation power intensity values, partition heating duration sequences, and inter-zone power switching timing schemes.

[0133] In an embodiment of the present invention, an adaptive partial derivative is obtained from step 302, which quantifies the actual impact intensity of a unit power change on the core temperature (for example, 0.03°C / W means that the core temperature will increase by 0.03°C for every 1W increase in power). The adaptive partial derivative is used as a reference for the gradient descent direction, that is, the target direction of power regulation: if the current core temperature is lower than the target temperature, the gradient direction is to increase power; if it is higher than the target temperature, it is to reduce power. In order to ensure the accuracy of the adjustment direction, the adaptive partial derivative is verified twice: compare the current temperature deviation trend (such as the temperature continues to rise or fall for three consecutive times of sampling) with the sign of the partial derivative; if they are inconsistent (for example, the temperature rises but the partial derivative indicates an increase in power), the backup adjustment logic is enabled, and the actual trend of the temperature deviation is used as the standard to avoid incorrect adjustment.

[0134] Starting with the current device operating parameters (such as current power, duty cycle, and period), an initial search path is constructed in a multidimensional parameter space that encompasses all adjustable variables, including partitioned radiant power intensity, heating duration, and power switching timing. The step weight defaults to a small value (corresponding to small power adjustments, such as 5W steps), and the correction strength is set to a medium level (with a thermal response inertia compensation coefficient of 0.5) to ensure stability during the initial adjustment and avoid temperature runaway due to sudden parameter changes.

[0135] Step 401: The device's rated maximum power limit (e.g., 2000W) must be met. The sum of the radiant power intensity in any partition must not exceed this threshold. To protect hardware (e.g., relays), the duration of a single heating or shutoff cycle must not be less than a safe value (e.g., 1 second) to prevent damage to the device due to high-frequency switching. A user-defined target temperature tolerance (e.g., ±2°C) ensures that the core temperature does not exceed this range during regulation. A surface temperature standard deviation limit (e.g., ≤3°C) calculated from infrared thermal imaging data prevents localized overheating or overcooling.

[0136] With the direction determined by the adaptive partial derivative as the axis, step-by-step detection is performed in the parameter space. Each detection adjusts one parameter dimension (such as adjusting the radiation power of partition 1 first, and then adjusting the heating duration), while other parameters remain unchanged to separate the impact of each parameter on the temperature. The step weight is converted into a power adjustment step. For example, a step weight of 0.2 corresponds to a 5W step, so the power will increase or decrease by 5W during the next detection; it is converted into a thermal response inertia compensation coefficient, which is used to adjust the prediction model of temperature changes. The higher the coefficient, the stronger the compensation for temperature inertia (for example, the temperature will not rise immediately after heating, and the delay effect needs to be considered). After each parameter adjustment, it is immediately verified whether the intermittent power constraint and temperature error tolerance are violated. If the new parameters cause the power to exceed the upper limit, the intermittent time is too short, or the core temperature exceeds the fluctuation range, the current detection direction is terminated, roll back to the previous valid parameter state, and reduce the step weight to re-detect.

[0137] Step 402: Continuously iteratively adjust parameters along the search direction, while satisfying all constraints. Each iteration records the current parameter combination and its corresponding temperature response (e.g., core temperature change and surface temperature uniformity). When approaching the constraint boundary, a suboptimal solution (e.g., a parameter combination that slightly decreases surface temperature uniformity but reaches the core temperature faster) is accepted with a certain probability to prevent the algorithm from falling into a local optimum and expand the search range. All parameter combinations within the constraint boundary, where no other parameter combination optimizes both core and surface temperature uniformity, are retained to form a Pareto front. For example, if parameter combination A achieves the core temperature target but poor surface uniformity, while parameter combination B does the opposite, both are retained. The optimal power allocation for each heating zone is determined (e.g., 800W for zone 1 and 600W for zone 2). The heating-off time for each zone is planned (e.g., zone 1 heating for 3 seconds, off for 2 seconds, and so on in a loop). The power switching sequence of multiple zones is coordinated to avoid current surges caused by simultaneous startup (e.g., zone 1 shuts down 0.5 seconds before zone 2 starts up again). The generated parameter set is simulated and run, and the temperature change trend is predicted through historical data and heat conduction model. If the prediction result does not meet the tolerance requirements, the suboptimal solution is selected from the solution set and fine-tuned until a feasible spatial power allocation parameter set is generated.

[0138] Combining the dual constraints of intermittent power and temperature error, it ensures that power allocation not only meets the safe operation of the equipment (such as avoiding overload and high-frequency switching), but also improves the heating quality of food. The directional search algorithm guided by adaptive partial derivatives can quickly locate the final power parameters and shorten the temperature adjustment time. It is especially suitable for the rapid heating stage of thick food. The Pareto solution generation mechanism takes into account both the core temperature compliance and surface uniformity optimization, avoiding the trade-off problem caused by over-optimization of a single indicator (such as the core temperature meets the standard but the surface is burnt), and achieving a dual improvement in energy saving and quality. Constraints such as the minimum intermittent duration protect the hardware life and prevent premature damage to components such as relays; simulated annealing and rollback strategies ensure the stable operation of the system under complex working conditions and extend the equipment maintenance cycle.

[0139] In a preferred embodiment of the present invention, the above step 5, associating the spatial power allocation parameter set with the corresponding morphological feature vector and the periodic power parameter and storing them in the heat diffusion knowledge base, and automatically matching the spatial power allocation parameter set when a new food is detected, may include:

[0140] Step 500: Establish a primary correlation mapping between the spatial power allocation parameter set and the morphological feature vector; establish a secondary adjustment mapping between the spatial power allocation parameter set and the periodic power parameter;

[0141] Step 501: Using the real-time geometric shape factor in the morphological feature vector as the thermal field distribution reference key, using the real-time duty cycle dynamic range in the periodic power parameter as the power adjustment reference key, and associating and storing a set of spatial power allocation parameters, a heat diffusion knowledge base storage structure is constructed;

[0142] Step 502: When new food is detected, the real-time morphological feature vector and periodic power parameters of the new food are extracted, and the degree of fit of the thermal diffusion characteristics with the thermal field distribution benchmark key is calculated based on the real-time geometric shape factor, and the dynamic compatibility with the power regulation benchmark key is verified based on the real-time duty cycle dynamic range; the spatial power allocation parameter set that has passed the double verification is retrieved, and the partitioned radiation power intensity value, partitioned heating duration sequence and cross-zone power switching timing plan are loaded.

[0143] In an embodiment of the present invention, the spatial power allocation parameter set generated in step 402 (including partitioned radiation power intensity, heating duration sequence, power switching timing) is directly associated with the morphological feature vector (recording information such as the geometric shape, thickness distribution, surface curvature, etc. of the food), and each spatial power allocation parameter set corresponds to a unique morphological feature vector identifier (such as a feature code generated by a hash algorithm), ensuring a one-to-one correspondence between the parameter set and the food morphology.

[0144] During the association process, the morphological feature vectors are compressed and stored, redundant data (such as repeated edge area features) are removed, key feature points (such as curvature extreme points and thickness mutations) are retained, and the vector dimension is compressed from 100 to about 20 dimensions to save storage space. At the same time, an index relationship between the morphological feature vector and the original complete data is established. A secondary mapping is established between the spatial power allocation parameter set and the periodic power parameters (duty cycle, period value), which is used to fine-tune the power allocation strategy:

[0145] The duty cycle in the periodic power parameter determines the average power density, and the period value affects the power switching frequency. Both of them work together to determine the specific execution of the spatial power allocation parameter set.

[0146] The secondary mapping uses a hierarchical association approach:

[0147] The first layer: classified by duty cycle range (such as low: less than 30%, medium: 30%-70%, high: greater than 70%);

[0148] Second level: Within each duty cycle range, it is further subdivided by the period value (e.g. fast: less than 5 seconds, medium: 5-10 seconds, slow: greater than 10 seconds).

[0149] For example, when the spatial power allocation parameter set corresponds to a duty cycle of 60% and a cycle value of 8 seconds, the system stores it in the "Medium Duty Cycle-Medium Speed ​​Cycle" subdirectory and records the adjustment rules of the two for the partition power intensity and heating time (such as for every 10% increase in duty cycle, the power of partition 1 increases by 50W).

[0150] Step 501: Extract the real-time geometric shape factor from the morphological feature vector as the core identifier to characterize the thermal field distribution characteristics of the food. The geometric shape factor includes the curvature distribution information of convex, concave and flat surfaces. Divide the curvature range into multiple intervals (e.g. convex: 0.05-0.1mm -1 , concave: -0.1-0.05mm-1), each interval corresponds to a unique code. For example, the surface of a certain food is mainly convex, with a curvature concentrated between 0.06-0.08mm -1 , then the thermal field distribution reference key is "convex - 0.06-0.08mm -1 The real-time duty cycle dynamic range is extracted from the periodic power parameters as the core indicator of the power adjustment strategy. The duty cycle range is divided into several segments (such as 20%-40%, 40%-60%), and the power adjustment characteristics corresponding to each range are annotated based on historical data (for example, a duty cycle of 40%-60% is suitable for the constant temperature stage of thick food).

[0151] The heat diffusion knowledge base is stored in a tree-like hierarchical structure:

[0152] Root node: Classified by ingredient type (such as beef, chicken, fish);

[0153] First-level subnode: thermal field distribution benchmark key (geometric shape factor classification);

[0154] Secondary subnode: power adjustment reference key (duty cycle dynamic range);

[0155] Leaf node: stores the corresponding spatial power allocation parameter set.

[0156] For example, a piece of geometric shape is convex (curvature 0.07mm -1 ), the parameter set storage path for steak with a duty cycle dynamic range of 40%-60% is: Knowledge Base → Beef → Convexity - 0.05-0.1mm -1 →40%-60% duty cycle →Specific parameter set file.

[0157] Hash index tables are constructed for the thermal field distribution benchmark key and the power regulation benchmark key respectively, and the storage location of the parameter set corresponding to each key value is recorded, so that the target node can be directly located during retrieval.

[0158] Step 502: When a new ingredient is detected, its morphological features are collected in real time through sensors (such as laser scanners, infrared thermal imagers), and a real-time morphological feature vector is generated. The periodic power parameters (such as an initial duty cycle of 50% and a period of 10 seconds) are set according to the current heating demand. The real-time geometric shape factor of the new ingredient is extracted and matched with the thermal field distribution reference key. Based on the curvature distribution of the geometric shape factor, the Euclidean distance between the new ingredient and the reference key is calculated, and the distance is ≤ 0.01mm. -1 The real-time duty cycle dynamic range of the new food is extracted and compared with the power adjustment reference key. The verification logic includes:

[0159] Whether the duty cycle range fully or partially covers the reference key interval;

[0160] Whether the power regulation characteristics corresponding to the current duty cycle match the food requirements (e.g., thick food requires a high duty cycle to maintain heating);

[0161] Only when the thermal diffusion characteristics and dynamic compatibility are verified, the corresponding spatial power allocation parameter set is retrieved from the knowledge base. After retrieving the matching parameter set, the partitioned radiation power intensity value, partitioned heating duration sequence, and cross-zone power switching timing plan are loaded. During the loading process, the parameters are double-checked:

[0162] Check whether the power intensity exceeds the maximum power limit of the device;

[0163] Verify that the heating duration meets the minimum intermittent constraint.

[0164] If there is a conflict in parameters, the parameter fine-tuning mechanism will be automatically activated: the power intensity will be scaled proportionally (such as reducing the power of all zones by 10%), or the heating time will be adjusted (such as extending the shutdown time by 0.5 seconds) until the parameters meet the equipment operation requirements, and the adjusted parameters will then be applied to the heating process.

[0165] Fast adaptation and efficient heating: The heat diffusion knowledge base reduces the retrieval time for power allocation plans for new ingredients to less than 1 second through dual matching of morphological characteristics and power parameters, thus reducing preheating waiting time. Parameter matching based on geometric shape and duty cycle ensures that the thermal field distribution of ingredients of different shapes is highly consistent with the power adjustment strategy, avoiding problems such as partial over- or undercooking. The parameter sets accumulated in the knowledge base can be used as historical experience to guide subsequent adjustments. Parameter plans for new ingredients are automatically added to the database after practical verification, forming a "learning-application-optimization" closed loop. The verification and fine-tuning mechanism before parameter loading effectively avoids equipment overload or high-frequency switching, extending hardware life.

[0166] like Figure 2 As shown, the embodiment of the present invention also provides an energy-saving and heat-insulating adaptive dynamic control system for frying and grilling equipment, including:

[0167] The feature extraction module is used to collect food surface temperature distribution, thickness gradient and ambient temperature and humidity data in real time, extract three-dimensional thermal diffusion morphological features and generate initial morphological feature vectors;

[0168] A power parameter calculation module is used to calculate the pseudo partial derivative estimate based on the real-time deviation between the set target temperature and the measured food core temperature, combined with the morphological feature vector, and generate periodic intermittent power parameters;

[0169] A dynamic correction module is used to input the morphological feature vector into the dynamic compensation function to output the morphological correction coefficient, dynamically correct the pseudo partial derivative estimate based on the periodic intermittent power parameter, and generate an adaptive partial derivative;

[0170] The parameter optimization module is used to input the adaptive partial derivatives into the dynamic power allocator, optimize the step weight and correction strength through a directed parameter search algorithm based on the intermittent power constraint and temperature error tolerance, and generate a spatial power allocation parameter set;

[0171] The matching module is used to associate the spatial power allocation parameter set with the corresponding morphological feature vector and periodic power parameter and store them in the heat diffusion knowledge base, and automatically match the spatial power allocation parameter set when new food is detected.

[0172] It should be noted that this system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0173] An embodiment of the present invention further provides a computing device comprising: a processor and a memory storing a computer program, wherein the computer program, when executed by the processor, performs the above-described method. All implementations in the above-described method embodiments are applicable to this embodiment and can achieve the same technical effects.

[0174] The embodiment of the present invention further provides a computer-readable storage medium storing instructions, which, when executed on a computer, causes the computer to execute the above-described method. All implementations in the above-described method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0175] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for adaptive dynamic control of energy saving and heat preservation of frying and grilling equipment, characterized in that: The method comprises: Step 1: Real-time data collection of food surface temperature distribution, thickness gradient, and ambient temperature and humidity are collected to extract three-dimensional thermal diffusion morphological features and generate an initial morphological feature vector. Step 2: Based on the real-time deviation between the set target temperature and the measured food core temperature, combined with the morphological feature vector, a pseudo partial derivative estimate is calculated, and based on the pseudo partial derivative estimate, periodic intermittent power parameters, including duty cycle and period, are generated; Step 3: Input the morphological feature vector into the dynamic compensation function, output the morphological correction coefficient, and dynamically correct the pseudo partial derivative estimate based on the periodic intermittent power parameter to generate an adaptive partial derivative; Step 4: Input the adaptive partial derivatives into the dynamic power allocator, combine the intermittent power constraint and the temperature error tolerance, optimize the step weight and correction strength through the directed parameter search algorithm, and generate the spatial power allocation parameter set; Step 5: Associate the spatial power allocation parameter set with the corresponding morphological feature vector and periodic power parameter and store them in the heat diffusion knowledge base. When new food is detected, the spatial power allocation parameter set is automatically matched.

2. The adaptive dynamic control method for energy saving and heat preservation of grilling equipment according to claim 1 is characterized in that: Real-time collection of food surface temperature distribution, thickness gradient, and ambient temperature and humidity data, extraction of three-dimensional thermal diffusion morphological features, and generation of initial morphological feature vectors, including: The infrared thermal imaging array is used to collect the surface temperature distribution data of the food in real time, the laser thickness sensor is used to obtain the thickness gradient data of the food, and the temperature and humidity sensor is used to collect the ambient temperature and humidity data; Extract three-dimensional thermal diffusion morphological features based on surface temperature distribution, thickness gradient, and ambient temperature and humidity data; The three-dimensional thermal diffusion morphological features are fused to generate the initial morphological feature vector.

3. The adaptive dynamic control method for energy saving and heat preservation of grilling equipment according to claim 2 is characterized in that: The three-dimensional thermal diffusion morphological characteristics include: Geometric shape factor, i.e. the thermal diffusion rate parameter of the food surface curvature distribution; Thickness distribution factor, which is the heat conduction efficiency parameter of the thickness difference of different areas of food; The surface curvature factor is the parameter of convective heat transfer efficiency on the concave and convex surface of food.

4. The adaptive dynamic control method for energy saving and heat preservation of grilling equipment according to claim 3 is characterized in that: Based on the real-time deviation between the set target temperature and the measured food core temperature, combined with the morphological eigenvector, the pseudo partial derivative estimate is calculated. Based on the pseudo partial derivative estimate, the periodic intermittent power parameters, including duty cycle and period, are generated. Obtain the initial morphological feature vector and calculate the deviation between the set target temperature and the measured food core temperature in real time; The deviation value and the initial morphological feature vector are input into the pseudo partial derivative estimator, the thickness distribution factor in the initial morphological feature vector is processed to generate a thickness weight coefficient, and the thickness weight coefficient is fused with the historical deviation data. The dynamic weighted least squares algorithm is used for rolling optimization and the pseudo partial derivative estimate is output. The pseudo partial derivative estimate represents the intensity of the impact of unit power change on the core temperature. The estimated value of the pseudo partial derivative is input into the intermittent power generator to calculate the basic duty cycle, wherein the basic duty cycle is inversely proportional to the estimated value of the pseudo partial derivative; the geometric shape factor in the initial morphological feature vector is obtained, the curvature compensation is performed on the period value according to the geometric shape factor, and the periodic intermittent power parameters are output, including the duty cycle optimized by the pseudo partial derivative and the period value compensated by the geometric shape factor.

5. The adaptive dynamic control method for energy saving and heat preservation of grilling equipment according to claim 4 is characterized in that: The morphological feature vector is input into the dynamic compensation function, the morphological correction coefficient is output, and the pseudo partial derivative estimation value is dynamically corrected based on the periodic intermittent power parameter to generate the adaptive partial derivative, including: The morphological feature vector is input into the dynamic compensation function, and the geometric shape factor and the surface curvature factor are nonlinearly mapped using Gaussian kernel to generate the curvature-diffusion coupling coefficient; Based on the interaction between the coupling coefficient and the thickness distribution factor, the morphology correction coefficient is output to quantify the intensity of the compensation demand for thermal diffusion on the food surface; The periodic intermittent power parameters are obtained, and the power density characteristics in the duty cycle and the switching frequency characteristics in the period value are analyzed. The morphology correction coefficient and the power switching frequency characteristics are integrated to construct a dynamic weight matrix. The morphology correction coefficient is used as the main correction factor to perform variable gain correction on the pseudo partial derivative estimate. The adaptive partial derivative is output to represent the actual impact intensity of the unit power on the core temperature after compensation.

6. The adaptive dynamic control method for energy saving and heat preservation of grilling equipment according to claim 5, characterized in that: The adaptive partial derivatives are input into the dynamic power allocator. Combined with the intermittent power constraint and temperature error tolerance, the step weight and correction strength are optimized through a directed parameter search algorithm to generate a spatial power allocation parameter set, including: Input the adaptive partial derivatives into the dynamic power allocator, use the adaptive partial derivatives as the gradient descent direction reference, and initialize the parameter search path; Combined with intermittent power constraints, including the maximum power threshold limit and the minimum intermittent duration constraint, and temperature error tolerance, including the core temperature fluctuation range and the surface temperature uniformity threshold, axial detection is performed along the adaptive partial derivative direction through a directional parameter search algorithm, where the step weight is mapped to the power adjustment step amount and the correction intensity is mapped to the thermal response inertia compensation coefficient. During the axial detection process, a Pareto final solution set is generated within the constraint boundary, which includes the partition radiation power intensity value, partition heating duration sequence and cross-zone power switching timing scheme.

7. The adaptive dynamic control method for energy saving and heat preservation of grilling equipment according to claim 6, characterized in that: The spatial power allocation parameter set is associated with the corresponding morphological feature vector and periodic power parameter and stored in the heat diffusion knowledge base. When a new food is detected, the spatial power allocation parameter set is automatically matched, including: Establishing a primary correlation mapping between the spatial power allocation parameter set and the morphological feature vector; establishing a secondary adjustment mapping between the spatial power allocation parameter set and the periodic power parameter; The real-time geometric shape factor in the morphological feature vector is used as the thermal field distribution benchmark key, the real-time duty cycle dynamic range in the periodic power parameter is used as the power regulation benchmark key, and the spatial power allocation parameter set is associated and stored to construct a heat diffusion knowledge base storage structure; When new food is detected, the real-time morphological feature vector and periodic power parameters of the new food are extracted, and the degree of fit of the thermal diffusion characteristics with the thermal field distribution benchmark key is calculated based on the real-time geometric shape factor, and the dynamic compatibility with the power regulation benchmark key is verified based on the real-time duty cycle dynamic range; the spatial power allocation parameter set that has passed the double verification is retrieved, and the partitioned radiation power intensity value, partitioned heating duration sequence and cross-zone power switching timing plan are loaded.

8. An adaptive dynamic control system for energy saving and heat preservation of grilling equipment, the system implementing the method according to any one of claims 1 to 7, characterized in that: include: The feature extraction module is used to collect food surface temperature distribution, thickness gradient and ambient temperature and humidity data in real time, extract three-dimensional thermal diffusion morphological features and generate initial morphological feature vectors; A power parameter calculation module is used to calculate the pseudo partial derivative estimate based on the real-time deviation between the set target temperature and the measured food core temperature, combined with the morphological feature vector, and generate periodic intermittent power parameters; A dynamic correction module is used to input the morphological feature vector into the dynamic compensation function to output the morphological correction coefficient, dynamically correct the pseudo partial derivative estimate based on the periodic intermittent power parameter, and generate an adaptive partial derivative; The parameter optimization module is used to input the adaptive partial derivatives into the dynamic power allocator, optimize the step weight and correction strength through a directed parameter search algorithm based on the intermittent power constraint and temperature error tolerance, and generate a spatial power allocation parameter set; The matching module is used to associate the spatial power allocation parameter set with the corresponding morphological feature vector and periodic power parameter and store them in the heat diffusion knowledge base, and automatically match the spatial power allocation parameter set when new food is detected.

9. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images