Full-automatic pet health care product processing system and method based on AI optimization

Abstract

The invention relates to the technical field of pet health care product production and processing, and discloses a full-automatic pet health care product processing system and method based on AI optimization, and the full-automatic pet health care product processing method based on AI optimization comprises the following steps: constructing a pulsed electric field assisted low-temperature sterilization system; establishing a microbial electric field sensitivity database to realize accurate customization of electric field parameters; thermal response characteristics of different active components in the pet health care product are analyzed, a thermal diffusion equation inversion algorithm is constructed, and accurate reconstruction of an internal temperature field of the material is achieved; a Raman spectrum technology and a thermal imaging system are combined to realize real-time monitoring of the content of the active ingredients, and an active ingredient thermal degradation kinetic model is established; in combination with a multi-target Pareto optimization method, the optimal balance between the sterilization effect and the retention rate of different active ingredients is realized; the method solves the technical problem of serious loss of active ingredients in the traditional heat sterilization process, and improves the retention rate of heat-sensitive ingredients in the pet health care product.

Description

Technical Field

[0001] The present invention relates to the technical field of pet health care product production and processing, and more specifically, to a fully automatic pet health care product processing system and method based on AI optimization. Background Art

[0002] With the expansion of the pet market and the increasing attention of pet owners to pet health, the demand for pet health products is growing. Pet health products usually contain a variety of active ingredients, such as vitamins, probiotics, enzyme preparations, etc. These ingredients are sensitive to temperature and are easily inactivated or degraded during traditional thermal processing.

[0003] Traditional thermal sterilization technology is commonly used in the production process of pet health products. Although it can effectively kill microorganisms, it will significantly reduce the content of active ingredients, resulting in reduced product efficacy. Different active ingredients in pet health products have different sensitivities to heat, and traditional single temperature control strategies cannot ensure the activity of multiple ingredients at the same time. Traditional thermal processing technology is difficult to accurately control local temperature distribution, and cannot monitor changes in active ingredient content in real time, resulting in serious loss of nutrients during processing.

[0004] Although non-thermal sterilization technologies such as high pressure treatment and ultraviolet irradiation can reduce the loss of active ingredients to a certain extent, they have disadvantages such as complex equipment, poor processing uniformity, high cost, and difficulty in precise regulation for pet health products with complex formulas. Therefore, a pet health product processing method is needed that can maximize the retention of multiple heat-sensitive active ingredients while ensuring the sterilization effect. Summary of the invention

[0005] The present invention provides a fully automatic pet health product processing system and method based on AI optimization, which solves the technical problem of the technical contradiction between sterilization and maintenance of active ingredients of pet health products in related technologies.

[0006] The present invention provides a fully automatic pet health product processing method based on AI optimization, comprising the following steps: Construct a pulse electric field assisted low temperature sterilization system, including a pulse electric field generating unit, a temperature control unit, a product flow unit and a control unit; Establish a microbial electric field sensitivity database to achieve precise customization of electric field parameters based on the electric field sensitivity characteristics of common microorganisms in pet health products; Based on the microbial electric field sensitivity database, the thermal response characteristics of different active ingredients in pet health products are analyzed, and the inversion algorithm of the heat diffusion equation is constructed to achieve accurate reconstruction of the internal temperature field of the material; Using the inversion results of the heat diffusion equation, combined with Raman spectroscopy technology and thermal imaging system, the active ingredient content is monitored in real time, and a thermal degradation kinetic model of the active ingredient is established; Based on the thermal degradation kinetics model of active ingredients and the electric field sensitivity data of microorganisms, computational intelligence algorithms are used to simultaneously optimize the electric field parameters and auxiliary temperature. Combined with the multi-objective Pareto optimization method, the optimal balance between the bactericidal effect and the retention rate of different active ingredients is achieved.

[0007] In a preferred embodiment, the construction of the pulse electric field assisted low temperature sterilization system includes: Design the electrode configuration, using parallel plate electrodes, coaxial cylindrical electrodes or finger-shaped electrode structures; Constructing a pulse electric field generating unit, including a high voltage pulse power supply and a pulse shaping circuit; Design a temperature control unit, including a circulating water bath, a temperature sensor array, and a PID controller; Build product flow units, including peristaltic pumps, piping, and agitation devices; Build a control unit based on an embedded system, integrate the data acquisition module, parameter adjustment module and safety monitoring module to achieve fully automatic operation of the system.

[0008] In a preferred embodiment, the pulse electric field generating unit includes a high voltage pulse power supply, a multi-stage electrode chamber and a parameter adjustment module; The temperature control unit includes a precision temperature control device and a temperature sensor array; The product flow unit includes a peristaltic pump, a fluid pipeline and a processing chamber; The control unit includes a main controller, a data acquisition module and a human-computer interaction interface; The control unit coordinates the work of each unit to achieve coordinated control of electric field parameters and temperature.

[0009] In a preferred embodiment, the pulse electric field assisted low temperature sterilization system selectively destroys microbial cell membranes through the electroporation effect of microbial cell membranes.

[0010] In a preferred embodiment, the establishment of a microbial electric field sensitivity database comprises: The sensitivity data of common microorganisms in pet health products to different electric field parameters were collected and analyzed, and an electric field sterilization kinetic model was established. The electric field sterilization kinetic model is expressed as: ; in, represents the electric field sterilization kinetic model, N represents the number of microorganisms after treatment, represents the initial number of microorganisms, k represents the microbial inactivation rate constant, t represents the treatment time, E represents the electric field strength, m and n represent the exponential parameters of time and electric field strength, respectively. The logarithmic ratio indicating the number of microorganisms decreases with increasing treatment time and electric field strength.

[0011] In a preferred embodiment, the thermal response characteristics analysis includes: Differential scanning calorimetry and thermogravimetric analysis were used to determine the thermal stability parameters of the active ingredients, and a model of the relationship between thermal degradation rate and temperature was established: ; in, The model representing the relationship between thermal degradation rate and temperature is: is the concentration of active ingredient, is the pre-exponential factor, is the activation energy, is the gas constant, is the absolute temperature, Indicates the degree to which the reaction depends on concentration; The heat diffusion equation inversion algorithm uses nonlinear least squares method combined with finite element analysis to invert the internal temperature field distribution based on surface temperature measurement data. The heat diffusion equation inversion algorithm is implemented by solving the equation: ; in, represents the inversion algorithm of the heat diffusion equation, is the thermal diffusion coefficient, is the internal heat source term, is the density, is the specific heat capacity, is the material temperature, is the Laplace operator.

[0012] In a preferred embodiment, the Raman spectroscopy technology and thermal imaging system realize real-time monitoring of active ingredient content, including: Use portable Raman spectrometer to conduct online detection of active ingredients in pet health products and collect spectral data; The multivariate calibration model was used to establish the quantitative relationship between Raman spectra and the content of active ingredients. The model expression is: ; in, is the active ingredient content, is the intercept, For the The regression coefficient at the wavelength is For the The Raman spectrum intensity at wavelengths is is the error term, is the number of spectral data points included in the model; Use infrared thermal imager to construct product surface temperature field distribution; Combining Raman spectroscopy analysis results with thermal field distribution information, a thermal degradation kinetic model of active ingredients is established: ; in, represents the natural logarithm of the active ingredient retention rate, is the initial active ingredient content, For time The active ingredient content after is the frequency factor, For processing time, is the activation energy, is the gas constant, is the absolute temperature, It means that the content of active ingredients decreases with the increase of treatment time; The retention rate of active ingredients under different process parameters is predicted in real time based on the thermal degradation kinetic model of active ingredients, providing feedback data for the optimization algorithm.

[0013] In a preferred embodiment, the computational intelligence algorithm simultaneously optimizes the electric field parameters and the auxiliary temperature, including: Construct a multi-objective optimization model, taking into account the following optimization objectives: Maximize microbial kill rate: ; Maximize active ingredient retention: ; Minimize energy consumption: ; in, represents the maximum microbial killing rate, It means maximizing the retention rate of active ingredients. represents minimizing energy consumption, represents the logarithmic function, and are the number of microorganisms before and after treatment, and After treatment and before treatment, respectively. The content of active ingredients, is the weight coefficient, is power, For processing time, Indicates the total amount of active ingredients; An improved non-dominated sorting genetic algorithm is used to solve multi-objective optimization problems; Based on non-dominated sorting and crowding distance calculation, the Pareto optimal solution set is obtained to provide a variety of options for decision-making; According to product characteristics and quality requirements, select the electric field parameter combination that best meets current needs from the Pareto frontier.

[0014] In a preferred embodiment, the optimization method further includes constructing an adaptive parameter optimization control system, and the adaptive parameter optimization control system includes: Based on the Pareto optimal solution set and combined with real-time monitoring data, the deep reinforcement learning method is used to achieve dynamic optimization of processing parameters; The state space of the control system includes current product characteristic parameters, real-time monitored microbial inactivation rate and active ingredient retention rate; The action space includes the adjustment amount of electric field parameters and auxiliary temperature; The reward function is defined as a weighted combination of microbial kill rate, active ingredient retention rate, and energy consumption; Use deep Q network to realize the mapping from state to action; Through experience replay and target network technology, the learning stability and efficiency are improved, and the adaptive optimization of parameters in the processing of pet health products is achieved.

[0015] In a preferred embodiment, a fully automatic pet health product processing system based on AI optimization is used to perform a fully automatic pet health product processing method based on AI optimization, including: A pulse electric field generating unit, used to generate an adjustable high-voltage pulse electric field, including a high-voltage pulse power supply, a multi-stage electrode chamber and a parameter adjustment module; a temperature control unit, used to accurately adjust and maintain the processing temperature, including a precision temperature control device and a temperature sensor array; A product flow unit, which is used to control the flow state of the pet health product in the processing chamber, and includes a peristaltic pump, a fluid pipeline and a processing chamber; Monitoring and analysis unit, including Raman spectrometer and infrared thermal imager, used to monitor the active ingredient content and temperature field distribution in real time; The control unit, including the main controller, data acquisition module and human-computer interaction interface, integrates deep reinforcement learning algorithm to achieve adaptive optimization control of system parameters.

[0016] The beneficial effects of the present invention are: Improve the retention rate of heat-sensitive active ingredients: Use pulsed electric field-assisted low-temperature sterilization instead of traditional heat sterilization to reduce the processing temperature. While ensuring that the microbial killing rate reaches 5 logarithmic levels, it also improves the retention rate of heat-sensitive active ingredients.

[0017] Achieve differentiated protection for multiple heat-sensitive ingredients: By constructing a thermal degradation kinetic model of active ingredients and a multi-objective optimization algorithm, the technical difficulty of simultaneously preserving multiple heat-sensitive ingredients is solved, reducing fluctuations in active ingredient content between product batches.

[0018] Improve processing efficiency and energy utilization: Based on the technical route combining real-time monitoring and precise control, dynamic optimization of processing parameters is achieved. Compared with traditional thermal processing technology, energy consumption is reduced and processing time is shortened.

[0019] Intelligent and precise processing: By introducing an adaptive parameter optimization control system built using deep reinforcement learning methods, precise processing technology with the goal of "ingredient retention" is realized, transforming the processing of pet health products from the traditional "sterilization-centered" to "active ingredient preservation-centered", thereby improving the effectiveness and quality stability of the products. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of a fully automatic pet health product processing method based on AI optimization of the present invention. DETAILED DESCRIPTION

[0021] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is only to enable those skilled in the art to better understand and implement the subject matter described herein, and the functions and arrangements of the elements discussed may be changed without departing from the scope of protection of the present specification. Various examples may omit, replace, or add various processes or components as needed. In addition, the features described in some examples may also be combined in other examples.

[0022] At least one embodiment of the present invention discloses a fully automatic pet health product processing method based on AI optimization, such as Figure 1 As shown, the following steps are included: Step 1, constructing a pulse electric field assisted low-temperature sterilization system, including a pulse electric field generating unit, a temperature control unit, a product flow unit and a control unit; Among them, the pulse electric field generating unit includes a high-voltage pulse power supply, a multi-stage electrode chamber and a parameter adjustment module; the temperature control unit includes a precision temperature control device and a temperature sensor array; the product flow unit includes a peristaltic pump, a fluid pipeline and a processing chamber; the control unit includes a main controller, a data acquisition module and a human-computer interaction interface.

[0023] The pulse electric field generating unit may also include an impedance matching circuit for improving energy transmission efficiency; In some embodiments, the temperature control unit may use a water circulation jacket or microwave-assisted heating method, which can be selected according to the thermal sensitivity of different products; for example, for high-viscosity paste pet health products, a spiral extrusion product flow unit can be used instead of a peristaltic pump to improve flow uniformity.

[0024] In this step, the electroporation effect of the microbial cell membrane under the action of the pulsed electric field is the core mechanism for achieving low-temperature sterilization. The electroporation process can be represented by the following mathematical model: ; in, It represents the membrane perforation density per unit area; represents the rate of change of perforation density over time; represents the applied electric field strength; represents the activation energy of cell membrane electroporation; represents the gas constant; represents absolute temperature; represents the rate constant; represents the electric field dependence index; represents the maximum possible perforation density; It represents the membrane perforation density per unit area; Represents an exponential function.

[0025] It includes the following sub-steps: Step 1.1, determine the optimal electric field parameter combination; According to one embodiment of the present application, by analyzing the electroporation characteristics of common microorganisms in pet health products, it is determined that the electric field strength range is 15-25 kV / cm, the pulse width is 1-5 μs, the frequency is 100-500 Hz, and the number of pulses is 20-50.

[0026] For certain types of pet health products, such as products containing high concentrations of fat, the electric field strength can be appropriately increased to 30kV / cm; while for products containing beneficial microorganisms such as probiotics, a lower electric field strength (10-15kV / cm) and a higher auxiliary temperature (50-55°C) can be used to achieve selective killing of harmful microorganisms.

[0027] Step 1.2, configuring a multi-level electrode structure; A multi-level electrode structure combining parallel plates and coaxial columns can be used to ensure uniformity of electric field distribution. Medical grade titanium alloy is used as the electrode material, and the electrode spacing can be adjusted within the range of 5-15mm to adapt to different product characteristics.

[0028] In some embodiments, the electrode surface may be coated with a conductive polymer to reduce the interfacial reaction between the electrode and the product; for example, for acidic pet health products, platinum-iridium alloy electrodes may be used instead of titanium alloys to improve corrosion resistance; the electrode structure may also use an array of staggered rod electrodes, which is suitable for processing granular or block pet health products.

[0029] Step 1.3, constructing an electric field-temperature coordinated control system; In this embodiment, the electric field parameters are integrated with the auxiliary temperature control system to form a synergistic control model to maintain the temperature in the range of 40-50°C during the treatment process, which is lower than the 70-90°C required for traditional thermal sterilization, thereby reducing the loss of heat-sensitive components.

[0030] For particularly heat-sensitive products, a pulsed temperature control strategy can be used, that is, a short-term temperature increase during the application of the electric field and rapid cooling after the electric field is removed to minimize the overall thermal history of the product.

[0031] In some embodiments, a flow field control module may be integrated into the temperature control system to further improve processing uniformity through swirl or directional shear flow.

[0032] Step 2: Establish a microbial electric field sensitivity database, and accurately customize electric field parameters based on the electric field sensitivity characteristics of common microorganisms in pet health products; It includes the following sub-steps: Step 2.1, establishing a microbial electric field sensitivity model; Collect and analyze the sensitivity data of common microorganisms in pet health products to different electric field parameters, and establish electric field sterilization kinetic models for various microorganisms: ; in, represents the electric field sterilization kinetic model, Indicates the number of microorganisms after treatment; represents the initial number of microorganisms; represents the microbial inactivation rate constant; Indicates processing time; Indicates the electric field strength; represents the time index parameter; represents the electric field intensity index parameter; The logarithmic ratio indicating the number of microorganisms decreases with increasing treatment time and electric field strength.

[0033] For microorganisms in different physiological states, environmental factor correction terms can be introduced, and the corrected model is: ; in, represents the modified electric field sterilization kinetic model, represents the microbial inactivation rate constant; Indicates processing time; Indicates the electric field strength; represents the time index parameter; represents the electric field intensity index parameter; Indicates the correction factor of environmental conditions on the sterilization effect of electric field; Indicates the pH of the treatment environment; Indicates water activity; Indicates the processing temperature; The logarithmic ratio indicating the number of microorganisms decreases with increasing treatment time and electric field strength.

[0034] For example, under acidic conditions, the sensitivity of most pathogens to electric fields increases, and the same bactericidal effect can be achieved by reducing the electric field intensity, thereby further reducing the impact on the active ingredients.

[0035] The implementation methods of the electric field sterilization kinetic model include: The target microbial strain is inoculated into a suitable culture medium and cultured to the logarithmic growth phase; The bacterial solution was divided into a special electroporation sample pool and subjected to pulsed electric field treatment with different intensities (15-25 kV / cm), different pulse widths (1-5 μs) and different frequencies (100-500 Hz); Viable bacteria were counted immediately after treatment, and the survival rate of microorganisms under different treatment conditions was recorded; Determine model parameters through nonlinear regression analysis , and The value of .

[0036] In practical applications, taking Salmonella, which is common in pet health products, as an example, the model parameters determined by the above method are: , , According to this parameter, it can be calculated that under the conditions of an electric field strength of 20 kV / cm, a pulse width of 3 μs, and a pulse number of 30, the killing rate of Salmonella can reach 4.5 logarithmic levels (i.e., 99.997%).

[0037] The model parameters are different for different microbial species. The database established in this application contains a complete parameter set of 20 common microorganisms, providing basic data support for subsequent parameter optimization and control.

[0038] In some embodiments, the database can be connected to a cloud data center to achieve data sharing and remote updates, enabling the system to respond to the challenges of new pathogens; for example, by introducing machine learning methods, the system can predict the electric field sensitivity parameters of new strains based on the basic characteristics of new strains and the electric field response data of existing strains, thereby quickly forming a preliminary treatment plan.

[0039] Step 2.2, optimizing the electrode structure to improve processing uniformity; Based on the finite element analysis method, the electric field distribution under different electrode geometries and arrangements is simulated, and an optimized electrode structure that can achieve uniform electric field distribution is constructed based on the rheological properties of pet health products. The optimized electrode structure can control the coefficient of variation of the electric field intensity in the treatment area within 5%, ensuring the consistency of the sterilization effect.

[0040] The specific implementation methods of the electrode structure optimization of the present application include: The three-dimensional electric field distribution model was established using COMSOL Multiphysics software to parameterize the characteristics of the electrode chamber, processing chamber, and product fluid. Set multiple electrode structure parameters as variables and run the electric field distribution simulation; The spatial distribution of the electric field intensity in the processing chamber is calculated to evaluate the electric field uniformity under different electrode configurations.

[0041] Different electrode structures can be used for different product forms: For example, for liquid pet health products with good fluidity, continuous flow tubular electrodes can be used; For viscous gel-like products, a flat plate electrode with pulsating flow can be used; For granular products, an electrode system that uses a vibrating bed and an electric field in conjunction can be used.

[0042] In some embodiments, the electrode surface may also be processed at the nanoscale to reduce bubble formation on the electrode surface and improve the uniformity of the electric field distribution.

[0043] In practical applications, the present application optimizes the above method to obtain an "interlaced coaxial double-ring electrode structure", the specific structure of which includes: titanium alloy ring electrodes with an inner ring diameter of 45mm and an outer ring diameter of 80mm, the two rings are fixed by an insulating bracket, the axial spacing between adjacent electrode pairs is 12mm, and the polarities of adjacent electrode pairs are interlaced.

[0044] Step 2.3, integrating bioimpedance analysis technology for real-time monitoring; The bioimpedance analysis module is integrated into the treatment system to monitor the microbial inactivation rate in real time by measuring the changes in impedance parameters of the product before and after treatment.

[0045] Impedance measurement uses a four-electrode method with a frequency range of 1kHz-1MHz. A prediction model for the bactericidal effect is established based on the impedance spectrum changes: ; in, Indicates the impedance value after treatment; represents the initial impedance value; is the proportionality coefficient; Indicates the number of microorganisms before treatment; Indicates the number of microorganisms after treatment; It represents the log reduction value of microorganisms.

[0046] Through the sterilization effect prediction model, the sterilization effect can be evaluated in real time without interrupting production, and the processing parameters can be dynamically adjusted according to the evaluation results.

[0047] The specific implementation methods of bioimpedance analysis technology include: A four-electrode impedance measurement probe is installed at the inlet and outlet of the processing chamber, two outer electrodes are used to apply a weak alternating current, and two inner electrodes are used to measure the voltage response; Use an impedance analyzer to scan and measure the impedance at different frequencies to obtain an impedance spectrum; The microbial inactivation rate is calculated by extracting the impedance value changes at the characteristic frequency points and combining the pre-established microbial quantity and impedance relationship model.

[0048] Temperature and conductivity compensation algorithms can be introduced into the bioimpedance analysis system to eliminate the influence of non-biological factors on impedance measurement; For example, by measuring the impedance change of the reference channel, baseline drift caused by temperature increase can be excluded.

[0049] In some embodiments, impedance analysis can also be combined with other online detection technologies to construct a multi-parameter fusion microbial activity assessment model to further improve monitoring accuracy.

[0050] In actual applications, by real-time monitoring of impedance changes, fluctuations in treatment effects can be detected within 2 seconds. For example, when the impedance change rate of a batch of products is detected to be lower than the preset threshold (indicating insufficient sterilization effect), the system will automatically increase the electric field strength or extend the treatment time to ensure that the expected sterilization effect is always achieved while avoiding energy waste and loss of active ingredients caused by excessive treatment.

[0051] Step 3: Based on the microbial electric field sensitivity database, analyze the thermal response characteristics of different active ingredients in pet health products, construct an inversion algorithm for the heat diffusion equation, and achieve accurate reconstruction of the internal temperature field of the material; It includes the following sub-steps: Step 3.1, constructing a pulse infrared thermal wave analysis system; The pulse infrared thermal wave analysis system includes a pulse heat source, an infrared thermal imager, a data acquisition unit and an analysis and processing unit.

[0052] By applying a pulsed heat source to the sample, a high-precision infrared thermal imager (temperature resolution ≤ 0.05 °C, spatial resolution ≤ 100 μm) was used to record the changes in sample surface temperature over time with a sampling frequency of 50-200 Hz.

[0053] Pulse heat sources can be excited in different ways, for example, infrared laser pulses can be used for liquid products, flash lamps or electromagnetic induction heating can be used for solid or semi-solid products; In some embodiments, the thermal imager can be equipped with detectors of different wavelength ranges to meet the detection requirements of different temperature ranges; For example, for products that need to be accurately monitored in the range of 40-60°C, a 3-5μm medium-wave infrared detector can be used to improve the detection accuracy of the temperature range.

[0054] Step 3.2, establish the inversion algorithm of heat diffusion equation; Based on the heat diffusion equation: ; in, represents the heat diffusion equation, Indicates material temperature; Indicates time; represents the thermal diffusion coefficient; represents the Laplace operator; represents the heat source function; Indicates the material density; Represents specific heat capacity.

[0055] The inverse heat conduction algorithm is used to infer the internal temperature distribution from the time evolution of the surface temperature field.

[0056] The specific implementation includes: Establish a forward heat conduction model to simulate the temperature field distribution under given thermal physical parameters and thermal boundary conditions; Construct the optimization objective function: ; in, represents the optimization objective function, represents the set of thermophysical parameters to be inverted, Indicates the measured temperature value. Indicates the calculated temperature value, and Respectively represent the position and time of the measurement point; The Levenberg-Marquardt algorithm is used to solve the parameter set that minimizes the objective function and realize the reconstruction of the internal temperature field. The Levenberg-Marquardt algorithm is an optimization method for solving nonlinear least squares problems, which combines the advantages of the gradient descent method and the Gauss-Newton method.

[0057] The specific implementation of the heat diffusion equation inversion algorithm is as follows: The heat diffusion equation is discretized by finite difference and the Crank-Nicolson implicit scheme is used to ensure numerical stability. Establish a sensitivity matrix between surface temperature and internal temperature to indicate the influence of internal temperature change on surface temperature measurement value; Based on the measured surface temperature time series data, the optimal thermophysical parameters and internal heat source distribution are found through iteration; The forward heat conduction calculation is performed again using the optimized parameters to obtain the temperature distribution in the entire spatial domain.

[0058] To handle products with complex geometries, the inversion algorithm can use the finite element method instead of the finite difference method to improve spatial resolution and boundary processing capabilities; For example, for irregularly shaped pet chews, tetrahedral meshing can more accurately describe the internal temperature distribution.

[0059] In some embodiments, a Bayesian inference framework may be introduced to consider measurement noise and model uncertainty, provide a confidence interval for temperature field reconstruction, and make control decisions more reliable.

[0060] For real-time control situations with high requirements for computing efficiency, a simplified inversion model can be selected. For example, the characteristic time method can be used to quickly estimate the effective thermal diffusion coefficient of the product, calculate the internal temperature distribution based on the analytical solution, and achieve millisecond-level temperature field reconstruction.

[0061] The heat diffusion equation inversion algorithm was used to analyze the internal temperature distribution of compound pet health products containing multiple vitamins during electric field treatment. By applying a 20ms infrared pulse heat source on the product surface and recording the surface temperature response curve within 400ms, the three-dimensional temperature field at a depth of 1mm to 5mm inside the product was successfully reconstructed using the above inversion algorithm, with a temperature reconstruction accuracy of ±0.2℃.

[0062] Step 3.3, establish product thermal characteristics database; For different pet health product formulas, the thermal physical properties such as thermal diffusivity, specific heat capacity and thermal conductivity are measured to establish a product thermal properties database.

[0063] By combining the product thermal characteristics database with the inversion algorithm, the internal temperature field of any new formula product during the processing can be predicted with an accuracy of ±0.5℃.

[0064] The specific implementation methods of the product thermal characteristics database in this application include: Laser Flash Method (LFM) was used to measure the thermal diffusivity of pet health care product samples with different composition ratios and water contents. The sample thickness was 2 mm and the measurement temperature range was 10-80°C. Differential Scanning Calorimetry (DSC) was used to measure the specific heat capacity of the samples; Calculate the thermal conductivity of the sample based on thermal diffusivity, specific heat capacity and density; A multivariate correlation model is established between the thermophysical property parameters and the component ratio, water content and temperature of the sample to form a parameterized thermal properties database.

[0065] In order to improve the accuracy and applicability of the database, artificial neural network (ANN) can be used to replace the traditional multiple regression model to better capture the nonlinear interactions between components; For example, a three-layer feedforward neural network is used to predict thermophysical property parameters with protein, fat, carbohydrate, moisture content and temperature as input features.

[0066] In some embodiments, the thermal characteristics database can also be adaptively updated according to changes in product batches, and the model parameters can be continuously optimized through online measured thermal response data to ensure the long-term stability of prediction accuracy.

[0067] Step 4, using the inversion results of the heat diffusion equation, combined with Raman spectroscopy technology and thermal imaging system to achieve real-time monitoring of the active ingredient content, and establish a thermal degradation kinetic model of the active ingredient; It includes the following sub-steps: Step 4.1, constructing an active ingredient detection system based on Raman spectroscopy; The active ingredient detection system based on Raman spectroscopy uses a Raman spectrometer with a laser wavelength of 785nm and a spectral resolution of ≤2cm - ¹, combined with fiber optic probe to achieve online detection.

[0068] For the key active ingredients in pet health products (such as proteins, peptides, vitamins, amino acids, etc.), a Raman characteristic peak database is established, and the quantitative relationship between the active ingredient content and the Raman spectrum intensity is established through multivariate analysis methods: ; in, Indicates The content of active ingredients, The wave number is The Raman spectrum intensity at represents the regression coefficient, represents the intercept, represents the number of characteristic wavenumber points used for modeling, Indicates the index of the wave number point.

[0069] For different types of pet health products, different spectral preprocessing methods can be used to improve prediction accuracy; for example, For products containing fat, the extended multivariate scattering calibration method can be used to eliminate the fluorescence interference of fat; For darker colored products, standard normal variable transformation methods can be used to eliminate scatter effects.

[0070] In some embodiments, a variable screening algorithm may be used to select the most representative spectral variables, reduce redundant information, and improve model robustness.

[0071] The specific implementation methods of the Raman spectroscopy active ingredient detection system include: Select representative pet health product samples covering different content ranges of target active ingredients, and use HPLC to accurately determine the content of each active ingredient in the samples as calibration standards; The sample was scanned using a 785 nm laser Raman spectrometer to obtain the 400-3000 cm - ¹Raman spectra in the range; Preprocess the spectral data, including baseline correction, smoothing, normalization and other operations; The partial least squares regression algorithm was used to establish a quantitative relationship model between the spectral intensity at the characteristic wavenumber and the active ingredient content.

[0072] In specific applications, this system focuses on several types of heat-sensitive active ingredients in pet health products: vitamin C (characteristic peak at 1690cm - ¹), collagen peptide (characteristic peak is located at 1242-1272cm - ¹), probiotics (characteristic peak at 1002cm - ¹) and trace element chelates (characteristic peaks at 550-600cm - ¹).

[0073] By installing the fiber optic probe at the observation window of the processing chamber, the system can collect the Raman spectrum signal of the product in real time during the electric field treatment process, complete a scan and analysis every 2 seconds, and calculate the current content of each active ingredient.

[0074] Step 4.2, establishing a thermal degradation kinetic model of the active ingredient; For various thermosensitive active ingredients in pet health products, thermal degradation experiments were conducted under different temperature conditions (40-90°C), the changes in the content of active ingredients over time were measured, and a thermal degradation kinetic model was established: ; in, represents the rate of change of the active ingredient concentration C with time t, Indicates the concentration of active ingredient, Indicates time, represents the frequency factor, represents the activation energy of the thermal degradation reaction, is the gas constant, represents absolute temperature, represents the reaction order, Represents an exponential function.

[0075] According to the degradation characteristics of different active ingredients, the retention rate of active ingredients under any temperature and time combination can be obtained by solving the above equation: ; in, Indicates the current active ingredient concentration, represents the initial active ingredient concentration, Indicates the retention rate of active ingredients, represents the frequency factor, represents the activation energy of the thermal degradation reaction, is the gas constant, represents absolute temperature, Indicates the duration of the thermal degradation process, It represents the effect of temperature on the reaction rate, which conforms to the Arrhenius equation.

[0076] For some active ingredients with complex degradation behaviors, a multi-stage degradation model can be used to more accurately describe their thermal degradation process; for example, the degradation of vitamin B12 in aqueous solution can be represented as a two-stage process: an initial rapid degradation stage and a subsequent slow degradation stage, each stage having different kinetic parameters.

[0077] In some embodiments, the effects of environmental factors such as pH value, ionic strength, and redox potential on the stability of the active ingredient may also be considered to establish an extended thermal degradation model: ; in, It represents the rate of change of the active ingredient concentration over time. Indicates the concentration of active ingredient, Indicates the duration of the thermal degradation process, represents the frequency factor, represents the activation energy of the thermal degradation reaction, is the gas constant, represents absolute temperature, represents the reaction order, is the environmental factor correction function, Indicates the pH of the solution. represents the ionic strength, represents the redox potential, Represents an exponential function.

[0078] The specific implementation of the thermal degradation kinetic model includes: Extracting the target active ingredients in pet health products and preparing them into solutions or suspensions; The samples were divided into sealed reaction tubes, placed in a precision thermostatic water bath, and kept warm at six temperatures: 40°C, 50°C, 60°C, 70°C, 80°C, and 90°C; Sampling is performed at predetermined time intervals, and the content of active ingredients in the samples is determined using liquid chromatography or enzyme-linked immunosorbent assay (ELISA); The data of the measured active ingredient content changing with time were substituted into the thermal degradation kinetic equation, and the kinetic parameters of each active ingredient were determined by nonlinear fitting.

[0079] The thermal degradation kinetic parameters of several key active ingredients in a pet nutritional supplement were determined. The results showed that the activation energy of vitamin C was 68.5 kJ / mol and the frequency factor was 5.6×10 6 min - ¹, the reaction order is 1; the activation energy of taurine is 75.3 kJ / mol, and the frequency factor is 2.3×10 7 min - ¹, the reaction order is 1.2; the activation energy of probiotic activity is 82.7 kJ / mol, and the frequency factor is 4.5×10 8 min - ¹, the reaction order is 1.5.

[0080] Based on these parameters, the system can accurately predict the retention rate of each active ingredient under different temperature-time combinations; Under the low-temperature electric field-assisted sterilization conditions (45°C, 2 minutes) of this application, the retention rate of vitamin C can reach 92%, the retention rate of taurine can reach 96%, and the retention rate of probiotics can reach 85%. This accurate prediction ability provides a theoretical basis for the system to optimize the processing parameters and maximize the retention rate of active ingredients.

[0081] Step 4.3, integrating thermal imaging and Raman spectroscopy data to achieve dynamic monitoring of active ingredients; The temperature field reconstruction results are combined with the thermal degradation model to calculate the spatial distribution and content changes of active ingredients in the product in real time. The system completes a full-field scan and calculation every 5 seconds, with a monitoring accuracy of ±3%, providing real-time feedback for subsequent processing parameter optimization.

[0082] For pet health products with complex formulas, data fusion algorithms can be used to integrate multi-source information; for example, Kalman filtering technology can be used to combine direct Raman spectroscopy measurements and predicted values ​​based on temperature fields to generate more accurate estimates of the distribution of active ingredients. In some embodiments, edge computing technology can also be introduced to migrate some data processing tasks to the acquisition end, reduce communication delays, and improve system response speed; for example, a small processor is integrated in the Raman spectroscopy acquisition module to complete data preprocessing and feature extraction, and only key feature values ​​are transmitted to the central control system.

[0083] The specific implementation methods of the active ingredient dynamic monitoring system in this application include: Import the reconstructed three-dimensional temperature field data and the established active ingredient thermal degradation kinetic model parameters into the real-time calculation module; The real-time calculation module calculates the current content of active ingredients at each spatial point based on the instantaneous temperature and cumulative thermal history of each spatial point, and generates a spatial distribution map of the active ingredients inside the product; The Raman spectroscopy system measures the active ingredient content in real time at several key points on the product surface. These measurements are used to correct and verify the temperature field calculation results. The system forms a more accurate active ingredient distribution model by integrating these two types of data and updates it in real time.

[0084] The active ingredient dynamic monitoring system is used in the production process of liquid multivitamin health products for dogs. When the health products flow through the electric field treatment area, the system calculates the concentration distribution of various water-soluble vitamins such as vitamin B1, B2, B6, and C in the product in real time. Through the three-dimensional visualization interface, operators can intuitively see which areas of the product have more active ingredient loss (usually displayed as red areas).

[0085] Step 5: Based on the active ingredient thermal degradation kinetics model and microbial electric field sensitivity data, the computational intelligence algorithm is used to simultaneously optimize the electric field parameters and the auxiliary temperature, and the multi-objective Pareto optimization method is combined to achieve the best balance between the bactericidal effect and the retention rate of different active ingredients; It includes the following sub-steps: Step 5.1, construct a multi-objective optimization model; In the embodiments of the present application, the following optimization goals are clearly defined for the pet health product processing process: Maximize microbial kill rate: ; Maximize active ingredient retention: ; Minimize energy consumption: ; in, represents the maximum microbial killing rate, It means maximizing the retention rate of active ingredients. represents minimizing energy consumption, represents the logarithmic function, and are the number of microorganisms before and after treatment, and After treatment and before treatment, respectively. The content of active ingredients, is the weight coefficient, is power, For processing time, Indicates the total amount of active ingredients.

[0086] The decision variables include: Electric field strength (15-25kV / cm) indicates the voltage applied per unit distance; Pulse Width (1-5μs) indicates the duration of each electrical pulse; Pulse frequency (100-500Hz) indicates the number of pulses generated per second; Number of pulses (20-50) indicates the total number of pulses applied during the treatment; Auxiliary temperature (40-55°C) indicates the ambient temperature maintained during the processing.

[0087] The optimization objectives can be adjusted appropriately according to the characteristics and quality requirements of different types of pet health products; for example, for high-end pet functional health products, the weight of active ingredient retention rate can be increased, and even a fourth optimization objective can be introduced: active ingredient uniformity to ensure the consistency of active ingredient content in all parts of the product.

[0088] In some embodiments, the objective function may also take into account the sensory properties of the product. For example, for certain non-enzymatic browning reactions that may affect the color or flavor of the product, their reaction extent may be incorporated into the optimization target to minimize quality losses caused by adverse reactions.

[0089] The specific implementation methods of the multi-objective optimization model include: Establish the relationship model between decision variables and objective functions, including: the relationship model between electric field parameters and microbial killing rate, the relationship model between temperature and active ingredient retention rate, and the relationship model between treatment parameters and energy consumption; Normalize each objective function so that its value range is unified between 0 and 1; Define the comprehensive objective function, which is the weighted sum of each sub-objective function: ; in, represents the maximum microbial killing rate, It means maximizing the retention rate of active ingredients. represents minimizing energy consumption, , , They are the weight coefficients for maximizing microbial killing rate, maximizing active ingredient retention rate and minimizing energy consumption, respectively.

[0090] In a specific application, for a pet liquid nutritional supplement containing multiple B vitamins and probiotics, the system automatically sets the weight coefficient based on the product characteristics: (Bactericidal effect weight), (Active ingredient retention weight), (Energy consumption weight).

[0091] The system also automatically adjusted the parameters in the sub-objective functions based on the characteristics of the product's main target pathogens (Escherichia coli and Salmonella) and key active ingredients (vitamin B12 and Bifidobacterium).

[0092] The optimized electric field treatment parameters are: electric field intensity 22kV / cm, pulse width 2.5μs, pulse frequency 300Hz, pulse number 35, auxiliary temperature 45℃. This set of parameters achieves a 5-logarithmic sterilization effect while ensuring that the retention rate of vitamin B12 reaches 91% and the retention rate of bifidobacterium activity reaches 85%, which are 30% and 70% higher than the retention rates of traditional thermal sterilization processes, respectively.

[0093] Step 5.2, realize multi-objective Pareto optimization; The improved non-dominated sorting genetic algorithm (NSGA-II) is used to solve the above multi-objective optimization problem. The steps include: Initialize the population: randomly generate parameter combinations that meet the constraints; Evaluate individual fitness: Use the model established in steps 2 to 4 to calculate the objective function value corresponding to each set of parameters; Non-dominated sorting: Sort individuals according to the objective function value to form a Pareto frontier; Selection, crossover and mutation: generating new generations of population; Repeat the above steps until the termination condition is met.

[0094] In this embodiment, the algorithm parameters are set as follows: population size is 50, maximum number of iterations is 100, crossover probability is 0.8, and mutation probability is 0.1.

[0095] For real-time optimization scenarios, a more computationally efficient optimization algorithm can be used; For example, use multi-objective particle swarm optimization or multi-objective differential evolution algorithms, which converge faster when dealing with high-dimensional decision spaces and are suitable for online optimization applications.

[0096] In some embodiments, constraint processing technology can also be introduced to better handle various practical constraints in the production of pet health products; for example, an improved constraint control principle can be used to handle hard constraints such as the maximum temperature not exceeding a specific threshold and the minimum sterilization rate must meet food safety standards.

[0097] The specific implementation of the improved NSGA-II algorithm in this application includes: Real number coding is used to represent decision variables, and special crossover and mutation operators are designed according to the characteristics of electric field parameters. Introducing a selection mechanism based on crowding distance to ensure the diversity of solutions on the Pareto front; An adaptive mutation rate strategy was designed, which uses a larger mutation rate in the early stage of iteration to increase the search range, and reduces the mutation rate in the later stage to increase the convergence speed; An elite retention strategy is adopted to ensure that the best individuals will not be lost during the evolution process.

[0098] In practical applications, the non-dominated sorting genetic algorithm optimizes parameters for different types of pet health products. For example, for liquid probiotic products that are sensitive to heat, the algorithm focuses more on the retention rate of active ingredients, and the optimal parameter combination is low temperature (42°C), high electric field strength (24kV / cm), and short processing time (60s). For the special nutrient solution for infants and puppies that requires higher sterilization requirements, the algorithm focuses more on the sterilization effect, and the optimal parameter combination is medium temperature (48°C), medium electric field strength (20kV / cm), and longer processing time (120s). Different solutions on the Pareto front can meet the specific needs of different products and provide flexible options for production.

[0099] Step 5.3, construct an adaptive parameter optimization control system; Based on the Pareto optimal solution set and combined with real-time monitoring data, a deep reinforcement learning method is used to construct an adaptive control system to achieve dynamic optimization of machining parameters.

[0100] The state space of the control system includes: current product characteristic parameters, real-time monitored microbial inactivation rate and active ingredient retention rate; The action space includes: the adjustment amount of electric field parameters and auxiliary temperature; The reward function is defined as a weighted combination of microbial killing rate, active ingredient retention rate, and energy consumption.

[0101] Different types of deep reinforcement learning algorithms can be selected for different control strategy requirements; for example, for scenarios that pursue stability, a conservative policy gradient algorithm can be used to limit the step size of each policy update to avoid drastic fluctuations in control parameters; for scenarios that need to quickly adapt to changes in product formulas, a model-based reinforcement learning method can be used to accelerate the policy learning process by building an environmental dynamics model. In some implementations, transfer learning technology can also be introduced to migrate the strategy trained on a certain type of pet health products to new products, greatly reducing the strategy optimization time for new products.

[0102] A deep Q network is used to realize the mapping from state to action. The network structure is: Input layer (state dimension) → fully connected layer with 256 neurons → fully connected layer with 128 neurons → fully connected layer with 64 neurons → output layer (action dimension). Improve learning stability and efficiency through experience replay and target network technology.

[0103] The specific implementation methods of the deep Q network include: The defined state space contains 12 features (including physical property parameters such as product temperature, conductivity, pH value, flow rate, as well as real-time monitored microbial population, content of four key active ingredients, current electric field parameters, and accumulated energy consumption); Discretize the action space and define five adjustment levels (large increase, small increase, unchanged, small decrease, large decrease) for the electric field intensity, pulse width, frequency, and auxiliary temperature respectively; The reward function is designed as: ; in, , , They represent the importance weights of bactericidal effect, active ingredient retention and energy consumption in the total reward respectively; Indicates the number of microorganisms before treatment; Indicates the number of microorganisms after treatment; It expresses the logarithmic value of microbial reduction; Indicates that after processing The content of active ingredients; Indicates the first The content of active ingredients; Indicates Retention rate of active ingredients; Indicates Weight coefficient of each active ingredient; Indicates the total amount of active ingredients; Indicates system power; Indicates processing time; It represents the total energy consumption of the process; By interacting with the simulated environment, the experience replay buffer is used to collect state transition samples and the Q-value estimate is updated through temporal difference learning.

[0104] In actual application, the Deep Q network control system was tested on the production line of a pet health products factory.

[0105] The system was pre-trained for 1 million steps in a simulation environment, and then put into use and continuously optimized in an actual production environment. Compared with fixed parameter control strategies, the deep Q network control system can dynamically adjust processing parameters based on real-time detected changes in product characteristics and processing effects.

[0106] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation mode. The above-mentioned specific implementation mode is merely illustrative and not restrictive. Under the guidance of this embodiment, ordinary technicians in this field can also make more forms of equivalent embodiments, all of which are within the protection of this embodiment.

Claims

1. A fully automatic pet health product processing method based on AI optimization, characterized in that: The following steps are involved: Construct a pulse electric field assisted low temperature sterilization system, including a pulse electric field generating unit, a temperature control unit, a product flow unit and a control unit; Establish a microbial electric field sensitivity database to achieve precise customization of electric field parameters based on the electric field sensitivity characteristics of common microorganisms in pet health products; Based on the microbial electric field sensitivity database, the thermal response characteristics of different active ingredients in pet health products are analyzed, and the inversion algorithm of the heat diffusion equation is constructed to achieve accurate reconstruction of the internal temperature field of the material; Using the inversion results of the heat diffusion equation, combined with Raman spectroscopy technology and thermal imaging system, the active ingredient content is monitored in real time, and a thermal degradation kinetic model of the active ingredient is established; Based on the thermal degradation kinetics model of active ingredients and the electric field sensitivity data of microorganisms, computational intelligence algorithms are used to simultaneously optimize the electric field parameters and auxiliary temperature. Combined with the multi-objective Pareto optimization method, the optimal balance between the bactericidal effect and the retention rate of different active ingredients is achieved.

2. The fully automatic pet health product processing method based on AI optimization according to claim 1 is characterized in that: The construction of the pulse electric field assisted low temperature sterilization system includes: Design the electrode configuration, using parallel plate electrodes, coaxial cylindrical electrodes or finger-shaped electrode structures; Constructing a pulse electric field generating unit, including a high voltage pulse power supply and a pulse shaping circuit; Design a temperature control unit, including a circulating water bath, a temperature sensor array, and a PID controller; Build product flow units, including peristaltic pumps, piping, and agitation devices; Build a control unit based on an embedded system, integrate the data acquisition module, parameter adjustment module and safety monitoring module to achieve fully automatic operation of the system.

3. The fully automatic pet health product processing method based on AI optimization according to claim 1 is characterized in that: The pulse electric field generating unit includes a high voltage pulse power supply, a multi-stage electrode chamber and a parameter adjustment module; The temperature control unit includes a precision temperature control device and a temperature sensor array; The product flow unit includes a peristaltic pump, a fluid pipeline and a processing chamber; The control unit includes a main controller, a data acquisition module and a human-computer interaction interface; The control unit coordinates the work of each unit to achieve coordinated control of electric field parameters and temperature.

4. The fully automatic pet health product processing method based on AI optimization according to claim 1 is characterized in that: The pulse electric field assisted low temperature sterilization system selectively destroys microbial cell membranes through the electroporation effect of microbial cell membranes.

5. The fully automatic pet health product processing method based on AI optimization according to claim 1 is characterized in that: The establishment of a microbial electric field sensitivity database comprises: The sensitivity data of common microorganisms in pet health products to different electric field parameters were collected and analyzed, and an electric field sterilization kinetic model was established. The electric field sterilization kinetic model is expressed as: ； in, represents the electric field sterilization kinetic model, N represents the number of microorganisms after treatment, represents the initial number of microorganisms, k represents the microbial inactivation rate constant, t represents the treatment time, E represents the electric field strength, m and n represent the exponential parameters of time and electric field strength, respectively. The logarithmic ratio indicating the number of microorganisms decreases with increasing treatment time and electric field strength.

6. The fully automatic pet health product processing method based on AI optimization according to claim 1 is characterized in that: The thermal response characteristics analysis includes: Differential scanning calorimetry and thermogravimetric analysis were used to determine the thermal stability parameters of the active ingredients, and a model of the relationship between thermal degradation rate and temperature was established: ； in, The model representing the relationship between thermal degradation rate and temperature is: is the concentration of active ingredient, is the pre-exponential factor, is the activation energy, is the gas constant, is the absolute temperature, Indicates the degree to which the reaction depends on concentration; The heat diffusion equation inversion algorithm uses nonlinear least squares method combined with finite element analysis to invert the internal temperature field distribution based on surface temperature measurement data. The heat diffusion equation inversion algorithm is implemented by solving the equation: ； in, represents the inversion algorithm of the heat diffusion equation, is the thermal diffusion coefficient, is the internal heat source term, is the density, is the specific heat capacity, is the material temperature, is the Laplace operator.

7. The fully automatic pet health product processing method based on AI optimization according to claim 1 is characterized in that: The Raman spectroscopy technology and thermal imaging system realize real-time monitoring of active ingredient content, including: Use portable Raman spectrometer to conduct online detection of active ingredients in pet health products and collect spectral data; The multivariate calibration model was used to establish the quantitative relationship between Raman spectra and the content of active ingredients. The model expression is: ； in, is the active ingredient content, is the intercept, For the The regression coefficient at the wavelength is For the The Raman spectrum intensity at wavelengths is is the error term, is the number of spectral data points included in the model; Use infrared thermal imager to construct product surface temperature field distribution; Combining Raman spectroscopy analysis results with thermal field distribution information, a thermal degradation kinetic model of active ingredients is established: ； in, represents the natural logarithm of the active ingredient retention rate, is the initial active ingredient content, For time The active ingredient content after is the frequency factor, For processing time, is the activation energy, is the gas constant, is the absolute temperature, It means that the content of active ingredients decreases with the increase of treatment time; The retention rate of active ingredients under different process parameters is predicted in real time based on the thermal degradation kinetic model of active ingredients, providing feedback data for the optimization algorithm.

8. The fully automatic pet health product processing method based on AI optimization according to claim 1 is characterized in that: The computational intelligence algorithm simultaneously optimizes the electric field parameters and the auxiliary temperature including: Construct a multi-objective optimization model, taking into account the following optimization objectives: Maximize microbial kill rate: ； Maximize active ingredient retention: ； Minimize energy consumption: ； in, represents the maximum microbial killing rate, It means maximizing the retention rate of active ingredients. represents minimizing energy consumption, represents the logarithmic function, and are the number of microorganisms before and after treatment, and After treatment and before treatment, respectively. The content of active ingredients, is the weight coefficient, is power, For processing time, Indicates the total amount of active ingredients; An improved non-dominated sorting genetic algorithm is used to solve multi-objective optimization problems; Based on non-dominated sorting and crowding distance calculation, the Pareto optimal solution set is obtained to provide a variety of options for decision-making; According to product characteristics and quality requirements, select the electric field parameter combination that best meets current needs from the Pareto frontier.

9. The fully automatic pet health product processing method based on AI optimization according to claim 1 is characterized in that: The optimization method further includes constructing an adaptive parameter optimization control system, wherein the adaptive parameter optimization control system includes: Based on the Pareto optimal solution set and combined with real-time monitoring data, the deep reinforcement learning method is used to achieve dynamic optimization of processing parameters; The state space of the control system includes current product characteristic parameters, real-time monitored microbial inactivation rate and active ingredient retention rate; The action space includes the adjustment amount of electric field parameters and auxiliary temperature; The reward function is defined as a weighted combination of microbial kill rate, active ingredient retention rate, and energy consumption; Use deep Q network to realize the mapping from state to action; Through experience replay and target network technology, the learning stability and efficiency are improved, and the adaptive optimization of parameters in the processing of pet health products is achieved.

10. A fully automatic pet health care product processing system based on AI optimization, used to execute a fully automatic pet health care product processing method based on AI optimization according to any one of claims 1 to 9, characterized in that: include: A pulse electric field generating unit, used to generate an adjustable high-voltage pulse electric field, including a high-voltage pulse power supply, a multi-stage electrode chamber and a parameter adjustment module; A temperature control unit for accurately regulating and maintaining the processing temperature, including a precision temperature control device and a temperature sensor array; A product flow unit, which is used to control the flow state of the pet health product in the processing chamber, and includes a peristaltic pump, a fluid pipeline and a processing chamber; Monitoring and analysis unit, including Raman spectrometer and infrared thermal imager, used to monitor the active ingredient content and temperature field distribution in real time; The control unit, including the main controller, data acquisition module and human-computer interaction interface, integrates deep reinforcement learning algorithm to achieve adaptive optimization control of system parameters.

Images