Intelligent fig wine brewing process parameter optimization system and method

Abstract

The invention discloses an intelligent fig wine brewing process parameter optimization system and method, and relates to the technical field of process parameter control, and the method comprises the following steps: collecting spectral data, rheological data, temperature data and dissolved oxygen data of fig slurry in real time through a multi-parameter sensing module; calculating a pectin component predicted value and a protein turbid potential value based on the collected data through a feature resolving module; inputting the data into a pectin memory neural network model through a physical gating neural control module, and outputting a control strategy; and inputting data collected at the current moment into the long-short-term memory network model corrected in the steps, generating a variable-temperature enzymolysis curve and a segmented enzyme adding strategy of the next stage, and controlling brewing equipment through an execution regulation and control module. The method provided by the invention solves the problem of control model response oscillation or incomplete enzymolysis of stubborn substrates caused by large fluctuation of pectin methyl esterification degrees of different batches of fig raw materials.

Description

Technical Field

[0001] This invention relates to the field of process parameter control technology, specifically to an intelligent fig wine brewing process parameter optimization system and method. Background Technology

[0002] Fig wine, a novel functional fruit wine combining nutritional value and unique flavor, has gained popularity in the consumer market in recent years. However, the complexity of the raw material matrix poses a serious challenge to the quality of the finished product during the industrial brewing process of fig wine. Fresh figs are rich in highly methylated pectin, heat-sensitive proteins, and polyphenol oxidases, which are prone to complex biochemical reactions during fermentation and aging. Specifically, on the one hand, the complex pectin structure leads to high viscosity of the liquid and low juice yield; on the other hand, residual proteins combine with oxidized polyphenols, easily causing irreversible turbidity and precipitation (post-turbidity) and browning during the shelf life after bottling, seriously affecting the sensory quality and value of the product.

[0003] To address the aforementioned issues, those skilled in the art typically employ a combination of enzyme preparation modification and automated temperature control. Specifically, to improve enzymatic hydrolysis efficiency, modern production lines often utilize PID controllers or expert systems based on general machine learning models to maintain the hydrolysis tank temperature within the optimal range for enzyme activity, or set fixed segmented temperature control curves based on historical experience.

[0004] However, the above-mentioned solutions still have significant limitations: traditional fixed-parameter control or general models lack the ability to perceive the "substrate specificity" and "multi-physics coupling risks" of figs. First, the pectin structure of different batches of figs varies greatly, and fixed processes often lead to incomplete or excessive enzymatic hydrolysis. Second, existing control logics often focus solely on pursuing "degradation efficiency," neglecting the catalytic effect of high temperatures on the oxidation of heat-sensitive proteins and polyphenols. This often results in the potential for "later browning and turbidity" while pursuing high juice yield, making it difficult to find a balance between efficiency and quality.

[0005] Therefore, the present invention provides an intelligent system and method for optimizing process parameters in fig wine brewing. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent system and method for optimizing the process parameters of fig wine brewing, so as to solve the existing problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent fig wine brewing process parameter optimization system, comprising: A multi-parameter sensing module is configured to collect spectral data, rheological data, temperature data, and dissolved oxygen data of fig slurry in real time. The feature calculation module is configured to calculate the predicted value of pectin components and the protein turbidity potential value based on the output of the multi-parameter sensing module. The physical gated neural control module is configured to construct a pectin memory neural network model, use the predicted values ​​of the pectin components to correct the bias parameters of the forgetting gate in real time, and use the protein turbidity potential value to reconstruct the penalty weights of the loss function in real time. The execution control module is configured to input real-time collected data into the corrected pectin memory neural network model, obtain the output temperature-varying enzymatic hydrolysis curve and segmented enzyme addition strategy, and drive the temperature control device and dosing device to perform corresponding actions.

[0008] A further improvement of the present invention is that the multi-parameter sensing module includes: a near-infrared spectrometer for acquiring spectral absorption characteristics; an online viscometer for acquiring the real-time shear viscosity of the slurry and its rate of change; and a thermal excitation unit configured to apply periodic micro-temperature pulses to the slurry during monitoring, so that the multi-parameter sensing module can collect transmittance response data.

[0009] A further improvement of the present invention is that, when calculating the predicted value of the pectin component, the feature calculation module is configured to perform the following operations: extracting the esterification degree characteristic peak area from the spectral data and identifying the first derivative inflection point of the viscosity decrease rate in the rheological data; and, based on the characteristic peak area and the time corresponding to the inflection point, calculating the normalized predicted value of the pectin component characterizing the degree of pectin methyl esterification through weighted mapping.

[0010] A further improvement of the present invention is that, when calculating the protein turbidity potential value, the feature calculation module is configured to perform the following operations: calculating the fluctuation amplitude of the transmittance data during the temperature pulse, and calculating the oxidative crosslinking factor in combination with the dissolved oxygen consumption rate; and using the oxidative crosslinking factor to perform an integral correction on the fluctuation amplitude to obtain the protein turbidity potential value.

[0011] A further improvement of this invention is that the pectin memory neural network model in the physical gated neural control module is configured to incorporate a forgetting gate bias adaptive strategy and a loss function reconstruction strategy. When the predicted value of the pectin component is greater than a preset structural threshold, the bias parameter of the forgetting gate is increased, causing the long short-term memory network model to increase the retention weight of the cell state at the previous time step, thereby smoothing the control strategy for recalcitrant pectin. When the protein turbidity potential value is greater than a preset risk threshold, the penalty weight for the temperature rise term in the loss function is increased, causing the long short-term memory network model to be forced to converge to the low-temperature range during gradient descent.

[0012] On the other hand, the present invention provides an intelligent method for optimizing the process parameters of fig wine brewing, comprising the following steps: Step S1: Real-time acquisition of spectral data, rheological data, temperature data, and dissolved oxygen data of fig slurry through a multi-parameter sensing module; Step S2: Calculate the predicted values ​​of pectin components and protein turbidity potential values ​​based on the collected data using the feature calculation module. Step S3: Through the physical gated neural control module, the data from step S2 is input into the pectin memory neural network model, and the control strategy is output to step S4. Step S4: Input the data collected at the current moment into the long short-term memory network model after the correction in step S3 to generate the temperature-varying enzymatic hydrolysis curve and segmented enzyme addition strategy for the next stage, and control the brewing equipment through the execution control module.

[0013] A further improvement of this invention is that the construction process of the pectin memory neural network model includes: Step S31: At time step t, receive the state vector. It is configured to describe the current physicochemical state of the slurry; and simultaneously receives a physical control vector. , are configured as characteristic parameters of the real-time modulation network structure; Step S32: Use the predicted value of the pectin component as the input of the forget gate bias adaptive strategy to obtain the corrected forget gate; Step S33: After step S32, the input data undergoes information filtering and transmission, utilizing the physically corrected forget gate. Update the long-term memory and output the output gate and hidden state vector at this time; Step S34: Input the hidden state vector into the fully connected layer for decoding to obtain the initial parameter control strategy; Step S35: Use the protein turbidity potential value as input to the loss function reconstruction strategy, and output the corrected loss function.

[0014] A further improvement of this invention is that the specific formula for adjusting the forget gate bias parameter in the forget gate bias adaptive strategy is expressed as follows: in, This indicates the corrected forget gate bias. Indicates reference bias. Represents the memory gain coefficient. This represents the normalized predicted value of the pectin component.

[0015] A further improvement of this invention is that the specific calculation logic of the penalty weight in the loss function reconstruction strategy is expressed as follows: Step S351: Define the basic loss function The mean squared error of the prediction; Step S352: Define security constraint terms It is an exponential function of the difference between the predicted temperature and the upper limit of the safe temperature. Step S353: Construct the reconstructed loss function ;in, Indicates the risk sensitivity coefficient. Indicates the protein turbidity potential value; when When the temperature rises, the model imposes an exponential penalty on predictions that exceed the preset safe temperature limit.

[0016] A further improvement of this invention lies in that the generation process of the variable-temperature enzymatic hydrolysis curve and the segmented enzyme addition strategy is mapped through an optimization path in a high-dimensional state space using a physically gated long short-term memory network, including utilizing a fully connected decoding layer to map the current hidden state vector. Mapped to the suggested temperature value for the next moment And the probability of enzyme addition action In this process, the protein turbidity potential value is configured as a soft constraint boundary, through the reconstructed loss function, to limit the suggested temperature value. The maximum value forces the model to output a control trajectory that converges to low temperature when the risk of turbidity increases; at the same time, based on learning from historical data, the model automatically increases the probability of enzyme addition when outputting a low temperature trajectory. This generates a segmented enzyme addition strategy that is discretely distributed along the time axis.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention firstly solves the problem of control model response oscillation or incomplete enzymatic hydrolysis of stubborn substrates caused by large fluctuations in the degree of pectin methyl esterification of different batches of fig raw materials by introducing a real-time physical correction mechanism for the forget gate bias of the LSTM model based on the predicted value of pectin components. This achieves long-term memory adaptive stable control for substrates with high structural impedance, effectively prevents the accumulation of bitter peptides caused by misjudgment of the enzymatic hydrolysis endpoint, and improves the process stability between batches. 2. By using the protein turbidity potential value calculated under thermal excitation perturbation to dynamically reconstruct the safety penalty weight in the neural network loss function, the endogenous coupling contradiction between improving enzymatic hydrolysis efficiency (requiring high temperature) and inhibiting browning of heat-sensitive proteins (requiring low temperature) in traditional brewing processes is resolved. This achieves feedforward active avoidance of potential turbidity and oxidation risks during aging, and solves the problems of clarification during brewing and turbidity recurrence after bottling.

[0018] 3. By using a physical-gated LSTM model to generate a variable-temperature enzymatic hydrolysis curve in a multi-dimensional state space and coordinating it with a segmented enzyme addition strategy, the problem of significant reduction in enzymatic hydrolysis kinetic efficiency under low-temperature risk avoidance mode in traditional static processes is solved. A dynamic balance between "temperature suppression" and "enzyme compensation" is achieved, which shortens the production cycle and improves the utilization rate of raw materials as much as possible while ensuring the complete preservation of heat-sensitive flavor substances and color. Attached Figure Description

[0019] Figure 1 This is a framework diagram of an intelligent fig wine brewing process parameter optimization system according to the present invention; Figure 2 This is a flowchart of an intelligent method for optimizing process parameters in fig wine brewing according to the present invention. Detailed Implementation

[0020] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0021] The term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three cases: A exists alone, A and B exist simultaneously, and B exists alone.

[0022] Example 1 Figure 1 This embodiment illustrates a framework diagram of an intelligent fig wine brewing process parameter optimization system disclosed in this embodiment. The system includes: A multi-parameter sensing module is configured to collect spectral data, rheological data, temperature data, and dissolved oxygen data of fig slurry in real time. The multi-parameter sensing module includes: a near-infrared spectrometer for acquiring wavenumbers in the range of 1600-1800. The system includes: spectral absorption characteristics within a specified range; an online viscometer for obtaining the real-time shear viscosity of the slurry and its rate of change; and a thermal excitation unit configured to apply periodic micro-temperature pulses to the slurry during monitoring, so that the multi-parameter sensing module can collect transmittance response data.

[0023] The feature calculation module is configured to calculate the predicted value of pectin components reflecting substrate characteristics and the protein turbidity potential value reflecting colloidal instability based on the output of the multi-parameter sensing module. The system collects basic data on the slurry at a frequency of 1 Hz. Simultaneously, a "micro-thermal shock" operation is performed every 20 minutes: for example, raising the temperature of the slurry in the bypass flow tank within 30 seconds. It was then allowed to cool naturally, and the transmittance change curve during the process was recorded. .

[0024] Conventional methods only monitor static indicators and cannot predict future turbidity risks. This embodiment introduces "thermal excitation," which is equivalent to giving the colloidal system a "stress test." This can actively induce instability in potential heat-sensitive proteins that are not visible under steady state, thereby capturing their precursor signals before turbidity actually occurs.

[0025] The physically gated neural control module includes a pre-built long short-term memory network (LSTM) model with a forgetting gate for controlling the degree of memory retention and a loss function for evaluating the cost of prediction. The pectin memory neural network model in the physical gated neural control module is configured to incorporate a forgetting gate bias adaptive strategy and a loss function reconstruction strategy. When the predicted value of the pectin component is greater than a preset structural threshold, the bias parameter of the forgetting gate is increased, causing the long short-term memory network model to increase the retention weight of the cell state at the previous time step, thereby smoothing the control strategy for recalcitrant pectin. When the protein turbidity potential value is greater than a preset risk threshold, the penalty weight for the temperature increase term in the loss function is increased, causing the long short-term memory network model to be forced to converge to the low-temperature range during gradient descent.

[0026] The execution control module is configured to input real-time collected data into the corrected pectin memory neural network model, obtain the output temperature-varying enzymatic hydrolysis curve and segmented enzyme addition strategy, and drive the temperature control device and dosing device to perform corresponding actions.

[0027] Example 2 Figure 2 This invention presents a flowchart of an intelligent method for optimizing process parameters in fig wine brewing. Based on the same inventive concept as Example 1, this invention provides an intelligent method for optimizing process parameters in fig wine brewing, comprising the following steps: Step S1: Real-time acquisition of spectral data, rheological data, temperature data, and dissolved oxygen data of fig slurry through a multi-parameter sensing module; Step S2: Calculate the predicted values ​​of pectin components and protein turbidity potential values ​​based on the collected data using the feature calculation module. The predicted value of the pectin component The construction process includes: Step 1: Real-time acquisition of near-infrared spectral data of fig sap, extracting wavenumbers between 1600-1800. The area of ​​the characteristic peak within the range is denoted as ; Step 2: Simultaneously monitor the shear viscosity of the slurry. When the viscosity decreases at a certain rate When the inflection point first appears, record the cumulative enzymatic hydrolysis time at that time. ; Step 3: Construct the feature mapping space, and... and Projected onto a pre-defined pectin structure database, the pectin component is predicted by calculating a weighted Euclidean distance and outputting a value between 0 and 1. Where 1 represents highly methylated, recalcitrant pectin, and 0 represents low-methylated, easily degradable pectin. The calculation formula is expressed as: ,in, express The spectral peak area at (ester carbonyl stretching vibration); For reference peak area; Represents the absolute value of the viscosity decrease rate; To prevent tiny constants with a denominator of zero; This is the weighting coefficient (in this embodiment, it can be 0.6).

[0028] Due to 1740 in the spectrum The peaks directly reflect the degree of methyl esterification (chemical structure) of pectin; the higher the degree of methyl esterification, the more difficult it is for pectinase to cleave it. The viscosity decrease rate reflects the kinetic resistance of actual enzymatic hydrolysis. Simply looking at the spectrum may be inaccurate due to solid interference, and simply looking at viscosity may be lagging. Therefore, combining the two... When the value approaches 1, the physical meaning clearly indicates that the current substrate is highly esterified and recalcitrant pectin, and that enzymatic hydrolysis is inhibited. This characterizes the degree of recalcitrant degradation and structural complexity of the current pectin.

[0029] The protein turbidity potential value The construction process includes: Step 1: During the enzymatic hydrolysis process, introduce a small amount of thermal disturbance, such as a momentary temperature increase. Continuously monitor light transmittance The response fluctuation amplitude is expressed as the fluctuation amplitude of transmittance during the thermal shock period. ; Step 2: Combine with the current dissolved oxygen consumption rate Calculate the oxidation-mediated crosslinking factor ; Step 3, when When the value exceeds a preset threshold, it is determined that a heat-sensitive protein is present, and the oxidation-mediated cross-linking factor is activated. After correcting it, the integral yields the total potential turbidity of the current system, which is... The calculation formula is expressed as: The unit is NTU, which represents the probability of protein turbidity occurring during future aging; among which, This indicates the fluctuation range of transmittance during the heat shock period; This indicates the rate of dissolved oxygen consumption.

[0030] transmittance fluctuation This represents the tendency of protein particles to aggregate under thermal conditions; dissolved oxygen consumption. This represents the rate at which phenolic substances oxidize to form "glue" (quinones). The turbidity mechanism in fig wine is the cross-linking of "heat-sensitive proteins" and "oxidized polyphenols." The risk of turbidity only increases exponentially when both are highly active. Therefore, the formula uses a product form.

[0031] This solves the problem of "clear during brewing, cloudy after bottling" in traditional processes. This embodiment utilizes protein turbidity potential values. It can quantify the probability of future turbidity in advance, providing a clear "risk avoidance signal" for subsequent control.

[0032] Due to high (Recalcitrant pectin) forms a colloidal protective layer that encapsulates proteins, leading to... The measured value was too low (false negative), which hindered protein precipitation; while the measured value was too high. (High protein content) will bind with polyphenols and inhibit pectinase activity, thus... The corresponding viscosity decrease curve is distorted, which affects the judgment of pectin components. Therefore, a pectin memory neural network model is constructed to use the predicted pectin component values ​​to correct the bias parameters of the forgetting gate in real time, and to use the protein turbidity potential values ​​to reconstruct the penalty weights of the loss function in real time. Step S3: Through the physical gated neural control module, the data from step S2 is input into the pectin memory neural network model, and the control strategy is output to step S4. The construction process of the pectin memory neural network model includes: Step S31: At time step t, receive the state vector. It is configured to describe the current physicochemical state of the slurry, and after normalization, it is obtained. ;in, Indicates the current real-time viscosity; This represents the rate of change of viscosity (first derivative). Indicates light transmittance; Indicates the current temperature inside the tank; Indicates dissolved oxygen concentration; This represents the cumulative amount of enzyme added; and the physical control vector received. , are configured as characteristic parameters of the real-time modulation network structure; in, This is the predicted value for the pectin component. This represents the protein turbidity potential value.

[0033] Step S32: Use the predicted value of the pectin component as the input of the forget gate bias adaptive strategy to obtain the corrected forget gate; in a standard LSTM unit, the forget gate determines how much historical information is retained.

[0034] The standard forget gate formula is expressed as follows: ; Fixed bias Replace with dynamic bias function : in, This indicates the corrected forget gate bias. Indicates reference bias. Represents the memory gain coefficient. This represents the normalized predicted value of pectin component; using this formula, a high predicted value of pectin component will delay the network's forgetting of historical viscosity trends.

[0035] The corrected forget gate is then represented as: ; when This indicates that the pectin is difficult to degrade and has a high reaction inertia. As the value increases, the Sigmoid function output approaches 1. This forces the network to ignore short-term fluctuations and retain long-term memory, making the output control strategy more stable; when... When indicating readily biodegradable pectin, As the size decreases, the network becomes more sensitive and can respond quickly to the end of the enzymatic digestion process.

[0036] Step S33: After step S32, the input data undergoes information filtering and transmission, utilizing the physically corrected forget gate. Update the long-term memory and output the output gate and hidden state vector at this time; After physical gating modulation, the data enters the core loop of the LSTM for information filtering and transmission, including: Input gate It is configured to determine the current new state vector. How much was written: Obtain candidate cell state : ; Then, the long-term memory is updated using a physically corrected forgetting gate $f_t^{phy}$: , This indicates element-wise multiplication; This leads to the output gate. With hidden state : , .

[0037] Step S34: Input the hidden state vector into the fully connected layer for decoding to obtain the initial parameter control strategy; Fully connected layer calculation: , ,in Recommend the temperature for the next moment. The recommended enzyme dosage is given.

[0038] Step S35: Use the protein turbidity potential value as input to the loss function reconstruction strategy, and output the corrected loss function; the specific calculation logic of the penalty weight in the loss function reconstruction strategy is expressed as follows: Step S351: Define the basic loss function The mean squared error of the prediction; Step S352: Define security constraint terms The value is an exponential function of the difference between the predicted temperature and the upper limit of the safe temperature; safety loss. Represented as: This indicates the set critical temperature for protein denaturation (e.g., ), This represents the activation function, which is applied only when the suggested temperature is... Non-zero penalties only occur at certain times; Indicates dynamic penalty gain; when the turbidity potential value At higher levels, this coefficient increases dramatically, leading to It becomes quite large.

[0039] Step S353: Construct the reconstructed loss function ;in, Indicates the risk sensitivity coefficient. Indicates the protein turbidity potential value; when When the temperature rises, the model imposes an exponential penalty on predictions that exceed the preset safe temperature limit.

[0040] Step S4: Input the data collected at the current moment into the long short-term memory network model after the correction in step S3 to generate the temperature-varying enzymatic hydrolysis curve and segmented enzyme addition strategy for the next stage, and control the brewing equipment through the execution control module.

[0041] The generation process of the variable-temperature enzymatic hydrolysis curve and the segmented enzyme addition strategy is mapped through an optimization path in a high-dimensional state space using a physically gated long short-term memory network (PG-LSTM). This includes using a fully connected decoding layer to map the current hidden state vector. Mapped to the suggested temperature value for the next moment And the probability of enzyme addition action In this process, the protein turbidity potential value is configured as a soft constraint boundary, through the reconstructed loss function, to limit the suggested temperature value. The maximum value forces the model to output a control trajectory that converges to low temperature when the risk of turbidity increases; at the same time, based on learning from historical data, the model automatically increases the probability of enzyme addition when outputting a low temperature trajectory. This generates a segmented enzyme addition strategy discretely distributed along the time axis to compensate for enzyme activity loss caused by low temperatures. For example, assuming the current time is t, the model calculates the optimal temperature and enzyme dosage at time t+1 based on the current state; then, at time t+1, it calculates the instruction for time t+2 based on new feedback. Ultimately, connecting these discrete points forms a continuous temperature-varying curve and a step-like segmented enzyme addition strategy, specifically represented as follows: Step S41: Generate variable-temperature enzymatic hydrolysis curve : Temperature control is a continuous variable, relying on the decoding of the hidden state by the fully connected output layer of the LSTM. Therefore, the input source is represented as the LSTM hidden state vector after physical gating. This includes high-dimensional information such as the current difficulty of enzymatic hydrolysis and whether the historical viscosity decreased rapidly. The calculation formula is mapped as follows: , and The weights and biases of the temperature decoding layer are determined, and the sigmoid function is used to limit the output to between 0 and 1. and These represent the upper and lower temperature limits of the physical equipment (e.g., 25℃ ~ 50℃).

[0042] Although the above formula is a forward propagation calculation, the protein turbidity potential value is used during the loss constraint phase of model training or online inference. To play a role, if the current protein turbidity potential value It is very high, and the model outputs a high value. (For example, 45℃), at this temperature, the safety penalty term in the loss function It will explode. To reduce the loss, the model will be forced to adjust the weights or search for lower values ​​in the gradient descent direction, ultimately outputting a lower value. (For example, 30℃). As time goes on, The fluctuations will directly compress the originally flat high temperature curve into a concave risk avoidance curve.

[0043] Step S42: Generate a segmented enzyme addition strategy : Input source: Represented as the LSTM hidden state vector after physical gating. Since enzyme activity decreases at low temperatures, it may be necessary to add more enzyme to compensate. Therefore, the formula for calculating the output is as follows: The output at this time This represents a probability distribution, corresponding to three actions: : This embodiment implements the functionality when the model is... High and forced to lower temperature In this embodiment, based on the associations learned internally from the training data, attention is paid to the fact that low temperature will lead to a decrease in enzymatic hydrolysis efficiency (viscosity). The decrease slows down); in order to maintain the target of viscosity decrease (minimize) The model will The output is compensated to trigger the "enzyme addition" action; this ultimately forms "segmented enzyme addition"—that is, at the cooling node or viscosity bottleneck point, a pulse-like enzyme addition operation automatically occurs.

[0044] For example: Ideal conditions (easily degradable pectin, low risk of turbidity): Predicted pectin component values Low protein turbidity potential value Low; Curve: It is a flat-topped trapezoid, rapidly heated to the optimal enzyme activity temperature (e.g., 45℃), and kept constant until the enzymatic hydrolysis is completed. Strategy: "Single Feeding": The standard enzyme amount is added all at once at the initial moment, and subsequently... All are 0.

[0045] High-risk avoidance state (high amount of heat-sensitive proteins): Protein turbidity potential value It rises sharply with increasing temperature. The curve exhibits a rapid decay pattern or a low-temperature plateau; initial attempts to increase the temperature were unsuccessful as potential protein turbidity was detected. The temperature spikes, then the curve immediately bends downwards and stabilizes at a safe temperature (e.g., 30°C).

[0046] The strategy is multi-pulse compensation: add enzyme at t=0, and at t=1h (because the temperature is suppressed, the enzyme efficiency is insufficient). A second pulse (adding 20%) occurs, and a third fine-tuning may occur at t=3h.

[0047] Stubborn substrate degradation status (pectin is extremely difficult to degrade): Predicted pectin component values Extremely high (large forgetting gate bias, long memory duration). The curve exhibits a gradual upward slope. Due to the forgetting gate effect, the model does not become agitated or erratic due to short-term viscosity not decreasing. Instead, it maintains a stable high temperature and even attempts slight overtemperature (within the protein turbidity potential range). Within permissible limits, to overcome stubborn pectin; The strategy is to maintain a high concentration, and it is recommended to keep the amount of enzyme added at a high level.

[0048] The threshold and weight settings involved in this embodiment can be set by default according to the present invention, or can be set by those skilled in the art.

[0049] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0050] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0051] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0052] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0053] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. An intelligent optimization system for process parameters of fig wine brewing, characterized in that: The system includes: A multi-parameter sensing module is configured to collect spectral data, rheological data, temperature data, and dissolved oxygen data of fig slurry in real time. The feature calculation module is configured to calculate the predicted value of pectin components and the protein turbidity potential value based on the output of the multi-parameter sensing module. The physical gated neural control module is configured to construct a pectin memory neural network model, use the predicted values ​​of the pectin components to correct the bias parameters of the forgetting gate in real time, and use the protein turbidity potential value to reconstruct the penalty weights of the loss function in real time. The execution control module is configured to input real-time collected data into the corrected pectin memory neural network model, obtain the output temperature-varying enzymatic hydrolysis curve and segmented enzyme addition strategy, and drive the temperature control device and dosing device to perform corresponding actions.

2. The intelligent parameter optimization system for fig wine brewing process according to claim 1, characterized in that: The multi-parameter sensing module includes: a near-infrared spectrometer for acquiring spectral absorption characteristics; an online viscometer for acquiring the real-time shear viscosity of the slurry and its rate of change; and a thermal excitation unit configured to apply periodic micro-temperature pulses to the slurry during monitoring, so that the multi-parameter sensing module can collect transmittance response data.

3. The intelligent fig wine brewing process parameter optimization system according to claim 2, characterized in that: When calculating the predicted value of the pectin component, the feature calculation module is configured to perform the following operations: extracting the esterification degree characteristic peak area from the spectral data and identifying the first derivative inflection point of the viscosity decrease rate in the rheological data; and calculating the normalized predicted value of the pectin component characterizing the degree of pectin methyl esterification through weighted mapping based on the characteristic peak area and the time corresponding to the inflection point.

4. The intelligent parameter optimization system for fig wine brewing process according to claim 3, characterized in that: When calculating the protein turbidity potential value, the feature calculation module is configured to perform the following operations: calculate the fluctuation range of transmittance data during the temperature pulse, and calculate the oxidative crosslinking factor in combination with the dissolved oxygen consumption rate. The fluctuation amplitude is integrally corrected using the oxidative cross-linking factor to obtain the protein turbidity potential value.

5. The intelligentized parameter optimization system for fig wine brewing process according to claim 4, characterized in that: The pectin memory neural network model in the physical gated neural control module is configured to carry a forget gate bias adaptive strategy and a loss function reconstruction strategy. When the predicted value of the pectin component is greater than the preset structural threshold, the bias parameter of the forget gate is increased, so that the long short-term memory network model increases the retention weight of the cell state at the previous time step, so as to smooth the control strategy for recalcitrant pectin. When the protein turbidity potential value is greater than the preset risk threshold, the penalty weight for the temperature rise term in the loss function is increased, which is a control strategy that forces the long short-term memory network model to converge to the low temperature range during the gradient descent process.

6. The intelligent optimization method of the process parameters of fig wine making, used for executing the intelligent optimization system of the process parameters of fig wine making as claimed in any one of claims 1-5, characterized in that: Includes the following steps: Step S1: Real-time acquisition of spectral data, rheological data, temperature data, and dissolved oxygen data of fig slurry through a multi-parameter sensing module; Step S2: Calculate the predicted values ​​of pectin components and protein turbidity potential values ​​based on the collected data using the feature calculation module. Step S3: Through the physical gated neural control module, the data from step S2 is input into the pectin memory neural network model, and the control strategy is output to step S4. Step S4: Input the data collected at the current moment into the long short-term memory network model after the correction in step S3 to generate the temperature-varying enzymatic hydrolysis curve and segmented enzyme addition strategy for the next stage, and control the brewing equipment through the execution control module.

7. The method for optimizing parameters of a fig wine brewing process according to claim 6, characterized in that: The construction process of the pectin memory neural network model includes: Step S31, at time step t, receiving a state vector configured to describe the physico-chemical state of the current slurry; while receiving a physical control vector configured to modulate in real time the characteristic parameters of the network structure; Step S32: Use the predicted value of the pectin component as the input of the forget gate bias adaptive strategy to obtain the corrected forget gate; Step S33, after the input data is processed by step S32, information screening and transmission are performed, and the forgetting gate is corrected physically The long-term memory is updated, and the output gate and hidden state vector at this time are output. Step S34: Input the hidden state vector into the fully connected layer for decoding to obtain the initial parameter control strategy; Step S35: Use the protein turbidity potential value as input to the loss function reconstruction strategy, and output the corrected loss function.

8. The method for optimizing parameters of a fig wine brewing process according to claim 7, characterized in that: The specific formula for adjusting the forget gate bias parameters in the aforementioned forget gate bias adaptive strategy is expressed as follows: in, This indicates the corrected forget gate bias. Indicates reference bias. Represents the memory gain coefficient. This represents the normalized predicted value of the pectin component.

9. The intelligent fig wine brewing process parameter optimization method according to claim 7, characterized in that: The specific calculation logic for the penalty weight in the loss function reconstruction strategy is as follows: Step S351: Define the basic loss function The mean squared error of the prediction; Step S352: Define security constraint terms It is an exponential function of the difference between the predicted temperature and the upper limit of the safe temperature. Step S353: Construct the reconstructed loss function ;in, Indicates the risk sensitivity coefficient. Indicates the protein turbidity potential value; when When the temperature rises, the model imposes an exponential penalty on predictions that exceed the preset safe temperature limit.

10. The intelligent fig wine brewing process parameter optimization method according to claim 6, characterized in that: The generation process of the variable-temperature enzymatic hydrolysis curve and the segmented enzyme addition strategy is mapped through an optimization path in a high-dimensional state space using a physically gated long short-term memory network. This includes utilizing a fully connected decoding layer to map the current hidden state vector. Mapped to the suggested temperature value for the next moment And the probability of enzyme addition action In this process, the protein turbidity potential value is configured as a soft constraint boundary, through the reconstructed loss function, to limit the suggested temperature value. The maximum value forces the model to output a control trajectory that converges to low temperature when the risk of turbidity increases; at the same time, based on learning from historical data, the model automatically increases the probability of enzyme addition when outputting a low temperature trajectory. This generates a segmented enzyme addition strategy that is discretely distributed along the time axis.

Images