Regulation and control method and device for fermentation process of dairy product and readable storage medium

Abstract

The embodiment of the invention provides a dairy product fermentation process regulation and control method, a dairy product fermentation process regulation and control device and a readable storage medium, production data related to the fermentation process of a target batch of dairy products is automatically obtained, and the acidity value of the target batch of dairy products is predicted through a quality prediction model. And according to the predicted acidity value, a control instruction is obtained and executed under the condition that constraint conditions such as production quality are met. Meanwhile, an execution result is obtained, when the adjusting effect is not ideal, re-adjustment is conducted in time, and the adjusting effect is guaranteed. The acidity value can be automatically, quickly and accurately predicted, the acidity value can be regulated and controlled in time, the automatic processing process integrating data collection, acidity value prediction and acidity value regulation and control is achieved, the response time is shortened, batch waste products are effectively prevented from being generated, and the production efficiency and the product quality stability are improved. Manual participation is reduced, and the production process is more intelligent and active. And production fluctuation and quality problems caused by changes of external factors are reduced.

Description

Technical Field

[0001] This application relates to the field of dairy product production technology, and in particular to a method, apparatus and readable storage medium for controlling the fermentation process of dairy products. Background Technology

[0002] In manufacturing, the production processes of many products involve time-delay responses between parameters. Taking the dairy industry as an example, processes such as yogurt fermentation exhibit significant "acidity time-delay response characteristics"—fluctuations in fermentation temperature and lactic acid bacteria inoculation amount take 2-3 hours to manifest as acidity (pH) anomalies.

[0003] Given this time-delay response characteristic of acidity, static rules are usually preset and expert experience is relied upon. Process parameters of production equipment are manually adjusted through on-site observation and subjective judgment. However, it is difficult to detect parameter fluctuations in dairy products, and even if fluctuations are observed, it is difficult to take effective measures to control acidity. By the time abnormal acidity values ​​are discovered, fermentation is nearing its end, and the quality of this batch of dairy products can no longer be salvaged.

[0004] Therefore, there is an urgent need for a solution to accurately and timely adjust the acidity value during the fermentation process of dairy products. Summary of the Invention

[0005] This application provides a method, apparatus, and readable storage medium for controlling the fermentation process of dairy products, so as to accurately and timely adjust the acidity value during the fermentation process of dairy products.

[0006] In a first aspect, embodiments of this application provide a method for regulating the fermentation process of dairy products, the method comprising: Obtain production data related to the fermentation process of the target batch of dairy products; the production data includes: process parameters; the process parameters include the collected fermentation tank temperature corresponding to the target batch of dairy products, the collection time corresponding to the fermentation tank temperature, and the inoculation amount of the microbial strain during the fermentation of the target batch of dairy products; The production data is input into the quality prediction model to obtain the predicted acidity value of the target batch of dairy products; If the predicted acidity value does not fall within the preset acidity value range, production constraints are determined based on the predicted acidity value and the preset acidity value range; under the constraints of the production constraints, control instructions are obtained based on the production data and the predicted acidity value. The parameters are adjusted according to the control instructions, and the execution results corresponding to the control instructions are obtained; If the execution result indicates that the adjustment has failed, then based on the execution result and the control instruction, the control instruction is re-determined; and the process returns to the step of adjusting the parameters according to the control instruction. If the execution result indicates successful adjustment, then after the fermentation of the target batch of dairy products is completed, fermentation-related data of the fermentation process of the target batch of dairy products are obtained; the fermentation-related data includes the production data and the quality index data of the target batch of dairy products. Based on the fermentation-related data, the quality prediction model is incrementally trained to update the quality prediction model. Secondly, embodiments of this application provide a control device for a dairy product fermentation process, the device comprising: The data acquisition module is used to acquire production data related to the fermentation process of the target batch of dairy products; the production data includes: process parameters; the process parameters include the acquired fermentation tank temperature corresponding to the target batch of dairy products, the acquisition time corresponding to the fermentation tank temperature, and the inoculation amount of the bacteria during the fermentation of the target batch of dairy products; The quality prediction module is used to input the production data into the quality prediction model to obtain the predicted acidity value of the target batch of dairy products. The process optimization module is used to determine production constraints based on the predicted acidity value and the preset acidity value range when the predicted acidity value does not fall within the preset acidity value range; and to obtain control instructions based on the production data and the predicted acidity value under the constraints of the production constraints. The execution feedback module is used to adjust parameters according to the control command and obtain the execution result corresponding to the control command; if the execution result indicates that the adjustment has failed, the execution result and the control command are fed back to the process optimization module so that the process optimization module can re-determine the control command. The quality prediction module is further configured to: if the execution result indicates successful adjustment, then after the fermentation of the target batch of dairy products is completed, acquire fermentation-related data of the fermentation process of the target batch of dairy products; the fermentation-related data includes the production data and the quality index data of the target batch of dairy products; and based on the fermentation-related data, incrementally train the quality prediction model to update the quality prediction model.

[0007] Thirdly, embodiments of this application provide a non-transitory machine-readable storage medium storing executable code, which, when executed by a processor of an electronic device, enables the processor to at least implement the method for controlling the dairy fermentation process as described in the first aspect.

[0008] Fourthly, embodiments of this application provide a computer program product, which includes a computer program that, when executed by a processor, can implement the method for regulating the fermentation process of dairy products as described in the first aspect.

[0009] The dairy fermentation process control scheme provided in this application automatically acquires production data related to the fermentation process of the target batch of dairy products, and predicts the acidity value of the target batch of dairy products through a quality prediction model. This fully considers production data that influences the acidity value of dairy products, improving the accuracy of the predicted acidity value. Based on the predicted acidity value, control commands are obtained and executed under constraints such as production quality. Simultaneously, the execution results are acquired, and if the adjustment effect is unsatisfactory, timely readjustment is performed to ensure the adjustment effect. This automatic, rapid, and accurate prediction of acidity values ​​and timely acidity value adjustment achieves an automated processing process integrating data acquisition, acidity value prediction, and acidity value adjustment. It can quickly complete the entire process from anomaly detection and strategy generation to parameter adjustment, significantly shortening the response time compared to traditional methods, effectively avoiding the generation of batches of defective products due to continuous anomalies, and significantly improving production efficiency and product quality stability. It reduces manual intervention, making the production process more intelligent and proactive. After each batch of dairy products has completed fermentation, the production data and quality indicators involved in the fermentation process are used to incrementally train the quality prediction model. This allows the model to adapt to changes in the performance of current production equipment, changes in the production environment, and fluctuations in raw material quality, resulting in more accurate acidity predictions and improved accuracy in acidity control. This automatic adaptive capability enables the production process to maintain efficient and stable operation in complex and changing environments, reducing production fluctuations and quality problems caused by changes in external factors. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A schematic diagram illustrating a control scenario for a dairy product fermentation process provided in an embodiment of this application; Figure 2 A flowchart illustrating a method for controlling a dairy product fermentation process, provided as an embodiment of this application; Figure 3 A flowchart of another method for controlling the fermentation process of dairy products provided in this application embodiment; Figure 4 A structural diagram of a control device for a dairy product fermentation process provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in this embodiment. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application. In addition, the timing of the steps in the following method embodiments is only an example and not a strict limitation.

[0013] It should be noted that, in the cases involving user information in the embodiments of this application, the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the embodiments of this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse. In addition, the various models involved in this application (including but not limited to large language models or other models) comply with relevant laws and standards.

[0014] First, the terms or concepts involved in the embodiments of this application will be explained: An intelligent agent is a proxy capable of perceiving its environment and taking actions to achieve specific goals. It can be software, hardware, or a system, possessing autonomy, adaptability, and interactivity. By perceiving changes in the environment (e.g., through sensors or data input), an intelligent agent makes judgments and decisions based on its learned knowledge and algorithms, and then executes actions to influence the environment or achieve predetermined goals.

[0015] Manufacturing Execution System (MES) is the core hub connecting an enterprise's ERP (Enterprise Resource Planning) system with the shop floor equipment control system. It focuses on the refined management of the "production execution layer" and aims to solve problems such as "information gaps," "low efficiency," and "difficulty in quality traceability" in the production process. It is a key component of the intelligent manufacturing system.

[0016] Dairy products, also known as dairy products, refer to products made primarily from fresh cow (or sheep) milk and its derivatives. For example, yogurt is a type of dairy product.

[0017] For ease of understanding, the following description uses dairy products, specifically yogurt. In yogurt production, there are specific requirements for the acidity of the finished product, which typically needs to be controlled within a preset acidity range. Yogurt fermentation and other processes exhibit significant "acidity time-lag response characteristics." Therefore, during yogurt fermentation, before the actual acidity exceeds the preset range, it is necessary to promptly and accurately adjust the acidity based on production fluctuations to ensure the finished yogurt's acidity remains within the preset range.

[0018] This application provides a method for controlling the fermentation process of dairy products, which involves pre-setting static rules and relying on expert experience, manually adjusting the process parameters of the production equipment through on-site observation and subjective judgment analysis. However, it is difficult to detect parameter fluctuations in dairy products, and even if fluctuations are observed, it is difficult to take effective measures to control acidity. By the time abnormal acidity values ​​are detected, fermentation is nearing its end, and the quality of this batch of dairy products can no longer be salvaged.

[0019] In addition, this application provides another method for controlling the fermentation process of dairy products, which trains a quality prediction model based on historical data and uses this model to predict acidity values ​​during dairy production. However, this method cannot capture the time lag characteristics of dairy fermentation, resulting in a high prediction inaccuracy rate when the composition of raw milk fluctuates. It also cannot adapt to dynamic changes in the production line, leading to prediction inaccuracies when equipment performance deteriorates or raw material quality fluctuates.

[0020] In summary, the above-mentioned methods for controlling the fermentation process of dairy products in this application all have their own shortcomings. To overcome these shortcomings, this application proposes a control scheme for the fermentation process of dairy products. The control scheme for the fermentation process of dairy products provided in this application automatically acquires production data related to the fermentation process of the target batch of dairy products and predicts the acidity value of the target batch of dairy products using a quality prediction model. This fully considers production data that affects the acidity value of dairy products, improving the accuracy of the predicted acidity value. Based on the predicted acidity value, control commands are obtained and executed under constraints such as production quality. Simultaneously, the execution results are acquired, and if the adjustment effect is not ideal, timely readjustment is performed to ensure the adjustment effect. This scheme can automatically, quickly, and accurately predict acidity values ​​and adjust them in a timely manner, thereby realizing an automated processing process integrating data collection, acidity value prediction, and acidity value adjustment. It can quickly complete the entire process from anomaly detection and strategy generation to parameter adjustment, significantly shortening the response time compared to traditional methods, effectively avoiding the generation of batches of defective products due to continuous anomalies, and significantly improving production efficiency and product quality stability. This reduces manual intervention, making the production process more intelligent and proactive. After each batch of dairy products has finished fermenting, the production data and quality indicators involved in the fermentation process are used to incrementally train the quality prediction model. This allows the model to adapt to changes in equipment performance, environmental conditions, and raw material quality fluctuations, resulting in more accurate acidity predictions and improved acidity control. This automatic adaptability enables the production process to maintain efficient and stable operation in complex and changing environments, reducing production fluctuations and quality problems caused by external factors.

[0021] The following describes in detail, with reference to the accompanying drawings, the control scheme for the dairy product fermentation process provided in the embodiments of this application.

[0022] Figure 1 This is a schematic diagram of a control scenario for a dairy product fermentation process provided in an embodiment of this application. The scenario for the dairy product fermentation process includes, but is not limited to: workshop equipment control cabinet 101, quality inspection system equipment 102, environmental monitoring point equipment 103, data acquisition equipment 104, quality prediction equipment 105, process optimization equipment 106, and execution feedback equipment 107.

[0023] One or more control devices are housed in the workshop equipment control cabinet 101. Optionally, a control system such as a MES (Manufacturing Execution System) can be deployed in the control devices in the workshop equipment control cabinet 101. In this application, dairy products that undergo a single fermentation process using one fermenter are referred to as a batch of dairy products. The types and inoculation amounts of lactic acid bacteria in each batch of dairy products can be obtained through the MES.

[0024] The quality inspection system equipment 102 deploys a quality inspection system to acquire relevant data obtained from quality inspection, such as quality-related parameters measured during fermentation, including the actual acidity value of the dairy product. Optionally, the acidity value can be the pH value.

[0025] The environmental monitoring point device 103 may include one or more devices for measuring environmental parameters. Optionally, the environmental monitoring point device 103 may include, but is not limited to, temperature sensors and humidity sensors. For example, a temperature sensor may be installed on the top or inside of each fermenter, and this temperature sensor can collect the fermenter temperature.

[0026] The data acquisition device 104 is connected to the workshop equipment control cabinet 101, the quality inspection system device 102 and the environmental monitoring point device 103 respectively. It is used to collect the data sent by the workshop equipment control cabinet 101, the quality inspection system device 102 and the environmental monitoring point device 103, and send the collected data to the quality prediction device 105 for fermentation process control.

[0027] In this embodiment, the data acquisition device 104, the quality prediction device 105, the process optimization device 106, and the execution feedback device 107 can be the same device or different devices.

[0028] The following describes a method for controlling the fermentation process of dairy products according to an embodiment of this application. Optionally, the following method is performed by the aforementioned control device for the fermentation process of dairy products, which includes... Figure 1 The data acquisition device 104, quality prediction device 105, process optimization device 106, and execution feedback device 107 are shown.

[0029] Figure 2 This is a schematic flowchart illustrating a method for controlling the fermentation process of dairy products, provided as an embodiment of this application. Figure 2 As shown, the methods for controlling the fermentation process of dairy products include the following steps: 201. Obtain production data related to the fermentation process of the target batch of dairy products; production data includes: process parameters; process parameters include the fermentation tank temperature corresponding to the target batch of dairy products, the collection time corresponding to the fermentation tank temperature, and the inoculation amount of bacteria during the fermentation of the target batch of dairy products.

[0030] 202. Input the production data into the quality prediction model to obtain the predicted acidity value of the target batch of dairy products.

[0031] 203. When the predicted acidity value does not fall within the preset acidity value range, production constraints are determined based on the predicted acidity value and the preset acidity value range; under the constraints of the production constraints, control instructions are obtained based on the production data and the predicted acidity value.

[0032] 204. Adjust parameters according to control commands and obtain the execution results corresponding to the control commands.

[0033] 205. Determine whether the execution result indicates successful adjustment.

[0034] If the execution result indicates that the adjustment failed, proceed to step 206; if the execution result indicates that the adjustment was successful, proceed to step 207.

[0035] 206. Based on the execution results and control instructions, redetermine the control instructions. Return to step 204.

[0036] 207. After the fermentation of the target batch of dairy products is completed, obtain fermentation-related data of the fermentation process of the target batch of dairy products; fermentation-related data include production data and quality index data of the target batch of dairy products.

[0037] 208. Based on fermentation-related data, incrementally train the quality prediction model to update the quality prediction model.

[0038] The data acquisition equipment obtains production data related to the fermentation process of the target batch of dairy products. This production data includes, but is not limited to, process parameters. Process parameters include the fermentation tank temperature corresponding to the target batch of dairy products, the acquisition time corresponding to the fermentation tank temperature, and the inoculum quantity during the fermentation of the target batch of dairy products.

[0039] The temperature of the fermentation tank corresponding to the target batch of dairy products can be measured by a temperature sensor installed on the top of the fermentation tank. Optionally, the accuracy of this temperature sensor is controlled within ±0.1℃. The temperature sensor can collect the fermentation tank temperature at a preset sampling period, for example, it can collect the fermentation tank temperature once every 10 seconds.

[0040] The sampling time corresponding to the fermenter temperature refers to the time during which the temperature of the fermenter is collected.

[0041] The inoculation amount can also be called the lactic acid bacteria inoculation amount. Usually, before fermentation begins, the required inoculation amount of the target batch of lactic acid bacteria is inoculated according to the prescribed inoculation amount. Optionally, this data can be obtained from the MES (Manufacturing Execution System).

[0042] In one possible embodiment, the process parameters further include at least one of the following data: fermentation time, uniformity of ultra-high temperature (UHT) sterilization temperature, and final acidity value at the fermentation endpoint. UHT sterilization temperature uniformity refers to whether the temperature of the milk is consistent at all points as it flows through the pipeline during ultra-high temperature instantaneous sterilization. Inconsistency will result in some dairy products being incompletely sterilized while others are overheated. The final acidity value at the fermentation endpoint refers to the specific acidity value that determines when fermentation should be stopped.

[0043] In one possible embodiment, the production data also includes at least one of the following: raw material parameters, equipment parameters, and environmental parameters.

[0044] Optionally, the raw material parameters may include at least one of the following: total bacterial count of raw milk, fat content of raw milk, protein content of raw milk, and lactose content of raw milk.

[0045] Optionally, the equipment parameters include at least one of the following: sterilization temperature, sterilization time, stirring speed of the fermenter, heat transfer efficiency of the fermenter, and homogenization pressure. The heat transfer efficiency of the fermenter is used to control the fermentation rate and precisely terminate fermentation. Homogenization pressure refers to using high pressure to break up the fat globules in the milk, preventing fat from rising to form a milk skin. Insufficient pressure will cause the product to separate into whey layers; excessive pressure may result in a metallic taste or excessive oxidation.

[0046] Optionally, environmental parameters may include at least one of the following: workshop temperature, humidity, air cleanliness, and water quality indicators for production water. Air cleanliness refers to the cleanliness of the environment surrounding the product during final filling, typically measured by the number of dust particles per cubic meter of air. Poor air cleanliness can lead to product contamination by airborne microorganisms in the final step, shortening shelf life. Water quality indicators for production water may include at least one of the following: acidity value, hardness value, and mineral content of the production water.

[0047] In one possible embodiment, the data acquisition device may be communicatively connected to multiple data acquisition sub-devices, each individually. For example, the data acquisition sub-devices may be... Figure 1 The workshop equipment control cabinet 101 shown is connected to the data acquisition device, quality inspection system device 102, and environmental monitoring point device 103. Regarding the time difference between real-time production data acquired by multiple data acquisition sub-devices and acidity detection 2-3 hours later, clock deviations exist between older and newer devices in the data acquisition sub-devices, potentially leading to incorrect time correlations between production data fluctuations and quality anomalies. Therefore, it is necessary to ensure the accuracy of the acquisition time for real-time production data. This can be achieved by calibrating the acquisition time as follows: multiple data acquisition sub-devices calibrate their respective local times according to the time source server. Production data related to the fermentation process of the target batch of dairy products is acquired, and the corresponding acquisition time for each production data point is determined based on the local time; the production data is then sent to the data acquisition device. Furthermore, the data acquisition device sends the production data to the quality prediction device according to the acquisition time and data priority. This ensures the accuracy of the production data acquisition time, guarantees a precise correlation between production data fluctuations and acidity anomalies, avoids misjudging later parameters as the root cause of anomalies due to data asynchrony, and reduces manual troubleshooting time.

[0048] The time source server can be deployed in the dairy production workshop. Multiple data acquisition sub-devices are each connected to the time source server. These sub-devices synchronize their clocks with the time source server to calibrate their respective local times.

[0049] Optionally, the specific process of multiple data acquisition sub-devices receiving and transmitting data can be as follows: After receiving data, the multiple data acquisition sub-devices can obtain the reception time based on the current local time. The reception time is used as the timestamp of the data, and the data along with the timestamp is sent to the data acquisition device. The data acquisition device will determine the data acquisition time based on the timestamp and the reception delay of the acquisition sub-devices.

[0050] The following explanation uses a temperature acquisition device as an example to illustrate the above process: The temperature acquisition device communicates with a time source server and performs local time calibration every preset time. For example, in a yogurt fermentation scenario, the preset time could be 5 minutes. After the temperature sensor acquires the temperature of the fermentation tank, it can transmit the data to the temperature acquisition device. Upon receiving the fermentation tank temperature, the temperature acquisition device obtains the reception time as a timestamp for the fermentation tank temperature and sends the fermentation tank temperature and timestamp to the data acquisition device. For example, a timestamp can be added to the header of each fermentation tank temperature data record. This timestamp can be accurate to microseconds and has the format: YYYY-MM-DD HH:MM:SS:ffffff, where YYYY represents the year, MM represents the month, DD represents the day, HH represents the hour, SS represents the minute, and ffffff represents the microsecond. The data acquisition device determines the acquisition time corresponding to the fermentation tank temperature based on the data transmission delay between the temperature sensor and the temperature acquisition device. Optionally, the data acquisition device can store a deviation table containing the transmission times of multiple data acquisition devices transmitting data to their respective corresponding data acquisition devices.

[0051] In practical applications, production data related to the fermentation process of the target batch of dairy products are obtained and input into the quality prediction model.

[0052] In one possible embodiment, due to the diverse types of production data, such as the frequent reception of data like fermenter temperature, the overall volume of received data is large, which may cause data transmission delays when inputting it into the quality prediction model. Therefore, different types of production data can be assigned preset data priorities. For example, fermenter temperature data has the highest priority; if data transmission congestion occurs, fermenter temperature data will be sent to the quality prediction model first.

[0053] In one possible embodiment, after acquiring the production data involved in the fermentation process of the target batch of dairy products, the production data can be preprocessed and then input into a quality prediction model. Optionally, preprocessing may include protocol conversion and / or data cleaning. Protocol conversion refers to converting the format of different types of production data into a unified format. Since different production workshops have diverse equipment types, heterogeneous communication protocols, and inconsistent data formats, protocol conversion is necessary to unify different types of heterogeneous communication protocols and different data formats. For example, a three-level data acquisition and cleaning process of "protocol identification - format conversion - cleaning" can be used. Specifically, nameplate information of each device can be acquired in advance, and the protocol can be automatically matched using the nameplate information to convert the received data. Based on the fermentation progress of the target batch, production data is collected at preset times, including fermenter temperature, stirring speed, lactic acid bacteria inoculation amount, and offline acidity detection results. Outliers in the production data are deleted to complete production data cleaning and standardization. For example, outliers such as "temperature jumping instantaneously from 43℃ to 50℃" caused by temperature sensor malfunction are removed.

[0054] Production data is input into a quality prediction model to obtain the predicted acidity value of the target batch of dairy products.

[0055] The quality prediction model is a trained neural network model. For example, it could be a model built using an online-learned Long Short-Term Memory network with a self-attention mechanism (LSTM-Transformer). Based on the input production data, the quality prediction model outputs a predicted acidity value.

[0056] When training a quality prediction model, historical data can be used as training samples to train the initially established model. The loss is calculated based on the loss function until the model converges, resulting in the quality prediction model. For example, historical data on "temperature and inoculum quantity sequence during fermentation 0-6 hours" can be used as input to the initially established model for training. The model outputs "predicted acidity values ​​for fermentation 3-6 hours," allowing the model to focus on learning the nonlinear mapping relationship between "parameters in the first 2 hours" and "acidity in the last 3 hours." For instance, if the fermenter temperature is >43.5℃ for 10 minutes, the acidity decrease rate increases by 0.05 pH / h after 3 hours.

[0057] In practical applications, during the fermentation process of each batch of dairy products, a quality prediction model is used to predict the acidity value and implement subsequent adjustments. Therefore, after the fermentation process of each batch of dairy products is completed, fermentation-related data for that batch can be obtained. This fermentation-related data is then used for incremental training of the quality prediction model.

[0058] In one possible embodiment, incremental training of the quality prediction model can be performed as follows: Based on quality index data, determine the production stability index of the fermentation process for the target batch of dairy products. Based on the production stability index, determine the loss weights of fermentation-related data in the loss function, thus obtaining the loss function. Based on the fermentation-related data, and using the loss function as the basis for training termination, incrementally train the quality prediction model to update it.

[0059] Furthermore, the loss weight of fermentation-related data in the loss function can be determined as follows: If the production stability index indicates that the current production is in a fluctuating state, then the loss value of the fermentation-related data and the loss value of historical fermentation-related data are weighted and summed to obtain the loss function, where the loss weight of the fermentation-related data is greater than the loss weight of historical fermentation-related data; if the production stability index indicates that the current production is in a stable state, then the loss value of the fermentation-related data and the loss value of historical fermentation-related data are weighted and summed to obtain the loss function, where the loss weight of the fermentation-related data is equal to the loss weight of historical fermentation-related data.

[0060] For example, the weights of the loss function during incremental training can be dynamically adjusted based on production stability. For instance, when fermentation fluctuates significantly, it indicates poor current production stability. If the difference in lactose content in the raw milk is greater than 1%, it suggests that the equipment or production conditions are volatile. Incremental training should then focus more on recent or current fermentation-related data; therefore, the loss weight for recent or current fermentation-related data should be increased. When production is stable, the weights of historical and current fermentation-related data should be balanced.

[0061] Furthermore, a version control mechanism can be employed, generating a new version of the quality prediction model with each training iteration. The performance of the new and old versions is then evaluated through cross-validation. For example, if the accuracy of the new version is improved compared to the old version, the new version is used to replace the old one. Through this version control mechanism, the adaptation speed of the quality prediction model to fluctuations in raw milk composition and decreased equipment temperature control precision is shortened from retraining over days to online updates on an hourly basis, effectively improving the accuracy of quality anomaly predictions on the production line.

[0062] In one possible embodiment, the quality prediction model can also determine the risk level to which the current predicted acidity value belongs based on a preset risk level and output that risk level. Furthermore, in the dairy fermentation process, acidity anomalies have different effects at different fermentation stages; for example, a rapid decrease in acidity in the first 2 hours of fermentation may lead to excessive acidity later. In one possible implementation, a unified warning threshold can be used. However, this can easily lead to false alarms or missed alarms. In another possible implementation, multi-level warning threshold ranges are set for quality indicators at different fermentation stages. For example, a three-level warning threshold range can be set. A risk heatmap is displayed in real-time through the system interface, using different colors to mark the risk level of each process, and a detailed data dashboard is provided, displaying risk indicators, influencing factors, and historical trends. When a warning is triggered, warning information is pushed through multiple channels. For example, a level 3 warning triggers a workshop audible and visual alarm, a level 2 warning is pushed to an engineer's application (APP), and a level 1 warning is only marked on the central control screen, reducing invalid interference and false alarm rates.

[0063] If the predicted acidity value does not fall within the preset acidity range, production constraints are determined based on both the predicted and preset acidity values. Under these constraints, control commands are generated by adjusting production data and the predicted acidity value. These commands, through parameter adjustments, ensure that the actual acidity value at the target fermentation time conforms to the preset acidity range.

[0064] The preset acidity range refers to the range of acidity values ​​corresponding to the fermentation stage to which the predicted acidity value belongs. Quality indicators for different fermentation stages of dairy products are governed by relevant industry or production regulations; therefore, quality indicators are preset according to these regulations for different fermentation stages. These preset quality indicators include preset acidity values. In the above embodiments, the multi-level warning threshold ranges set for the quality indicators of different fermentation stages are typically smaller than the preset acidity range for that fermentation stage.

[0065] Production constraints refer to the conditions that must be followed when predicting and controlling acidity values. Production constraints may include a preset acidity value range. Optionally, production constraints may also include, but are not limited to, at least one of the following: equipment operation constraints, production completion time, and unit product energy consumption standard value. Equipment operation constraints refer to limiting factors related to the equipment itself during the production process, such as a minimum fermentation tank temperature of 41°C. Production completion time refers to the time before which current production needs to be completed. Unit product energy consumption standard value refers to the energy consumption required to produce one unit of product that needs to be controlled below this value. Optionally, the unit product energy consumption standard value can be obtained through an energy management system.

[0066] In practical applications, after obtaining the predicted acidity value, it is necessary to determine whether the predicted acidity value falls within the preset acidity value range. If the predicted acidity value falls within the preset acidity value range, no further processing is required. If the predicted acidity value does not fall within the preset acidity value range, production constraints are determined based on the predicted acidity value and the preset acidity value range. Under the constraints of the production constraints, control commands are obtained by making adjustments and predictions based on production data and the predicted acidity value.

[0067] In an optional embodiment, determining whether the predicted acidity value falls within a preset acidity range can be achieved using a Proximal Policy Optimization (PPO) algorithm. For example, the input layer of the policy network in the PPO algorithm receives production constraints, production data, and the predicted acidity value. For instance, the fermenter temperature is 43.2°C after 2 hours of fermentation, and the predicted pH after 4 hours is 4.1. The output layer is the probability distribution of fermentation temperature adjustments. Optionally, the output layer may also include, but is not limited to, the fermenter's stirring speed. During training, considering the time lag characteristics of dairy products, a time decay coefficient is introduced when collecting training data in the production environment. This time decay coefficient means that the current temperature adjustment has a weight of 0.8 on the acidity after 2 hours and 0.3 on the acidity after 1 hour. The dominance value is calculated using generalized dominance estimation. The policy network is updated by pruning importance sampling to optimize the dominance function. Pareto front calculation is performed: through multiple iterations of training, multiple optimization schemes satisfying different objective priorities are generated, and the optimal solution is provided by plotting the Pareto front curve. This optimal solution is the control command obtained from the regulation prediction. Typically, control commands cannot be adjusted drastically immediately; for example, the temperature cannot be drastically reduced immediately to avoid insufficient acidity later. For instance, a control command might be to ferment for 2-3 hours, with the fermentation temperature decreasing from 43℃ to 42.7℃. Furthermore, the generated commands undergo multi-parameter conflict detection; if they exceed equipment limits, the control commands are automatically corrected and recalculated.

[0068] Then, the control command is executed, and the execution result corresponding to the control command is obtained; if the execution result indicates that the adjustment has failed, the control command is re-determined based on the execution result and the control command.

[0069] In practical applications, the response characteristics of equipment on the production site vary greatly, and the heat transfer efficiency of different fermenters differs. This leads to different temperature adjustment response speeds, which may result in problems such as chaotic command execution and blurred protocol security boundaries. Furthermore, traditional processes rely on manual intervention, leading to low efficiency and high error rates. To address these issues on the production site, this embodiment, specifically for dairy product fermentation, automatically executes control commands.

[0070] Optionally, before executing control commands, protocol conversion can be performed to convert them into the protocol corresponding to the control command execution device. Furthermore, an industrial protocol conversion engine can be used to convert control commands from a general format to a device-specific protocol. For example, "adjust the temperature to 42.7℃" can be converted into a proportional-integral-derivative (PID) control signal for the fermenter (accuracy ±0.05℃).

[0071] In practical applications, after executing control commands, it is also necessary to monitor the execution results, which reflect the effectiveness of the execution. The execution results can indicate whether the adjustment was successful or failed. The basis for indicating the execution results can include two dimensions: the actual acidity value adjustment and the status of other quality indicators. In the dairy fermentation scenario of this application, the adjustment parameters are used to regulate the subsequent actual acidity value. However, due to the interrelationships between parameters, adjusting the parameters may lead to changes in other quality indicators besides the acidity value. Therefore, after executing control commands, it is necessary to monitor the execution effect to ensure product quality. Furthermore, in dairy production, simultaneously optimizing objectives such as "qualified acidity," "reduced energy consumption," and "shortened fermentation time" avoids sacrificing other objectives for the sake of pursuing one, such as causing excessive acidity due to rapid fermentation. It can combine objectives with the current production situation to adjust decision-making strategies, thereby achieving synergistic optimization of multiple objectives.

[0072] In one possible embodiment, the execution result is obtained by detecting whether the quality indicator data of the target batch of dairy products meets the quality requirements at and before reaching the target fermentation time. This requires detecting whether the actual acidity value of the target batch of dairy products meets the preset range, and whether other quality indicator data besides the acidity value meets the quality requirements.

[0073] For example, monitoring points can be inserted into the fermenter control loop to collect data from each data acquisition sub-device in real time after the control commands are executed. For instance, the actual temperature value of the fermenter can be recorded every 30 seconds. The actual execution result is compared with the target temperature value specified in the quality regulations, and the error rate between the two is calculated. If the temperature fluctuation exceeds the specified range, for example, a fluctuation exceeding 0.1℃, a fine-tuning is triggered; for example, if the temperature briefly rises to 42.8℃, readjustment is performed. In addition to equipment parameters and other quality indicator data, offline acidity detection data, such as the pH value after 3 hours of fermentation, is collected simultaneously to verify whether it meets the preset acidity value range. Thus, the effectiveness of the control commands is verified from a quality perspective.

[0074] If the execution result indicates adjustment failure, i.e., the actual acidity value is detected to be outside the preset acidity range and / or other quality indicators do not meet the specified range of the quality indicators, then the control instructions are redefined based on the execution result and control instructions. The redefined control instructions are then executed. Optionally, if the execution result indicates adjustment failure, the parameters that specifically do not meet the specified range of the quality indicators should also be considered when redefined the control instructions.

[0075] If the execution result indicates successful adjustment, fermentation-related data for the target batch of dairy products will be acquired upon completion of fermentation. This fermentation-related data includes production data and quality indicator data for the target batch of dairy products. Based on this fermentation-related data, the quality prediction model will be incrementally trained to update the model.

[0076] Among them, quality indicator data refers to relevant data reflecting the quality of the target batch of dairy products. Quality indicator data includes the acidity value of the target batch of dairy products.

[0077] Optionally, the quality indicator data may also include at least one of the following: color value, flavor and odor value, texture analysis index, viscosity value, fat content, protein content, lactose content, viable lactic acid bacteria count, yeast count, mold count, and pathogenic bacteria count. Among them, the texture analysis index includes at least one of the following: hardness value, elasticity value, and adhesiveness value.

[0078] Alternatively, if the execution result indicates successful adjustment, one possible implementation may not require special processing. Another possible implementation stores the first optimized data in a knowledge base. The first optimized data includes: production data, predicted acidity values, control instructions, and execution results.

[0079] In subsequent regulation, the current target production scenario is determined, and second optimized data with a similarity threshold reaching the target production scenario is obtained from the knowledge base. Target scenarios include: scenarios using new equipment and / or scenarios inoculating new bacterial strains; when making regulation predictions under the target production scenario, the second optimized data is used as a reference for regulation prediction.

[0080] Optionally, the data in the knowledge base can be stored in a database.

[0081] In practical applications, production knowledge suffers from reduced timeliness due to the introduction of new processes and equipment, making it difficult to transform tacit experience into structured knowledge. Furthermore, significant differences exist across production lines, hindering the reuse of experience. In dairy fermentation scenarios, the time lag characteristics of new lactic acid bacteria strains, with fermentation speeds potentially around 10% faster than traditional strains, can render existing parameters inapplicable, requiring lengthy debugging periods for traditional methods. Therefore, the first optimized data is stored in a knowledge base. Optionally, key information can be extracted from the first optimized data to construct a knowledge graph and save it to the knowledge base. Optionally, the knowledge graph can also include the causal chain of this adjustment, for example, an inoculum amount of 3.2% and a temperature of 43℃ → an acidity drop from 6.5 to 4.3 requiring 4 hours. When a new strain needs optimization, a search is performed in the knowledge base to find the most similar historical cases for parameter adaptation. The adaptation process could involve comparing the time lag characteristics of the new strain with those of a historical strain. When the similarity value is ≥80%, the second optimized data from the historical strain can be directly reused. When the similarity value is between 60% and 80%, the second optimization data of the historical strain can be fine-tuned using an adapter, for example, by shortening the fermentation time by 10%. When optimizing a new production line, the adaptation process can be similar to the treatment process for the new strain, or the optimization strategy can be adjusted based on the equipment characteristics and raw material differences of the new production line to generate a customized solution. The optimization results are then fed back to the knowledge base to automatically update the knowledge graph.

[0082] By deploying a protocol conversion engine and security verification mechanism compatible with devices and mitigating risks, a fully automated closed-loop process is ultimately formed, from command reception, conversion, execution to verification, requiring no manual intervention throughout. This solution effectively improves parameter adjustment accuracy, process continuity, and device compatibility, reduces human error, and significantly enhances production efficiency and management effectiveness.

[0083] During model training and dynamic optimization, effective parameter adjustment strategies and optimal parameter combinations under different production conditions are stored as rules in a rule base, constructing a reusable "process parameter-quality result" knowledge base. When encountering similar production scenarios later, relevant rules can be directly retrieved from the rule base to quickly formulate optimization strategies without the need for complex re-analysis and calculations. For example, when changing lactic acid bacteria strains, there's no need to re-explore parameters, and the time for new users to master fermentation optimization skills is significantly shortened—for instance, from 3 months to about 1 week. Based on meta-learning, process strategy transfer is constructed. A shared feature extractor captures common features, and an adapter fine-tunes the parameters of specific processes for rapid optimization. This not only effectively accumulates process optimization knowledge, avoiding knowledge loss due to personnel turnover, but also allows for rapid reuse of knowledge in scenarios such as new production line construction and process improvement, reducing the cost and time of process optimization and improving the company's process innovation capabilities and competitiveness.

[0084] In summary, the method provided in this embodiment monitors various data in the dairy production process in real time, including the lactose content of raw milk and the heat transfer efficiency of the fermentation tank. It continuously learns and analyzes data using deep learning algorithms and reinforcement learning mechanisms. When fluctuations in the lactose content of raw milk or equipment performance degradation are detected, the method automatically adjusts process parameters to optimize strategies and maintain stable product acidity. The device can automatically adapt to dynamic variables such as fluctuations in raw material quality and equipment performance degradation. Based on preset optimization goals and real-time monitored production data, it proactively analyzes various situations in the production process. In dairy fermentation, when a temperature fluctuation exceeding 0.5°C within 1 hour of fermentation is detected, it proactively initiates an optimization process without waiting for external instructions, quickly formulates optimization strategies (such as fine-tuning the temperature), and executes parameter adjustments. This prevents problems from occurring or escalating, effectively preventing taste deterioration and whey separation caused by excessive acidity, reducing manual intervention, and making the production process more intelligent and proactive. The method analyzes the impact weight of parameter adjustments on quality indicators, automatically generates a visual decision path, clearly presenting the complete causal chain from parameter adjustment to quality improvement, achieving transparency and interpretability in production optimization decisions, and facilitating rapid traceability.

[0085] In some embodiments, Figure 1 In the control scenario of the dairy product fermentation process shown, intelligent agents can be deployed in the workshop equipment control cabinet 101, the quality inspection system equipment 102, and the environmental monitoring point equipment 103 to automatically collect relevant data and send it to the data acquisition device 104. For example, a quality data acquisition intelligent agent can be deployed in the quality inspection system equipment 102 to acquire quality-related parameters.

[0086] Intelligent agents are deployed in the quality prediction device 105, the process optimization device 106, and the execution feedback device 107, respectively. Specifically, a quality prediction intelligent agent is deployed in the quality prediction device 105; a process optimization intelligent agent is deployed in the process optimization device 106; and an execution feedback intelligent agent is deployed in the execution feedback device 107. Optionally, each of these intelligent agents can perform its corresponding function through preset prompts.

[0087] In this embodiment, the data acquisition device 104, quality prediction device 105, process optimization device 106, and execution feedback device 107 can be the same server or different servers. Correspondingly, the data acquisition agent deployed in the data acquisition device 104, the quality prediction agent deployed in the quality prediction device 105, the process optimization agent deployed in the process optimization device 106, and the execution feedback agent deployed in the execution feedback device 107 can be the same agent or different agents. Figure 1The example shows the data acquisition device 104, quality prediction device 105, process optimization device 106, and execution feedback device 107 as different servers.

[0088] The following describes a method for regulating the dairy fermentation process through the coordinated action of various intelligent agents. Optionally, the data acquisition agent in the following embodiments is... Figure 1 The intelligent agent deployed in the data acquisition device 104 shown. The quality prediction intelligent agent is... Figure 1 The intelligent agent deployed in the quality prediction device 105 shown. The process optimization intelligent agent is... Figure 1 The intelligent agent deployed in the process optimization equipment 106 shown. The execution feedback intelligent agent is... Figure 1 The intelligent agent deployed in the execution feedback device 107 shown.

[0089] Figure 3 This is a schematic flowchart illustrating another method for controlling the fermentation process of dairy products, provided as an embodiment of this application. Figure 3 As shown, the methods for controlling the fermentation process of dairy products include the following steps: 301. The data acquisition agent acquires production data related to the fermentation process of the target batch of dairy products; the production data includes: process parameters; the process parameters include the fermentation tank temperature corresponding to the target batch of dairy products, the acquisition time corresponding to the fermentation tank temperature, and the inoculation amount of bacteria during the fermentation of the target batch of dairy products.

[0090] 302. The quality prediction agent inputs production data into the quality prediction model to obtain the predicted acidity value of the target batch of dairy products.

[0091] 303. When the predicted acidity value does not fall within the preset acidity value range, the process optimization intelligent agent determines production constraints based on the predicted acidity value and the preset acidity value range; and under the constraints of the production constraints, it obtains control instructions based on the production data and the predicted acidity value.

[0092] 304. The execution feedback agent adjusts parameters according to control commands and obtains the execution results corresponding to the control commands.

[0093] 305. The execution feedback agent determines whether the execution result indicates successful adjustment.

[0094] If not, proceed to step 306; if yes, proceed to step 307.

[0095] 306. The execution feedback agent feeds back the execution results and control instructions to the process optimization agent so that the process optimization agent can redetermine the control instructions.

[0096] 307. At the end of the fermentation of the target batch of dairy products, the quality prediction agent obtains fermentation-related data of the fermentation process of the target batch of dairy products; the fermentation-related data includes production data and quality index data of the target batch of dairy products.

[0097] 308. The quality prediction agent incrementally trains the quality prediction model based on fermentation-related data to update the quality prediction model.

[0098] Among them, with the above Figure 2 Similar descriptions to the illustrated embodiments will not be repeated in this embodiment.

[0099] In this embodiment, the data acquisition agent obtains production data related to the fermentation process of the target batch of dairy products.

[0100] In one possible embodiment, the data acquisition agent can communicate with multiple acquisition sub-agents. Regarding the time difference between the real-time production data acquired by the multiple acquisition sub-agents and the acidity detection 2-3 hours later, clock discrepancies exist between older and newer equipment, potentially leading to incorrect time correlations between production data fluctuations and quality anomalies. Therefore, it is necessary to ensure the accuracy of the acquisition time for real-time data in the production data. Acquisition time calibration can be achieved as follows: multiple acquisition sub-agents are each used to calibrate their local time according to the time source server. They acquire the production data related to the fermentation process of the target batch of dairy products and determine the corresponding acquisition time for each piece of production data based on their local time; then, the production data is sent to the data acquisition agent. Specifically, the data acquisition agent sends the production data to the quality prediction agent according to the acquisition time and data priority.

[0101] The time source server can be deployed in the dairy production workshop. Multiple data acquisition sub-agents communicate with the time source server. Each sub-agent synchronizes its local time with the time source server.

[0102] Optionally, the specific process of multiple data acquisition sub-agents receiving and transmitting data can be as follows: After receiving data, the multiple data acquisition sub-agents can obtain the reception time based on their current local time. They then use the reception time as the timestamp of the data and send the data along with the timestamp to the data acquisition agent. The data acquisition agent will then determine the data acquisition time based on the timestamp and the reception delay of the data acquisition sub-agents.

[0103] Optionally, the multiple data collection sub-agents may include the above-mentioned... Figure 1 The intelligent agents deployed in the workshop equipment control cabinet 101, the quality inspection system equipment 102, and the environmental monitoring point equipment 103 in the illustrated embodiment.

[0104] The following explanation uses a temperature acquisition agent as an example to illustrate the above process: The temperature acquisition agent communicates with the time source server and performs local time calibration every preset time. For example, in a yogurt fermentation scenario, the preset time could be 5 minutes. After the temperature sensor acquires the temperature of the fermentation tank, it can transmit the data to the temperature acquisition agent. Upon receiving the fermentation tank temperature, the temperature acquisition agent obtains the reception time as a timestamp for the fermentation tank temperature and sends the fermentation tank temperature and timestamp to the data acquisition agent. For example, a timestamp can be added to the header of each fermentation tank temperature data record. This timestamp can be accurate to microseconds and has the format: YYYY-MM-DD HH:MM:SS:ffffff, where YYYY represents the year, MM represents the month, DD represents the day, HH represents the hour, SS represents the minute, and ffffff represents the microsecond. The data acquisition agent determines the acquisition time corresponding to the fermentation tank temperature based on the data transmission delay between the temperature sensor and the temperature acquisition agent. Optionally, the data acquisition agent can store a deviation table, which contains the transmission times of multiple acquisition sub-agents transmitting data to their respective acquisition sub-agents.

[0105] In practical applications, the data acquisition agent obtains production data related to the fermentation process of the target batch of dairy products and sends it to the quality prediction agent.

[0106] In one possible embodiment, the data acquisition agent directly forwards the received production data to the quality prediction agent. However, since the data acquisition agent receives various types of data, such as fermenter temperature data which is received frequently, the overall amount of data received is large, which may cause data transmission delays when sending it to the quality prediction agent. Therefore, different types of production data can be preset with data priorities. For example, fermenter temperature data has the highest priority, and if data transmission congestion occurs, fermenter temperature data is sent to the quality prediction agent first.

[0107] In one possible embodiment, the data acquisition agent can preprocess the production data before sending it to the quality prediction agent, and then send the preprocessed production data to the quality prediction agent. Optionally, preprocessing may include protocol conversion and / or data cleaning. Protocol conversion refers to converting the format of different types of production data into a unified format. Since different production workshops have diverse equipment types, heterogeneous communication protocols, and inconsistent data formats, protocol conversion is necessary to unify different types of heterogeneous communication protocols and different data formats. For example, the data acquisition agent can use a three-level data acquisition and cleaning process: protocol identification, format conversion, and cleaning. Specifically, the nameplate information of each device can be pre-acquired, and the protocol can be automatically matched using the device nameplate information to perform protocol conversion on the received data. Based on the fermentation progress of the target batch, production data is collected at preset times, including fermenter temperature, stirring speed, lactic acid bacteria inoculation amount, and offline acidity detection results. Outliers in the production data are deleted to complete production data cleaning and standardization. For example, outliers such as "temperature jumping instantaneously from 43℃ to 50℃" caused by temperature sensor malfunction are removed.

[0108] The quality prediction agent is used to input production data into the quality prediction model to obtain the predicted acidity value of the target batch of dairy products.

[0109] The quality prediction agent is also used to acquire fermentation-related data of the target batch of dairy products at the end of fermentation. This fermentation-related data includes production data and quality indicator data of the target batch of dairy products. Based on this fermentation-related data, the quality prediction model is incrementally trained to update the model.

[0110] In practical applications, during the fermentation process of each batch of dairy products, a quality prediction model is used to predict the acidity value and implement subsequent adjustments. Therefore, after the fermentation process of each batch of dairy products is completed, fermentation-related data for that batch can be obtained. This fermentation-related data is then used for incremental training of the quality prediction model.

[0111] In one possible embodiment, the quality prediction agent incrementally trains the quality prediction model as follows: Based on quality indicator data, the quality prediction agent determines the production stability index of the fermentation process for the target batch of dairy products. Based on the production stability index, it determines the loss weights of fermentation-related data in the loss function, thus obtaining the loss function. Based on the fermentation-related data, and using the loss function as the basis for training termination, the quality prediction model is incrementally trained to update the model.

[0112] Furthermore, the quality prediction agent can determine the loss weight of fermentation-related data in the loss function in the following ways: if the production stability index indicates that the current production is in a fluctuating state, then the loss value of the fermentation-related data and the loss value of historical fermentation-related data are weighted and summed to obtain the loss function, and the loss weight of the fermentation-related data is greater than the loss weight of historical fermentation-related data; if the production stability index indicates that the current production is in a stable state, then the loss value of the fermentation-related data and the loss value of historical fermentation-related data are weighted and summed to obtain the loss function, and the loss weight of the fermentation-related data is equal to the loss weight of historical fermentation-related data.

[0113] Furthermore, a version control mechanism can be employed, generating a new version of the quality prediction model each time it is trained. The performance of the new and old versions is then evaluated through cross-validation. For example, if the accuracy of the new version is improved compared to the old version, the new version is used to replace the old one. Moreover, after the replacement of the old and new versions, the process optimization agent can be notified. Through this version control mechanism, the adaptation speed of the quality prediction model to fluctuations in raw milk composition and decreased equipment temperature control accuracy is shortened from retraining over days to online updates on an hourly basis, effectively improving the accuracy of quality anomaly prediction on the production line.

[0114] In one possible embodiment, the quality prediction model can also determine the risk level to which the current predicted acidity value belongs based on a preset risk level and output that risk level. Furthermore, in the dairy fermentation process, acidity anomalies have different effects at different fermentation stages; for example, a rapid decrease in acidity in the first 2 hours of fermentation may lead to excessive acidity later. In one possible implementation, a unified warning threshold can be used. However, this can easily lead to false alarms or missed alarms. In another possible implementation, multi-level warning threshold ranges are set for quality indicators at different fermentation stages. For example, a three-level warning threshold range can be set. A risk heatmap is displayed in real-time through the system interface, using different colors to mark the risk level of each process, and a detailed data dashboard is provided, displaying risk indicators, influencing factors, and historical trends. When a warning is triggered, warning information is pushed through multiple channels. For example, a level 3 warning triggers a workshop audible and visual alarm, a level 2 warning is pushed to an engineer's application (APP), and a level 1 warning is only marked on the central control screen, reducing invalid interference and false alarm rates.

[0115] In this embodiment, the optimization results are fed back to the quality prediction model, enabling online iteration of the decision-making strategy. After the process parameters are adjusted, the agent can continuously perceive environmental changes and adjust its behavior based on feedback, forming a closed loop of "perception-decision-action-feedback," thus achieving iterative updates to the quality prediction model. The online iterative optimization method targeting changes in the fermentation time lag characteristics of dairy products allows the agent to continuously improve its optimization capabilities as production data accumulates and the production environment dynamically changes, maintaining optimal decision-making performance. This significantly enhances the model's flexibility and adaptability, whereas ordinary machine learning models are often trained offline and applied only once, making it difficult to cope with dynamic environments.

[0116] The process optimization agent is used to: determine production constraints based on the predicted acidity value and the preset acidity value range when the predicted acidity value does not fall within the preset range; under the constraints of the production constraints, adjust and predict based on production data and the predicted acidity value to obtain control commands; and use the control commands to adjust parameters to ensure that the actual acidity value at the target fermentation time conforms to the preset acidity value range.

[0117] The preset acidity range refers to the range of acidity values ​​corresponding to the fermentation stage to which the predicted acidity value belongs. Quality indicators for different fermentation stages of dairy products are governed by relevant industry or production regulations; therefore, quality indicators are preset according to these regulations for different fermentation stages. These preset quality indicators include preset acidity values. In the above embodiments, the multi-level warning threshold ranges set for the quality indicators of different fermentation stages are typically smaller than the preset acidity range for that fermentation stage.

[0118] Production constraints refer to the conditions that must be followed when predicting and controlling acidity values. Production constraints may include a preset acidity value range. Optionally, production constraints may also include, but are not limited to, at least one of the following: equipment operation constraints, production completion time, and unit product energy consumption standard value. Equipment operation constraints refer to limiting factors related to the equipment itself during the production process, such as a minimum fermentation tank temperature of 41°C. Production completion time refers to the time before which current production needs to be completed. Unit product energy consumption standard value refers to the energy consumption required to produce one unit of product that needs to be controlled below this value. Optionally, the unit product energy consumption standard value can be obtained through an energy management system.

[0119] In practical applications, after receiving the predicted acidity value, the process optimization agent determines whether the predicted acidity value falls within the preset acidity range. If the predicted acidity value falls within the preset range, no further action is required. If the predicted acidity value does not fall within the preset range, production constraints are determined based on the predicted acidity value and the preset acidity range. Under the constraints, control predictions are made based on production data and the predicted acidity value to obtain control commands. The process optimization agent sends the control commands to the execution feedback agent. The execution feedback agent adjusts the parameters according to the control commands to ensure that the actual acidity value at the target fermentation time conforms to the preset acidity range.

[0120] In an optional embodiment, the process optimization agent can be implemented using the Proximal Policy Optimization (PPO) algorithm. For example, the input layer of the policy network in the PPO algorithm receives production constraints, production data, and predicted acidity values. For instance, the fermenter temperature is 43.2°C after 2 hours of fermentation, and the predicted pH after 4 hours is 4.1. The output layer is the probability distribution of fermentation temperature adjustments. Optionally, the output layer may also include, but is not limited to, the fermenter's stirring speed. During training, considering the time lag characteristics of dairy products, a time decay coefficient is introduced when collecting training data in the production environment. This time decay coefficient means that the current temperature adjustment has a weight of 0.8 on the acidity after 2 hours and 0.3 on the acidity after 1 hour. The advantage value is calculated using generalized advantage estimation. The advantage function is optimized by updating the policy network through importance sampling. Pareto front calculation is performed: through multiple iterations of training, multiple optimization schemes satisfying different objective priorities are generated, and the optimal solution is provided by plotting the Pareto front curve. This optimal solution is the control command obtained through regulation prediction. Typically, control commands cannot be adjusted drastically immediately; for example, the temperature cannot be drastically reduced immediately to avoid insufficient acidity later. For instance, a control command might be to ferment for 2-3 hours, reducing the fermentation temperature from 43℃ to 42.7℃. Furthermore, the generated commands undergo multi-parameter conflict detection; if they exceed equipment limits, the control commands are automatically corrected and recalculated. Optionally, control commands can be sent to the execution feedback agent according to their urgency.

[0121] The execution feedback agent is used to: execute control commands and obtain the corresponding execution results; if the execution result indicates that the adjustment has failed, the execution result and control command are fed back to the process optimization agent so that the process optimization agent can re-determine the control command.

[0122] In practical applications, the response characteristics of equipment on the production site vary greatly, and the heat transfer efficiency of different fermenters differs. This leads to different temperature adjustment response speeds, which may result in problems such as chaotic command execution and blurred protocol security boundaries. Furthermore, traditional processes rely on manual intervention, leading to low efficiency and high error rates. To address these issues on the production site, this embodiment, specifically for dairy product fermentation, utilizes an execution feedback intelligent agent to automatically execute control commands.

[0123] Optionally, the execution feedback agent can perform protocol conversion on control commands, transforming them into protocols corresponding to the control command execution device. Furthermore, the execution feedback agent can incorporate an industrial protocol conversion engine, which converts control commands from a general format to a device-specific protocol. For example, "adjust the temperature to 42.7℃" can be converted into a proportional-integral-derivative (PID) control signal for the fermenter (accuracy ±0.05℃).

[0124] In practical applications, after the intelligent agent executes control commands, it is also necessary to detect the execution results reflecting the execution effectiveness. The execution results can indicate whether the adjustment was successful or failed. The basis for indicating the execution results can include two dimensions: the actual acidity value adjustment and the status of other quality indicators. In the dairy product fermentation scenario of this application, the adjustment parameters are used to regulate the subsequent actual acidity value. However, due to the interrelationship between parameters, adjusting the parameters may lead to changes in other quality indicators besides the acidity value. Therefore, after executing control commands, it is necessary to detect the execution effectiveness to ensure product quality. Furthermore, in dairy product production, through intelligent agent collaboration, objectives such as "acceptable acidity," "reduced energy consumption," and "shortened fermentation time" can be optimized simultaneously, avoiding situations where other objectives are sacrificed in pursuit of a single goal, such as exceeding acidity limits due to rapid fermentation. The intelligent agents achieve information sharing and interaction through communication methods such as message queues and API interfaces. The intelligent agents can adjust their decision-making strategies based on their own objectives and current production conditions, thereby achieving synergistic optimization of multiple objectives.

[0125] In one possible embodiment, the execution feedback agent obtains the execution result by detecting whether the quality indicator data of the target batch of dairy products meets the quality requirements at and before reaching the target fermentation time. This requires detecting whether the actual acidity value of the target batch of dairy products meets the preset range, and whether other quality indicator data besides the acidity value meets the quality requirements.

[0126] For example, monitoring points can be inserted into the fermenter control loop to collect real-time feedback data from the equipment after the control commands are executed. For instance, the actual temperature value of the fermenter can be recorded every 30 seconds. The actual execution result is compared with the target temperature value specified in the quality guidelines, and the error rate between the two is calculated. If the temperature fluctuation exceeds the specified range, for example, a fluctuation exceeding 0.1℃, the process optimization agent is triggered for fine-tuning. For example, if the temperature briefly rises to 42.8℃, the process optimization agent is returned for further adjustment. In addition to quality indicators such as equipment parameters, offline acidity detection data, such as the pH value after 3 hours of fermentation, is collected simultaneously to verify whether it meets the preset acidity range. Thus, the effectiveness of the control commands is verified from a quality perspective.

[0127] If the execution result indicates adjustment failure, i.e., the actual acidity value is detected to be outside the preset acidity range and / or other quality indicators do not meet the specified quality indicator range, the execution result and control instructions are fed back to the process optimization agent. Upon receiving the execution result indicating adjustment failure, the process optimization agent re-determines the control instructions. The process optimization agent sends the re-determined control instructions to the execution feedback agent for execution. Optionally, if the execution result indicates adjustment failure, the parameters that specifically do not meet the specified quality indicator range are also simultaneously sent to the process optimization agent.

[0128] If the execution result indicates successful adjustment, in one possible implementation, no special processing is required. In another possible implementation, the first optimization data is sent to the process optimization agent. The first optimization data includes: production data, predicted acidity value, control instructions, and execution results. The process optimization agent is also used to: store the first optimization data in a knowledge base. The process optimization agent is also used to: when determining that the current production scenario is a target scenario, retrieve second optimization data from the knowledge base that has a similarity to the target production scenario that reaches a preset similarity threshold. The target scenario includes: scenarios using new equipment and / or scenarios inoculating new microbial strains; when making regulation predictions under the target production scenario, the second optimization data is used as a reference for regulation predictions.

[0129] Optionally, data in the knowledge base can be managed through a knowledge management agent. The knowledge management agent is connected to the process optimization agent.

[0130] Optionally, the data in the knowledge base can be stored in a database.

[0131] In practical applications, production knowledge suffers from reduced timeliness due to the introduction of new processes and equipment, making it difficult to transform tacit experience into structured knowledge. Furthermore, significant differences exist across production lines, hindering the reuse of experience. In dairy fermentation scenarios, the time lag characteristics of new lactic acid bacteria strains, with fermentation speeds potentially around 10% faster than traditional strains, can render existing parameters inapplicable, requiring lengthy debugging periods with traditional methods. Therefore, each process optimization agent stores the first optimization data in a knowledge base. Optionally, key information can be extracted from the first optimization data to construct a knowledge graph and save it to the knowledge base. Optionally, the knowledge graph can also include the causal chain of the adjustment, for example, an inoculum amount of 3.2% and a temperature of 43℃ → an acidity drop from 6.5 to 4.3 requiring 4 hours. When a new strain needs optimization, the knowledge management agent searches the knowledge base, identifying the most similar historical cases for parameter adaptation. The adaptation process could involve comparing the time lag characteristics of the new strain with those of a historical strain. When the similarity value is ≥80%, the second optimization data from the historical strain can be directly reused. When the similarity value is between 60% and 80%, the second optimization data of the historical strain can be fine-tuned using an adapter, for example, by shortening the fermentation time by 10%. When optimizing a new production line, the adaptation process can be similar to the treatment process for the new strain, or the optimization strategy can be adjusted based on the equipment characteristics and raw material differences of the new production line to generate a customized solution. The optimization results are then fed back to the knowledge base to automatically update the knowledge graph.

[0132] By deploying a protocol conversion engine and security verification mechanism compatible with devices and mitigating risks, a fully automated closed-loop process is ultimately formed, from command reception, conversion, execution to verification, requiring no manual intervention throughout. This solution effectively improves parameter adjustment accuracy, process continuity, and device compatibility, reduces human error, and significantly enhances production efficiency and management effectiveness.

[0133] During model training and dynamic optimization, the agent stores effective parameter adjustment strategies and optimal parameter combinations under different production conditions in the form of rules in a rule base, constructing a reusable "process parameter-quality result" knowledge base. When encountering similar production scenarios later, relevant rules can be directly retrieved from the rule base to quickly formulate optimization strategies without the need for complex re-analysis and calculation. For example, when changing lactic acid bacteria strains, there is no need to re-explore parameters, and the time for new users to master fermentation optimization skills is significantly shortened, for example, from 3 months to about 1 week. Based on meta-learning, process strategy transfer is constructed. Common features are captured through a shared feature extractor, and combined with an adapter to fine-tune parameters for specific processes for rapid optimization strategies. This not only achieves effective accumulation of process optimization knowledge, avoiding knowledge loss due to personnel turnover, but also enables rapid reuse of knowledge in scenarios such as new production line construction and process improvement, reducing the cost and time of process optimization and improving the enterprise's process innovation capabilities and competitiveness.

[0134] In summary, the method provided in this embodiment uses an intelligent agent to monitor various data in the dairy production process in real time, including the lactose content of raw milk and the heat transfer efficiency of the fermentation tank. It continuously learns and analyzes data using deep learning algorithms and reinforcement learning mechanisms. When fluctuations in the lactose content of raw milk or equipment performance degradation are detected, the agent automatically adjusts process parameters to optimize strategies and maintain stable product acidity. The device can automatically adapt to dynamic variables such as fluctuations in raw material quality and equipment performance degradation. Based on preset optimization goals and real-time monitored production data, the intelligent agent proactively analyzes various situations in the production process. In dairy fermentation, when a temperature fluctuation exceeding 0.5°C within 1 hour of fermentation is detected, the agent will proactively initiate an optimization process without waiting for external instructions, quickly formulate optimization strategies (such as fine-tuning the temperature), and execute parameter adjustments. This prevents the occurrence or escalation of problems, effectively preventing taste deterioration and whey separation caused by excessive acidity, reducing human intervention, and making the production process more intelligent and proactive. The system analyzes the impact weight of parameter adjustments on quality indicators, automatically generates a visual decision path, and clearly presents the complete causal chain from parameter adjustment to quality improvement, thereby achieving transparency and interpretability in production optimization decisions and facilitating rapid traceability.

[0135] Figure 4 A structural diagram of a control device for a dairy product fermentation process provided in this application embodiment is shown below. Figure 4 As shown, the device includes the following modules: a data acquisition agent 401, a quality prediction agent 402, a process optimization agent 403, and an execution feedback agent 404.

[0136] The data acquisition agent 401 is used to acquire production data related to the fermentation process of the target batch of dairy products. The production data includes process parameters, which include the fermentation tank temperature corresponding to the target batch of dairy products, the acquisition time corresponding to the fermentation tank temperature, and the inoculation amount of bacteria during the fermentation of the target batch of dairy products.

[0137] The quality prediction agent 402 is used to input production data into the quality prediction model to obtain the predicted acidity value of the target batch of dairy products.

[0138] The process optimization agent 403 is used to determine production constraints based on the predicted acidity value and the preset acidity value range when the predicted acidity value does not fall within the preset acidity value range; and under the constraints of the production constraints, it performs regulation and prediction based on production data and the predicted acidity value to obtain control instructions; the control instructions are used to adjust parameters so that the actual acidity value at the target fermentation time meets the preset acidity value range.

[0139] The execution feedback agent 404 is used to adjust parameters according to control commands and obtain the execution results corresponding to the control commands. If the execution result indicates that the adjustment has failed, the execution result and control command are fed back to the process optimization agent so that the process optimization agent can re-determine the control command.

[0140] The quality prediction agent 402 is also used to acquire fermentation-related data of the target batch of dairy products at the end of fermentation, if the execution result indicates successful adjustment; the fermentation-related data includes production data and quality index data of the target batch of dairy products. Based on the fermentation-related data, the quality prediction model is incrementally trained to update the quality prediction model.

[0141] Figure 4 The device shown is similar in principle and technical effect to the method in the foregoing embodiments, as described in the foregoing embodiments, and will not be repeated here.

[0142] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 5 As shown, in practice, this electronic device includes a memory 21 and a processor 22.

[0143] Memory 21 is used to store computer programs and can be configured to store various other data to support operation on the electronic device. Examples of this data include instructions for any application or method used to operate on the electronic device, data structures, contact data, phone book data, messages, pictures, videos, etc.

[0144] The processor 22, coupled to the memory 21, is used to execute the computer program in the memory 21 to implement the method for controlling the dairy fermentation process provided in the foregoing embodiments.

[0145] Furthermore, such as Figure 5 As shown, the electronic device also includes other components such as a communication component 23, a display 24, a power supply component 25, and an audio component 26. Figure 5 The diagram only shows some components and does not mean that the electronic device includes only these components. Figure 5 The components shown are as follows. The electronic device in this embodiment can be a terminal device such as a desktop computer, laptop computer, smartphone, or IoT device, or a server device such as a conventional server, cloud server, or server array.

[0146] The aforementioned memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0147] The aforementioned communication component is configured to facilitate wired or wireless communication between the device containing the communication component and other devices. The device containing the communication component can access wireless networks based on communication standards, such as 2G, 3G, 4G / LTE, 5G, or combinations thereof. In one exemplary embodiment, the communication component receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel.

[0148] The aforementioned display includes a screen, which may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a Touch Panel, the screen can be implemented as a touchscreen to receive input signals from the user. The Touch Panel includes one or more touch sensors to sense touches, swipes, and gestures on the Touch Panel. The touch sensors can sense not only the boundaries of touch or swipe actions but also the duration and pressure associated with the touch or swipe operation.

[0149] The aforementioned power supply components provide power to various components within the device in which they reside. These power supply components may include a power management system, one or more power sources, and other components associated with generating, managing, and distributing power to the device in which they reside.

[0150] The aforementioned audio component can be configured to output and / or input audio signals. For example, the audio component includes a microphone (MIC) configured to receive external audio signals when the device containing the audio component is in an operating mode, such as call mode, recording mode, or voice recognition mode. The received audio signals can be further stored in memory or transmitted via a communication component. In some embodiments, the audio component also includes a speaker for outputting audio signals.

[0151] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed by a processor, enables the processor to implement the steps in the above-described method embodiments. The computer-readable storage medium includes volatile or non-volatile components, or a combination thereof, and can be removable or non-removable. Examples of computer-readable storage media include, but are not limited to, phase-change random access memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random-access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), flash memory or other memory technologies, CD-ROM, Digital Video Disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium. Accordingly, this application also provides a computer program product, which includes a computer program or instructions. When the computer program or instructions are executed by a processor, the processor is able to implement the steps in the above method embodiments. It should be understood that each step or combination of steps in the above method flow can be implemented by the computer program or instructions. In addition, these computer programs or instructions can be applied to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device, so that the processor of the general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device can be implemented as a means to implement the corresponding functions in the above method embodiments.

[0152] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for controlling the fermentation process of dairy products, characterized in that, The method includes: Obtain production data related to the fermentation process of the target batch of dairy products; the production data includes: process parameters; the process parameters include the collected fermentation tank temperature corresponding to the target batch of dairy products, the collection time corresponding to the fermentation tank temperature, and the inoculation amount of the microbial strain during the fermentation of the target batch of dairy products; The production data is input into the quality prediction model to obtain the predicted acidity value of the target batch of dairy products; If the predicted acidity value does not fall within the preset acidity value range, production constraints are determined based on the predicted acidity value and the preset acidity value range; under the constraints of the production constraints, control instructions are obtained based on the production data and the predicted acidity value. The parameters are adjusted according to the control instructions, and the execution results corresponding to the control instructions are obtained; If the execution result indicates that the adjustment has failed, then based on the execution result and the control instruction, the control instruction is re-determined; and the process returns to the step of adjusting the parameters according to the control instruction. If the execution result indicates successful adjustment, then after the fermentation of the target batch of dairy products is completed, fermentation-related data of the fermentation process of the target batch of dairy products are obtained; the fermentation-related data includes the production data and the quality index data of the target batch of dairy products. Based on the fermentation-related data, the quality prediction model is incrementally trained to update the quality prediction model.

2. The method according to claim 1, characterized in that, The step of incrementally training the quality prediction model based on the fermentation-related data to update the quality prediction model includes: Based on the quality index data, the production stability index of the fermentation process of the target batch of dairy products is determined; Based on the production stability index, the loss weight of the fermentation-related data in the loss function is determined, and the loss function is obtained; Based on the fermentation-related data, and using the loss function as the basis for training termination, the quality prediction model is incrementally trained to update the quality prediction model.

3. The method according to claim 2, characterized in that, The step of determining the loss weight of the fermentation-related data in the loss function based on the production stability index, and obtaining the loss function, includes: If the production stability index indicates that the current production is in a state of fluctuation, then the loss value of the fermentation-related data and the loss value of the historical fermentation-related data are weighted and summed to obtain the loss function, wherein the loss weight of the fermentation-related data is greater than the loss weight of the historical fermentation-related data. If the production stability index indicates that the current production is in a stable state, then the loss value of the fermentation-related data and the loss value of the historical fermentation-related data are weighted and summed to obtain the loss function, and the loss weight of the fermentation-related data is equal to the loss weight of the historical fermentation-related data.

4. The method according to claim 1, characterized in that, The quality index data includes: acidity value; the quality index data also includes at least one of the following: color value, flavor and odor index value, texture analysis index, viscosity value, fat content, protein content, lactose content, number of viable lactic acid bacteria, number of yeasts, number of molds and number of pathogenic bacteria; the texture analysis index includes at least one of the following: hardness value, elasticity value, and adhesiveness value.

5. The method according to claim 1, characterized in that, The execution result is obtained in the following way: The quality index data of the target batch of dairy products are tested to ensure that they meet the quality requirements at and before the target fermentation time. The execution result indicates adjustment failure, including: The quality index data of the target batch of dairy products were detected to be inconsistent with the quality regulations.

6. The method according to claim 5, characterized in that, The method further includes: If the execution result indicates successful adjustment, the first optimized data is stored in the knowledge base; the first optimized data includes: the production data, the predicted acidity value, the control command, and the execution result; When it is determined that the current production scenario is a target scenario, second optimized data with a similarity to the target production scenario reaching a preset similarity threshold is obtained from the knowledge base; the target scenario includes: scenarios of using new equipment and / or scenarios of inoculating new strains; when making regulation predictions under the target production scenario, the second optimized data is used as a reference for regulation predictions.

7. The method according to claim 1, characterized in that, The process parameters also include at least one of the following data: fermentation time, uniformity of UHT sterilization temperature, and acidity value at the fermentation endpoint. The production data also includes at least one of the following: raw material parameters, equipment parameters, and environmental parameters; The raw material parameters include at least one of the following: total bacterial count of raw milk, fat content of raw milk, protein content of raw milk, and lactose content of raw milk; The equipment parameters include at least one of the following: sterilization temperature, sterilization time, stirring speed of the fermenter, heat transfer efficiency of the fermenter, and homogenization pressure; The environmental parameters include at least one of the following: workshop temperature, humidity, air cleanliness, and water quality indicators for production water.

8. The method according to claim 1, characterized in that, The production data includes the collection time; The production data involved in obtaining the fermentation process of the target batch of dairy products includes: The system receives production data sent by multiple acquisition modules, wherein the acquisition time is determined by the multiple acquisition modules according to their respective local times, and the local times of the multiple acquisition modules are calibrated according to the time source server.

9. A device for controlling the fermentation process of dairy products, characterized in that, The device includes: The data acquisition module is used to acquire production data related to the fermentation process of the target batch of dairy products; the production data includes: process parameters; the process parameters include the acquired fermentation tank temperature corresponding to the target batch of dairy products, the acquisition time corresponding to the fermentation tank temperature, and the inoculation amount of the bacteria during the fermentation of the target batch of dairy products; The quality prediction module is used to input the production data into the quality prediction model to obtain the predicted acidity value of the target batch of dairy products. The process optimization module is used to determine production constraints based on the predicted acidity value and the preset acidity value range when the predicted acidity value does not fall within the preset acidity value range; and to obtain control instructions based on the production data and the predicted acidity value under the constraints of the production constraints. The execution feedback module is used to adjust parameters according to the control command and obtain the execution result corresponding to the control command; if the execution result indicates that the adjustment has failed, the execution result and the control command are fed back to the process optimization module so that the process optimization module can re-determine the control command. The quality prediction module is further configured to: if the execution result indicates successful adjustment, then after the fermentation of the target batch of dairy products is completed, acquire fermentation-related data of the fermentation process of the target batch of dairy products; the fermentation-related data includes the production data and the quality index data of the target batch of dairy products; and based on the fermentation-related data, incrementally train the quality prediction model to update the quality prediction model.

10. A non-transitory machine-readable storage medium, characterized in that, The non-transitory machine-readable storage medium stores executable code that, when executed by a processor of an electronic device, causes the processor to perform the method for controlling the fermentation process of dairy products as described in any one of claims 1-8.

Images