Intelligent monitoring and upgrading optimization system for dairy beverage production equipment

Abstract

The invention relates to the technical field of equipment monitoring analysis, in particular to an intelligent monitoring and upgrading optimization system for dairy beverage production equipment, which comprises a raw material fluctuation monitoring and identifying module for monitoring and identifying the quality fluctuation of raw materials in the dairy beverage production process in real time and automatically detecting the quality fluctuation data of the raw materials; the production process self-adaptive adjustment module is used for adjusting production process parameters in real time through a self-adaptive algorithm on the basis of the raw material quality fluctuation data so as to compensate the influence caused by raw material fluctuation; and the production process feedback and optimization module is used for feeding back the execution effect of the production process according to the product quality data collected in real time in the production process in combination with the adjustment of the production process parameters, and adjusting and optimizing the production process parameters based on the execution effect. According to the invention, product quality fluctuation caused by unstable quality of raw materials in the production process is avoided, so that the production stability and the product consistency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment monitoring and analysis, and in particular to an intelligent monitoring and upgrading optimization system for dairy beverage production equipment. Background Art

[0002] As the consumer market's requirements for the quality of dairy beverages continue to increase, the traditional dairy beverage production process faces a series of challenges. The production process of dairy beverages involves multiple links, including raw material procurement, production process, equipment operation, etc. The fluctuation or instability of any link will have a direct impact on the quality of the final product. The traditional production process often relies on manual monitoring and experience adjustment, which makes it difficult to achieve precise control of raw material fluctuations, production processes and product quality, and cannot respond to environmental changes in real time. This not only leads to low production efficiency, but may also cause waste of resources and unstable product quality.

[0003] On the one hand, the quality of the main raw materials of dairy beverages (such as milk, syrup, spices, etc.) has a certain volatility. This volatility may show different characteristics in each batch of production, directly affecting the key quality indicators of the final product such as taste, mouthfeel, color, etc. When dealing with these fluctuations, existing technologies usually rely on fixed production process parameters, which are difficult to make real-time adjustments based on the specific fluctuations of the raw materials, resulting in inconsistency in product quality.

[0004] On the other hand, product quality data (such as taste, color, density, etc.) involved in the production process of dairy beverages is crucial for monitoring and optimizing the production process. However, the feedback mechanism in existing technologies is often relatively simple, failing to effectively combine the equipment operation status with the product quality feedback system, and failing to make real-time process parameter adjustments based on data feedback. Therefore, the existing system cannot accurately respond to changes caused by raw material fluctuations during the production process, nor can it achieve intelligent optimization of the production process. Summary of the invention

[0005] The invention provides an intelligent monitoring and upgrading optimization system for dairy beverage production equipment.

[0006] Intelligent monitoring and upgrading optimization system for dairy beverage production equipment, including:

[0007] Raw material fluctuation monitoring and identification module: real-time monitoring and identification of raw material quality fluctuations in the production process of dairy beverages, including milk, syrup, and flavoring ingredients. Real-time collection of raw material quality data is performed through a sensor network, and raw material quality fluctuation data is automatically detected based on data analysis algorithms.

[0008] Production process adaptive adjustment module: Based on raw material quality fluctuation data, the production process parameters, including temperature, pressure, stirring speed, and heating time, are adjusted in real time through adaptive algorithms to compensate for the impact of raw material fluctuations;

[0009] Production process feedback and optimization module: Based on the product quality data (such as taste and color) collected in real time during the production process, combined with the adjustment of production process parameters, feedback is provided on the execution effect of the production process, and the production process parameters are adjusted and optimized based on the execution effect to ensure that the product quality of each production cycle remains consistent.

[0010] Optionally, the raw material quality data includes milk fat concentration, sugar concentration and flavor chemical component concentration, and the sensor network specifically includes:

[0011] Using chemical sensors to detect fat concentration in milk;

[0012] Measure the sugar concentration in syrup using a refractive index sensor;

[0013] Using spectral sensors to measure the concentration of chemical components in spices;

[0014] The data analysis algorithm automatically detects raw material quality fluctuation data through a sliding average method.

[0015] Optionally, the sliding average method is expressed as: Among them, SMA t is the sliding average at time t, O L is the raw material quality data value at the Lth moment, and n is the window size.

[0016] Optionally, the adaptive algorithm in the production process adaptive adjustment module includes multi-task learning (MTL) and reinforcement learning, wherein;

[0017] The multi-task learning part learns multiple target tasks at the same time. The target tasks include adjusting temperature, pressure, stirring speed and heating time. Each target task learns parameters through the shared layer of the neural network.

[0018] Reinforcement learning maximizes rewards by interacting with the environment and learning strategies. In the production process of dairy beverages, reinforcement learning optimizes production process parameters based on the output of multi-task learning. Multi-task learning provides preliminary process parameters adjusted based on raw material quality fluctuations, while reinforcement learning optimizes and adjusts production process parameters by interacting with the production process and using real-time feedback.

[0019] Optionally, in the multi-task learning stage, the target task is set represents the adjustment of process parameters i, including temperature, pressure, stirring speed and heating time. The goal of each target task is to learn an adjustment model that can output the corresponding process parameters according to the raw material quality fluctuation;

[0020] Input data: The input vector is the raw material quality fluctuation data, expressed as:

[0021] X=[C milk-fat ,C syrup-sugar ,C flavor ], where C milk-fat is the milk fat concentration, C syrup-sugar is the sugar concentration of the syrup, C flavor It is the concentration of spice;

[0022] Target task output: Each process parameter adjustment task has an independent objective function. The objective function is the core of multi-task learning. The goal is to adjust the corresponding process parameters according to the input data, including temperature y temp , pressure pressure , stirring speed y speed And heating time time ;

[0023] The multi-task learning is based on a shared underlying neural network for multiple tasks. The shared underlying neural network adopts a multi-layer feedforward neural network with a shared layer. The input vector is X and the output is different process parameters. The multi-task learning learns the relationship between the input data and the output target through the shared layer (shared intermediate layer), and at the same time allocates an independent output layer to each task to adjust different process parameters respectively.

[0024] Optionally, the multi-layer feedforward neural network structure includes:

[0025] Input layer: input vector X;

[0026] Hidden layer: that is, the shared neural network layer (shared layer);

[0027] Output layer: The output layer corresponding to each process parameter target task outputs temperature, pressure, stirring speed, and heating time respectively;

[0028] The loss function of multi-task learning is the weighted loss sum of multiple tasks:

[0029] in, For the task The loss function is For network tasks The predicted value of the jth sample, y ij For the task The true label of the jth sample, λ i is the weight coefficient of the task, and N is the number of samples.

[0030] Optionally, the reinforcement learning uses the output of the multi-task learning as the initialization input, and the production process parameters output by the multi-task learning will be used as part of the environment state. The reinforcement learning will start from the initial state. The reinforcement learning is defined as follows:

[0031] State S: current environment description, including current raw material quality data and production process parameters adjusted by multi-task learning;

[0032] Action A: The operation selected according to state S. In the production of dairy beverages, the action is to adjust the specific value of the production process parameters;

[0033] Reward R: As a feedback signal, it is used to evaluate whether the current state and actions taken contribute to the achievement of the goal;

[0034] Strategy π: The strategy is the rule for the agent to choose actions in a certain state. The goal is to maximize the cumulative reward by optimizing the strategy.

[0035] Optionally, the specific learning of the reinforcement learning includes:

[0036] Initialization: Set the initial state S0 to the production process parameters output by multi-task learning and the current raw material quality data;

[0037] Action selection: At each time step, the reinforcement learning agent selects the action based on the current state S at time t. t Select an action A t , that is, adjusting the production process parameters;

[0038] Environmental feedback: The production process produces a new state S according to the selected action t+1 , and generates a reward R based on this state t , that is, the actual results in the production process (such as product quality, efficiency, etc.);

[0039] Strategy update: Based on the rewards obtained, the reinforcement learning algorithm updates the strategy and optimizes future action selection;

[0040] Iteration: Repeat the action selection to strategy update, and finally obtain the optimal production process parameter adjustment plan through multiple interactive optimization strategies.

[0041] Optionally, the product quality data in the production process feedback and optimization module includes taste and color, wherein:

[0042] Taste data Q 口感 (t): reflects the taste quality of dairy beverages, obtained through sensory testing, instrumental measurement or survey feedback;

[0043] Color data Q 色泽(t): reflects the appearance color of dairy beverages, measured using a spectral sensor;

[0044] Compare with the preset target mass value and calculate the mass error ΔQ 口感 and ΔQ 色泽 :

[0045]

[0046] Mass error ΔQ 口感 and ΔQ 色泽 As the execution effect, the production process parameters are adjusted and optimized based on the execution effect. The adjustment and optimization includes adjusting the reward function R t :

[0047] When the quality error is large: reduce the reward to encourage the agent to adjust the production process;

[0048] When the quality error is small: increase the reward to encourage the agent to maintain the current process parameters.

[0049] Beneficial effects of the present invention:

[0050] The present invention collects raw material quality data such as milk fat concentration, syrup sugar concentration, and flavor chemical component concentration in real time, and combines the sliding average method to effectively monitor and identify quality fluctuations of raw materials. By accurately capturing raw material quality fluctuations, an important basis for subsequent production process adjustments is provided. Raw material fluctuations are instantly identified during the production process, and measures are taken in advance to avoid product quality fluctuations caused by unstable raw material quality during the production process, thereby improving production stability and product consistency.

[0051] The present invention realizes adaptive production process adjustment based on real-time monitored raw material quality fluctuation data through the combination of multi-task learning and reinforcement learning. The multi-task learning part uses the raw material quality fluctuation data to predict and adjust the production process parameters, such as temperature, pressure, stirring speed and heating time, in the preliminary stage to ensure that the production process can respond to the raw material fluctuation in time. On the basis of multi-task learning, reinforcement learning further optimizes the production process according to the real-time feedback of the production process, continuously adjusts and optimizes the process parameters through interactive learning and reward mechanism, so that the product quality of each production cycle meets the target requirements, and flexibly adjusts the production process parameters dynamically in a complex production environment, effectively compensates for the impact of raw material fluctuations, and improves the consistency of production efficiency and product quality.

[0052] The present invention collects product quality data (such as taste and color) in real time during the production process, and feeds back the execution effect of the production process based on these data, further adjusts and optimizes the production process, calculates the quality error through real-time comparison with the product quality data (such as taste and color), and feeds back the error as a reward signal to the reinforcement learning system, thereby adjusting the reward function to ensure that the product quality of each production cycle is stable and consistent, thereby improving the intelligence level of the production line and the controllability of product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0054] Figure 1 A schematic diagram of a system module according to an embodiment of the present invention;

[0055] Figure 2 Schematic diagram of reinforcement learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0057] It should be noted that the references to "one embodiment", "an embodiment", "an exemplary embodiment", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0058] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0059] like Figure 1-Figure 2 As shown in the figure, the intelligent monitoring and upgrading optimization system for dairy beverage production equipment includes:

[0060] Raw material fluctuation monitoring and identification module: real-time monitoring and identification of raw material quality fluctuations in the production process of dairy beverages, including milk, syrup, and flavoring ingredients. Real-time collection of raw material quality data is performed through a sensor network, and raw material quality fluctuation data is automatically detected based on data analysis algorithms.

[0061] Production process adaptive adjustment module: Based on raw material quality fluctuation data, the production process parameters, including temperature, pressure, stirring speed, and heating time, are adjusted in real time through adaptive algorithms to compensate for the impact of raw material fluctuations;

[0062] Production process feedback and optimization module: Based on the product quality data (such as taste and color) collected in real time during the production process, combined with the adjustment of production process parameters, feedback is provided on the execution effect of the production process, and the production process parameters are adjusted and optimized based on the execution effect to ensure that the product quality of each production cycle remains consistent.

[0063] Raw material quality data includes milk fat concentration, sugar concentration, and flavor chemical component concentration. The sensor network specifically includes:

[0064] Using chemical sensors to detect fat concentration in milk;

[0065] Measure the sugar concentration in syrup using a refractive index sensor;

[0066] Use spectral sensors to measure the concentration of chemical components in spices, such as vanilla and coffee, and perform quantitative analysis based on absorbance or reflectance;

[0067] The data analysis algorithm automatically detects raw material quality fluctuation data through the sliding average method.

[0068] The moving average method is expressed as: Among them, SMA t is the sliding average at time t, O L is the raw material quality data value at the Lth moment, and n is the window size.

[0069] The adaptive algorithms in the production process adaptive adjustment module include multi-task learning (MTL) and reinforcement learning, among which;

[0070] The multi-task learning part learns multiple target tasks at the same time. The target tasks include adjusting temperature, pressure, stirring speed and heating time. Each target task learns parameters through the shared layer of the neural network.

[0071] Reinforcement learning maximizes rewards by interacting with the environment and learning strategies. In the production process of dairy beverages, reinforcement learning optimizes production process parameters based on the output of multi-task learning. Multi-task learning provides preliminary process parameters adjusted based on raw material quality fluctuations, while reinforcement learning optimizes and adjusts production process parameters by interacting with the production process and using real-time feedback.

[0072] In the multi-task learning stage, set the target task represents the adjustment of process parameters i, including temperature, pressure, stirring speed and heating time. The goal of each target task is to learn an adjustment model that can output the corresponding process parameters according to the raw material quality fluctuation;

[0073] Input data: The input vector is the raw material quality fluctuation data, expressed as:

[0074] X=[C milk-fat ,C syrup-sugar ,C flavor ], where C milk-fat is the milk fat concentration, C syrup-sugar is the sugar concentration of the syrup, C flavor It is the concentration of spice;

[0075] Target task output: Each process parameter adjustment task has an independent objective function. The objective function is the core of multi-task learning. The goal is to adjust the corresponding process parameters according to the input data, including temperature y temp , pressure pressure , stirring speed y speed And heating time time ;

[0076] Multi-task learning is based on multiple tasks sharing the underlying neural network. The shared underlying neural network uses a multi-layer feedforward neural network with a shared layer. The input vector is X and the output is different process parameters. Multi-task learning learns the relationship between input data and output targets through shared layers (shared intermediate layers), and at the same time assigns an independent output layer to each task to adjust different process parameters respectively.

[0077] The multi-layer feedforward neural network structure includes:

[0078] Input layer: input vector X;

[0079] Hidden layer: that is, the shared neural network layer (shared layer);

[0080] Output layer: The output layer corresponding to each process parameter target task outputs temperature, pressure, stirring speed, and heating time respectively;

[0081] The loss function of multi-task learning is the weighted loss sum of multiple tasks:

[0082] in, For the task The loss function is For network tasks The predicted value of the jth sample, y ij For the task The true label of the jth sample, λ i is the weight coefficient of the task, and N is the number of samples.

[0083] Reinforcement learning uses the output of multi-task learning as the initialization input. The production process parameters output by multi-task learning will be used as part of the environment state. Reinforcement learning will start from the initial state. Reinforcement learning is defined as follows:

[0084] State S: current environment description, including current raw material quality data and production process parameters adjusted by multi-task learning;

[0085] Action A: The operation selected according to state S. In the production of dairy beverages, the action is to adjust the specific value of the production process parameters;

[0086] Reward R: As a feedback signal, it is used to evaluate whether the current state and actions taken contribute to the achievement of the goal. The reward can be defined based on the final result of the production process, such as the stability and consistency of product quality, or the operating efficiency of the equipment.

[0087] Strategy π: The strategy is the rule for the agent to choose actions in a certain state. The goal is to maximize the cumulative reward by optimizing the strategy.

[0088] The specific learning of reinforcement learning includes:

[0089] Initialization: Set the initial state S0 to the production process parameters output by multi-task learning and the current raw material quality data;

[0090] Action selection: At each time step, the reinforcement learning agent selects the action based on the current state S at time t. t Select an action A t , that is, adjusting the production process parameters;

[0091] Environmental feedback: The production process produces a new state S according to the selected action t+1 , and generates a reward R based on this state t , that is, the actual results in the production process (such as product quality, efficiency, etc.);

[0092] Strategy update: Based on the rewards obtained, the reinforcement learning algorithm updates the strategy and optimizes future action selection;

[0093] Iteration: Repeat the action selection to strategy update, and finally obtain the optimal production process parameter adjustment plan through multiple interactive optimization strategies.

[0094] The update strategy updates the Q value based on deep Q learning:

[0095] The strategy is optimized by updating the Q value. The formula for updating the Q value is:

[0096] Among them, Q(S t ,A t ) indicates the current state S t Next, select Action A t Q value, R t+1 Indicates taking action A t After that, the reward of environmental feedback, γ is the discount factor, which controls the impact of future rewards, Indicates the next state S t+1 The maximum possible Q value represents the expected benefit of the agent choosing the best action in the future. α is the learning rate, which determines the speed of Q value update.

[0097] During training, the agent selects actions using an ε-greedy strategy: it randomly selects an action with probability ∈ (exploration) and selects the action with the highest current Q value with probability 1-∈ (exploitation).

[0098] The product quality data in the production process feedback and optimization module include taste and color, among which:

[0099] Taste data Q 口感 (t): reflects the taste quality of dairy beverages, obtained through sensory testing, instrumental measurement or survey feedback;

[0100] Color data Q 色泽 (t): reflects the appearance color of dairy beverages, measured using a spectral sensor;

[0101] Compare with the preset target mass value and calculate the mass error ΔQ 口感 and ΔQ 色泽 :

[0102]

[0103] Mass error ΔQ 口感 and ΔQ 色泽 As the execution effect, the production process parameters are adjusted and optimized based on the execution effect. The adjustment and optimization includes adjusting the reward function R t :

[0104] When the quality error is large: reduce the reward to encourage the agent to adjust the production process;

[0105] When the quality error is small: increase the reward to encourage the agent to maintain the current process parameters;

[0106] The reward function adjusted by product quality feedback can be expressed as:

[0107] R t =-(α 口感 ΔQ 口感 +α 色泽 ΔQ 色泽 ), where α 口感 and α 色泽 It is to adjust the weight to indicate the importance of taste and color errors in the reward function.

[0108] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0109] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. Intelligent monitoring and upgrading optimization system for dairy beverage production equipment, characterized by: include: Raw material fluctuation monitoring and identification module: real-time monitoring and identification of raw material quality fluctuations in the production process of dairy beverages, including milk, syrup, and flavoring ingredients. Real-time collection of raw material quality data is performed through a sensor network, and raw material quality fluctuation data is automatically detected based on data analysis algorithms. Production process adaptive adjustment module: Based on raw material quality fluctuation data, the production process parameters, including temperature, pressure, stirring speed, and heating time, are adjusted in real time through adaptive algorithms to compensate for the impact of raw material fluctuations; Production process feedback and optimization module: Based on the product quality data collected in real time during the production process and combined with the adjustment of production process parameters, the execution effect of the production process is fed back, and the production process parameters are adjusted and optimized based on the execution effect to ensure that the product quality of each production cycle remains consistent.

2. The intelligent monitoring and upgrading optimization system for dairy beverage production equipment according to claim 1 is characterized in that: The raw material quality data includes milk fat concentration, sugar concentration and flavor chemical component concentration, and the sensor network specifically includes: Using chemical sensors to detect fat concentration in milk; Measure the sugar concentration in syrup using a refractive index sensor; Using spectral sensors to measure the concentration of chemical components in spices; The data analysis algorithm automatically detects raw material quality fluctuation data through a sliding average method.

3. The intelligent monitoring and upgrading optimization system for dairy beverage production equipment according to claim 2 is characterized in that: The sliding average method is expressed as: Among them, SMA t is the sliding average at time t, O L is the raw material quality data value at the Lth moment, and n is the window size.

4. The intelligent monitoring and upgrading optimization system for dairy beverage production equipment according to claim 1 is characterized in that: The adaptive algorithm in the production process adaptive adjustment module includes multi-task learning and reinforcement learning, wherein; The multi-task learning part learns multiple target tasks at the same time. The target tasks include adjusting temperature, pressure, stirring speed and heating time. Each target task learns parameters through the shared layer of the neural network. Reinforcement learning maximizes rewards by interacting with the environment and learning strategies. Reinforcement learning optimizes production process parameters based on the output of multi-task learning. Multi-task learning provides preliminary process parameters based on raw material quality fluctuations, while reinforcement learning optimizes and adjusts production process parameters by interacting with the production process and using real-time feedback.

5. The intelligent monitoring and upgrading optimization system for dairy beverage production equipment according to claim 4 is characterized in that: In the multi-task learning stage, the target task is represents the adjustment of process parameters i, including temperature, pressure, stirring speed and heating time. The goal of each target task is to learn an adjustment model that can output the corresponding process parameters according to the raw material quality fluctuation; Input data: The input vector is the raw material quality fluctuation data, expressed as: X=[C milk-fat ,C syrup-sugar ,C flavor ], where C milk-fat is the milk fat concentration, C syrup-sugar is the sugar concentration of the syrup, C flavor It is the concentration of spice; Target task output: Each process parameter adjustment task has an independent objective function. The objective function is the core of multi-task learning. The goal is to adjust the corresponding process parameters according to the input data, including temperature y temp , pressure pressure , stirring speed y speed And heating time time ; The multi-task learning is based on a shared underlying neural network for multiple tasks. The shared underlying neural network adopts a multi-layer feedforward neural network with a shared layer. The input vector is X and the output is different process parameters. The multi-task learning learns the relationship between the input data and the output target through the shared layer, and at the same time allocates an independent output layer to each task to adjust different process parameters respectively.

6. The intelligent monitoring and upgrading optimization system for dairy beverage production equipment according to claim 5 is characterized in that: The multi-layer feedforward neural network structure comprises: Input layer: input vector X; Hidden layer: the shared neural network layer; Output layer: The output layer corresponding to each process parameter target task outputs temperature, pressure, stirring speed, and heating time respectively; The loss function of multi-task learning is the weighted loss sum of multiple tasks: in, For the task The loss function is For network tasks The predicted value of the jth sample, y ij For the task The true label of the jth sample, λ i is the weight coefficient of the task, and N is the number of samples.

7. The intelligent monitoring and upgrading optimization system for dairy beverage production equipment according to claim 6 is characterized in that: The reinforcement learning uses the output of multi-task learning as the initialization input. The production process parameters output by the multi-task learning will be used as part of the environment state. The reinforcement learning will start from the initial state. The reinforcement learning is defined as follows: State S: current environment description, including current raw material quality data and production process parameters adjusted by multi-task learning; Action A: The operation selected according to state S. In the production of dairy beverages, the action is to adjust the specific value of the production process parameters; Reward R: As a feedback signal, it is used to evaluate whether the current state and actions taken contribute to the achievement of the goal; Strategy π: The strategy is the rule for the agent to choose actions in a certain state. The goal is to maximize the cumulative reward by optimizing the strategy.

8. The intelligent monitoring and upgrading optimization system for dairy beverage production equipment according to claim 7 is characterized in that: The specific learning of the reinforcement learning includes: Initialization: Set the initial state S0 to the production process parameters output by multi-task learning and the current raw material quality data; Action selection: At each time step, the reinforcement learning agent selects the action based on the current state S at time t. t Select an action A t , that is, adjusting the production process parameters; Environmental feedback: The production process produces a new state S according to the selected action t+1 , and generates a reward R based on this state t , that is, the actual results of the production process; Strategy update: Based on the rewards obtained, the reinforcement learning algorithm updates the strategy and optimizes future action selection; Iteration: Repeat the action selection to strategy update, and finally obtain the optimal production process parameter adjustment plan through multiple interactive optimization strategies.

9. The intelligent monitoring and upgrading optimization system for dairy beverage production equipment according to claim 8, characterized in that: The product quality data in the production process feedback and optimization module include taste and color, among which: Taste data Q 口感 (t): reflects the taste quality of dairy beverages; Color data Q 色泽 (t): reflects the appearance color of dairy beverages; Compare with the preset target mass value and calculate the mass error ΔQ 口感 and ΔQ 色泽 : Mass error ΔQ 口感 and ΔQ 色泽 As the execution effect, the production process parameters are adjusted and optimized based on the execution effect. The adjustment and optimization includes adjusting the reward function R t : When the quality error is large: reduce the reward to encourage the agent to adjust the production process; When the quality error is small: increase the reward to encourage the agent to maintain the current process parameters.

Images