Dyeing process PH value self-optimization system based on multi-agent intelligent control

Abstract

The invention relates to a dyeing process PH value self-optimization system based on multi-agent intelligent control. The dyeing process PH value self-optimization system comprises a sensing unit, a control decision unit and an execution unit. The sensing unit collects key data in the cloth dyeing process, and the key data comprise the pH value, temperature, dye concentration and flow velocity of dye liquor. The control decision-making unit adopts a multi-agent architecture and comprises a data fusion agent, a modeling agent, an optimization decision-making agent and a cooperative control agent; and the execution unit is used for regulating and controlling the pH value in the dye liquor in real time according to the final decision scheme of the control decision unit and based on PID (Proportion Integration Differentiation) control. The intelligent level of dyeing process production is enhanced while the control precision of the pH value of the dye liquor is ensured.

Description

Technical Field

[0001] The invention relates to the field of intelligent control, and in particular to a dyeing process pH value self-optimization system based on multi-agent intelligent control. Background Art

[0002] With the rapid development of the textile industry, higher and higher requirements are placed on the precise control of the dyeing process. In the dyeing process, pH value is one of the core parameters that affect the dyeing quality. Its fluctuation will not only lead to uneven dyeing and unstable color, but may also cause problems such as reprocessing or dye waste, directly affecting production efficiency and economic benefits. In traditional dyeing processes, the adjustment of pH value mainly depends on manual intervention, and the dosage of alkaline or acidic agents is adjusted based on experience. However, this method not only has a slow response speed, but also has difficulty adapting to the complex and dynamic dye liquor reaction process, resulting in dyeing accuracy and consistency. It is difficult to ensure. In response to the above problems, the intelligent control system of the dyeing process based on industrial automation and artificial intelligence technology has gradually become a hot spot for research and application. Summary of the invention

[0003] In order to solve the above problems, the purpose of the present invention is to provide a dyeing process pH self-optimization system based on multi-agent intelligent control, which not only ensures the control accuracy of the pH value of the dye solution, but also enhances the intelligence level of the dyeing process production.

[0004] To achieve the above object, the present invention adopts the following technical solutions: A dyeing process pH value self-optimization system based on multi-agent intelligent control, comprising a sensing unit, a control decision unit and an execution unit; the sensing unit collects key data in the cloth dyeing process, including pH value, temperature, dye concentration and flow rate of the dye solution; the control decision unit adopts a multi-agent architecture, including a data fusion agent, a modeling agent, an optimization decision agent and a collaborative control agent; The data fusion agent fuses the multi-dimensional data collected by the sensing unit into the real state of the current dye solution; Modeling agents, based on machine learning algorithms and physical models, build dynamic prediction models to characterize the relationship between pH value and process parameters and make predictions based on the actual state of the current dye solution; Optimize the decision-making agent, combined with the reinforcement learning algorithm, and dynamically adjust the industrial operation parameters based on the prediction results and the deviation of the pH value during the dyeing process as the objective function; Collaborative control agent: coordinate the data and control information between various agents to form a final decision-making plan; The execution unit adjusts the pH value of the dye solution in real time based on PID control according to the final decision plan of the control decision unit.

[0005] Furthermore, the perception unit includes a sensor module, an edge computing unit and a data transmission module; the sensor module includes a pH sensor, a temperature sensor, an optical concentration sensor and an ultrasonic flow meter; after the sensor data is collected, it will be transmitted to the edge computing unit for preprocessing, and after data cleaning and abnormal point elimination, it will be transmitted to the control decision unit through the data transmission module.

[0006] Furthermore, the data fusion agent fuses the multi-dimensional data collected by the perception unit into the real state of the current dye solution, as follows: Time-align the collected data of each sensor to ensure the timing consistency of the data; The sensor observation data set is D={D1,D2,D3,D4} Among them, D1 comes from the pH value of the pH sensor; D2 comes from the temperature data of the temperature sensor; D3 comes from the dye concentration of the optical concentration sensor; D4 comes from the flow rate data of the ultrasonic flow meter; Each sensor measurement contains noise, which is distributed as ; ; in, is the noise variance of the ith sensor; θ represents the true value to be estimated; D is the sensor observation data set; D i For the i The observation values ​​of the sensors; is the likelihood function, which means that the sensor i obtains the measurement value D when the true value is assumed to be θ i probability; According to the Bayesian formula, multiple sensors are integrated: ; in, is the posterior probability, is the prior probability, n is the number of sensors; Assuming that the prior distribution and the observation noise of each sensor obey Gaussian distribution, the result of the posterior estimation is: ; in, u 0 and are the mean and variance of the prior distribution respectively; Output fusion result , the variance of the posterior distribution Indicates the credibility of the fusion result, specifically: ; The fused data represents the real state of the current dye solution.

[0007] Furthermore, a dynamic prediction model is constructed based on machine learning algorithms and physical models, as follows: Obtained through the physical model, the current hydrogen ion concentration change Δ[H+], the dynamic reaction rate constant k; the physically predicted pH value pH physical (t+1), the physical model is as follows: Neutralization reaction kinetic equation: ; Where r is the reaction rate; is the hydrogen ion concentration; is the hydroxide ion concentration; pH value conversion formula: ; Among them, K w is the dissociation equilibrium constant of water; PH is the pH value; The equation for the change of hydrogen ion concentration is:

[0008] in, is the Arrhenius formula; is the dosing rate of the alkaline or acidic agent; V is the total volume of the dye solution; Physical prediction pH value prediction formula: ; The output of the physical model is used as input features, combined with the current perception data, and input into the machine learning model to predict a more accurate pH value at the next moment.

[0009] Furthermore, the machine learning model is built based on the LSTM model, as follows: Input: time series features X=[x1,x2,…,xt], where each x t Include ;Wherein, T is temperature; Output: Predicted pH value at the next moment ; The LSTM model is as follows: Forget Gate: ; Input Gate: ; Candidate values: ; Cell status update: ; Output Gate: ; Hide status updates: ; Output layer: ; in, is the hidden state at time step t; For the Gate of Oblivion; is the input gate; is the candidate value; that is, the candidate memory cell at the current moment; is the cell state; it indicates the memory cell state at the current moment; is the output gate; is the hidden state at the current moment; , , , They are the weight matrices of the forget gate, input gate, candidate value, and output gate respectively; , , , are the bias vectors of the forget gate, input gate, candidate value, and output gate, respectively. is the activation function.

[0010] Further, the decision-making agent is optimized as follows: The environmental state St contains all process state parameters of the dyeing process at the current moment, including the physical collection data of the perception unit and the output of the prediction model; ; Among them, pH t is the current pH value; T t , C t , R add,t They are the current temperature, dye concentration, and alkali solution addition rate; ΔpH t and are the pH value change rate and hydrogen ion concentration change rate output by the prediction model respectively; The agent is based on the state S t Select Action A t , used to control the process parameters of dyeing process: A t =[ΔR add ,ΔT,ΔC]; Among them, ΔR add To adjust the rate of adding alkali solution; ΔT is to adjust the temperature; ΔC is to adjust the initial concentration of the dye; Reward function R t It is the core of the optimized reinforcement learning model. The design goal is to guide the agent to minimize the pH deviation in the dyeing process: ; Among them, pH targetis the target pH value; λ1 and λ2 are weight coefficients; strategy π is the strategy of the optimization decision agent, which is used to select the best action; The DDPG algorithm is used in the continuous action space.

[0011] Furthermore, the DDPG algorithm includes a policy network and a value network, as follows: The policy network is responsible for selecting actions: ;in are the strategy parameters; The value network evaluates the value of a state-action pair ;in is the value parameter; Update the target Q value using the Bellman equation: ; Among them, γ is the discount factor; The policy network is updated by maximizing the expected reward of the action: ; in, Represents the objective function of the policy parameters The gradient of Respectively represent The state and action of samples; N is the number of samples; Indicates action The gradient of Represents the value parameter The gradient of Introducing the target network to alleviate the low stability of policy updates: Use an experience pool D during training c Cache history interaction (S t ,A t ,R t ,S t+1 ), random sampling training to avoid overfitting caused by time correlation; Adam optimizer is used to optimize the policy network and value network separately.

[0012] Furthermore, the collaborative control agents coordinate the data and control information between the agents and finally make a decision, as follows: Obtain the actual status and predicted results of the current dye solution from the data fusion agent and the modeling agent; According to the dyeing process goal, decompose the global control goal into sub-goals of each sub-agent; Coordinate the flow of information between data fusion, modeling and optimization decision agents to ensure information sharing and decision consistency; Integrate and optimize the control strategies generated by the decision-making agent to form the final global control action; Adjust the weights and strategies of each sub-agent based on real-time feedback data.

[0013] Furthermore, the execution unit, according to the final decision plan of the control decision unit, controls the pH value in the dye solution in real time based on PID control, as follows: Real-time pH control closed loop structure: Input: Target pH target The final decision from the collaborative control agent; Feedback: pH Actual pH actual The pH sensor in the dye solution measures the current pH value of the dye solution in real time; Real-time calculation error: ; Perform proportional, integral and differential calculations on the error e(t) to obtain the control signal u(t); Adjust the dosing flow rate R of the alkaline or acidic agent according to u(t) add , control the pH dynamics of the dye solution: For pH adjustment of dye liquor, the control operation is to adjust the flow rate R of the dosing device. add (t) is realized by converting the PID output signal u(t) into the flow regulation formula: R add (t) = R add,0 +u(t); Among them, R add,0 is the initial injection acceleration rate; u(t) is the control signal generated by the PID controller, which is used to adjust the injection flow rate in real time; Dynamic model of the system with injection flow rate as input: ΔpH t =f(R add ); Discrete PID controller, the output signal u(t) is: ; Among them, K p is the proportional gain; is the integral gain; K d is the differential gain; is the error at the tth moment.

[0014] The present invention has the following beneficial effects: 1. The present invention can accurately characterize the complex nonlinear relationship between pH value and process parameters through filtering and simplifying multidimensional data by the data fusion agent and building a dynamic prediction model by the modeling agent using machine learning algorithms and physical models, thereby significantly improving the accuracy of pH value control; 2. The present invention can adjust the control parameters in the dyeing process in real time to minimize the pH value deviation, significantly improve the process automation level and efficiency, and through the collaborative control agent, can efficiently coordinate data fusion, modeling and optimization decision-making agents to form an accurate dyeing process control strategy, and realize the intelligent and automated optimization of the industrial process; 3. The present invention combines PID control to execute the final decision plan of the control decision unit in real time, accurately regulates the pH value in the dye solution, maintains it within the target range, and responds to dynamic changes in process parameters. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a system architecture diagram of the present invention. DETAILED DESCRIPTION

[0016] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments: refer to Figure 1 In this embodiment, a dyeing process pH value self-optimization system based on multi-agent intelligent control is provided, including a sensing unit, a control decision unit and an execution unit; the sensing unit collects key data in the cloth dyeing process, including the pH value, temperature, dye concentration and flow rate of the dye solution; the control decision unit adopts a multi-agent architecture, including a data fusion agent, a modeling agent, an optimization decision agent and a collaborative control agent; The data fusion agent fuses the multi-dimensional data collected by the sensing unit into the real state of the current dye solution; Modeling agents build dynamic prediction models based on machine learning algorithms and physical models to characterize the relationship between pH value and process parameters (temperature, dye concentration, reaction time, etc.) and make predictions based on the actual state of the current dye solution; Optimize the decision-making agent, combined with the reinforcement learning algorithm, based on the prediction results, take the deviation of pH value in the dyeing process as the objective function, and dynamically adjust the industrial operation parameters (such as the dosage of alkaline or acidic agents); Collaborative control agent: coordinate the data and control information between various agents to form a final decision-making plan; The execution unit adjusts the pH value of the dye solution in real time based on PID control according to the final decision plan of the control decision unit.

[0017] In this embodiment, the perception unit includes a sensor module, an edge computing unit and a data transmission module; the sensor module includes a pH sensor, a temperature sensor, an optical concentration sensor and an ultrasonic flow meter; after the sensor data is collected, it will be transmitted to the edge computing unit for preprocessing, and after data cleaning and abnormal point elimination, it will be transmitted to the control decision unit through the data transmission module.

[0018] In this embodiment, the data fusion agent fuses the multi-dimensional data collected by the perception unit into the real state of the current dye solution, as follows: Time-align the collected data of each sensor to ensure the timing consistency of the data; The sensor observation data set is D={D1,D2,D3,D4} Among them, D1 comes from the pH value of the pH sensor; D2 comes from the temperature data of the temperature sensor; D3 comes from the dye concentration of the optical concentration sensor; D4 comes from the flow rate data of the ultrasonic flow meter; Each sensor measurement contains noise, which is distributed as ; ; in, is the noise variance of the ith sensor; θ represents the true value to be estimated; D is the sensor observation data set; D i For the i The observed value of the sensor; is the likelihood function, which means that the sensor i obtains the measured value D under the assumption that the true value is θ. i The probability of According to the Bayesian formula, multiple sensors are integrated: ; in, is the posterior probability, is the prior probability, n is the number of sensors; Assuming that the prior distribution and the observation noise of each sensor obey Gaussian distribution (normal distribution), the result of the posterior estimation is: ; in, u 0 and are the mean and variance of the prior distribution respectively; Output fusion result , the variance of the posterior distribution Indicates the credibility of the fusion result, specifically: ; The fused data represents the real state of the current dye solution.

[0019] In this embodiment, a dynamic prediction model is constructed based on a machine learning algorithm and a physical model, as follows: Obtained through the physical model, the current hydrogen ion concentration change Δ[H+], the dynamic reaction rate constant k; the physically predicted pH value pH physical (t+1), the physical model is as follows: Neutralization reaction kinetic equation: ; Where r is the reaction rate; is the hydrogen ion concentration; is the hydroxide ion concentration; pH value conversion formula: ; Among them, K w is the dissociation equilibrium constant of water; PH is the pH value; The equation for the change of hydrogen ion concentration is:

[0020] in, is the Arrhenius formula; is the dosing rate of the alkaline or acidic agent; V is the total volume of the dye solution; Physical prediction pH value prediction formula: ; The output of the physical model is used as input features, combined with the current perception data, and input into the machine learning model to predict a more accurate pH value at the next moment.

[0021] In this embodiment, the machine learning model is constructed based on the LSTM model, as follows: Input: time series features X=[x1,x2,…,xt], where each x t Include ;Wherein, T is temperature; Output: Predicted pH value at the next moment ; The LSTM model is as follows: Forget Gate: ; Input Gate: ; Candidate values: ; Cell status update: ; Output Gate: ; Hide status updates: ; Output layer: ; in, is the hidden state at time step t; For the Gate of Oblivion; is the input gate; is the candidate value; that is, the candidate memory cell at the current moment; is the cell state; it indicates the memory cell state at the current moment; is the output gate; is the hidden state at the current moment; , , , They are the weight matrices of the forget gate, input gate, candidate value, and output gate respectively; , , , are the bias vectors of the forget gate, input gate, candidate value, and output gate, respectively. is the activation function.

[0022] In this embodiment, the optimization decision agent is as follows: The environmental state St contains all process state parameters of the dyeing process at the current moment, including the physical collection data of the perception unit and the output of the prediction model; ; Among them, pH t is the current pH value; T t , C t , R add,t They are the current temperature, dye concentration, and alkali solution addition rate; ΔpH t and are the pH value change rate and hydrogen ion concentration change rate output by the prediction model respectively; The agent is based on the state S t Select Action A t , used to control the process parameters of dyeing process: A t =[ΔR add ,ΔT,ΔC]; Among them, ΔR add To adjust the rate of alkali solution addition (such as increasing / decreasing a specific rate); ΔT is to adjust the temperature; ΔC is to adjust the initial concentration of the dye; Reward function R t It is the core of the optimized reinforcement learning model. The design goal is to guide the agent to minimize the pH deviation in the dyeing process: ; Among them, pH targetis the target pH value; λ1 and λ2 are weight coefficients; strategy π is the strategy of the optimization decision agent, which is used to select the best action; The DDPG algorithm is used in the continuous action space.

[0023] In this embodiment, the DDPG algorithm includes a policy network and a value network, as follows: The policy network is responsible for selecting actions: ,in is the strategy parameter; The value network evaluates the value of a state-action pair ,in is the value parameter; Update the target Q value using the Bellman equation: ; Among them, γ is the discount factor; The policy network is updated by maximizing the expected reward of the action: ; in, Represents the objective function of the policy parameters The gradient of Respectively represent The state and action of samples; N is the number of samples; Indicates action The gradient of Represents the value parameter The gradient of Introducing the target network to alleviate the low stability of policy updates: Use an experience pool D during training c Cache history interaction (S t ,A t ,R t ,S t+1 ), random sampling training to avoid overfitting caused by time correlation; Adam optimizer is used to optimize the policy network and value network separately.

[0024] In this embodiment, the collaborative control agent coordinates the data and control information between the agents and finally forms a decision, as follows: Obtain the actual status and predicted results of the current dye solution from the data fusion agent and the modeling agent; According to the dyeing process goal, decompose the global control goal into sub-goals of each sub-agent; Coordinate the flow of information between data fusion, modeling and optimization decision agents to ensure information sharing and decision consistency; Integrate and optimize the control strategies generated by the decision-making agent to form the final global control action; Adjust the weights and strategies of each sub-agent based on real-time feedback data.

[0025] In this embodiment, the execution unit, according to the final decision plan of the control decision unit, controls the pH value in the dye solution in real time based on PID control, as follows: Real-time pH control closed loop structure: Input: Target pH target The final decision from the collaborative control agent; Feedback: pH Actual pH actual The pH sensor in the dye solution measures the current pH value of the dye solution in real time; Real-time calculation error: ; Perform proportional, integral and differential calculations on the error e(t) to obtain the control signal u(t); Adjust the dosing flow rate R of the alkaline or acidic agent according to u(t) add , control the pH dynamics of the dye solution: For pH adjustment of dye liquor, the control operation is to adjust the flow rate R of the dosing device. add (t) is realized by converting the PID output signal u(t) into the flow regulation formula: R add (t) = R add,0 +u(t); Among them, R add,0 is the initial injection acceleration rate; u(t) is the control signal generated by the PID controller, which is used to adjust the injection flow rate in real time; Dynamic model of the system with injection flow rate as input: ΔpH t =f(R add ); Discrete PID controller, the output signal u(t) is: ; Among them, K p is the proportional gain; is the integral gain; K d is the differential gain; is the error at the tth moment.

[0026] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0027] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0028] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0029] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0030] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A dyeing process pH self-optimization system based on multi-agent intelligent control, characterized in that: It includes a perception unit, a control decision unit and an execution unit; the perception unit collects key data in the cloth dyeing process, including pH value, temperature, dye concentration and flow rate of the dye solution; the control decision unit adopts a multi-agent architecture, including a data fusion agent, a modeling agent, an optimization decision agent and a collaborative control agent; The data fusion agent fuses the multi-dimensional data collected by the sensing unit into the real state of the current dye solution; Modeling agents, based on machine learning algorithms and physical models, build dynamic prediction models to characterize the relationship between pH value and process parameters and make predictions based on the actual state of the current dye solution; Optimize the decision-making agent, combined with the reinforcement learning algorithm, and dynamically adjust the industrial operation parameters based on the prediction results and the deviation of the pH value during the dyeing process as the objective function; Collaborative control of intelligent agents, coordinating data and control information between various intelligent agents to form a final decision-making plan; The execution unit adjusts the pH value of the dye solution in real time based on PID control according to the final decision plan of the control decision unit.

2. The dyeing process pH self-optimization system based on multi-agent intelligent control according to claim 1 is characterized in that: The perception unit includes a sensor module, an edge computing unit and a data transmission module; the sensor module includes a pH sensor, a temperature sensor, an optical concentration sensor and an ultrasonic flow meter; after the sensor data is collected, it will be transmitted to the edge computing unit for preprocessing, and after data cleaning and abnormal point elimination, it will be transmitted to the control decision unit through the data transmission module.

3. The dyeing process pH self-optimization system based on multi-agent intelligent control according to claim 1 is characterized in that: The data fusion agent fuses the multi-dimensional data collected by the perception unit into the real state of the current dye solution, as follows: Time-align the collected data of each sensor to ensure the timing consistency of the data; The sensor observation data set is D={D1,D2,D3,D4} Among them, D1 comes from the pH value of the pH sensor; D2 comes from the temperature data of the temperature sensor; D3 comes from the dye concentration of the optical concentration sensor; D4 comes from the flow rate data of the ultrasonic flow meter; Each sensor measurement contains noise, which is distributed as ; ； in, is the noise variance of the ith sensor; θ represents the true value to be estimated; D is the sensor observation data set; D i For the i The observation values ​​of the sensors; is the likelihood function, which means that the sensor i obtains the measurement value D when the true value is assumed to be θ i probability; According to the Bayesian formula, multiple sensors are integrated: ； in, is the posterior probability, is the prior probability, n is the number of sensors; Assuming that the prior distribution and the observation noise of each sensor obey Gaussian distribution, the result of the posterior estimation is: ； in, u 0 and are the mean and variance of the prior distribution respectively; Output fusion result , the variance of the posterior distribution Indicates the credibility of the fusion result, specifically: ； The fused data represents the real state of the current dye solution.

4. The dyeing process pH self-optimization system based on multi-agent intelligent control according to claim 3 is characterized in that: The dynamic prediction model is constructed based on the machine learning algorithm and the physical model, as follows: Obtained through the physical model, the current hydrogen ion concentration change Δ[H+], the dynamic reaction rate constant k; the physically predicted pH value pH physical (t+1), the physical model is as follows: Neutralization reaction kinetic equation: ; Where r is the reaction rate; is the hydrogen ion concentration; is the hydroxide ion concentration; pH value conversion formula: ； Among them, K w is the dissociation equilibrium constant of water; PH is the pH value; The equation for the change of hydrogen ion concentration is: in, is the Arrhenius formula; is the addition rate of alkaline or acidic agent; V is the total volume of dye solution; Physical prediction pH value prediction formula: ； The output of the physical model is used as input features, combined with the current perception data, and input into the machine learning model to predict a more accurate pH value at the next moment.

5. The dyeing process pH self-optimization system based on multi-agent intelligent control according to claim 4 is characterized in that: The machine learning model is built based on the LSTM model, as follows: Input: time series features X=[x1,x2,…,xt], where each x t Include ;in, T is temperature; Output: Predicted pH value at the next moment ; The LSTM model is as follows: Forget Gate: ; Input Gate: ; Candidate values: ; Cell status update: ; Output Gate: ; Hide status updates: ; Output layer: ; in, is the hidden state at time step t; For the Gate of Oblivion; is the input gate; is the candidate value; that is, the candidate memory cell at the current moment; is the cell state; it indicates the memory cell state at the current moment; is the output gate; is the hidden state at the current moment; , , , They are the weight matrices of the forget gate, input gate, candidate value, and output gate respectively; , , , are the bias vectors of the forget gate, input gate, candidate value, and output gate, respectively. is the activation function.

6. The dyeing process pH self-optimization system based on multi-agent intelligent control according to claim 1 is characterized in that: The optimization decision agent is as follows: The environmental state St contains all process state parameters of the dyeing process at the current moment, including the physical collection data of the perception unit and the output of the prediction model; ； Among them, pH t is the current pH value; T t , C t , R add,t They are the current temperature, dye concentration, and alkali solution addition rate; ΔpH t and are the pH value change rate and hydrogen ion concentration change rate output by the prediction model respectively; The agent is based on the state S t Select Action A t , used to control the process parameters of dyeing process: A t =[ΔR add ,ΔT,ΔC]; Among them, ΔR add To adjust the rate of adding alkali solution; ΔT to adjust the temperature; ΔC to adjust the initial concentration of the dye; Reward function R t It is the core of the optimized reinforcement learning model. The design goal is to guide the agent to minimize the pH deviation in the dyeing process: ； Among them, pH target is the target pH value; λ1 and λ2 are weight coefficients; strategy π is the strategy of the optimization decision agent, which is used to select the best action; The DDPG algorithm is used in the continuous action space.

7. The dyeing process pH self-optimization system based on multi-agent intelligent control according to claim 6 is characterized in that: The DDPG algorithm includes a policy network and a value network, as follows: The policy network is responsible for selecting actions: ,in is the strategy parameter; The value network evaluates the value of a state-action pair ,in is the value parameter; Update the target Q value using the Bellman equation: ； Among them, γ is the discount factor; The policy network is updated by maximizing the expected reward of the action: ； in, Represents the objective function of the policy parameters The gradient of Respectively represent The state and action of samples; N is the number of samples; Indicates action The gradient of Represents the value parameter The gradient of Introducing the target network to alleviate the low stability of policy updates: Use an experience pool D during training c Cache history interaction (S t ,A t ,R t ,S t+1 ), random sampling training to avoid overfitting caused by time correlation; Adam optimizer is used to optimize the policy network and value network separately.

8. The dyeing process pH self-optimization system based on multi-agent intelligent control according to claim 1 is characterized in that: The collaborative control agent coordinates the data and control information between the agents to form a final decision plan, as follows: Obtain the actual status and predicted results of the current dye solution from the data fusion agent and the modeling agent; According to the dyeing process goal, decompose the global control goal into sub-goals of each sub-agent; Coordinate the flow of information between data fusion, modeling and optimization decision agents to ensure information sharing and decision consistency; Integrate and optimize the control strategies generated by the decision-making agent to form the final global control action; Adjust the weights and strategies of each sub-agent based on real-time feedback data.

9. The dyeing process pH self-optimization system based on multi-agent intelligent control according to claim 1 is characterized in that: The execution unit, according to the final decision plan of the control decision unit, based on PID control, controls the pH value in the dye solution in real time, as follows: Real-time pH control closed loop structure: Input: Target pH target The final decision from the collaborative control agent; Feedback: pH Actual pH actual The pH sensor in the dye solution measures the current pH value of the dye solution in real time; Real-time calculation error: ； Perform proportional, integral and differential calculations on the error e(t) to obtain the control signal u(t); Adjust the dosing flow rate R of the alkaline or acidic agent according to u(t) add , control the dynamic changes of pH of dye solution: For pH adjustment of dye liquor, the control operation is to adjust the flow rate R of the dosing device. add (t) is realized by converting the PID output signal u(t) into the flow regulation formula: R add (t)=R add,0 +u(t); Among them, R add,0 is the initial injection acceleration rate; u(t) is the control signal generated by the PID controller, which is used to adjust the injection flow rate in real time; Dynamic model of the system with injection flow rate as input: ΔpH t =f(R add ); Discrete PID controller, the output signal u(t) is: ； Among them, K p is the proportional gain; is the integral gain; K d is the differential gain; is the error at time t.

Images