A paper pulp mixing control system and method based on multi-sensor

Abstract

The present invention relates to a multi-sensor based paper pulp mixing control system and method, comprising a multi-sensor module that continuously monitors key parameters in the pulp mixing process, and transmits the data to an intelligent analysis module after preprocessing by a data acquisition module; the intelligent analysis module is provided with a pulp quality prediction model, and predicts the pulp quality based on the preprocessed data; the control module performs real-time parameter adjustment according to the analysis result of the intelligent analysis module, and uses a multi-objective optimization technology to balance between improving pulp quality and reducing energy consumption, and obtains an initial parameter optimization scheme, and based on the initial parameter optimization scheme, an MPC-based optimization engine is used to optimize the mixing parameters in real time under process constraints, and perform fine-grained control optimization; the data management and visualization module obtains the data of each module and performs visualization. The present invention effectively improves the overall efficiency and product quality of the paper pulp mixing process, and realizes effective control of energy consumption and cost.

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing, and in particular to a multi-sensor based paper pulp mixing control system and method. Background Art

[0002] With the continuous development of industrial automation and intelligent manufacturing technology, the manufacturing industry has higher and higher requirements for fine control and quality of the production process. In the roll paper industry, precise control of the slurry mixing process is a key link to ensure product quality and production efficiency. Roll paper products need to have specific softness, strength and water absorption, which puts strict requirements on the uniformity and composition of the slurry. Traditional slurry mixing control usually relies on manual operation and simple feedback mechanism, which is difficult to cope with complex and changing process conditions and increasingly stringent quality standards. Summary of the invention

[0003] In order to solve the above problems, the purpose of the present invention is to provide a multi-sensor based paper pulp mixing control system and method, which effectively improves the overall efficiency and product quality of the paper pulp mixing process and realizes effective control of energy consumption and cost.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] A paper pulp mixing control system based on multiple sensors, comprising a multiple sensor module, a data acquisition module, an intelligent analysis module, a control module and a data management and visualization module;

[0006] The multi-sensor module continuously monitors the key parameters in the slurry mixing process and transmits them to the intelligent analysis module after pre-processing by the data acquisition module;

[0007] The intelligent analysis module is provided with a slurry quality prediction model, which predicts the slurry quality based on the preprocessed data;

[0008] The control module performs real-time parameter adjustment according to the analysis results of the intelligent analysis module, uses multi-objective optimization technology to balance between improving slurry quality and reducing energy consumption, obtains an initial parameter optimization scheme, and based on the initial parameter optimization scheme, uses an MPC-based optimization engine to optimize the mixing parameters in real time under process constraints and perform fine-grained control optimization;

[0009] The data management and visualization module is connected to the data acquisition module, the intelligent analysis module, and the control module respectively to obtain the data of each module and perform visualization.

[0010] Furthermore, the multi-sensor module includes a flow sensor, a concentration sensor, a temperature sensor, a pH sensor, a pressure sensor and a spectral sensor; the flow sensor is used to monitor the flow of the slurry to ensure that the slurry is mixed at a set ratio; the concentration sensor is used to detect the fiber concentration in the slurry in real time; the temperature sensor is used to monitor the temperature change of the slurry; the pH sensor is used to monitor the acidity and alkalinity of the slurry to control the chemical properties; the pressure sensor is used to detect pressure fluctuations in the system to ensure safe operation; the spectral sensor is used to evaluate the composition and uniformity of the slurry.

[0011] Further, preprocessing includes using moving average filtering to remove sensor noise, using interpolation to fill in missing data, and performing data standardization on the sensor data to scale the data to a standard normal distribution.

[0012] Furthermore, the slurry quality prediction model is constructed based on the XGBOOST model, as follows:

[0013] Obtain relevant historical data in the process of mixing slurry, including feature data and target variables, and perform feature importance evaluation, screen out features with importance greater than a threshold, and construct a training data set; the feature data includes flow, concentration, temperature and pH value; the target variables include viscosity, uniformity and strength;

[0014] Define XGBoost parameters, including learning rate, maximum depth, subsampling rate, and regularization parameters;

[0015] The XGBoost model is trained based on the training data set. The XGBoost model predicts the target value through G trees:

[0016] ;

[0017] in For the feature data, For the The actual target variable, For the predict target variables; f g is the g-th tree;

[0018] The objective function of XGBoost is:

[0019] ;

[0020] in, is the loss function; is the regularization term of the g-th tree;

[0021] ;

[0022] in, is the weight of the number of leaf nodes; is the L2 regularization coefficient of the leaf weight; is the prediction score of the jth leaf node; T is the number of leaf nodes in the tree;

[0023] The model is constructed incrementally, that is, one tree is added each time to reduce the error of the previous prediction. For the tth iteration, the model is updated as follows:

[0024] ;

[0025] in For the added trees; For the The predicted value of the sample at the tth iteration;

[0026] For each node split, a greedy method is used to find the best split point.

[0027] Furthermore, for each node split, a greedy method is used to find the best split point, as follows:

[0028] For each feature, XGBoost performs the following steps to find the best split point:

[0029] Sort all samples of feature q;

[0030] Enumerate all possible splitting points of feature values ​​one by one. For continuous features, select the middle value as the potential splitting point;

[0031] For each split point, calculate the gain:

[0032]

[0033] in, , Respectively represent the gradient sum of the left child node and the right child node after splitting, and Respectively represent the second-order gradient sum of the left child node and the right child node after splitting; is the regularization parameter; is the split penalty parameter of the leaf node;

[0034] The splitting point that maximizes the gain is selected as the optimal splitting point for the feature.

[0035] Furthermore, a multi-objective optimization technique is used to strike a balance between improving slurry quality and reducing energy consumption, as follows:

[0036] Optimize the target definition, improve the slurry quality, and reduce energy consumption at the same time, then the multi-objective function is as follows:

[0037] min(f1(u),f2(u));

[0038] ;

[0039] ;

[0040] in, f 1 (u) is the slurry quality error: is the energy consumption estimation; the goal is to find the set of parameters u; To predict the slurry quality, TU is the target slurry quality; v is the flow rate, is the temperature, and is the coefficient related to energy consumption;

[0041] Based on genetic algorithm, process parameters are dynamically adjusted to improve slurry uniformity and reduce energy consumption to solve the above multi-objective function;

[0042] Randomly generate a population consisting of process parameters;

[0043] Combine the output of the prediction model and the definition of the multi-objective function to evaluate the fitness of each individual;

[0044] Using the non-dominated sorting method of NSGA-II, individuals are classified into different dominant layers. The goal is to find the Pareto frontier between reducing quality deviation and energy consumption, and calculate the crowding distance of individuals in each non-dominated layer; based on non-dominated sorting and crowding distance, tournament selection is performed to select better individuals to enter the next generation;

[0045] Generate new individuals using simulated binary crossover operation :

[0046] ;

[0047] in, is the control factor, , There are two parent individuals;

[0048] Use polynomial mutation to make small adjustments to individual genes:

[0049] ;

[0050] in, is a gene value of an individual; x max and x minare the upper and lower limits of the value respectively; δ is the random variation;

[0051] The population is updated during the iteration process, and the above steps are repeated until convergence or the maximum number of iterations is reached.

[0052] Furthermore, the non-dominated sorting method of NSGA-II is used to classify individuals into different dominant layers. The goal is to find the Pareto frontier between reducing quality deviation and energy consumption, and calculate the crowding distance of individuals in each non-dominated layer; based on non-dominated sorting and crowding distance, the tournament selection is carried out to select better individuals to enter the next generation, as follows:

[0053] (1) For the objective function f 1 (u),f 2 (u) , sorted according to non-dominated relations:

[0054] Definition of non-domination: An individual a is dominated by another individual b if and only if:

[0055] Individual b is not inferior to individual a in all objectives;

[0056] And is superior to individual a in at least one goal;

[0057] Sorting process:

[0058] Calculate the dominating set and dominated count for each individual;

[0059] All individuals whose dominated counts are 0 constitute the first frontier;

[0060] Remove the individuals on the first frontier and recalculate the dominated counts of the remaining individuals to generate the next frontier;

[0061] (2) The crowding distance is used to measure the sparsity of individuals in their dominant layer, with the aim of maintaining the diversity of solutions on the Pareto front during the selection process:

[0062] For each individual in the same non-dominated layer, initialize its crowding distance to 0;

[0063] Sort individuals by each objective function separately;

[0064] For each target, the crowding value of the individuals at both ends is calculated to be infinite;

[0065] For the middle individual, the crowding distance increment is the absolute value of the difference between the objective function values ​​of the two adjacent individuals;

[0066] The crowding distance increments of all targets are accumulated to obtain the total crowding distance;

[0067] (3) Tournament Selection:

[0068] Perform tournament selection based on non-dominated sorting and crowding distance to select individuals that will enter the next generation;

[0069] Randomly select two individuals from the population to compare:

[0070] Prioritize sorting by non-dominated layer, and prioritize individuals with lower non-dominated layers;

[0071] Secondly, we compare the crowding distances: if they are in the same layer, we select individuals with larger crowding distances. The goal is to increase the diversity of frontier solutions while maintaining evolutionary pressure.

[0072] Iteratively: Repeat tournament selection until the next generation of the complete population is constructed.

[0073] Furthermore, the MPC-based optimization engine optimizes the mixing parameters in real time under process constraints, performs fine-grained control optimization, and obtains the final parameter optimization solution, as follows:

[0074] Establish a state space model, and input the initial parameter optimization plan; predict the model equation:

[0075] ;

[0076] Among them, x(k) is the state vector; u(k) is the control input vector; y(k) is the output vector; A, B, C, D are system matrices; w(k) and v(k) are process noise and measurement noise respectively;

[0077] Define the prediction time domain N p and control time domain N c The system calculates the optimal control action in real time on the rolling time domain and p In the example above, we predict the future behavior of the system based on its current state and control inputs:

[0078]

[0079] Among them, Y is the future N p The output sequence of the step; U is the prediction model matrix, and It is derived through the state space equation;

[0080] Define the objective function of the optimization problem to solve the future N c The optimal control sequence of steps is:

[0081] ;

[0082] Among them, yref (k+i) is the reference output of the i-th prediction step; is the actual output of the i-th prediction step; is the penalty coefficient for the change of control action; u(k+i) represents the change of control action at the i-th prediction step;

[0083] Process constraints:

[0084] ;

[0085] Among them, u min、 u max are the minimum and maximum values ​​of the control input respectively; y min ,y max Output the minimum value of the variable;

[0086] Solve the optimization problem to obtain the optimal control action at the current moment, apply it to the process, monitor the output slurry quality in real time, update the status according to the feedback of MPC, and correct the control strategy;

[0087] Continuously obtain real-time slurry quality data and run MPC cycles;

[0088] Update prediction models and optimization parameters to ensure stable system performance and achieve optimization goals.

[0089] Furthermore, the optimization problem is solved to obtain the optimal control action at the current moment, as follows:

[0090] The objective function is transformed into a quadratic programming problem (QP) and implemented by the solver QuadProg:

[0091] ;

[0092] Among them, H is a quadratic matrix, which reflects the measurement of the control input change in the objective function; f is a linear term vector, involving the relationship between the current state, reference output and prediction model; Minimize means minimization;

[0093] From the optimal control sequence U ∗ Extract the first step u ∗ (k) and apply it to the system.

[0094] A paper pulp mixing control method based on multiple sensors comprises the following steps:

[0095] S1: Continuously monitor the key parameters in the slurry mixing process through the multi-sensor module, and transmit them to the intelligent analysis module after pre-processing through the data acquisition module;

[0096] S2: The intelligent analysis module predicts the slurry quality based on the pre-processed data through the slurry quality prediction model;

[0097] S3: The control module makes real-time parameter adjustments based on the analysis results of the intelligent analysis module, and uses multi-objective optimization technology to strike a balance between improving slurry quality and reducing energy consumption to obtain an initial parameter optimization solution. Based on the initial parameter optimization solution, the MPC-based optimization engine is used to optimize the mixing parameters in real time under process constraints and perform fine-grained control optimization.

[0098] The present invention has the following beneficial effects:

[0099] 1. The present invention effectively improves the overall efficiency and product quality of the pulp mixing process of the paper roll, and realizes effective control of energy consumption and cost;

[0100] 2. The genetic algorithm of the present invention effectively maintains the balance between population diversity and selection pressure through non-dominated sorting and crowding distance calculation. NSGA-II can ensure convergence to the global optimal solution and maintain the diversity of solutions when searching for Pareto frontier solutions that meet complex multi-objective optimization requirements. It can further achieve flexible adjustment to the optimal working conditions and effectively respond to changing process requirements and constraints.

[0101] 3. The present invention combines MPC with genetic algorithm, and provides an initial parameter optimization scheme through the genetic algorithm. MPC performs fine-grained control optimization according to this scheme, realizes fine regulation and comprehensive optimization of the slurry mixing process, thereby achieving the process goals of improving uniformity and reducing energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] Figure 1 This is a system architecture diagram of the present invention;

[0103] Figure 2 It is a flow chart of a control method in one embodiment of the present invention. DETAILED DESCRIPTION

[0104] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0105] refer to Figure 1 , in this embodiment, a multi-sensor based paper pulp mixing control system is provided, including a multi-sensor module, a data acquisition module, an intelligent analysis module, a control module and a data management and visualization module;

[0106] The multi-sensor module continuously monitors the key parameters in the slurry mixing process and transmits them to the intelligent analysis module after pre-processing by the data acquisition module;

[0107] The intelligent analysis module is provided with a slurry quality prediction model, which predicts the slurry quality based on the preprocessed data;

[0108] The control module performs real-time parameter adjustment according to the analysis results of the intelligent analysis module, uses multi-objective optimization technology to balance between improving slurry quality and reducing energy consumption, obtains an initial parameter optimization scheme, and based on the initial parameter optimization scheme, uses an MPC-based optimization engine to optimize the mixing parameters in real time under process constraints and perform fine-grained control optimization;

[0109] The data management and visualization module is connected to the data acquisition module, the intelligent analysis module, and the control module respectively to obtain the data of each module and perform visualization.

[0110] In this embodiment, the multi-sensor module includes a flow sensor, a concentration sensor, a temperature sensor, a pH sensor, a pressure sensor and a spectral sensor; the flow sensor is used to monitor the flow of the slurry to ensure that the slurry is mixed according to the set ratio; the concentration sensor is used to detect the fiber concentration in the slurry in real time; the temperature sensor is used to monitor the temperature change of the slurry; the pH sensor is used to monitor the acidity and alkalinity of the slurry to control the chemical properties; the pressure sensor is used to detect pressure fluctuations in the system to ensure safe operation; the spectral sensor is used to evaluate the composition and uniformity of the slurry.

[0111] In this embodiment, preprocessing includes using moving average filtering to remove sensor noise, using interpolation to fill in missing data, and performing data standardization on the sensor data to scale the data to a standard normal distribution.

[0112] In this embodiment, the slurry quality prediction model is constructed based on the XGBOOST model, as follows:

[0113] Obtain relevant historical data in the process of mixing slurry, including feature data and target variables, and perform feature importance evaluation, screen out features with importance greater than a threshold, and construct a training data set; the feature data includes flow, concentration, temperature and pH value; the target variables include viscosity, uniformity and strength;

[0114] Define XGBoost parameters, including learning rate: adjusted to a small value between 0.01 and 0.3, so that the model's high-level degree is gradually optimized. Maximum depth: set to 3-10 to control the complexity of the single tree model. Subsampling rate: set between 0.5 and 1.0 to prevent overfitting. Regularization parameters, defined as L1 and L2 regularization terms, control the complexity of the tree;

[0115] The XGBoost model is trained based on the training data set. The XGBoost model predicts the target value through G trees:

[0116] ;

[0117] in For the feature data, For the The actual target variable, For the predict target variables; f g is the g-th tree;

[0118] The objective function of XGBoost is:

[0119] ;

[0120] in, is the loss function; is the regularization term of the g-th tree;

[0121] ;

[0122] in, is the weight of the number of leaf nodes; is the L2 regularization coefficient of the leaf weight; is the prediction score of the jth leaf node; T is the number of leaf nodes in the tree;

[0123] The model is constructed incrementally, that is, one tree is added each time to reduce the error of the previous prediction. For the tth iteration, the model is updated as follows:

[0124] ;

[0125] in For the added trees; For the The predicted value of the sample at the tth iteration;

[0126] For each node split, a greedy method is used to find the best split point.

[0127] In this embodiment, for the splitting of each node, a greedy method is used to find the best splitting point, as follows:

[0128] For each feature, XGBoost performs the following steps to find the best split point:

[0129] Sort all samples of feature q;

[0130] Enumerate all possible splitting points of feature values ​​one by one. For continuous features, select the middle value as the potential splitting point;

[0131] For each split point, calculate the gain:

[0132]

[0133] in, , Respectively represent the gradient sum of the left child node and the right child node after splitting, and Respectively represent the second-order gradient sum of the left child node and the right child node after splitting (usually the derivative of the Hessian matrix); is the regularization parameter; is the split penalty parameter of the leaf node;

[0134] The splitting point that maximizes the gain is selected as the optimal splitting point for the feature.

[0135] In this embodiment, a multi-objective optimization technique is used to balance between improving slurry quality and reducing energy consumption, as follows:

[0136] Optimize the target definition, improve the slurry quality, and reduce energy consumption at the same time, then the multi-objective function is as follows:

[0137] min(f1(u),f2(u));

[0138] ;

[0139] ;

[0140] in, f 1 (u) is the slurry quality error: is the energy consumption estimation; the goal is to find the set of parameters u; To predict the slurry quality, TU is the target slurry quality; v is the flow rate, is the temperature, and is the coefficient related to energy consumption;

[0141] Based on genetic algorithm, process parameters are dynamically adjusted to improve slurry uniformity and reduce energy consumption to solve the above multi-objective function;

[0142] A population consisting of process parameters (flow, concentration, temperature, etc.) is randomly generated, and each individual can be represented as: Chromosome;

[0143] Combine the output of the prediction model and the definition of the multi-objective function to evaluate the fitness of each individual;

[0144] Using the non-dominated sorting method of NSGA-II, individuals are classified into different dominant layers. The goal is to find the Pareto frontier between reducing quality deviation and energy consumption, and calculate the crowding distance of individuals in each non-dominated layer; based on non-dominated sorting and crowding distance, tournament selection is performed to select better individuals to enter the next generation;

[0145] Generate new individuals using simulated binary crossover (SBX) operation :

[0146] ;

[0147] in, is the control factor, , There are two parent individuals;

[0148] Use polynomial mutation to make small adjustments to individual genes:

[0149] ;

[0150] in, is a gene value of an individual; x max and x min are the upper and lower limits of the value respectively; δ is the random variation;

[0151] The population is updated during the iteration process, and the above steps are repeated until convergence or the maximum number of iterations is reached.

[0152] In this embodiment, the non-dominated sorting method of NSGA-II is used to classify individuals into different dominant layers. The goal is to find the Pareto frontier between reducing quality deviation and energy consumption, and calculate the crowding distance of individuals in each non-dominated layer; based on the non-dominated sorting and crowding distance, the tournament selection is performed to select better individuals to enter the next generation, as follows:

[0153] (1) For the objective function f 1 (u),f 2 (u) , sorted according to non-dominated relations:

[0154] Definition of non-domination: An individual a is dominated by another individual b if and only if:

[0155] Individual b is no worse than individual a in all objectives;

[0156] And is superior to individual a in at least one goal;

[0157] Sorting process:

[0158] For each individual, calculate the dominating set (the set of individuals it dominates) and the dominated count (how many individuals it dominates);

[0159] All individuals whose dominated counts are 0 constitute the first frontier (non-dominated layer);

[0160] Remove the individuals on the first frontier and recalculate the dominated counts of the remaining individuals to generate the next frontier;

[0161] (2) The crowding distance is used to measure the sparsity of individuals in their dominant layer, with the aim of maintaining the diversity of solutions on the Pareto front during the selection process:

[0162] For each individual in the same non-dominated layer, initialize its crowding distance to 0;

[0163] Sort individuals by each objective function separately;

[0164] For each target, the crowding value of the individuals at both ends is calculated to be infinite (characterizing the boundary solution);

[0165] For the middle individual, the crowding distance increment is the absolute value of the difference between the objective function values ​​of the two adjacent individuals (after normalization);

[0166] The crowding distance increments of all targets are accumulated to obtain the total crowding distance;

[0167] (3) Tournament Selection:

[0168] Perform tournament selection based on non-dominated sorting and crowding distance to select individuals that will enter the next generation;

[0169] Randomly select two individuals from the population to compare:

[0170] Prioritize sorting by non-dominated layer, and prioritize individuals with lower non-dominated layers;

[0171] Secondly, we compare the crowding distances: if they are in the same layer, we select individuals with larger crowding distances. The goal is to increase the diversity of frontier solutions while maintaining evolutionary pressure.

[0172] Iteratively: Repeat tournament selection until the next generation of the complete population is constructed.

[0173] In this embodiment, an MPC-based optimization engine is used to optimize the mixing parameters in real time under process constraints, and fine-grained control optimization is performed to obtain the final parameter optimization solution, as follows:

[0174] Establish a state space model, and input the initial parameter optimization plan; predict the model equation:

[0175] ;

[0176] Among them, x(k) is the state vector; u(k) is the control input vector; y(k) is the output vector; A, B, C, D are system matrices; w(k) and v(k) are process noise and measurement noise respectively;

[0177] Define the prediction time domain N p and control time domain N c The system calculates the optimal control action in real time on the rolling time domain and p In the example above, we predict the future behavior of the system based on its current state and control inputs:

[0178]

[0179] Among them, Y is the future N p The output sequence of the step; U is the prediction model matrix, and It is derived through the state space equation;

[0180] Define the objective function of the optimization problem to solve the future N c The optimal control sequence of steps is:

[0181] ;

[0182] Among them, y ref (k+i) is the reference output of the i-th prediction step; is the actual output of the i-th prediction step; is the penalty coefficient for the change of control action; u(k+i) represents the change of control action at the i-th prediction step;

[0183] Process constraints:

[0184] ;

[0185] Among them, u min、 u max are the minimum and maximum values ​​of the control input respectively; y min ,y max Output the minimum value of the variable;

[0186] Solve the optimization problem to obtain the optimal control action at the current moment, apply it to the process, monitor the output slurry quality in real time, update the status according to the feedback of MPC, and correct the control strategy;

[0187] Continuously obtain real-time slurry quality data and run MPC cycles;

[0188] Update prediction models and optimization parameters to ensure stable system performance and achieve optimization goals.

[0189] In this embodiment, the optimization problem is solved to obtain the optimal control action at the current moment, as follows:

[0190] The objective function is transformed into a quadratic programming problem (QP) and implemented by the solver QuadProg:

[0191] ;

[0192] Among them, H is a quadratic matrix, which reflects the measurement of the control input change in the objective function; f is a linear term vector, involving the relationship between the current state, reference output and prediction model; Minimize means minimization;

[0193] From the optimal control sequence U ∗ Extract the optimal control input vector u from ∗ (k) and apply it to the system.

[0194] In this embodiment, the data management and visualization module is provided with a data acquisition module interface, an intelligent analysis module interface, a control module interface, a data management and storage module, and a data visualization module;

[0195] Data acquisition module interface: Use standardized data protocols (such as OPC UA, MQTT, Modbus, etc.) to connect to the data acquisition module to ensure stable data exchange. Receive real-time collected sensor and equipment status data, such as temperature, flow, concentration, etc.

[0196] Intelligent analysis module interface: Use API interfaces or message queues (such as Kafka, RabbitMQ) to obtain analysis results, such as predicted quality indicators (viscosity, uniformity, etc.). Structured data formats (such as JSON, XML) are easy to parse and process.

[0197] Control module interface: Exchange data with the control module through a low-latency, high-reliability communication protocol, such as Ethernet / IP. Obtain information such as device control status and current execution configuration.

[0198] The data management and storage module uses the database time series database InfluxDB to store data of different time scales: including real-time data submodule and historical data submodule. The real-time data submodule is used for short-term working condition monitoring and analysis and stores high-frequency data; the historical data submodule is used for long-term trend analysis and modeling and stores summarized low-frequency data.

[0199] Data visualization module, using the data visualization tool Grafana to create interactive visualization dashboards.

[0200] Provides three views: real-time monitoring, historical trend analysis, and analysis result display: Real-time monitoring: Displays the current status of each parameter through instruments, line charts, and heat maps. Historical trends: Displays long-term data trend changes, such as time series charts. Analysis results: Displays intelligent analysis results in the form of charts, such as statistical quality trends.

[0201] refer to Figure 2 In this embodiment, a paper pulp mixing control method based on multiple sensors is also provided, comprising the following steps:

[0202] S1: Continuously monitor the key parameters in the slurry mixing process through the multi-sensor module, and transmit them to the intelligent analysis module after pre-processing through the data acquisition module;

[0203] S2: The intelligent analysis module predicts the slurry quality based on the pre-processed data through the slurry quality prediction model;

[0204] S3: The control module makes real-time parameter adjustments based on the analysis results of the intelligent analysis module, and uses multi-objective optimization technology to strike a balance between improving slurry quality and reducing energy consumption to obtain an initial parameter optimization solution. Based on the initial parameter optimization solution, the MPC-based optimization engine is used to optimize the mixing parameters in real time under process constraints and perform fine-grained control optimization.

[0205] 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.

[0206] 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.

[0207] 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.

[0208] 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.

[0209] 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 paper pulp mixing control system based on multiple sensors, characterized in that: It includes multi-sensor module, data acquisition module, intelligent analysis module, control module and data management and visualization module; The multi-sensor module continuously monitors the key parameters in the slurry mixing process and transmits them to the intelligent analysis module after pre-processing by the data acquisition module; The intelligent analysis module is provided with a slurry quality prediction model, which predicts the slurry quality based on the preprocessed data; The control module performs real-time parameter adjustment according to the analysis results of the intelligent analysis module, uses multi-objective optimization technology to balance between improving slurry quality and reducing energy consumption, obtains an initial parameter optimization scheme, and based on the initial parameter optimization scheme, uses an MPC-based optimization engine to optimize the mixing parameters in real time under process constraints and perform fine-grained control optimization; The data management and visualization module is connected to the data acquisition module, the intelligent analysis module, and the control module respectively to obtain data from each module and visualize it; The multi-sensor module includes a flow sensor, a concentration sensor, a temperature sensor, a pH sensor, a pressure sensor and a spectrum sensor; the flow sensor is used to monitor the flow of the slurry to ensure that the slurry is mixed according to the set ratio; The concentration sensor is used to detect the fiber concentration in the slurry in real time; the temperature sensor is used to monitor the temperature change of the slurry; pH sensors, which monitor the acidity and alkalinity of the slurry to control the chemical properties; pressure sensors, which detect pressure fluctuations within the system to ensure safe operation; and spectral sensors, which assess the composition and uniformity of the slurry; The slurry quality prediction model is constructed based on the XGBOOST model, as follows: Obtain relevant historical data in the mixing slurry process, including feature data and target variables, and perform feature importance evaluation, filter out features with importance greater than the threshold, and construct a training data set; The characteristic data include flow rate, concentration, temperature and pH value; Target variables include viscosity, uniformity, and strength; Define XGBoost parameters, including learning rate, maximum depth, subsampling rate, and regularization parameters; The XGBoost model is trained based on the training data set. The XGBoost model predicts the target value through G trees: ; in For the feature data, For the The actual target variable, For the predict target variables; f g is the g-th tree; The objective function of XGBoost is: ； in, is the loss function; is the regularization term of the g-th tree; ； in, is the weight of the number of leaf nodes; is the L2 regularization coefficient of the leaf weight; is the prediction score of the jth leaf node; T is the number of leaf nodes in the tree; The model is constructed incrementally, that is, one tree is added each time to reduce the error of the previous prediction. For the tth iteration, the model is updated as follows: ； in For the added trees; For the The predicted value of the sample at the tth iteration; For each node split, a greedy method is used to find the best split point.

2. A multi-sensor based paper pulp mixing control system according to claim 1, characterized in that: The preprocessing includes using moving average filtering to remove sensor noise, using interpolation to fill in missing data, and performing data standardization on the sensor data to scale the data to a standard normal distribution.

3. The multi-sensor based pulp mixing control system of claim 1, characterized in that: For the splitting of each node, a greedy method is used to find the best splitting point, as follows: For each feature, XGBoost performs the following steps to find the best split point: Sort all samples of feature q; Enumerate all possible splitting points of feature values ​​one by one. For continuous features, select the middle value as the potential splitting point; For each split point, calculate the gain: in, , Respectively represent the gradient sum of the left child node and the right child node after splitting, and Respectively represent the second-order gradient sum of the left child node and the right child node after splitting; is the regularization parameter; is the split penalty parameter of the leaf node; The splitting point that maximizes the gain is selected as the optimal splitting point for the feature.

4. The multi-sensor based paper pulp mixing control system according to claim 1, characterized in that: The multi-objective optimization technology is used to balance between improving slurry quality and reducing energy consumption, as follows: Optimize the target definition, improve the slurry quality, and reduce energy consumption at the same time, then the multi-objective function is as follows: min(f1(u),f2(u)); ； ； in, f 1 (u) is the slurry quality error: is the energy consumption estimation; the goal is to find the set of parameters u; To predict the slurry quality, TU is the target slurry quality; v is the flow rate, is the temperature, and is the coefficient related to energy consumption; Based on genetic algorithm, process parameters are dynamically adjusted to improve slurry uniformity and reduce energy consumption to solve the above multi-objective function; Randomly generate a population consisting of process parameters; Combine the output of the prediction model and the definition of the multi-objective function to evaluate the fitness of each individual; Using the non-dominated sorting method of NSGA-II, individuals are classified into different dominant layers. The goal is to find the Pareto frontier between reducing quality deviation and energy consumption, and calculate the crowding distance of individuals in each non-dominated layer; based on non-dominated sorting and crowding distance, tournament selection is performed to select better individuals to enter the next generation; Generate new individuals using simulated binary crossover operation : ； in, is the control factor, , There are two parent individuals; Use polynomial mutation to make small adjustments to individual genes: ； in, is a gene value of an individual; x max and x min are the upper and lower limits of the value respectively; δ is the random variation; The population is updated during the iteration process, and the above steps are repeated until convergence or the maximum number of iterations is reached.

5. A multi-sensor based paper pulp mixing control system according to claim 4, characterized in that: The non-dominated sorting method of NSGA-II is used to classify individuals into different dominant layers. The goal is to find the Pareto frontier between reducing quality deviation and energy consumption, and calculate the crowding distance of individuals in each non-dominated layer; based on non-dominated sorting and crowding distance, the tournament selection is performed to select better individuals to enter the next generation, as follows: (1) For the objective function f 1 (u),f 2 (u) , sorted according to non-dominated relations: Definition of non-domination: An individual a is dominated by another individual b if and only if: Individual b is no worse than individual a in all objectives; And is superior to individual a in at least one goal; Sorting process: Calculate the dominating set and dominated count for each individual; All individuals whose dominated counts are 0 constitute the first frontier; Remove the individuals on the first frontier and recalculate the dominated counts of the remaining individuals to generate the next frontier; (2) The crowding distance is used to measure the sparsity of individuals in their dominant layer, with the aim of maintaining the diversity of solutions on the Pareto front during the selection process: For each individual in the same non-dominated layer, initialize its crowding distance to 0; Sort individuals by each objective function separately; For each target, the crowding value of the individuals at both ends is calculated to be infinite; For the middle individual, the crowding distance increment is the absolute value of the difference between the objective function values ​​of the two adjacent individuals; The crowding distance increments of all targets are accumulated to obtain the total crowding distance; (3) Tournament Selection: Perform tournament selection based on non-dominated sorting and crowding distance to select individuals that will enter the next generation; Randomly select two individuals from the population to compare: Prioritize sorting by non-dominated layer, and prioritize individuals with lower non-dominated layers; Secondly, we compare the crowding distances: if they are in the same layer, we select individuals with larger crowding distances. The goal is to increase the diversity of frontier solutions while maintaining evolutionary pressure. Iteratively: Repeat tournament selection until the next generation of the complete population is constructed.

6. The multi-sensor based pulp mixing control system of claim 1, characterized in that: The MPC-based optimization engine optimizes the mixing parameters in real time under process constraints, performs fine-grained control optimization, and obtains the final parameter optimization solution, as follows: Establish a state space model, and input the initial parameter optimization plan; predict the model equation: ； Among them, x(k) is the state vector; u(k) is the control input vector; y(k) is the output vector; A, B, C, D are system matrices; w(k) and v(k) are process noise and measurement noise respectively; Define the prediction time domain N p and control time domain N c The system calculates the optimal control action in real time on the rolling time domain and p In the example above, we predict the future behavior of the system based on its current state and control inputs: Among them, Y is the future N p The output sequence of the step; U is the prediction model matrix, and It is derived through the state space equation; Define the objective function of the optimization problem to solve the future N c The optimal control sequence of steps is: ； Among them, y ref (k+i) is the reference output of the i-th prediction step; is the actual output of the i-th prediction step; is the penalty coefficient for the change of control action; u(k+i) represents the change of control action at the i-th prediction step; Process constraints: ； Among them, u min、 u max are the minimum and maximum values ​​of the control input respectively; y min ,y max Output the minimum value of the variable; Solve the optimization problem to obtain the optimal control action at the current moment, apply it to the process, monitor the output slurry quality in real time, update the status according to the feedback of MPC, and correct the control strategy; Continuously obtain real-time slurry quality data and run MPC cycles; Update prediction models and optimization parameters to ensure stable system performance and achieve optimization goals.

7. A multi-sensor based pulp mixing control system for paper reel according to claim 6, characterized in that: The optimization problem is solved to obtain the optimal control action at the current moment, as follows: The objective function is transformed into a quadratic programming problem QP and implemented by the solver QuadProg: ； Among them, H is a quadratic matrix, which reflects the measurement of the control input change in the objective function; f is a linear term vector, involving the relationship between the current state, reference output and prediction model; Minimize means minimization; From the optimal control sequence U ∗ Extract the first step u ∗ (k) and apply it to the system.

8. A control method for a paper pulp mixing control system based on a multi-sensor according to any one of claims 1 to 7, characterized in that: The following steps are involved: S1: Continuously monitor the key parameters in the slurry mixing process through the multi-sensor module, and transmit them to the intelligent analysis module after pre-processing through the data acquisition module; S2: The intelligent analysis module predicts the slurry quality based on the pre-processed data through the slurry quality prediction model; S3: The control module makes real-time parameter adjustments based on the analysis results of the intelligent analysis module, and uses multi-objective optimization technology to strike a balance between improving slurry quality and reducing energy consumption to obtain an initial parameter optimization solution. Based on the initial parameter optimization solution, the MPC-based optimization engine is used to optimize the mixing parameters in real time under process constraints and perform fine-grained control optimization.

Images