A network intelligent measurement and control system for canned food production line based on cloud platform

Abstract

The present application relates to the technical field of intelligent measurement and control of can production line network, in particular to a kind of intelligent measurement and control system of can production line network based on cloud platform;It contains data acquisition, processing, strategy generation, execution, evaluation and adaptive tuning module;System constructs equipment health index, and combines commercial data set dynamically adjusts utility conversion coefficient;Its core is to build a global objective function that integrates output, energy consumption, material consumption and equipment health, and call cloud model to solve the optimal working condition strategy;Adaptive tuning is carried out based on the combined state of health level, transient overshoot, strategy compliance and performance deviation;The present application realizes the dynamic balance between economic benefit and equipment safety.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent measurement and control of can production line network, in particular to a can production line network intelligent measurement and control system based on a cloud platform. BACKGROUND

[0002] With the improvement of modern process industry automation level, the working condition of production system is increasingly complex, which brings challenges of multi-objective collaborative optimization and long-term stable operation of system; traditional measurement and control system mainly focuses on the stability of local process parameters, lacks overall comprehensive consideration of equipment health status and real-time business data, which leads to difficulty in achieving dynamic balance among output, energy consumption and equipment loss; in addition, the upper optimization model is prone to drift in long-term operation, and the compliance degree and transient process stability of cloud strategy at the physical execution level are difficult to be effectively guaranteed; therefore, how to construct an intelligent measurement and control system capable of dynamically integrating multi-source heterogeneous information, balancing global benefits and physical constraints, and having multi-level closed-loop adaptive tuning capability, becomes a technical problem to be solved in the field. SUMMARY

[0003] To solve the above technical problems, the present application provides a can production line network intelligent measurement and control system based on a cloud platform, in particular, the technical scheme of the present application comprises:

[0004] A data acquisition module is configured to acquire process data sets, equipment state data sets, batch data sets and business data sets in real time.

[0005] A first processing module is configured to construct an equipment health degree index based on the equipment state data sets.

[0006] The first processing module is further configured to determine the health status grade corresponding to the equipment health degree index according to a preset health degree threshold.

[0007] A second processing module is configured to dynamically adjust an utility conversion coefficient based on the business data sets.

[0008] The second processing module is further configured to calculate an equipment degradation index based on the equipment health degree index.

[0009] The second processing module is further configured to determine a global objective function value in combination with the utility conversion coefficient, the equipment degradation index and real-time production data.

[0010] A strategy generation module is configured to call a pre-trained cloud optimization model, to maximize the global objective function value, to fuse the process data sets, the batch data sets, the equipment health degree index and the utility conversion coefficient, and to solve and generate a working condition strategy vector.

[0011] A strategy execution module is configured to issue the working condition strategy vector to a distributed measurement and control node for execution.

[0012] The strategy execution module is further configured to monitor a transient overshoot of the key process parameters;

[0013] The performance evaluation module is configured to calculate a node strategy compliance degree of the measurement and control node;

[0014] The performance evaluation module is further configured to calculate a global performance deviation;

[0015] The adaptive tuning module is configured to perform adaptive closed-loop tuning based on a combined state of the health state level, the transient overshoot, the node strategy compliance degree, and the global performance deviation.

[0016] Preferably, the first processing module constructs a device health index, including:

[0017] Based on real-time values in the device state dataset, preset reference values, and failure threshold values, a normalized degradation factor is calculated;

[0018] Based on the pre-calibrated degradation weight and the normalized degradation factor, a device health index is determined using a weighted subtraction model.

[0019] Preferably, the second processing module determines a global objective function value, including:

[0020] In response to changes in electricity prices or order priorities in the business data, the utility conversion coefficient is dynamically adjusted;

[0021] The device health index is converted into a device degradation index;

[0022] A linear utility function is used to weight the qualified product output, total energy consumption, total material consumption, and device degradation index to determine the global objective function value.

[0023] Preferably, the operating condition strategy vector includes a sub-strategy package issued to the measurement and control node;

[0024] The sub-strategy package includes target set values of the key process parameters, parameter change slopes at mode switching, and coordination delay times of the execution strategy.

[0025] Preferably, the strategy execution module monitors the transient overshoot, including:

[0026] The measurement and control node executes parameter adjustment according to the parameter change slope and the coordination delay time in the sub-strategy package;

[0027] The peak value, the new target steady-state value, and the old steady-state value in the parameter switching process are obtained;

[0028] Based on the peak value, the new target steady-state value, and the old steady-state value, a transient overshoot is calculated using a percentage overshoot definition.

[0029] Preferably, the performance evaluation module calculates the node strategy compliance degree, including:

[0030] Comparing the actual value of the local parameter of the measurement and control node with the target set value within a specified time;

[0031] When the error between the actual value of the local parameter and the target set value is within the preset tolerance, the node compliance state is determined to be 1;

[0032] When the error between the actual value of the local parameter and the target set value is greater than the preset tolerance, the node compliance state is determined to be 0;

[0033] Using a statistical average model, the average value of the node compliance state of all measurement and control nodes is calculated to determine the node strategy compliance degree.

[0034] Preferably, the performance evaluation module calculates the global performance deviation, including:

[0035] Obtaining the predicted target value predicted by the strategy generation module;

[0036] Using the formula of the global target function and substituting the real-time measured production data and the equipment health degree index, the actual utility value is calculated;

[0037] Based on the predicted target value and the actual utility value, the error signal calculation is used to determine the global performance deviation.

[0038] Preferably, the adaptive tuning module performs adaptive closed-loop tuning, including:

[0039] When the node strategy compliance degree is lower than the preset compliance degree threshold and the health state level is healthy, it is determined that there is a local node control logic conflict, and a local correction factor is issued to the measurement and control node;

[0040] When the node strategy compliance degree is greater than or equal to the preset compliance degree threshold or the health state level is unhealthy, the local correction factor is not issued.

[0041] Preferably, the adaptive tuning module performs adaptive closed-loop tuning, further including:

[0042] When the node strategy compliance degree is lower than the preset compliance degree threshold and the health state level is unhealthy, or when the transient overshoot exceeds the preset overshoot threshold, the strategy generation module is triggered to recalculate to generate a more conservative working condition strategy vector;

[0043] When the node strategy compliance degree is greater than or equal to the preset compliance degree threshold or the health state level is healthy, and the transient overshoot is less than or equal to the preset overshoot threshold, the strategy generation module is not triggered to recalculate.

[0044] Preferably, the adaptive tuning module performs adaptive closed-loop tuning, further including:

[0045] When the global performance deviation continues to exceed the preset model drift threshold, it is determined that the cloud optimization model is inaccurate, and a retraining process of the cloud optimization model is triggered;

[0046] When the global performance deviation does not continue to exceed the preset model drift threshold, the retraining process of the cloud optimization model is not triggered.

[0047] Compared with the prior art, the present application has the following beneficial effects:

[0048] 1. The system builds a global optimization target that integrates output, energy consumption, material consumption and equipment health status; it can also dynamically adjust the weights of each target according to real-time electricity prices or order priority and other business data, thereby achieving dynamic balance between pursuing economic benefits and ensuring equipment safety;

[0049] 2. The system solves the problem of long-term operation of the optimization model prone to inaccuracy; by continuously comparing the global benefits predicted by the model with the actual benefits generated by the production line, once a persistent large deviation is found between the two, the model retraining process is automatically triggered, ensuring the long-term accuracy of optimization decisions;

[0050] 3. The system improves the stability of the production process during working condition switching; the strategy issued not only includes the final target setting value, but also finely defines the parameter change rate and coordination time between devices; at the same time, the system monitors the transient overshoot in the process in real time, and once a large fluctuation is found, it will automatically generate a more stable conservative strategy, ensuring the smoothness and safety of production switching;

[0051] 4. The system realizes a multi-level adaptive tuning closed loop; it can intelligently diagnose the root cause of the deviation, distinguish whether it is a local execution problem, a global strategy risk or a top-level model inaccuracy, and automatically call the corresponding correction measures such as correcting local control, recalculating strategy or retraining model, ensuring the robustness and efficiency of the system during long-term operation. BRIEF DESCRIPTION OF DRAWINGS

[0052] The present application will be further explained in conjunction with the accompanying drawings and examples:

[0053] Figure 1 is a structural diagram of the system of the present application. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical scheme and advantages of the present application clearer and more apparent, the present application will be further described in detail below in conjunction with specific examples.

[0055] Example 1:

[0056] Please refer to Figure 1The application discloses a cloud platform-based network intelligent measurement and control system for a can production line, comprising:

[0057] a data acquisition module configured to acquire process data sets, equipment state data sets, batch data sets and commercial data sets in real time;

[0058] a first processing module configured to construct an equipment health index based on the equipment state data sets;

[0059] The first processing module is further configured to determine a health state grade corresponding to the equipment health index according to a preset health index threshold;

[0060] a second processing module configured to dynamically adjust a utility conversion coefficient based on the commercial data sets;

[0061] The second processing module is further configured to calculate an equipment degradation index based on the equipment health index;

[0062] The second processing module is further configured to determine a global target function value by combining the utility conversion coefficient, the equipment degradation index and real-time production data;

[0063] a strategy generation module configured to call a pre-trained cloud optimization model, to maximize the global target function value, to fuse the process data sets, the batch data sets, the equipment health index and the utility conversion coefficient, and to solve a working condition strategy vector;

[0064] a strategy execution module configured to send the working condition strategy vector to a distributed measurement and control node for execution;

[0065] The strategy execution module is further configured to monitor a transient overshoot of a key process parameter;

[0066] a performance evaluation module configured to calculate a node strategy compliance degree of the measurement and control node;

[0067] The performance evaluation module is further configured to calculate a global performance deviation;

[0068] an adaptive tuning module configured to perform adaptive closed-loop tuning based on a combined state of the health state grade, the transient overshoot, the node strategy compliance degree and the global performance deviation.

[0069] The embodiment provides a cloud platform-based network intelligent measurement and control system for a can production line;

[0070] In particular, the system comprises a data acquisition module, which aims to acquire multi-source heterogeneous working condition information of the production line in real time and comprehensively; in this embodiment, the module acquires four types of key data sets in the can production line in real time through integrated distributed measurement and control nodes such as PLC, sensors, equipment management system EMS, manufacturing execution system MES, and cloud platform interface:

[0071] Process data set : contains high-frequency process data such as temperature and pressure of the sterilization kettle and ;

[0072] Equipment state data set : contains medium-frequency equipment state data such as vibration and current of the sealing machine motor and ;

[0073] Batch data set : contains low-frequency batch attribute data such as material acidity and size and ;

[0074] Business data set : contains external economic data from the cloud platform, such as real-time electricity price and order priority and ;

[0075] The system comprises a first processing module, which aims to convert the equipment state data at the physical layer into quantified health indicators that can be used for optimization at the upper layer; in this embodiment, the module constructs an equipment health index based on the equipment state data set , for example, by using vibration and current data and implementing a specific quantification model ; the equipment health index is a normalized, dimensionless value ranging from [0, 1], which is used to represent the degree of deviation of the equipment from the ideal health state ;

[0076] The first processing module is also used to determine the health status grade corresponding to the equipment health index according to the health threshold preset by statistical analysis of historical data, for example and ; for example, it is divided into three grades of health , sub-health , and fault warning ; this grading result will be used as a key input for adaptive tuning

[0077] The system comprises a second processing module, which aims to construct a unified optimization objective capable of reflecting global benefits; in the present embodiment, the module is based on commercial data sets For example, changes in electricity prices Or order priorities Dynamic adjustment of utility conversion coefficients Utility conversion coefficients Refers to a set of weight coefficients used to convert different physical quantities such as output, energy consumption, and material consumption into utility dimensions, such as The source is dynamically determined according to For example, when electricity prices Rise, increase the utility cost coefficient of energy consumption ;

[0078] The second processing module is also used to calculate the equipment degradation index Based on the equipment health index Calculated by the first processing module; the equipment degradation index Refers to A simple transformation such as Used to represent the utility loss caused by equipment degradation;

[0079] The second processing module is also used to determine the global objective function value Combining the utility conversion coefficient , the equipment degradation index And real-time production data such as qualified product output , total energy consumption , total material consumption ; the implementation of this Value represents the comprehensive utility under the current working condition;

[0080] The system comprises a strategy generation module, which aims to solve the optimal control strategy under the current working condition; in the present embodiment, the module calls a pre-trained cloud optimization model ; the cloud optimization model Refers to a deep reinforcement learning model or model predictive control (MPC) model that has been pre-trained using historical data, which learns the nonlinear mapping relationship between input conditions and optimal Value; the pre-training process is as follows: collect the complete data set during the historical production period As input features, and use the actual utility value Ja in the historical data as the training label or reward signal, and iteratively train using supervised learning or offline reinforcement learning algorithms until the model Converges; in the present embodiment, the deep reinforcement learning model Specifically, a deep deterministic gradient algorithm is adopted, with both the Actor and Critic networks using fully connected neural networks containing three hidden layers. The number of neurons in the hidden layers are 256, 128, and 64, respectively, and the activation function is ReLU. The model is trained based on historical datasets and is completed offline in batches. When implemented using a deep reinforcement learning model, its state space... At least including Its action space Corresponding to the working condition strategy vector The setting range of each parameter; its reward function Defined during training That is, the global objective function value calculated in Example 3;

[0081] This module aims to maximize the global objective function value. To achieve the goal of merging process datasets Batch dataset Equipment health index and utility conversion factor Through the model Solving for the generation of working condition policy vectors ;Right now To process multi-rate input data, in At the start of each decision cycle of the model, the module uses a pre-defined aggregation strategy to aggregate the high-frequency process dataset. Perform feature extraction, such as calculating its mean over the period. and maximum value Simultaneously, obtain the low-frequency batch dataset for the current period. Ultimately, the actual input state of the model. Constructed as This ensures that the dimension of the input vector is fixed and the timing is aligned;

[0082] The system includes a policy execution module, the purpose of which is to accurately execute the macro-level policies generated in the cloud on the physical device; in this embodiment, this module executes the operating condition policy vector. This vector contains sub-policy packets distributed to each telemetry and control node. The command is then sent to distributed monitoring and control nodes such as PLCs and frequency converters for execution.

[0083] The strategy execution module is also used to monitor transient overshoot of key process parameters. Transient overshoot This refers to process parameters such as temperature. The percentage by which the actual peak value exceeds the new target value during the transition from the old steady-state value to the new target setpoint; this is a key process stability indicator.

[0084] The system comprises a performance evaluation module, which aims to evaluate the overall performance of the system from two dimensions of execution accuracy and model accuracy; in this embodiment, the module is used to calculate the node strategy compliance degree of the measurement and control node ; node strategy compliance degree refers to a value in the range of [0, 1], which quantifies the degree to which all measurement and control nodes accurately execute the target setting values in the sub-strategy package issued by the cloud;

[0085] The performance evaluation module is also used to calculate the global performance deviation ; global performance deviation refers to the difference between the optimal target value predicted by the strategy generation module and the actual utility value calculated based on real-time data ;

[0086] The system comprises an adaptive tuning module, which aims to make multi-level closed-loop corrections to the system according to the results of performance evaluation; in this embodiment, the module is used to perform adaptive closed-loop tuning based on the combined state of the health state level from the first processing module, the transient overshoot from the strategy execution module, the node strategy compliance degree and the global performance deviation all from the performance evaluation module; the tuning logic includes arbitration at the node level, recalculation at the strategy level, and retraining at the model level;

[0087] By constructing a complete closed-loop intelligent measurement and control system that integrates multi-source data acquisition, quantitative health modeling, dynamic utility optimization, transient suppression execution, multi-dimensional performance evaluation, and three-level adaptive tuning, the present application realizes unprecedented depth optimization of the can production line; it not only can generate a globally optimal working condition strategy according to real-time business targets such as electricity price, orders and physical constraints such as equipment health degree, but also can dynamically correct execution deviation, strategy risk and model drift through real-time monitoring of execution process compliance , overshoot and model performance deviation , thereby maximizing the overall economic benefit and operational resilience of the production line while ensuring the stability of the production process and the safety of the equipment.

[0088] Embodiment 2:

[0089] The first processing module constructs an equipment health index, including:

[0090] Based on the real-time values in the equipment state data set, the preset reference and the failure threshold, a normalized degradation factor is calculated;

[0091] ​Based on the pre-calibration degradation weight and the normalized degradation factor, the device health index is determined by using a weighted subtraction model calculation.

[0092] This embodiment is a detailed description of the specific implementation of the first processing module in the system of embodiment 1 to construct the device health index ; the purpose is to provide a standardized and reproducible device health quantification method

[0093] In this embodiment, the first processing module constructs by the following steps:

[0094] Based on the real-time values in the device state data set , such as the real-time vibration values collected in , the preset reference , and the failure threshold , the normalized degradation factor is calculated; the preset reference refers to the parameter value of the device in the ideal healthy state, which is derived from the factory calibration or historical best working condition data of the device; the failure threshold refers to the maximum tolerance value that the device parameter reaches to determine that it is close to failure, which is derived from industry standards or historical failure database The normalized degradation factor refers to the standardized deviation degree of the first

[0095] monitoring parameter, which serves to eliminate the influence of different parameter dimensions and unify the degradation degree in the range of [0, 1]; its calculation formula in this embodiment is as follows: To ensure the robustness of the calculation and prevent the denominator from being zero, it is agreed that when , if , then , if , then ; when : ;

[0096] ;

[0097] wherein, : the real-time measurement value of the first state parameter;

[0098] : the ideal reference value of the first state parameter;

[0099] : the failure threshold of the first state parameter;

[0100] ​ : the normalized degradation factor of the : the normalized degradation factor of the

[0101] The calculation logic ensures that when the real-time value is at the ideal benchmark , it is 0; when reaches the failure threshold , it is 1;

[0102] The module calculates the equipment health index based on the pre-calibrated degradation weight and the normalized degradation factor calculated in the previous step using a weighted subtraction model; ;

[0103] The degradation weight refers to the influence weight of the : the influence weight of the : the influence weight of the ;

[0104] The calculation formula of the equipment health index in this embodiment is as follows:

[0105] ;

[0106] wherein, : the normalized equipment health index;

[0107] : the total number of equipment state parameters for evaluating health;

[0108] : the degradation weight of the : the degradation weight of the

[0109] : the normalized degradation factor of the : the normalized degradation factor of the

[0110] The calculation logic subtracts the unhealthiness determined by the degradation of each parameter and its importance from the ideal state through a weighted subtraction model, thereby obtaining the final quantitative health index ​​​​The weighted subtraction model adopted in this embodiment is a simplified and easy-to-implement calculation method. In other embodiments, a more complex nonlinear model can also be used to construct The relationship between to capture the complex coupling and nonlinear effects between various degradation factors. Although the weighted subtraction model is a linear simplification, the inventors have found through experimental data analysis that during the conventional wear stage of the main equipment of the canning production line, the influence of each degradation factor on is relatively small compared to the coupling effect before failure warning occurs. Therefore, using a linear model is an effective balance between ensuring engineering precision and greatly improving the computational efficiency and interpretability of the model . The system compensates for the long-term drift of the linear model through the model level retraining loop in embodiment 8.

[0111] By introducing normalized degradation factors and degradation weights , the present application provides an objective and standardized equipment health quantification model. It not only converts raw equipment data such as vibration and current of different physical dimensions into dimensionless health indexes , but also allows the index to accurately reflect the true impact of different component degradation on overall health. This provides stable, reliable, and clearly physically meaningful equipment constraint inputs for upper-level optimization models such as , greatly improving the accuracy and safety of global strategies.

[0112] Embodiment 3:

[0113] The second processing module determines the global objective function value, including:

[0114] Adjusting the utility conversion coefficient dynamically in response to changes in electricity prices or order priority in the business data set;

[0115] Converting the equipment health index into an equipment degradation index;

[0116] Using a linear utility function, the qualified product output, total energy consumption, total material consumption, and equipment degradation index are weighted to determine the global objective function value.

[0117] This embodiment is a detailed description of the specific implementation of the second processing module in embodiment 1 to determine the global objective function value . The purpose is to construct an optimization function that can uniformly measure multiple conflicting target outputs, energy consumption, material consumption, and equipment wear and can be dynamically guided by business objectives; ​

[0118] In this embodiment, the second processing module determines the utility conversion coefficient by the following steps :

[0119] The module dynamically adjusts the utility conversion coefficient in response to changes in the electricity price or order priority in the business data set ; i.e. ;

[0120] The utility conversion coefficient here specifically refers to the dynamic weight for converting physical quantities into a unified utility dimension; its adjustment logic is preset in the business rules of the cloud platform, for example: when the displayed electricity price increases, the system automatically increases the value of utility / kWh, meaning that the penalty for energy consumption is increased; when the order priority increases, the system automatically increases the value of utility / quantity, meaning that the reward for output is increased; the specific mathematical form of this adjustment logic is, for example: a piecewise linear relationship with the electricity price , where the coefficient takes different values in different electricity price intervals; an exponential relationship with the order priority , to significantly increase the output utility of high-priority orders; these functional relationships are preset in the business rules of the cloud platform; among them, the coefficients , and are determined by multivariate regression analysis on historical business data and historical actual utility values ;

[0121] The module converts the equipment health index calculated by the first processing module into an equipment degradation index ;

[0122] The equipment degradation index here specifically refers to ; it is a dimensionless value that quantifies the cost or risk caused by equipment degradation; The lower ,the higher ;

[0123] The module uses a linear utility function to calculate the qualified product output , total energy consumption , and total material consumption​and equipment degradation index The weighting is performed to determine the global objective function value ;

[0124] The purpose of this linear utility function is to convert all optimization objectives into a unified utility dimension for summation and comparison; its calculation formula in this embodiment is as follows:

[0125] ;

[0126] wherein, : global objective function value; The greater the value, the higher the global benefit;

[0127] : qualified product output;

[0128] : total energy consumption of the production line;

[0129] : total material consumption;

[0130] : equipment degradation index;

[0131] : dynamic utility conversion coefficient;

[0132] This calculation logic clearly shows that the system pursues the maximization of positive utility, while suppressing , and negative utility or cost; and the dynamic nature of the weight makes this optimization objective not fixed, but can reflect changes in the external business environment in real time; those skilled in the art should understand that this linear utility function (71) can also be extended to a nonlinear function, for example, replacing the energy consumption term with a nonlinear cost function to account for the impact of tiered electricity prices; or replacing the output term with a bounded benefit function to account for the order saturation effect;

[0133] By constructing a linear utility function that is dynamically driven by business data , the present application successfully unifies multiple originally conflicting physical production objectives of output, energy consumption, material consumption, and abstract equipment health objectives into a single, quantifiable utility index; this makes the upper-level optimization model With a clear, definite optimization direction directly linked to commercial value, namely maximization... This ensures the final generated operating condition strategy. It can achieve the best balance between economic benefits and engineering constraints.

[0134] Example 4:

[0135] The operating condition policy vector includes sub-policy packages that are sent to the measurement and control nodes;

[0136] The sub-strategy package includes target setpoints for key process parameters, the slope of parameter changes during mode switching, and the coordination delay time for execution strategies.

[0137] This embodiment is based on the working condition strategy vector in the system of Embodiment 1. A detailed description of its specific composition; its purpose is to define a strategy data structure that can carry complex control logic, especially one containing transient control information;

[0138] In this embodiment, the working condition policy vector generated by the policy generation module It is a set of strategies issued to N measurement and control nodes on the production line, that is... ;

[0139] This vector includes sub-policy packets sent to each telemetry and control node. ;

[0140] Sub-strategy package This refers to the first Each monitoring and control node, such as the specific control instruction set of a PLC for a sterilization autoclave, serves to ensure the macro-level strategy implemented in the cloud. It can be broken down into precise parameters that can be executed by local nodes;

[0141] In this embodiment, the sub-policy package Includes the following three key parameters:

[0142] Target settings for key process parameters ;

[0143] Target set value This refers to the new steady-state process parameter values ​​that the node needs to achieve, such as a target sterilization temperature of 121°C or a target pressure of 0.15 MPa.

[0144] Parameter change slope during mode switching ;

[0145] Parameter change slope This refers to the parameter changing from its old value. Adjust to new value The rate of change of the parameter value over time, e.g. unit: °C / s; its role is to control the smoothness of the parameter change, to suppress the transient overshoot of the process;

[0146] Coordination delay time of the execution strategy ;

[0147] Coordination delay time Refers to how long the node should wait after receiving the instruction before starting execution, e.g. unit: seconds And The defined parameter change; its role is to ensure the coordination of actions between upstream and downstream devices, such as sterilization tanks and conveyors, to avoid timing errors;

[0148] That is ;

[0149] By defining the working condition strategy vector As a set of sub-strategy packages , And The invention greatly enriches the connotation of the control strategy; it not only issues the steady-state target More importantly, it issues the transient path And timing coordination ; This design enables the cloud optimization strategy to have fine transient suppression and timing coordination capabilities at the execution level, ensuring smoothness and safety during complex working condition switching from the source, which is not available in traditional control strategies that only issue set values .

[0150] Example 5:

[0151] The strategy execution module monitors the transient overshoot, including:

[0152] The measurement and control node executes parameter adjustment according to the parameter change slope in the sub-strategy package and the coordination delay time;

[0153] Obtain the peak value, new target steady-state value and old steady-state value during the parameter switching process;

[0154] Based on the peak value, new target steady-state value and old steady-state value, the transient overshoot is calculated using the percentage overshoot definition.

[0155] This embodiment is a detailed description of the specific implementation method of the strategy execution module monitoring the transient overshoot In the system of embodiment 1; its purpose is to provide a method for quantitatively evaluating the stability of the system during the execution of the strategy, especially during the working condition switching;

[0156] ​In this embodiment, the policy execution module monitors the parameter adjustment process while executing the policy by the following steps :

[0157] The measurement and control node, such as a PLC, strictly adjusts the parameter according to the parameter change slope and coordination delay time in the sub-policy package received from the cloud ; For example, the PLC controls the heater to strictly increase the temperature from the old steady-state value to the new target set value at a slope of °C / s ;

[0158] During the parameter adjustment process, the policy execution module or its commissioned measurement and control node obtains the peak value , the new target steady-state value , and the old steady-state value during the parameter switching process ;

[0159] The peak value refers to the maximum or minimum instantaneous value actually measured during the change of the parameter from to ;

[0160] The new target steady-state value is derived from the sub-policy package ;

[0161] The old steady-state value is derived from the actual steady-state value of the system before switching ;

[0162] Based on this, the module calculates the transient overshoot based on the peak value , the new target steady-state value , and the old steady-state value using the percentage overshoot definition ;

[0163] This calculation formula takes the heating process as an example in this embodiment: to ensure the effectiveness of the calculation, it is defined that the calculation is only enabled when ; if , then ; ;

[0164] ;

[0165] wherein : transient overshoot ;

[0166] : measured parameter peak value during switching ;

[0167] : new target steady state value setting value

[0168] : old steady state value

[0169] The calculation logic applies the standard overshoot definition in control engineering, quantifies the relative degree to which the system response exceeds the desired target , and the denominator of which is the changing reference; The value of is the core index to measure the stability of the system transient response, and the smaller the value is, the more stable it is;

[0170] By calculating the transient overshoot in real time during the execution of the strategy , the present application provides a key physical process feedback index for the system; Directly measure the execution effect of the transient control strategy guided by and parameters from Example 4; this makes the system not only know whether the strategy is executed by evaluation, but also know whether the strategy is executed well by evaluation, which provides a key basis for judging whether the strategy is too aggressive for policy-level re-tuning decision.

[0171] Example 6:

[0172] The performance evaluation module calculates the node strategy compliance degree, including:

[0173] Compare the actual value of the local parameter of the measurement and control node with the target setting value within a specified time;

[0174] When the error between the actual value of the local parameter and the target setting value is within the preset tolerance, the node compliance state is determined to be 1;

[0175] When the error between the actual value of the local parameter and the target setting value is greater than the preset tolerance, the node compliance state is determined to be 0;

[0176] Using a statistical average model, the average value of the node compliance state of all measurement and control nodes is calculated to determine the node strategy compliance degree.

[0177] This embodiment is a detailed description of the specific implementation method of the performance evaluation module calculating the node strategy compliance degree in the system of Example 1; the purpose is to quantitatively evaluate the overall execution accuracy of all measurement and control nodes on the production line to the working condition strategy vector issued by the cloud;

[0178] In this embodiment, the performance evaluation module calculates by the following steps:

[0179] The module within the specified time Inside, compare the first Actual values ​​of local parameters of each monitoring and control node With sub-strategy package Target setting value ; Specified time It is a pre-set evaluation time window based on the device's response characteristics; actual values ​​of local parameters. Its source is the feedback value from the local sensors at the measurement and control node;

[0180] Based on the comparison results, the module determines the compliance status of the node. :

[0181] When the actual value of the local parameter Compared with the target set value The error is within the preset tolerance. Inner time Determine the node's dependency state A value of 1 indicates compliance; default tolerance. It is determined based on the process requirements and control accuracy experimental calibration of the corresponding measurement and control nodes;

[0182] When the actual value of the local parameter Compared with the target set value The error is greater than the preset tolerance At that time Determine the node's dependency state A value of 0 indicates non-compliance;

[0183] Node dependency state It is a boolean value of 0 or 1, indicating whether a single node has successfully executed the strategy;

[0184] The module uses a statistical average model to calculate all The node compliance status of each monitoring and control node The average value is used to determine the node policy compliance degree. ;

[0185] The calculation formula is as follows in this embodiment:

[0186] ;

[0187] in, Node policy compliance degree;

[0188] The total number of monitoring and control nodes participating in the evaluation; in this embodiment, it is assumed that... ;like ,but The default value is 1;

[0189] : No. The dependency state of each node;

[0190] The computational logic calculates the discrete dependency states of all N nodes. It converges into a continuous value within the range [0,1]. This indicates that all nodes executed the strategy perfectly. The lower the value, the greater the deviation at the execution level;

[0191] By introducing node policy compliance This metric enables the present invention to provide a quantitative assessment of the accuracy of strategy execution; As a system-level macro indicator, it enables the cloud platform to clearly grasp the overall implementation status of global strategies at the physical level; this allows the system to clearly distinguish whether the strategy itself is problematic or the strategy is not being executed properly, providing a core basis for judgment for adaptive tuning, especially in distinguishing between node-level conflicts and strategy-level problems.

[0192] Example 7:

[0193] The performance evaluation module calculates global performance deviations, including:

[0194] Obtain the predicted target value from the strategy generation module;

[0195] The actual utility value is calculated by using the formula of the global objective function and substituting real-time measured production data and equipment health index.

[0196] Based on the predicted target value and the actual utility value, the global performance deviation is determined by using the error signal calculation.

[0197] This embodiment calculates the global performance deviation in the performance evaluation module of the system in Embodiment 1. A detailed explanation of the specific implementation method; its purpose is to quantitatively evaluate the cloud optimization model from the highest level of the system, i.e., the overall benefit. The accuracy of the prediction;

[0198] In this embodiment, the performance evaluation module calculates through the following steps. :

[0199] The module obtains the strategy generation module in the strategy generation module. The predicted target value of the time prediction Predicted target value This refers to the cloud-based optimization model. In solving At that time, the estimated, in the execution The optimal global objective function value that can be achieved after the strategy;

[0200] Meanwhile, the module adopts the same formula as the global target function in Embodiment 3 and substitutes the real-time measured production data such as and the equipment health index calculated by the first processing module to obtain , the actual utility value ;

[0201] The actual utility value refers to the real global target function value reached by the production line after the strategy is actually executed; the calculation formula of Ja is: wherein all are real-time measured or calculated values;

[0202] Based on the above two values, the module determines the global performance deviation based on the error signal calculation based on the predicted target value and the actual utility value ;

[0203] The calculation formula in this embodiment is as follows:

[0204] ;

[0205] wherein, : global performance deviation;

[0206] : optimal utility value predicted by the model;

[0207] : actual utility value calculated by real-time data;

[0208] The calculation logic of is the standard error signal in control theory, which directly compares the gap between the model predicted and the physical world realized; a continuously increasing strongly indicates that the cloud model has lost its description of the physical reality, i.e., model drift has occurred;

[0209] By calculating the global performance deviation , the application constructs the highest level evaluation closed loop for the performance of the cloud optimization model itself; No longer focusing on the underlying execution details such as or Instead, it directly measures the predictive capability of the model on the ultimate goal of global benefit; this enables the system to timely discover and quantify model drift phenomenon, providing the most direct and critical trigger signal for model-level retraining tuning, thus ensuring the system's optimality even when production characteristics change slowly in the long run.

[0210] Embodiment 8:

[0211] The adaptive tuning module performs adaptive closed-loop tuning, comprising:

[0212] When the node strategy compliance degree is lower than the preset compliance threshold, and the health state level is healthy, it is determined that there is a local node control logic conflict, and a local correction factor is issued to the control node;

[0213] When the node strategy compliance degree is greater than or equal to the preset compliance threshold, or the health state level is unhealthy, the local correction factor is not issued.

[0214] The adaptive tuning module performs adaptive closed-loop tuning, further comprising:

[0215] When the node strategy compliance degree is lower than the preset compliance threshold and the health state level is unhealthy, or when the transient overshoot exceeds the preset overshoot threshold, the strategy generation module is triggered to recalculate and generate a more conservative working condition strategy vector;

[0216] When the node strategy compliance degree is greater than or equal to the preset compliance threshold or the health state level is healthy, and the transient overshoot is less than or equal to the preset overshoot threshold, the strategy generation module is not triggered to recalculate.

[0217] The adaptive tuning module performs adaptive closed-loop tuning, further comprising:

[0218] When the global performance deviation continuously exceeds the preset model drift threshold, it is determined that the cloud optimization model is inaccurate, and the retraining process of the cloud optimization model is triggered;

[0219] When the global performance deviation does not continuously exceed the preset model drift threshold, the retraining process of the cloud optimization model is not triggered.

[0220] This embodiment is a detailed description of the combined implementation of the adaptive tuning module of the adaptive closed-loop tuning in the system of Embodiment 1; the purpose is to build a multi-level, multi-resolution tuning mechanism that can intelligently diagnose different types of system deviation, deviation, strategy deviation, and model deviation, and automatically execute the most efficient correction measures;

[0221] The adaptive tuning module integrates the key state indicators calculated in the previous steps: the health state level comes from the first processing module of Embodiment 2, the transient overshoot Policy enforcement module from example 5, node policy compliance Performance evaluation module and global performance bias from example 6 Performance evaluation module from example 7

[0222] The adaptive closed-loop tuning logic of this module is divided into three levels:

[0223] Node-level arbitration: this tuning is used to solve local execution conflict problems;

[0224] When the node policy compliance is lower than the preset compliance threshold , and the health status level from the first processing module is healthy; is a threshold representing the minimum acceptable execution fidelity of the system, for example, set to 0.9 based on historical running data statistics;

[0225] The system determines that this case is low but high is a local node control logic conflict; that is, the device itself is in good condition, and the policy cannot be executed due to the mismatch of the internal logic or parameters of the local controller such as the PLC, such as the PID parameters;

[0226] A local correction factor causing a decrease is issued to the measurement and control node; the local correction factor refers to a correction coefficient used to forcibly adjust the local control loop parameters of the node, such as the PID gain, to ensure that it can effectively track the cloud-end issued again; The value of the local correction factor is determined according to a preset tuning rule library, for example, if the error continues to be greater than , then a small step search is performed based on the error size in the range of [0.8, 1.2], or a fuzzy logic reasoning-based determination is applied;

[0227] When the node policy compliance is greater than or equal to the preset compliance threshold , it indicates that the execution is good, or the health status level is unhealthy, which means that the problem may be in the device itself rather than the control logic, and the issued local correction factor is not executed;

[0228] This tuning achieves accurate attack on local execution bias; it avoids the huge calculation overhead and system fluctuations caused by triggering global policy recalculation due to local problems, and restores the policy compliance of the node with the highest efficiency;

[0229] Policy-level recalculation: this tuning is used to solve the problem of mismatch between the policy and reality;

[0230] When any of the following occurs:

[0231] Node policy compliance Below a preset compliance threshold And the health status level is unhealthy, such as sub-health or failure warning;

[0232] Transient overshoot Exceeds a preset overshoot threshold ; Is a safety upper limit set according to the strict requirements of process stability of the can production process, such as retort, for example, 5%;

[0233] Condition 1 Low and Low is determined as: the original policy Is too aggressive for the current degraded equipment, which cannot physically execute the policy; Condition 2 High is determined as: the original policy Transient parameters in Come from the unreasonable settings in Example 4, resulting in process instability;

[0234] Trigger the policy generation module, i.e. Recalculate to generate a more conservative working condition policy vector ; For example, the new Reduces production load based on the current low Value, or uses a more gentle Value to reduce ;

[0235] When the node policy compliance Is greater than or equal to a preset compliance threshold Or the health status level is healthy, and the transient overshoot Is less than or equal to a preset overshoot threshold , It indicates that the current policy can be executed and is stable enough, and the trigger policy generation module does not recalculate;

[0236] This tuning realizes closed-loop control of policy risk; it ensures that the global policy can dynamically adapt to the actual decline of equipment health and the actual stability of transient processes, while pursuing benefits At the same time, it actively avoids the risk of equipment damage or production fluctuation caused by aggressive policies;

[0237] Model-level retraining: this tuning is used to solve the long-term problem of inaccurate optimization models;

[0238] When the global performance deviation I.e. Lasts, for example, continuously The number of control cycles exceeded the preset model drift threshold. At that time; among them It is the maximum tolerable systematic deviation between the model predictions and actual utility, determined through offline simulation and historical data analysis. It is a duration window set to prevent false triggering caused by transient disturbances;

[0239] The system determines this situation. Continuously optimize models for cloud computing Inaccuracy; that is, the model Mathematical descriptions of production processes no longer accurately reflect physical reality. For example, equipment aging causes changes in energy consumption models, affecting their predictions. With reality This leads to systematic bias;

[0240] Trigger cloud optimization model The retraining process; this process will use the recently acquired dataset. and the corresponding actual utility value To update the model Internal parameters;

[0241] When global performance deviation The model drift threshold was not exceeded continuously. When this occurs, it indicates that the model is still accurate and the retraining process that triggers cloud-based model optimization is not executed.

[0242] This tuning constitutes the highest-level and longest-period adaptive closed loop of the system, solving the drift problem of the core optimization model; it ensures that the optimization model of the intelligent measurement and control system of this invention remains stable during long-term operation. It can always keep in line with the actual working conditions of the physical production line, thus ensuring the long-term effectiveness and accuracy of the system's optimization decisions;

[0243] By integrating these three levels of tuning logic, the adaptive tuning module of this invention achieves a hierarchical, efficient closed-loop control with diagnostic intelligence; it can adjust according to... The system accurately diagnoses the root cause of system deviations based on the combined states—whether it's a level one problem with local actuators, a level two problem with global strategies, or a level three problem with the optimization model—and automatically invokes the corresponding, most cost-effective corrective measures. Correction, Recalculation Retraining; this hierarchical tuning mechanism ensures that the system can achieve robust, efficient, and precise adaptive adjustment when facing different types of disturbances, which is the core guarantee for the long-term, stable, and optimal operation of this invention.

[0244] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A cloud platform-based network intelligent measurement and control system for a can production line, characterized in that, The method comprises the following steps: a data acquisition module is used to collect process data sets, equipment state data sets, batch data sets and commercial data sets in real time; a first processing module is used to construct an equipment health index based on the equipment state data sets; the first processing module is further used to determine a health state level corresponding to the equipment health index according to a preset health index threshold; a second processing module is used to dynamically adjust a utility conversion coefficient based on the commercial data sets; the second processing module is further used to calculate an equipment degradation index based on the equipment health index; the second processing module is further used to determine a global objective function value by combining the utility conversion coefficient, the equipment degradation index and real-time production data; a strategy generation module is used to call a pre-trained cloud optimization model to solve a working condition strategy vector by fusing the process data sets, the batch data sets, the equipment health index and the utility conversion coefficient, with the goal of maximizing the global objective function value; a strategy execution module is used to issue the working condition strategy vector to a distributed measurement and control node for execution; the strategy execution module is further used to monitor the transient overshoot of the key process parameters; a performance evaluation module is used to calculate the node strategy compliance degree of the measurement and control node; the performance evaluation module is further used to calculate the global performance deviation; an adaptive tuning module is used to perform adaptive closed-loop tuning based on the combined state of the health state level, the transient overshoot, the node strategy compliance degree and the global performance deviation; the adaptive tuning module performing adaptive closed-loop tuning comprises: when the node strategy compliance degree is lower than a preset compliance threshold and the health state level is healthy, it is determined that there is a local node control logic conflict, and a local correction factor is issued to the measurement and control node; when the node strategy compliance degree is greater than or equal to the preset compliance threshold or the health state level is unhealthy, the local correction factor is not issued; the adaptive tuning module performing adaptive closed-loop tuning further comprises: when the node strategy compliance degree is lower than the preset compliance threshold and the health state level is unhealthy, or when the transient overshoot exceeds a preset overshoot threshold, the strategy generation module is triggered to recalculate to generate a more conservative working condition strategy vector; when the node strategy compliance degree is greater than or equal to the preset compliance threshold or the health state level is healthy, and the transient overshoot is less than or equal to the preset overshoot threshold, the strategy generation module is not triggered to recalculate; the adaptive tuning module performing adaptive closed-loop tuning further comprises: when the global performance deviation continuously exceeds a preset model drift threshold, it is determined that the cloud optimization model is inaccurate, and a retraining process of the cloud optimization model is triggered; when the global performance deviation does not continuously exceed the preset model drift threshold, the retraining process of the cloud optimization model is not triggered.

2. The cloud platform-based network intelligent measurement and control system for canned food production lines according to claim 1, characterized in that, The first processing module constructs the equipment health index, comprising: based on the real-time values in the equipment state data sets, a preset reference and a failure threshold, a normalized degradation factor is calculated; based on the pre-calibrated degradation weight and the normalized degradation factor, an equipment health index is determined by using a weighted subtraction model.

3. The cloud platform-based network intelligent measurement and control system for canned food production lines according to claim 1, characterized in that, The second processing module determines the global objective function value, comprising: in response to changes in the electricity price or order priority in the commercial data sets, the utility conversion coefficient is dynamically adjusted; Converting the equipment health index into an equipment degradation index; Using a linear utility function, the qualified product output, total energy consumption, total material consumption, and equipment degradation index are weighted to determine the global objective function value.

4. The cloud platform-based network intelligent measurement and control system for canned food production lines according to claim 1, characterized in that, The working condition strategy vector includes a sub-strategy package issued to the measurement and control node; The sub-strategy package includes target set values of key process parameters, parameter change slopes at mode switching, and coordination delay time of the execution strategy.

5. The cloud platform-based network intelligent measurement and control system for canned food production lines according to claim 1, characterized in that, The strategy execution module monitors the transient overshoot, including: The measurement and control node executes parameter adjustment according to the parameter change slope and coordination delay time in the sub-strategy package; Obtain the peak value, new target steady-state value, and old steady-state value during the parameter switching process; Based on the peak value, new target steady-state value, and old steady-state value, the transient overshoot is calculated using the percentage overshoot definition.

6. The cloud platform-based network intelligent measurement and control system for canned food production lines according to claim 1, characterized in that, The performance evaluation module calculates the node strategy compliance degree, including: Within a specified time, compare the local parameter actual value of the measurement and control node with the target set value; When the error between the local parameter actual value and the target set value is within the preset tolerance, the node compliance state is determined to be 1; When the error between the local parameter actual value and the target set value is greater than the preset tolerance, the node compliance state is determined to be 0; Using a statistical average model, the average value of the node compliance state of all measurement and control nodes is calculated to determine the node strategy compliance degree.

7. The cloud platform-based network intelligent measurement and control system for canned food production lines according to claim 1, characterized in that, The performance evaluation module calculates the global performance deviation, including: Obtain the predicted target value predicted by the strategy generation module; Using the formula of the global objective function, and substituting the real-time measured production data and equipment health index, the actual utility value is calculated; Based on the predicted target value and the actual utility value, the global performance deviation is determined using error signal calculation.

Images