Kjeldahl determination full-process automatic control and regulation system based on artificial intelligence

Abstract

The invention relates to a Kjeldahl determination full-process automatic control and regulation system based on artificial intelligence, and belongs to the technical field of automatic control and regulation systems. Comprising a multi-modal particle coding module which is used for disassembling a particle position vector into a continuous process parameter sub-vector, a discrete strategy sub-vector and a scheduling sub-vector, and defining the physical meaning and the value constraint of each sub-vector; the particle swarm optimization and updating module is used for constructing a dynamic constraint agent model; establishing a particle speed updating equation of strategy constraint; developing a process response adaptive inertia weight; constructing a multi-target elite archiving strategy; constructing migration optimization of matrix feature embedding; the closed-loop online optimization execution module is used for executing closed-loop iteration of parameter sampling, process execution, signal acquisition, model updating and parameter optimization; and a Kjeldahl determination full-process automatic control and regulation module. According to the method, the optimization precision can be improved, and full-process automation and self-adaptive adjustment of the Kjeldahl determination experiment are realized.

Description

Technical Field

[0001] This invention belongs to the field of automated control and regulation system technology, specifically relating to an artificial intelligence-based Kjeldahl nitrogen determination full-process automated control and regulation system. Background Technology

[0002] The Kjeldahl method, a classic method for determining protein content, is widely used in food testing, agriculture, animal husbandry, and feed quality control. This method determines nitrogen content through steps such as acid digestion, alkaline distillation, and acid titration. The accuracy of the analytical results directly affects the reliability of the detection conclusions.

[0003] In practical applications, the Kjeldahl nitrogen determination experiment faces several technical bottlenecks. First, the experimental procedure involves various operational parameters, such as acid digestion temperature, catalyst dosage, distillation conditions, and titration methods. These parameters not only differ in their continuous, discrete, and scheduled nature but also exhibit significant coupling relationships; slight inappropriate adjustments can lead to decreased experimental efficiency or even distorted results. Second, existing optimization methods largely rely on human experience or traditional single-objective optimization algorithms, often requiring manual parameter adjustments to optimize experimental conditions. This is not only time-consuming and labor-intensive but also struggles to achieve experimental schemes that balance recovery rate and repeatability in complex multi-objective scenarios. Furthermore, conventional constraint handling methods are outdated, only revealing the infeasibility of parameter combinations after the experiment is completed, resulting in numerous invalid experiments and significant resource waste. With the expansion of experimental scale and the increasing complexity of detection scenarios, the dynamic nature of the experimental process and the differences in sample matrices further exacerbate the optimization difficulty. To address the aforementioned issues, there is an urgent need for an optimization and control method that can simultaneously handle multiple types of parameters, possess dynamic constraint avoidance capabilities, and efficiently search and adaptively adjust under multi-objective conditions. This method would support the automation and intelligent development of the Kjeldahl nitrogen determination experiment, thereby achieving efficient, reliable, and transferable experimental optimization and scheduling control. Summary of the Invention

[0004] To achieve the above objectives, the present invention employs the following technical solution: This invention provides an artificial intelligence-based fully automated control and regulation system for Kjeldahl nitrogen determination, comprising: Multimodal particle coding module: Decomposes the particle position vector into three orthogonal subspaces: continuous process parameter subvector, discrete strategy subvector, and scheduling subvector, and defines the physical meaning and value constraints of each subvector; This invention constructs a hierarchical heterogeneous coding strategy, decomposing the particle position vector into three orthogonal subspaces: a continuous process parameter subvector, a discrete strategy subvector, and a scheduling subvector. This achieves dimensional decoupling, ensuring that each subspace independently represents different types of parameters, thereby preserving the physical semantics of the parameters and eliminating search failures caused by heterogeneous variable coupling. Furthermore, the particle position vector represents the set of optimization parameters for the Kjeldahl nitrogen determination process; the elements of the continuous process parameter subvector correspond to specific control variables, the elements of the discrete strategy subvector correspond to strategy selection, and the elements of the scheduling subvector correspond to scheduling variables.

[0005] Particle Swarm Optimization and Update Module: Constructs a dynamic constraint proxy model to avoid quality violations in real time; establishes a particle velocity update equation with policy constraints, defines a feasible solution set and policy rules; develops process response adaptive inertia weights; constructs a multi-objective elite archiving strategy to obtain an elite solution archiving set; incorporates matrix features into the optimization framework by constructing matrix feature embedding vectors and similarity calculation mechanisms to achieve optimization knowledge transfer between different matrix samples; Furthermore, this invention constructs a real-time constraint surrogate model, uses process signals to predict quality indicators such as recovery rate and repeatability relative standard deviation in real time, calculates constraint violation degree, and triggers a parameter adjustment mechanism when a violation is predicted, thereby avoiding quality violations online, including: Define the process signal vector: Real-time acquisition of multi-source sensor signals during the Kjeldahl nitrogen determination process, including steam temperature, condensation power, distillate flow rate change rate, and the second derivative of the receiving liquid potential, to construct the process signal vector; Establish a quality index prediction model: Using a lightweight gradient boosting tree model, a mapping relationship is established from the process signal vector to recovery rate and repeatability relative standard deviation. The prediction model is obtained through training with historical data, realizing real-time prediction of recovery rate and repeatability, and obtaining the predicted values ​​of recovery rate and repeatability relative standard deviation; Calculate the real-time constraint violation degree: Based on the predicted recovery rate and repeatability relative standard deviation, the constraint violation degree is calculated; Trigger real-time parameter adjustment: Monitor the constraint violation degree. When a constraint violation degree greater than zero is detected, the parameter adjustment mechanism is immediately triggered to guide the particle swarm away from the violation area, realizing online avoidance of quality constraints.

[0006] Furthermore, this invention proposes a velocity update equation with policy constraints, defines a feasible solution set and policy rules, and ensures that the particle update direction simultaneously satisfies mass constraints and policy compatibility, thereby avoiding parameter conflicts and improving optimization performance, including: Define the velocity update equation for policy constraints: Based on the velocity update of the standard particle swarm optimization algorithm, the velocity vector of the particle is updated by combining the feasible historical best position and the policy-compatible global best position as guiding terms; Construct feasible solution set screening conditions: Define the screening conditions for the feasible solution set, requiring the particle position vector to simultaneously satisfy the constraint violation degree of zero, the policy rule function being true, and the physical constraint function being true; Implement rule engine filtering: During the particle update process, infeasible solutions are filtered in real time through the rule engine; Update particle position vector: Update the particle position vector according to the updated velocity vector to obtain the particle position vector for the next iteration.

[0007] Furthermore, the sensitivity to parameters varies significantly at different stages of the nitrogen determination process. For example, it is sensitive to temperature parameters in the early stages of distillation, while it is sensitive to threshold parameters near the endpoint. Using a fixed particle swarm optimization with inertial weights cannot adapt to this dynamic change, easily leading to low algorithm convergence efficiency and an imbalance between global exploration and local exploitation. This invention dynamically calculates adaptive inertial weights based on the norm of the gradient vector of the real-time process signal. When the process response is drastic, the inertial weights are reduced to enhance local search; when the process response is stable, the inertial weights are increased to accelerate global exploration, thereby balancing global exploration and local exploitation capabilities.

[0008] Furthermore, this invention, by combining a dynamic elite archiving mechanism, uses non-dominated ranking and density estimation to select uniformly distributed Pareto optimal solutions from the feasible solution set, thereby providing optimization options for multi-objective decision-making. A multi-objective elite archiving strategy is constructed to obtain an elite solution archiving set, including: Calculate the multi-objective vector of each particle: For each feasible particle, calculate its multi-objective vector, including the completion time of a single Kjeldahl nitrogen determination process, the energy consumption of a single nitrogen determination process, and the degree of human intervention; Perform non-dominated sorting: Perform Pareto non-dominated sorting on all particles in the feasible solution set to identify non-dominated solution sets of different levels; Elite solution screening and archiving: Based on the non-dominated sorting results and the spatial distribution density of the solutions, select elite solutions with sparsity greater than the density threshold from the first-level non-dominated solutions for archiving, and obtain the elite solution archive set.

[0009] Furthermore, this invention incorporates matrix features into the optimization framework by constructing matrix feature embedding vectors and a similarity calculation mechanism, thereby enabling the transfer of optimization knowledge between different matrix samples and significantly accelerating the optimization process for new samples, including: Constructing matrix feature embedding vectors: Define matrix feature embedding vectors, including percentage fat content, percentage sugar content, and log viscosity; Extending particle encoding structure: Embed the matrix feature embedding vectors as an additional dimension into the particle position vectors to form an extended particle representation; Calculating matrix similarity weights: Calculate the matrix similarity weights between samples based on the Euclidean distance between matrix feature embedding vectors; Incorporating elite guidance from similar samples: During particle velocity update, incorporate the historical optimal parameters of similar matrix samples as an additional guiding term, and sum them using similarity weights.

[0010] Closed-loop online optimization execution module: performs closed-loop iteration of execution parameter sampling, process execution, signal acquisition, model update and parameter optimization, combined with multi-station scheduling to optimize batch processing; Furthermore, this invention achieves end-to-end online adaptive optimization by constructing a real-time optimization-execution closed-loop circuit, performing continuous iterative parameter sampling, process execution, signal acquisition, model updating, and parameter optimization, including: S31. Parameter sampling and nitrogen determination process execution: Select a particle from the particle swarm optimization algorithm and send its corresponding parameters to the Kjeldahl nitrogen determination equipment to execute the nitrogen determination process. S32. Signal Acquisition and Objective Function Calculation: Acquire process signals in real time and calculate the objective function value. The objective function value is the weighted sum of the completion time of a single Kjeldahl nitrogen determination process, the energy consumption of a single nitrogen determination process, and the degree of human intervention. S33. Particle swarm parameter update mechanism: The particle swarm is updated using the objective function value, which is divided into individual update and swarm update. The individual update is the individual historical best update: if the particle position vector satisfies the feasibility condition, then the feasible historical best position of the particle is updated. The group update is a global optimal and elite solution archive set update: The first step is to perform non-dominated solution screening by adding the particle position vectors to the feasible solution set and performing non-dominated sorting. The second step is to update the elite archive. If the Pareto level of the particle position vector is the first level and the sparsity of the particle position vector in the target space is greater than the density threshold, then the particle position vector is added to the elite solution archive set, and the solutions dominated by the particle position vector in the elite solution archive set are removed. The third step is to perform a global optimal selection by polling the elite solution archive set to select the globally optimal position that is compatible with the strategy.

[0011] S34. Multi-station scheduling optimization: Use elite dearchived archive set to drive multi-station scheduling, optimize batch processing, and the scheduling objective is to minimize the maximum single Kjeldahl nitrogen determination process completion time of samples in the batch. The constraint condition is that the difference in matrix characteristics of samples in the batch is less than a threshold. S35. Online Model Update: After each nitrogen determination process is executed, the system collects the process signal vector and actual quality index data for each nitrogen determination process. The process signal vector includes steam temperature, condensation power, distillate flow rate change rate, and the second derivative of the receiving liquid potential. The actual quality index data includes recovery rate and repeatability relative standard deviation. The system uses the newly collected data to fully retrain the lightweight gradient boosting tree model.

[0012] The Kjeldahl nitrogen determination process automation control and regulation module selects matching parameters from the elite solution archive set and sends them to the Kjeldahl nitrogen determination equipment. It monitors process signals in real time and adjusts adaptively to achieve full process automation.

[0013] The advantages of this invention are: This invention decomposes continuous, discrete, and scheduling parameters into orthogonal subspaces to avoid search failures caused by coupling, preserves physical semantics, and improves optimization accuracy. It utilizes process signals to predict recovery rate and repeatability, calculates constraint violation in real time, and triggers parameter adjustments, replacing the conventional delay penalty function and significantly reducing invalid searches. It constructs a Pareto front elite solution set to ensure uniform solution distribution and accelerates optimization of new samples by transferring optimization knowledge through matrix feature embedding and similarity calculation. A closed-loop mechanism of "parameter sampling-execution-signal acquisition-model update-re-optimization" is established, combined with multi-station scheduling and elite solution distribution, to achieve full-process automation and adaptive adjustment of the Kjeldahl nitrogen determination experiment. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0015] Figure 1 is a flow chart of the steps of the present invention; Figure 2 This is a comparison of the process completion time optimization trajectories of the present invention and existing technologies; Figure 3 A comparison of the energy consumption optimization trajectories of this invention and existing technologies; Figure 4 This invention compares the degree of human intervention in optimizing trajectories with existing technologies; Figure 5 This is a typical PSO search path diagram; Figure 6 This is a search path diagram for the method of the present invention. Detailed Implementation

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments derived by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0017] Example 1 In this embodiment, as Figure 1 As shown, this invention provides an automated control and regulation system for the entire Kjeldahl nitrogen determination process based on artificial intelligence, the specific steps of which include: S1. Construct a multimodal particle coding structure The parameters of the Kjeldahl nitrogen determination process include continuous execution variables, discrete strategy selection variables, and hybrid scheduling variables. These parameters have heterogeneous coupling relationships, and conventional real-number encoding methods cannot effectively represent this heterogeneous coupling. Conventional particle swarm optimization algorithms use unified real-number encoding, which will destroy the physical meaning of discrete parameters, such as the endpoint determination mode, thereby distorting the search space and reducing optimization efficiency.

[0018] This invention constructs a hierarchical heterogeneous coding strategy, decomposing the particle position vector into three orthogonal subspaces: a continuous process parameter subvector, a discrete strategy subvector, and a scheduling subvector. This achieves dimensional decoupling, ensuring that each subspace independently represents different types of parameters, thereby preserving the physical semantics of the parameters and eliminating search failures caused by heterogeneous variable coupling. The specific steps are as follows: 1) Define the hierarchical structure of particle position vectors The particle position vector is decomposed into three orthogonal subspaces: a continuous process parameter subspace, a discrete strategy subspace, and a scheduling subspace. This achieves dimensional decoupling, ensuring that each subspace independently represents different types of parameters and avoiding search failures caused by coupling. Specifically... Define the particle position vector as The set of optimal parameters characterizing the Kjeldahl nitrogen determination process, i.e. ,in, This represents a subvector of continuous process parameters, belonging to a 10-dimensional real number space, corresponding to steam, condensate, and liquid circuit control quantities; This represents a discrete policy subvector, which belongs to a 4-dimensional integer space and corresponds to the policy selection variable; This represents a scheduling subvector, consisting of three real variables and one integer variable. The three real variables belong to a 3-dimensional real space, and the one integer variable belongs to an integer space, corresponding to the scheduling variable; (symbol) This represents a vector concatenation operation, which joins three sub-vectors into a single complete vector.

[0019] 2) Specify the physical meaning and constraints of each sub-vector. Define the physical meaning and value constraints of the parameters for each sub-vector to ensure that the encoding preserves the physical semantics and meets the practical requirements of the Kjeldahl nitrogen determination process. Specifically, Subvectors of parameters for continuous processes The elements correspond to specific control quantities, discrete strategy sub-vectors. The element corresponds to the strategy selection, and the scheduling sub-vector. The elements correspond to scheduling variables.

[0020] Furthermore, the element index of each subvector is defined as... , It is a positive integer, that is Represents the parameter subvector of a continuous process The Similarly, for each element... Represents discrete strategy subvectors The One element, Represents discrete strategy sub-vectors The Given an element, the specific constraints for each subvector are as follows: a) Subvectors of parameters for continuous processes The 10 elements correspond to the following control variables, with specific constraints as follows: This is a steam temperature control quantity, in °C, with constraints. ,like ; This is a percentage control quantity for steam flow, dimensionless, with constraints. ; This is the condensate flow rate control quantity, in L / min, with the following constraints: ; This is the distillation time control quantity, in minutes, with the following constraints: ; This is the condensing power control quantity, in watts (W), with the following constraints: ; This is the flow rate control quantity for the liquid circuit, in mL / min, with the following constraints: ; The receiving liquid flow rate is controlled in mL / min, and the constraint is... ; The temperature ramp rate is expressed in °C / min, with the following constraints: ; This is the pressure control quantity, in kPa, with the following constraint: ; This is a pH control quantity, dimensionless, with constraints. .

[0021] b) Discrete strategy subvectors The four elements correspond to the following policy variables, with specific constraints as follows: This is the endpoint determination mode, and its value is a discrete set. , where 0 corresponds to conventional titration, 1 corresponds to potentiometric titration, and 2 corresponds to sensitive potentiometric titration; The discretization threshold parameter takes the value of a set of integers. These correspond to different thresholds, such as 0 corresponding to 0.1mV, 1 corresponding to 0.2mV, 2 corresponding to 0.5mV, and 3 corresponding to 1.0mV; For condensation mode, the value is a discrete set. Where 0 corresponds to standard condensation and 1 corresponds to strong condensation; This is a distillation mode, and its value is a discrete set. , where 0 corresponds to standard distillation, 1 corresponds to rapid distillation, and 2 corresponds to energy-saving distillation.

[0022] c) Scheduling subvector The four elements correspond to the following scheduling variables, with specific constraints as follows: The time allocation for the distillation stage is in minutes, with the following constraints: ; Cooling time, in minutes, with the following constraints: ; The sample interval time is in minutes, and the constraint is... ; The sample ranking strategy takes values ​​from a discrete set. , where 0 corresponds to sequential processing, 1 corresponds to matrix processing, and 2 corresponds to priority processing.

[0023] S2, Particle Swarm Optimization and Update S201. Constructing a dynamic constraint proxy model Quality constraints for Kjeldahl nitrogen determination, such as recovery rate and relative standard deviation of repeatability, can only be verified after the experiment is completed. However, the optimization process needs to avoid violation areas online. Conventional penalty function methods lead to a large number of invalid searches due to the delay in constraint verification, thereby reducing optimization efficiency.

[0024] This invention constructs a real-time constraint surrogate model, uses process signals to predict quality indicators such as recovery rate and repeatability relative standard deviation in real time, calculates constraint violation degree, and triggers a parameter adjustment mechanism when a violation is predicted, thereby avoiding quality violations online. The specific steps are as follows: 1) Define the process signal vector Multi-source sensor signals, including steam temperature, condensation power, distillate flow rate change rate, and the second derivative of the received liquid potential, are acquired in real time during the Kjeldahl nitrogen determination process. A process signal vector is constructed and used as the input feature of the surrogate model to reflect the process state in real time, expressed as: , In the formula, The process signal vector at time t is used to characterize the state of the Kjeldahl nitrogen determination process in real time. This indicates the steam temperature, expressed in degrees Celsius, and reflects the heating intensity. This indicates condensing power, measured in watts, and reflects the operating status of the condensing system. This indicates the rate of change in distillate flow rate, expressed in milliliters per minute, reflecting the distillation rate. It represents the second derivative of the receiving liquid potential, with units of millivolts per square second. It reflects the acceleration of the potential change near the endpoint and is used to determine the endpoint of the reaction.

[0025] 2) Establish a quality indicator prediction model A lightweight gradient boosting tree model is used to establish a mapping relationship from the process signal vector to the relative standard deviation of recovery rate and repeatability. The prediction model is trained using historical data to achieve real-time prediction of recovery rate and repeatability, yielding the following predicted values: a) Define the predicted recovery rate as... The range of values ​​is The acquisition method is represented as , This represents the recovery rate prediction model, which is a preset first lightweight gradient boosting tree model. b) Define the predicted value of the repeatability relative standard deviation as The unit is percentage, and the method of acquisition is expressed as follows: , This represents a repetitive prediction model, which is a pre-defined second lightweight gradient boosting tree model.

[0026] In one implementation, when the preset first lightweight gradient boosting tree model performs recovery rate prediction, the input is a process signal vector. And the corresponding recovery rate, through the process signal vector After normalization preprocessing, the gradient boosting tree algorithm is used to iteratively construct decision trees, minimizing the mean squared error. Each decision tree is fitted with residuals, and the final model is a set of decision trees that outputs the predicted recovery rate. The number of decision trees is preset by the user, for example, 100 decision trees can be set. Similarly, when the pre-defined second lightweight gradient boosting tree model performs repetitive predictions, the model is constructed in the same way as the first lightweight gradient boosting tree model, with the input being the process signal vector. Given the corresponding repeatability relative standard deviation, output the predicted value of the repeatability relative standard deviation. .

[0027] It should be noted that the relative standard deviation used to measure repeatability is calculated as the ratio of the standard deviation to the mean of the elements in the vector. In conventional calculations, the relative standard deviation is obtained from the results of multiple repeated trials, i.e., the ratio of the standard deviation to the mean of multiple measurements, used to measure repeatability. However, in this invention, the repeatability prediction model is derived from the process signal vector. The predictive model forecasts the repeatability relative standard deviation, thus eliminating the need for real-time repeated experiments. Instead, it uses a predictive model trained on historical data to output the predicted value of the repeatability relative standard deviation. .

[0028] 3) Calculate the real-time constraint violation degree Based on the predicted recovery rate and the relative standard deviation of repeatability, the constraint violation degree is calculated. The constraint violation degree is positive when the predicted recovery rate is below the lower constraint limit or the predicted repeatability is above the upper constraint limit; otherwise, it is zero. This characterizes the degree of quality violation and is expressed as: , In the formula, This represents the degree of constraint violation at time t. It is a dimensionless quantity, and the larger the value, the more severe the violation. This represents the maximum value function, which selects the largest value from the parameters.

[0029] It should be noted that 99.5% is the lower limit of the recovery rate constraint, and 0.5% is the upper limit of the repeatability constraint. This indicates a violation of the recovery rate constraint, when the predicted value... A positive value is generated when the percentage is below 99.5%. This indicates a component that violates the repeatability constraint, when the predicted value... A positive value is generated when it is above 0.5%.

[0030] It should also be noted that setting quality constraints... The constraint violation degree is required to be 0, only if the predicted recovery rate is zero. And the predicted value of the relative standard deviation of repeatability hour, This indicates that the quality constraint is satisfied.

[0031] 4) Trigger real-time parameter adjustment Monitoring constraint violation rate When detected Upon activation, a parameter adjustment mechanism is immediately triggered to guide the particle swarm away from potentially violating regions, achieving online avoidance of mass constraints. The parameter adjustment mechanism is as follows: a) when At that time, the system triggers an adjustment and calculates the constraint violation degree. For particle position vector gradient Adjust the parameters along the negative gradient direction, i.e. ; in, This indicates an assignment operation; Adjust the learning rate in the direction of the negative gradient, for example, preferably set to 0.01; To constrain the degree of violation For particle position vector The gradient of is calculated using the chain rule, specifically through automatic differentiation or numerical difference, and is expressed as . .

[0032] b) Then adjust and recalculate the constraint violation degree. ,if Then continue adjusting the parameters along the negative gradient direction until... .

[0033] S202. Establish a policy-constrained particle update mechanism. The globally optimal particle in a conventional particle swarm optimization algorithm may violate mass constraints, leading to population degradation. Conventional constraint handling methods ignore the policy mutual exclusion between parameters. For example, high steam flow requires matching a strong condensation mode, which can cause conflicts in parameter combinations during actual execution, thereby reducing the optimization effect.

[0034] This invention proposes a velocity update equation with policy constraints, defines a feasible solution set and policy rules, and ensures that the particle update direction simultaneously satisfies mass constraints and policy compatibility, thereby avoiding parameter conflicts and improving optimization performance. The specific steps are as follows: 1) Define the velocity update equation with policy constraints Based on the velocity update of the standard particle swarm optimization algorithm, this algorithm combines the feasible historical optimal position and the policy-compatible global optimal position as guiding terms to ensure that the particle update direction satisfies the mass constraint and policy compatibility, expressed as: , In the formula, Indicates that the i-th particle is in The velocity vector at each instant is used to update the particle position; express The adaptive inertia weight at each moment is used to control the inertial effect of particle velocity updates in the particle swarm optimization algorithm, balancing global exploration and local exploitation capabilities. The initial value of the adaptive inertia weight is... Set to 0.1; Indicates that the i-th particle is in The velocity vector at any given moment; This represents the individual learning factor, which modulates the influence of the individual's historical best performance; the optimal setting is 2.0. express Random numbers within a range increase the randomness of the search; Let the feasible historical optimal position of the i-th particle satisfy the following condition: Quality constraints; Indicates that the i-th particle is in The position vector at time, and the particle position vector The content is the same, including The vector formed; This represents the social learning factor, which modulates the optimal group influence; the preferred setting is 2.5. express A random number within a given range; For time indexing, in the particle swarm optimization algorithm, time represents the number of iterations in the particle swarm optimization algorithm; The globally optimal position that is policy-compatible is represented from the set of feasible solutions. Choose from the options, and set them randomly during the first iteration.

[0035] It should be noted that the quality constraints means The position is the optimal position encountered by the i-th particle in the historical iterations, and the corresponding constraint violation degree is... That is, the predicted recovery rate. And the predicted value of the relative standard deviation of repeatability .

[0036] 2) Constructing feasible solution set selection criteria Define the selection criteria for the feasible solution set, requiring the particle position vector to simultaneously satisfy the following conditions: zero constraint violation, true policy rule function, and true physical constraint function. This ensures that the selected solution is feasible in actual execution, expressed as: , In the formula, This represents the set of feasible solutions, containing the positions of all particles that satisfy the constraints. This represents a policy rule function that returns a boolean value. This represents a physical constraint function that returns a Boolean value. This represents a logical AND operation, requiring all conditions to be met simultaneously. Indicates "such that" or "satisfies", used to define conditions in a set. The term represents the position vector of all particles that satisfy the constraints. A set of.

[0037] In practical implementation, The term represents the discrete strategy subvector. The condition is true when the policy rule is satisfied. Based on domain knowledge of Kjeldahl nitrogen determination, the compatibility of parameter combinations is ensured, and parameter combination conflicts are avoided. For example, when... At this time, it is in fast mode, which requires , indicating strong condensation, when At that time, for sensitive endpoint determination, it is required that This indicates a medium steam flow rate; it returns a Boolean value by checking whether these conditions are met. Similarly, The term represents the subvector of parameters in a continuous process. and discrete strategy subvectors The condition is true when physical constraints are met. By defining physical constraints, we ensure that parameter values ​​are within the allowable range of the device and that the parameters satisfy physical relationships. For example, we set... This indicates that the upper limit of steam temperature varies with the mode. This indicates that the minimum fluid flow rate is related to the threshold parameter. The system checks whether these inequalities are true and returns a Boolean value.

[0038] 3) Implement rule engine filtering During particle update, infeasible solutions are filtered in real time by a rule engine to ensure... and All are from the feasible solution set This ensures that the search direction simultaneously meets both quality and strategy constraints. The rules engine calculates the constraint violation rate in real time. Strategy rule function and physical constraint functions The candidate solutions are then filtered.

[0039] In practice, after each particle update, the constraint violation degree is calculated for each particle. Strategy rule function and physical constraint functions If a particle does not meet the conditions, the feasible historical best position is not updated. Or from the feasible solution set Exclusion criteria are checked in real time by the rule engine to ensure... and From the set of feasible solutions .

[0040] 4) Particle position vector update The particle position vector is updated based on the updated velocity vector to obtain the particle position vector for the next iteration, expressed as: , In the formula, Indicates that the i-th particle is in Position vector at any given time; Indicates that the i-th particle is in Position vector at any given time; Indicates that the i-th particle is in The velocity vector at time t.

[0041] Furthermore, the updated position vector is subjected to boundary constraints and discrete value set constraints to ensure that the position vector satisfies the basic parameter constraints. Specifically, For continuous control parameters in the continuous process parameter subvector, their values ​​are forcibly limited to a preset physical range. If the calculated value is lower than the minimum value allowed by the equipment, it will be automatically raised to the lower limit; if it exceeds the maximum value, it will be lowered to the upper limit. For discrete policy parameters in discrete policy subvectors, the calculated value is first rounded to the nearest integer, and then the integer value is verified to see if it belongs to the predefined set of legal options. If it does not belong, it is randomly replaced with one of the allowed option values ​​for the parameter. The scheduling parameters in the scheduling subvector are handled differently: continuous time parameters are subject to boundary constraints, while discrete sorting strategy parameters are subject to option matching.

[0042] S203, Development Process Response Adaptive Weights The sensitivity of parameters varies significantly at different stages of the nitrogen determination process. For example, it is sensitive to temperature parameters in the early stage of distillation, while it is sensitive to threshold parameters near the endpoint. Using fixed particle swarm optimization with inertial weights cannot adapt to such dynamic changes, which can easily lead to low algorithm convergence efficiency and an imbalance between global exploration and local development.

[0043] Based on the norm of the gradient vector of the real-time process signal, an adaptive inertia weight is dynamically calculated. When the process response is drastic, the inertia weight is reduced to enhance local search; when the process response is stable, the inertia weight is increased to accelerate global exploration, thereby balancing global exploration and local development capabilities. This is expressed as: , Where, express The adaptive inertia weight at any time is used to control the inertial effect of particle velocity updates in the particle swarm optimization algorithm, balancing global exploration and local exploitation capabilities. The minimum value of the inertia weight is set to ensure the minimum exploration capability of the algorithm; a value of 0.4 is preferred. The maximum value of the inertia weight is set to ensure the maximum exploration capability of the algorithm; a value of 0.9 is preferred. The attenuation coefficient controls the sensitivity of the inertia weight to changes in gradient; it is preferably set to 0.5. Represents the natural exponential function; This represents the L2 norm, which is equivalent to the Euclidean norm and is used to calculate the magnitude of the gradient vector. The gradient vector of the key indicator with respect to the parameter represents the severity of the process response. It is calculated through online sensitivity analysis, and the calculation method is expressed as follows: ; The completion time of the Kjeldahl nitrogen determination process is measured during execution. The recovery rate was calculated from the experimental results. For steam temperature, the value is related to the steam temperature control quantity. same; The endpoint determination parameter has the same value as the endpoint determination mode. same; The symbol represents the partial derivative.

[0044] It should be noted that when the process response is violent, The term value decreases, and the adaptive inertia weights are adjusted. Reduce and enhance local search; when the process is stable, Increase the term, adaptive inertia weight The addition of global exploration enhances the particle swarm optimization algorithm, improving local search and preventing oscillations when the process response is drastic, and enhancing global exploration and accelerating convergence when the process is stable, thereby improving optimization efficiency and stability.

[0045] S204, Constructing a Multi-Objective Elite Archiving Strategy The Kjeldahl nitrogen determination process requires the simultaneous optimization of multiple conflicting objectives such as time, energy consumption, and human intervention. The conventional weighted summation method requires pre-setting the weights of each objective, which is highly subjective and cannot obtain the Pareto front solution set that represents the optimal trade-off relationship. This can easily lead to the optimization results being biased towards a single objective and cannot meet the needs of multi-objective balanced optimization.

[0046] This invention combines a dynamic elite archiving mechanism with non-dominated ranking and density estimation to select uniformly distributed Pareto optimal solutions from the feasible solution set, thereby providing optimization options for multi-objective decision-making. The specific steps are as follows: 1) Calculate the multi-target vector of particles For each feasible particle, its multi-objective vector is calculated, including the completion time of a single Kjeldahl nitrogen determination process, the energy consumption of a single nitrogen determination process, and the degree of human intervention. This vector is used to comprehensively evaluate the performance of the parameter settings corresponding to that particle, and is expressed as follows: , In the formula, This represents the multi-objective vector of the i-th particle, used to comprehensively evaluate the performance of the parameter settings corresponding to that particle. The time taken to complete a single Kjeldahl nitrogen determination process using the parameters of the i-th particle is expressed in minutes and is obtained through actual execution or simulation; that is, the time taken from start to finish. This represents the energy consumption of a single nitrogen determination process when using the parameters of the i-th particle, in kilowatt-hours. It is calculated by integrating the power of the complete process, which is obtained by integrating the real-time power of the complete process. The real-time power comes from the measurement value of the power sensor. This represents the degree of human intervention required when using the parameters of the i-th particle. It is a dimensionless index and is quantified by the number of interventions and the intervention time. For example, the degree of human intervention is defined as the sum of the number of interventions and the total intervention time.

[0047] 2) Perform non-dominated sorting For feasible solution set All particles are sorted using the Pareto non-dominated algorithm to identify non-dominated solution sets of different ranks. The set of particles with rank=1 is the current Pareto front. The specific process is as follows: The input is the set of feasible solutions. The multi-objective vectors of all particles in the dataset, taking the i-th example as an example, are the input multi-objective vectors of the i-th particle. ; For the set of feasible solutions For each particle in the array, calculate its dominance relationship. Taking the i-th example, define the position vector of the i-th particle as... The position vector of the j-th particle is ,but: like All target values ​​are not inferior to If at least one objective value is better, then it is considered... Dominate ; Particles that are not dominated by any other particles are classified as the first rank, denoted as rank=1, which is the Pareto optimal solution set; Then, from the remaining particles, the undominated particles are selected as the second rank, denoted as rank=2, and so on, to complete the rank division of all particles. The output is the Pareto level assigned to each particle. ,definition express The Pareto rank, which is the rank value in a non-dominant order.

[0048] It should be noted that in the particle swarm optimization algorithm, Pareto optimality means that at least one objective cannot be improved without making any objective worse. Rank represents the non-dominated level, with rank=1 being the Pareto optimal solution set, rank=2 being the suboptimal solution set, and so on.

[0049] It should also be noted that when determining the dominance relationship between particles, "not inferior to" and "better" mean that for all targets, the value is not worse, that is, less than or equal to. For example, for the position vector of the i-th particle... and the position vector of the j-th particle After calculating the multi-objective vectors of both, if , ,and And it's better to have at least one objective that is strictly enforced; Let be the completion time of a single Kjeldahl nitrogen determination process when using the parameters of the j-th particle. The energy consumption of a single nitrogen determination process when using the parameters of the j-th particle is given. The degree of human intervention required when using the parameters of the j-th particle.

[0050] 3) Elite solution screening and archiving Based on the non-dominated ranking results and the spatial distribution density of the solutions, elite solutions with sparsity greater than the density threshold are selected from the first-level non-dominated solutions and archived to obtain an elite solution archive set. This ensures the uniformity of the Pareto front distribution, denoted as: , In the formula, express An archive of elite solutions at each time step, used to store the current best and evenly distributed Pareto solution; Represents the particle position vector; Represents the particle position vector Pareto grades, such as, This indicates that the particle is a first-order non-dominated solution; Represents the particle position vector Sparsity in the target space measures the distribution density of other solutions around the given solution, and is calculated as follows: ; Particle position vector Multi-objective vector; Dearchive the current elite collection Middle A multi-objective vector of solutions; Dearchive the current elite collection The particle index in the current elite solution archive set. Index of the solution; This represents the current elite solution archive set. Indicates the size of the current elite solution archive set; Density threshold Indicates when sparsity Greater than the density threshold When the solution is considered to be in a sparse region, it should be retained.

[0051] It should be noted that the globally optimal position The particles are archived from the elite solution set. The middle polling selection, that is, sequentially starting from the elite solution archive set. Choose the solution as the global optimal position This ensures that different target directions are explored, avoiding bias towards a single target.

[0052] It should also be noted that, Calculate the particle position vector With elite dearchived collection The reciprocal of the sum of distances to all solutions; the larger the sum of distances, the greater the value. The fewer the surrounding solutions, the higher the sparsity, which should be preserved to maintain the diversity of the Pareto front.

[0053] S205, Construction of matrix feature embedding transfer optimization Different sample matrices, such as high-fat or high-sugar samples, require different process parameter settings. However, it is costly to re-optimize and search for each new sample. Conventional optimization methods ignore the parameter migration patterns between different matrix samples and cannot utilize historical optimization experience, resulting in the need to start the search from scratch each time, leading to low optimization efficiency.

[0054] This invention incorporates matrix features into the optimization framework by constructing matrix feature embedding vectors and a similarity calculation mechanism, thereby enabling the transfer of optimization knowledge between different matrix samples and significantly accelerating the optimization process for new samples. The specific steps are as follows: 1) Constructing matrix feature embedding vectors Define a matrix feature embedding vector, including percentage fat content, percentage sugar content, and logarithm of viscosity, to quantitatively describe the matrix properties of the sample, expressed as: , In the formula, This represents the matrix feature embedding vector, used to quantitatively describe the matrix properties of the sample; This represents the percentage of fat content in a sample, with a value between 0 and 100. It is preset by the user or obtained through chemical analysis during sample pretreatment and is an external input quantity. This represents the percentage of sugar content in a sample, with a value between 0 and 100. It is preset by humans or obtained through chemical analysis and is an external input quantity. The value representing the logarithm of the sample viscosity, where The viscosity value of the sample is obtained by measuring with a viscometer, and the unit is mPa·s. Taking the logarithm makes the viscosity value distribution more uniform and facilitates similarity calculation. It is a logarithmic function, with the default base being the natural constant.

[0055] 2) Extended particle coding structure The matrix feature embedding vector is incorporated as an additional dimension into the particle position vector to form an extended particle representation. This allows the particles to simultaneously encode process parameter information and sample matrix information. The extended particle position vector is defined as follows: It is equivalent to ,symbol This represents the vector concatenation operation, which joins two subvectors into a single complete vector.

[0056] 3) Calculate matrix similarity weights Based on the Euclidean distance between matrix feature embedding vectors, the matrix similarity weight between samples is calculated. The larger the weight value, the more similar the matrix characteristics of the two samples are, as expressed as: , Where, Indicates the first The first sample and the second The matrix similarity weight between samples is in the range of (0,1]. The larger the weight, the more similar the matrix characteristics of the two samples are. This is the similarity decay coefficient, which controls the rate at which the weight decays with increasing distance; it is preferably set to 0.1. Indicates the first The matrix feature embedding vector of each sample contains the same content as the matrix feature embedding vector m, including... The vector formed; Indicates the first Matrix feature embedding vectors of each sample; This serves as a sample index, used to identify samples that are being optimized. To distinguish from The sample index is used to identify existing samples in the database.

[0057] 4) Integrating similar sample elite guidance During the particle velocity update process, the historical optimal parameters of matrix-similar samples are used as an additional guiding term. Through weighted summation based on similarity weights, the optimization process for new samples draws upon historical optimization experience, improving the convergence speed. This is expressed as: , Where, Indicates that the i-th particle is in The velocity vector at any given moment; This indicates an assignment operation; The social learning factor controls the strength of the similarity sample guidance term, and is preferably set to 1.0; To increase the randomness of the search, use random numbers in the range [0,1]. Indicates the relationship with the first A set of samples with similar matrix, specifically selecting matrix similarity weights. For samples exceeding a preset threshold, for example, if the threshold is set to 0.8, if the first... The first sample and the second The matrix characteristics of the samples are similar. Then the first One sample was included in the sample set. In, and, with the Each sample represents the sample currently being optimized during this velocity vector update process; Indicates the first The first sample and the second Matrix similarity weights between samples; Indicates the first The historical best parameter vector for a sample, i.e. the parameter settings that performed best on that sample, is specifically selected by Pareto sorting to select the best parameter vector.

[0058] S3. Implement closed-loop online optimization execution. Conventional offline optimization methods cannot respond to equipment state drift, resulting in a mismatch between optimized parameters and the actual process, which affects the accuracy and efficiency of the Kjeldahl nitrogen determination process.

[0059] This invention achieves end-to-end online adaptive optimization by constructing a real-time optimization-execution closed-loop circuit, performing continuous iterative parameter sampling, process execution, signal acquisition, model updating, and parameter optimization. The specific steps are as follows: 1) Parameter sampling and nitrogen determination process execution Select a particle from the particle swarm optimization algorithm, and send its corresponding parameters to the Kjeldahl nitrogen determination device for execution. Assume that the position vector is sampled from the particle swarm. And use this parameter to control the Kjeldahl nitrogen analyzer to perform the nitrogen determination process, defining Indicates that the particle is in The position vector at time , This indicates the number of closed-loop iterations, i.e., the index of the optimization-execution loop.

[0060] 2) Signal acquisition and objective function calculation The process signals are acquired in real time, and the objective function value is calculated. The objective function value is a weighted sum of the completion time of a single Kjeldahl nitrogen determination process, the energy consumption of a single nitrogen determination process, and the degree of human intervention. It is used to comprehensively evaluate the performance of the nitrogen determination process and guide particle swarm updates, and is expressed as: , In the formula, This represents the objective function value of the k-th iteration, used to comprehensively evaluate the performance of the nitrogen determination process; The weighting coefficient for the time objective is preferably set to 0.5; The weighting coefficient for the energy consumption target is preferably set to 0.3; The weighting coefficient for the target of human intervention is preferably set to 0.2; The time taken to complete a single Kjeldahl nitrogen determination process is measured in minutes and is obtained through actual execution or simulation; that is, the time taken from start to finish. The energy consumption for a single nitrogen determination process is expressed in kilowatt-hours. It is calculated using the integrated power of the complete process. The integrated power is calculated by integrating the real-time power of the complete process. The real-time power is obtained from the measurement value of the power sensor. The degree of human intervention is a dimensionless indicator that is quantified by factors such as the number of interventions and the duration of intervention. For example, the degree of human intervention can be defined by summing the number of interventions and the total intervention time.

[0061] It should be noted that, A fitness function characterizing scalarized multi-objectives is used for particle swarm optimization. Representing multi-objective vectors for Pareto sorting. Guide the search direction Maintain the diversity of solution sets.

[0062] 3) Particle swarm parameter update mechanism Using the objective function value Updating the particle swarm involves two steps: individual update and swarm update, as detailed below: a) Individual historical best update If the particle is Position vector at time Feasibility conditions are met: ,and This means that updates are made only if the new solution is feasible and better than the individual's historical best. .

[0063] b) Update of the archive set of global optimal and elite solutions: First, perform a screening of non-dominated solutions, and... Add to feasible solution set Perform non-dominated sorting; Then, perform elite archiving updates, if the particle position vector Pareto level And the particle position vector sparsity in the target space Then add it to the elite solution archive set. And remove the elite decryption archive. China The solution of domination; Finally, a globally optimal selection is performed, starting from the elite decomposition archive set. The polling strategy is compatible with the globally optimal position. This ensures balanced exploration across multiple objectives.

[0064] 4) Multi-station scheduling optimization Elite set-driven multi-station scheduling is used to optimize batch processing. The scheduling objective is to minimize the maximum single Kjeldahl nitrogen determination time for samples in a batch, denoted as: Meanwhile, to ensure that the difference in matrix characteristics among samples within a batch is less than a threshold, the constraint condition is as follows: ; in, This represents the maximum single Kjeldahl nitrogen determination process completion time for a sample in a batch that minimizes the total time. The time taken to complete a single Kjeldahl nitrogen determination process using the parameters of the i-th particle is expressed in minutes and is obtained through actual execution or simulation; that is, the time taken from start to finish. Indicates the sample index in the batch; Let represent the absolute value of the difference in matrix features between the i-th sample and j, and let represent the absolute value of the difference in matrix features between the i-th sample and j. Each sample refers to the other samples in a batch that are different from the i-th sample; The matrix similarity threshold is preferably set to 0.1.

[0065] It should be noted that this invention uses an elite dearchived collection. To drive batch processing, Elite Dearch Archive Collection Includes the completion time of a single Kjeldahl nitrogen determination process Energy consumption of a single nitrogen determination process and the degree of human intervention The scheduler optimizes parameter settings for balanced optimization across multiple objectives by archiving elite solutions from a set of resources. Select appropriate parameter settings and assign them to each workstation to optimize batch processing. Specific relationships include: Elite Solution Archive Collection It provides a diverse set of optimal parameter solutions, each with a different objective trade-off; The scheduling goal is to achieve this through Minimize the maximum single Kjeldahl nitrogen determination process completion time for samples in a batch, while simultaneously... Constrain the matrix similarity of samples within a batch; By dearchiving the elite collection By selecting parameters, the scheduler can quickly adapt to the sample characteristics of different workstations, improving overall processing efficiency.

[0066] 5) Online model updates To ensure that the fully automated control and regulation system for Kjeldahl nitrogen determination continuously adapts to changes in equipment status and differences in sample matrix, the model is updated online. As the nitrogen determination process continues, equipment components may drift or age, and sample characteristics may also change, causing the initially trained prediction model to gradually become ineffective, affecting the accuracy of the prediction of recovery rate and relative standard deviation of repeatability. Therefore, the system collects new data in real time and updates the model dynamically to maintain optimized performance.

[0067] In practice, after each nitrogen determination process is executed, the system collects the process signal vector and actual quality index data for each nitrogen determination process. The process signal vector includes steam temperature, condensation power, distillate flow rate change rate and the second derivative of the receiving liquid potential, while the actual quality index is obtained through experimental verification, including recovery rate and repeatability relative standard deviation. This new data is stored in a rolling historical database that maintains a certain number of recent samples to balance the impact of new and old data. The system periodically initiates the model update process every 10 experiments. During the update, the system uses the newly collected data to fully retrain the lightweight gradient boosting tree model. The updated model is immediately deployed to the real-time prediction module to calculate constraint violation rates and guide parameter adjustments, thereby ensuring accurate avoidance of quality constraints and the correctness of optimization directions.

[0068] S4, Fully Automated Control and Regulation of Kjeldahl Nitrogen Determination Process After completing parameter optimization and constructing a multi-objective elite solution set, the process enters the stage of fully automated control and regulation of the Kjeldahl nitrogen determination process. Relying on the optimal parameter configuration in the elite solution archive set, combined with real-time process monitoring and dynamic scheduling mechanisms, precise control and adaptive regulation of equipment execution are achieved.

[0069] In the specific implementation, the scheduler first dearchives the elite archive set. The parameter configuration that best matches the matrix characteristics of the current sample to be tested and meets the batch scheduling objective is selected. The scheduler calculates the similarity based on the sample matrix feature embedding vector (percentage of fat content, percentage of sugar content, and logarithm of viscosity). Then, based on the multi-objective optimization result of minimizing the maximum single Kjeldahl nitrogen determination process completion time, specific parameter settings are assigned to each station. When assigning, it is necessary to ensure that the matrix difference of samples within the same batch is less than the matrix similarity threshold in order to maintain process stability. After selecting the parameters, the system will assign the particle position vector The breakdown involves issuing specific control commands to the Kjeldahl nitrogen analyzer, including: Continuous process control: Continuous process parameter subvectors The 10 elements are directly converted into equipment setpoints, namely, steam temperature control values. As the steam temperature setpoint, the percentage control quantity for steam flow rate As a percentage instruction for steam flow, other parameters follow the same pattern; Discrete policy activation: Discrete policy subvector The four elements trigger the corresponding operation mode, namely, the endpoint determination mode. Select the endpoint determination mode and discretize the threshold parameter. As a set threshold, other parameters follow the same principle; Execution of scheduling variables: scheduling subvectors The four elements control the timing logic, namely, the time allocation of the distillation stage. Set the distillation stage time allocation and cooling time. The cooling time for the sample is determined accordingly, and other parameters are calculated similarly.

[0070] Example 2 In this embodiment, a multi-objective optimization performance comparison is conducted to evaluate the effectiveness of the proposed Kjeldahl nitrogen determination fully automated control and regulation system in optimizing multi-objective performance, and it is compared with conventional particle swarm optimization methods and genetic algorithms. The experiment simulates an iterative process, such as... Figure 2 , Figure 3 , Figure 4As shown, the optimization trajectories of three methods on three key objectives—process completion time, energy consumption, and degree of human intervention—are illustrated. The horizontal axis represents the number of iterations (dimensionless), and the vertical axes represent process completion time (minutes), energy consumption (kilowatt-hours), and degree of human intervention (dimensionless), respectively. Process completion time refers to the time required for a single Kjeldahl nitrogen determination process from start to finish; energy consumption refers to the total energy consumption of a single process; and degree of human intervention is a comprehensive indicator reflecting the frequency and intensity of operator intervention. The figure shows that the curve of the method described in this invention exhibits a faster decline trend and a lower stable value across all three objectives. Specifically, regarding process completion time, the curve of the method described in this invention has a higher initial value but declines rapidly, reaching a lower plateau in the middle of the iteration. In contrast, the curves of conventional particle swarm optimization and genetic algorithms decline slowly, and their final stable values ​​are significantly higher than those of the method described in this invention. Experimental results demonstrate that this invention, through a hierarchical heterogeneous coding strategy and decoupling dimensions, effectively avoids search failures caused by parameter coupling, thereby accelerating convergence. Regarding energy consumption, the curve of the method of this invention also shows a steeper decline and a lower final value, indicating that the system optimizes energy utilization efficiency through real-time constraint agent model and policy constraint update mechanism. In terms of the degree of human intervention, the curve of the method of this invention decreases rapidly and remains near zero, while the comparative method remains at a high level throughout, demonstrating that the present invention reduces the need for human intervention and improves operational efficiency through automated control and multi-objective elite archiving. Overall, this experiment, through the comparison of iterative processes, demonstrates the comprehensive advantages of the present invention in multi-objective optimization, exhibiting faster convergence, lower energy consumption, and a higher degree of automation.

[0071] Example 3 In this embodiment, the differences in search performance and convergence characteristics between the particle swarm optimization method based on hierarchical heterogeneous coding and dynamic constraint processing proposed in this invention and conventional particle swarm optimization methods are compared. The experiment visualizes the search paths of the two methods in the two-dimensional parameter space, such as... Figure 5 , Figure 6As shown, this invention demonstrates its advantages in optimization efficiency and search accuracy. In the experiment, the horizontal and vertical axes represent two key optimization parameters (parameter one and parameter two), both dimensionless quantities, representing the key control variables after normalization in the Kjeldahl nitrogen determination process. The red pentagrams in the figure indicate the location of the true optimal solution, i.e., the theoretically optimal parameter configuration. The experimental results show that the search path of conventional particle swarm optimization methods exhibits significant randomness and dispersion. The movement trajectories of particles in the parameter space are chaotic, with numerous invalid explorations and detours. Many particles linger in regions far from the optimal solution, resulting in slow convergence. This search pattern reflects the limitations of conventional real-number encoding methods when dealing with heterogeneous parameters. The inability to effectively represent the coupling relationship between discrete and continuous parameters leads to a distorted search space and low optimization efficiency. In contrast, the search path of the method in this invention exhibits high directionality and concentration. The particle swarm rapidly converges from its initial random position towards the optimal solution region, exhibiting clear and orderly path trajectories and a stable and efficient convergence process. This demonstrates the effective implementation of the hierarchical heterogeneous encoding strategy. By decoupling the continuous process parameters, discrete strategy selection, and scheduling variables, the physical semantics of each parameter are preserved, eliminating search failures caused by heterogeneous variable coupling. Simultaneously, the dynamic constraint surrogate model and real-time parameter adjustment mechanism ensure that particles can promptly avoid quality violation regions during the search process, further improving search efficiency. Experimental results show that the particle swarm of the proposed method ultimately clusters in a dense region around the true optimal solution, while the particle distribution of conventional methods is relatively dispersed. This proves the significant advantages of the proposed method in terms of search accuracy and convergence. In the optimization process of the Kjeldahl nitrogen determination process, it can quickly find parameter configurations that satisfy both quality constraints and achieve multi-objective optimization, thereby improving the efficiency and reliability of the nitrogen determination process.

[0072] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An automated control and regulation system for the entire Kjeldahl nitrogen determination process based on artificial intelligence, characterized in that, include: Multimodal particle coding module: Decomposes the particle position vector into three orthogonal subspaces: continuous process parameter subvector, discrete strategy subvector, and scheduling subvector, and defines the physical meaning and value constraints of each subvector; Particle Swarm Optimization and Update Module: Constructs a dynamic constraint proxy model to avoid quality violations in real time; establishes a particle velocity update equation with policy constraints, defines a feasible solution set and policy rules; develops process response adaptive inertia weights; constructs a multi-objective elite archiving strategy to obtain an elite solution archiving set; incorporates matrix features into the optimization framework by constructing matrix feature embedding vectors and similarity calculation mechanisms to achieve optimization knowledge transfer between different matrix samples; Closed-loop online optimization execution module: performs closed-loop iteration of execution parameter sampling, process execution, signal acquisition, model update and parameter optimization, combined with multi-station scheduling to optimize batch processing; The Kjeldahl nitrogen determination process automation control and regulation module selects matching parameters from the elite solution archive set and sends them to the Kjeldahl nitrogen determination equipment. It monitors process signals in real time and adjusts adaptively to achieve full process automation.

2. The fully automated control and regulation system for Kjeldahl nitrogen determination based on artificial intelligence according to claim 1, characterized in that, The particle position vector represents the set of optimization parameters for the Kjeldahl nitrogen determination process; the elements of the continuous process parameter subvector correspond to specific control quantities, the elements of the discrete strategy subvector correspond to strategy selection, and the elements of the scheduling subvector correspond to scheduling variables.

3. The fully automated control and regulation system for Kjeldahl nitrogen determination based on artificial intelligence according to claim 1, characterized in that, Constructing a dynamic constraint proxy model to avoid quality violations in real time: Define the process signal vector: Real-time acquisition of multi-source sensor signals during the Kjeldahl nitrogen determination process, including steam temperature, condensation power, distillate flow rate change rate, and the second derivative of the receiving liquid potential, to construct the process signal vector; Establish a quality index prediction model: Using a lightweight gradient boosting tree model, a mapping relationship is established from the process signal vector to recovery rate and repeatability relative standard deviation. The prediction model is obtained through training with historical data, realizing real-time prediction of recovery rate and repeatability, and obtaining the predicted values ​​of recovery rate and repeatability relative standard deviation; Calculate the real-time constraint violation degree: Based on the predicted recovery rate and repeatability relative standard deviation, the constraint violation degree is calculated; Trigger real-time parameter adjustment: Monitor the constraint violation degree. When a constraint violation degree greater than zero is detected, the parameter adjustment mechanism is immediately triggered to guide the particle swarm away from the violation area, realizing online avoidance of quality constraints.

4. The fully automated control and regulation system for Kjeldahl nitrogen determination based on artificial intelligence according to claim 1, characterized in that, Establish the particle velocity update equation with policy constraints, and define the feasible solution set and policy rules: Define the velocity update equation for policy constraints: Based on the velocity update of the standard particle swarm optimization algorithm, the velocity vector of the particle is updated by combining the feasible historical best position and the policy-compatible global best position as guiding terms; Construct feasible solution set screening conditions: Define the screening conditions for feasible solution set, requiring the particle position vector to simultaneously satisfy the constraint violation degree of zero, the policy rule function being true, and the physical constraint function being true; Implement rule engine filtering: During the particle update process, infeasible solutions are filtered in real time through the rule engine; Particle position vector update: Update the particle position vector based on the updated velocity vector to obtain the particle position vector for the next iteration.

5. The fully automated control and regulation system for Kjeldahl nitrogen determination based on artificial intelligence according to claim 1, characterized in that, Develop adaptive inertia weights for process response: dynamically calculate adaptive inertia weights based on the norm of the gradient vector of the real-time process signal.

6. The fully automated control and regulation system for Kjeldahl nitrogen determination based on artificial intelligence according to claim 1, characterized in that, Construct a multi-objective elite archiving strategy to obtain an elite solution archiving set: Calculate the multi-objective vector of each particle: For each feasible particle, calculate its multi-objective vector, including the completion time of a single Kjeldahl nitrogen determination process, the energy consumption of a single nitrogen determination process, and the degree of human intervention; Perform non-dominated sorting: Perform Pareto non-dominated sorting on all particles in the feasible solution set to identify non-dominated solution sets of different levels; Elite solution screening and archiving: Based on the non-dominated sorting results and the spatial distribution density of the solutions, select elite solutions with sparsity greater than the density threshold from the first-level non-dominated solutions for archiving, and obtain the elite solution archive set.

7. The fully automated control and regulation system for Kjeldahl nitrogen determination based on artificial intelligence according to claim 1, characterized in that, By constructing matrix feature embedding vectors and similarity calculation mechanisms, matrix features are incorporated into the optimization framework, enabling optimization knowledge transfer between samples with different matrices: Constructing matrix feature embedding vectors: Define matrix feature embedding vectors, including percentage fat content, percentage sugar content, and log viscosity; Extending particle encoding structure: Embed the matrix feature embedding vectors as an additional dimension into the particle position vectors to form an extended particle representation; Calculating matrix similarity weights: Calculate the matrix similarity weights between samples based on the Euclidean distance between matrix feature embedding vectors. Integrating similar sample elite guidance: During the particle velocity update process, the historical optimal parameters of matrix similar samples are combined as an additional guiding term and summed by similarity weight.

8. The fully automated control and regulation system for Kjeldahl nitrogen determination based on artificial intelligence according to claim 1, characterized in that, The closed-loop online optimization execution module specifically includes: S31. Parameter sampling and nitrogen determination process execution: Select a particle from the particle swarm optimization algorithm and send its corresponding parameters to the Kjeldahl nitrogen determination equipment to execute the nitrogen determination process. S32. Signal Acquisition and Objective Function Calculation: Acquire process signals in real time and calculate the objective function value. The objective function value is the weighted sum of the completion time of a single Kjeldahl nitrogen determination process, the energy consumption of a single nitrogen determination process, and the degree of human intervention. S33. Particle swarm parameter update mechanism: The particle swarm is updated using the objective function value, which is divided into individual update and swarm update. S34. Multi-station scheduling optimization: Use elite dearchived archive set to drive multi-station scheduling, optimize batch processing, and the scheduling objective is to minimize the maximum single Kjeldahl nitrogen determination process completion time of samples in the batch. The constraint condition is that the difference in matrix characteristics of samples in the batch is less than a threshold. S35. Online Model Update: After each nitrogen determination process is executed, the system collects the process signal vector and actual quality index data for each nitrogen determination process. The process signal vector includes steam temperature, condensation power, distillate flow rate change rate, and the second derivative of the receiving liquid potential. The actual quality index data includes recovery rate and repeatability relative standard deviation. The system uses the newly collected data to fully retrain the lightweight gradient boosting tree model.

9. The fully automated control and regulation system for Kjeldahl nitrogen determination based on artificial intelligence according to claim 8, characterized in that, The individual update is the individual historical best update: if the particle position vector satisfies the feasibility condition, then the feasible historical best position of the particle is updated.

10. The fully automated control and regulation system for Kjeldahl nitrogen determination based on artificial intelligence according to claim 8, characterized in that, The group update is a global optimal and elite solution archive set update: The first step is to perform non-dominated solution screening by adding the particle position vectors to the feasible solution set and performing non-dominated sorting. The second step is to update the elite archive. If the Pareto level of the particle position vector is the first level and the sparsity of the particle position vector in the target space is greater than the density threshold, then the particle position vector is added to the elite solution archive set, and the solutions dominated by the particle position vector in the elite solution archive set are removed. The third step is to perform a global optimal selection by polling the elite solution archive set to select the globally optimal position that is compatible with the strategy.

Images