Closed-loop regulation and control system and method based on oxygen-carbon and process parameter coupling model

Abstract

The invention discloses a closed-loop regulation and control system and method based on an oxygen-carbon and process parameter coupling model, and relates to the technical field of industrial automation, and the system comprises a scene adaptive modeling module, a digital twin mapping module, a multi-target collaborative decision module, and an execution and feedback module. The scene adaptive modeling module is used for constructing a cross-scene oxygen carbon-process parameter dynamic coupling model based on a scene feature library and an online parameter self-tuning algorithm; and the digital twin mapping module is electrically connected with the scene adaptive modeling module. According to the closed-loop regulation and control system and method based on the oxygen-carbon and process parameter coupling model provided by the invention, the limitation of scene simplification in the prior art is broken through, the adaptation of the same system to different industrial scenes is realized by constructing a cross-scene self-adaptive oxygen-carbon-process parameter coupling modeling system, and the implementation of the closed-loop regulation and control system is facilitated. The generalization ability and the application range of the system are greatly improved, and the control precision and the response speed under the dynamic working condition are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of industrial automation, in particular to a closed-loop control system and method based on oxygen-carbon and process parameter coupling model. BACKGROUND

[0002] Under the trend of intelligent upgrading of industrial production, the coordinated control of oxygen-carbon parameters (oxygen, carbon monoxide) and process parameters (temperature, pressure, air volume, etc.) is a key link to ensure production safety, improve efficiency and product quality. This type of control technology is widely used in boiler combustion, chemical reaction, metallurgical processing and other fields. Its core logic is to realize precise control of the production process through real-time monitoring and dynamic adjustment of key parameters. With the development of industrial internet and data-driven technology, closed-loop control based on coupling model has become an important direction of technological upgrading. It strengthens the correlation between multi-dimensional parameters, reduces manual intervention through the closed-loop mechanism of "perception-analysis-decision-execution", and improves the stability and economy of system operation, meeting the production needs of modern industry for high efficiency, low carbon and safety.

[0003] Current related technologies have been applied in industrial control to a certain extent, and can realize the monitoring and adjustment of single or partial parameters, providing basic protection for the production process. However, under complex working conditions, there is still room for optimization in existing technologies: some solutions do not fully consider the coupling relationship between oxygen-carbon parameters and process parameters, and fail to form a systematic collaborative control mechanism; the adaptability of some control models is limited, and it is difficult to respond accurately when facing working condition fluctuations, changes in material properties, etc.; at the same time, some systems still have room for improvement in real-time data transmission, anti-interference ability and dynamic adaptability of parameter adjustment, making it difficult to fully meet the fine control needs in complex industrial scenarios. Therefore, we propose a closed-loop control system and method based on oxygen-carbon and process parameter coupling model. SUMMARY

[0004] To solve the above technical problems, a closed-loop control system and method based on oxygen-carbon and process parameter coupling model are provided. This technical solution solves the problem of not fully considering the coupling relationship between oxygen-carbon and process parameters, lacking a systematic collaborative control mechanism; the model has limited adaptability, making it difficult to respond to changes in working conditions; and the data real-time performance, anti-interference ability and parameter dynamic adaptability are insufficient, making it difficult to meet the fine control needs in complex industrial scenarios.

[0005] To achieve the above purposes, the technical solution adopted by the present application is as follows:

[0006] The closed-loop control system based on oxygen-carbon and process parameter coupling model comprises:

[0007] a scene self-adaptive modeling module, a digital twin mapping module, a multi-objective collaborative decision module, an execution and feedback module.

[0008] The scene adaptive modeling module is configured to construct a cross-scene oxygen-carbon-process parameter dynamic coupling model based on a scene feature library and an online parameter self-tuning algorithm.

[0009] The digital twin mapping module is electrically connected with the scene adaptive modeling module, and is configured to map a physical system state in real time, predict a working condition fluctuation, and provide real-time input for the coupling model.

[0010] The multi-objective collaborative decision-making module is electrically connected with the digital twin mapping module, and is configured to output optimal control instructions by using an improved multi-objective optimization algorithm in combination with real-time carbon flow data.

[0011] The execution and feedback module is electrically connected with the multi-objective collaborative decision-making module, and is configured to execute the control instructions and feed back real-time parameters to the scene adaptive modeling module.

[0012] Preferably, the scene feature library is constructed by the following method:

[0013] Based on oxygen-carbon reaction data of cross-industry industrial scenes, typical process parameter sequences are obtained, and oxygen-carbon concentration, temperature, pressure and material flow data are collected in real time, which are subjected to outlier cleaning and normalization processing.

[0014] A reference scene feature template is set, and process parameter ranges, oxygen-carbon reaction mechanism threshold values and typical interference factor types of the reference scene are extracted.

[0015] Each scene process parameter sequence is matched and analyzed with the reference feature template, a feature difference vector is extracted, and a scene feature identifier is generated.

[0016] The distance measurement of each scene and the reference scene in the oxygen-carbon response characteristics is calculated, the feature clustering is performed in combination with a mechanism knowledge base, and a scene feature classification set is formed.

[0017] The scene feature identifier, the feature difference vector and the clustering result are integrated to construct a structured multi-scene feature library.

[0018] Preferably, the online parameter self-tuning algorithm specifically includes the following steps:

[0019] Based on the real-time collected oxygen-carbon concentration, temperature, pressure and material flow data, a dynamic input vector is constructed and subjected to standardization processing.

[0020] The model weight parameters are initialized, and the initial value of the forgetting factor and the update step are set.

[0021] An improved forgetting factor recursive least squares method is used to calculate the parameter correction amount according to the real-time input and output deviation, and the model weight is updated.

[0022] An adaptive forgetting factor adjustment mechanism is introduced to adjust the size of the forgetting factor according to the scene change rate and prediction error history;

[0023] The updated model weight is output and synchronized to the scene feature library, completing the parameter self-tuning closed loop.

[0024] Preferably, the construction of the cross-scene oxygen-carbon-process parameter dynamic coupling model is specifically:

[0025] Based on the matching of the feature identifier corresponding to the current working condition and the process parameter range in the scene feature library, the preset oxygen-carbon reaction mechanism equation is called to construct a basic oxygen-carbon-process parameter dynamic coupling model framework;

[0026] The model parameters are initialized, and the initial value and dynamic adjustment interval of the model weight are set according to the typical process parameter range and oxygen-carbon reaction mechanism threshold value in the scene feature library;

[0027] An online parameter self-tuning algorithm is introduced, and the real-time collected oxygen-carbon concentration, temperature, pressure and material flow data are input into the model, and the model weight parameters are dynamically updated through recursive calculation;

[0028] A dynamic feedback structure of the coupling model is established, the model output is compared with the actual oxygen-carbon state of the physical system, and the model structure parameters and response relationship are corrected in real time according to the deviation;

[0029] Integrate the scene adaptation layer, according to the scene classification and difference vector in the feature library, correct the model output for scene specificity, and generate a cross-scene usable oxygen-carbon-process parameter dynamic coupling model.

[0030] Preferably, the real-time mapping of the physical system state is specifically:

[0031] Real-time oxygen-carbon concentration data, process parameter data and equipment state data of each monitoring point in the physical system are collected to form a multi-source heterogeneous real-time data stream;

[0032] Based on the real-time data stream, a digital twin body consistent with the topology structure of the physical system is constructed, which includes an oxygen-carbon distribution layer, a process parameter layer and an equipment state layer, and a dynamic association relationship between the layers is established;

[0033] Through the time series data synchronization engine, the real-time data stream of the physical system is mapped to the corresponding layer of the digital twin body;

[0034] A real-time carbon flow calculation unit is embedded in the digital twin body, based on the oxygen-carbon distribution data and the material flow topology, the carbon flow intensity of each node is calculated and updated in real time, and a dynamic carbon flow map is generated;

[0035] The updated digital twin body data is output to the scene adaptive modeling module in real time.

[0036] Preferably, the method for predicting operating condition fluctuations and self-calibrating the model based on the digital twin mapping module is as follows:

[0037] A dynamic disturbance prediction unit is integrated into the digital twin. Based on historical operating data and the current real-time status, a time-series prediction model is used to predict the fluctuation trend of oxygen and carbon concentration and the range of process parameter changes within a future preset time window.

[0038] The dynamic disturbance prediction unit uses a reinforcement learning algorithm to generate a set of pre-control strategies for potential operating condition fluctuations based on the prediction results and historical control effects.

[0039] The set of pre-regulation strategies and the predicted fluctuation trend are sent to the multi-objective collaborative decision-making module as the basis for forward-looking regulation decisions.

[0040] A model self-calibration unit is set up in the digital twin to compare the measured data of the physical system with the output data of the digital twin in real time and calculate the deviation index between the two.

[0041] When the deviation index exceeds the set threshold, the parameter adaptive correction mechanism is triggered. Based on the deviation sequence, the state mapping relationship and prediction model parameters in the digital twin are dynamically adjusted, and the corrected parameters are synchronized to the scene adaptive modeling module.

[0042] Preferably, the specific method for generating the optimal control command is as follows:

[0043] Receive real-time oxygen and carbon status data, process parameter data, predicted operating condition fluctuation trends, and dynamic carbon flow intensity sequences from the digital twin mapping module;

[0044] A multi-objective optimization function is constructed, with the core optimization objective set as minimizing the oxygen-carbon ratio deviation, and the constraints set as the upper limit of carbon flow intensity, the industry benchmark value of energy consumption, and the fluctuation range of equipment operating load.

[0045] An improved multi-objective optimization algorithm is used to solve the core objective and the carbon flow constraint conditions in a coordinated manner to generate a preliminary optimized solution set.

[0046] Preferably, the step of outputting the optimal control command is as follows:

[0047] Based on the scene classification identifiers in the scene feature library, dynamic weights are assigned to different objectives in the multi-objective optimization function;

[0048] The digital twin constructed using the aforementioned digital twin mapping module is used to virtually execute and verify the multi-dimensional effects of the strategies in the preliminary optimized solution set.

[0049] Based on the verification results, the control strategy with the best overall performance is selected from the preliminary optimized solution set, and the final control command is generated and output.

[0050] Preferably, the method of executing control commands and feeding back real-time parameters to the scene adaptive modeling module is as follows:

[0051] Receive the final control instruction from the multi-objective collaborative decision-making module, and parse the specific actuator action type, adjustment range, and execution sequence contained in the instruction;

[0052] The parsed action commands are converted into control signals for the underlying physical actuators and sent synchronously to the corresponding regulating valves, frequency converters, heaters or material conveying devices.

[0053] During execution, the action status and response feedback of each actuator are monitored in real time, and dynamic coordination is performed to prevent execution conflicts or overshoot;

[0054] The actual system control quantity after execution is recorded and fed back to the digital twin mapping module for updating the state of the digital twin;

[0055] After the control command is executed, the updated oxygen and carbon concentration data, process parameter data, and equipment operating status data are collected in real time through the sensor network deployed in the physical system.

[0056] The collected feedback data is time-aligned and noise-filtered to form a structured feedback data packet;

[0057] The feedback data packet is synchronously sent to the digital twin mapping module for updating the twin's state, and simultaneously sent to the scene adaptive modeling module;

[0058] Based on the received feedback data, the scenario adaptive modeling module triggers the online parameter self-tuning algorithm to update the weights and correct the parameters of the oxygen-carbon process parameter dynamic coupling model, thereby completing closed-loop control.

[0059] The closed-loop control method based on the coupling model of oxygen, carbon, and process parameters includes the following steps:

[0060] S1. Collect process parameters and oxygen-carbon reaction data from different industrial scenarios, construct a multi-scenario feature library, and establish a cross-scenario applicable dynamic coupling model of oxygen-carbon and process parameters through parameter self-tuning algorithm.

[0061] S2. Based on real-time collected oxygen and carbon concentration, process parameters and equipment status data, construct a digital twin synchronized with the physical system to predict future operating condition fluctuations and calculate dynamic carbon flow intensity.

[0062] S3. Combining real-time oxygen and carbon data, predicted fluctuation trends and carbon flow intensity, a multi-objective optimization function is constructed with minimizing the oxygen-carbon ratio deviation as the core and carbon flow and energy consumption as constraints. The optimal control strategy is solved through optimization algorithms.

[0063] S4. Parse and execute control commands, collect system feedback data in real time, update the state of the digital twin and correct the parameters of the dynamic coupling model to form a closed-loop control.

[0064] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0065] The closed-loop control system and method based on the oxygen-carbon and process parameter coupling model proposed in this invention overcomes the limitation of single-scenario application in existing technologies. By constructing a cross-scenario adaptive oxygen-carbon-process parameter coupling modeling system, the same system can be adapted to different industrial scenarios, greatly improving the system's generalization ability and application scope. By integrating digital twin technology, real-time mapping of the physical system state and prediction of operating condition fluctuations are achieved. Combined with reinforcement learning algorithms, the control accuracy and response speed under dynamic operating conditions are significantly improved. The introduction of a multi-objective collaborative optimization intelligent decision-making framework not only focuses on oxygen-carbon balance but also takes into account low carbon emissions, energy consumption optimization, and equipment life extension, thereby improving the overall performance of the system. These factors together constitute the technical advantages of the system, enabling it to perform excellently in complex and ever-changing industrial environments, with significant economic and social benefits, demonstrating high innovation and practicality. Attached Figure Description

[0066] Figure 1 This is a system module diagram of the present invention;

[0067] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0068] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0069] Reference Figure 1 As shown, the closed-loop control system based on the oxygen-carbon and process parameter coupling model includes: a scenario adaptive modeling module, a digital twin mapping module, a multi-objective collaborative decision-making module, and an execution and feedback module.

[0070] The scenario adaptive modeling module is used to construct a cross-scenario oxygen-carbon process parameter dynamic coupling model based on a scenario feature library and an online parameter self-tuning algorithm.

[0071] The method for constructing the scene feature library is as follows:

[0072] Based on oxygen and carbon reaction data from cross-industry industrial scenarios, typical process parameter sequences are obtained, and oxygen and carbon concentration, temperature, pressure, and material flow data are collected in real time. Outlier cleaning and normalization processing are performed on these data.

[0073] Set a baseline scenario feature template and extract the process parameter range, oxygen-carbon reaction mechanism threshold, and typical interference factor types of the baseline scenario;

[0074] The process parameter sequences for each scenario are matched and analyzed with the baseline feature template to extract feature differences and generate scenario feature identifiers. The feature differences are calculated using a combination of Euclidean distance and cosine similarity. Euclidean distance measures the absolute difference between the process parameter sequences for each scenario and the baseline feature template in terms of parameter numerical range. The calculation formula is as follows:

[0075]

[0076] In the formula For Euclidean distance, Let be the process parameter value for the i-th scenario. The value of the process parameter in the baseline feature template is n, where n is the dimension of the process parameter. Cosine similarity is used to measure the relative difference between the two in the trend of parameter change, and the calculation formula is:

[0077]

[0078] In the formula For cosine similarity, The angle between the two vectors is given; the feature difference is obtained by combining the Euclidean distance and cosine similarity results.

[0079] The distance metric between each scene and the baseline scene in terms of oxygen and carbon response characteristics is calculated. Feature clustering is then performed using a mechanistic knowledge base to form a scene feature classification set. The feature clustering employs the density-based DBSCAN algorithm. This algorithm defines core points by setting an ε-neighborhood and a minimum number of points, and forms clusters based on the connectivity between core points. This algorithm effectively handles noisy scene feature data and identifies irregularly shaped cluster structures. During clustering, the feature difference vectors of each scene and the distance metric of oxygen and carbon response characteristics are used as input data. The ε-neighborhood and minimum number of points threshold are set based on the intrinsic correlation of oxygen and carbon reactions in different scenes in the mechanistic knowledge base.

[0080] By integrating scene feature identifiers, feature difference vectors, and clustering results, a structured multi-scene feature library is constructed. The specific method for converting clustering results into structured feature library entries is as follows: each cluster is treated as a basic classification unit and assigned a unique classification identifier. Each entry includes a cluster identifier, a list of feature identifiers for all scenes within the cluster, feature difference vectors for each scene relative to the baseline scene, typical process parameter ranges for scenes within the cluster, and oxygen-carbon reaction mechanism thresholds. At the same time, hierarchical relationships between clusters are established to form a structured multi-scene feature library.

[0081] The online parameter self-tuning algorithm is specifically as follows:

[0082] Based on real-time collected data on oxygen and carbon concentration, temperature, pressure, and material flow rate, a dynamic input vector is constructed and standardized.

[0083] Initialize the model weight parameters, set the initial value of the forgetting factor and the update step size;

[0084] An improved forgetting factor recursive least squares method is adopted, which calculates parameter corrections based on real-time input-output deviations and updates model weights. The specific improvement of the improved forgetting factor recursive least squares method lies in introducing an L2 regularization term to enhance numerical stability and optimizing the covariance matrix update method. The covariance matrix update formula of the standard forgetting factor recursive least squares method is:

[0085]

[0086] The improved formula for updating the covariance matrix by introducing an L2 regularization term is as follows:

[0087]

[0088] In the formula Let be the covariance matrix at time k. Forgetting factor, Let k be the observation vector at time k. is the regularization coefficient, and I is the identity matrix;

[0089] The introduction of L2 regularization can effectively avoid singular values ​​in the covariance matrix and enhance the numerical stability of the algorithm.

[0090] An adaptive forgetting factor adjustment mechanism is introduced, which adjusts the magnitude of the forgetting factor based on the rate of change of the scene and the history of prediction errors. The specific mathematical expression of the adaptive forgetting factor adjustment mechanism is as follows:

[0091]

[0092] In the formula Let k be the forgetting factor at time k. This represents the minimum forgetting factor, ranging from 0.8 to 0.9. and For adjustment coefficients, Let k be the difference between the prediction error at time k and the prediction error at the previous time. The rate of change of the scene;

[0093] When the prediction error increases, that is When the exponential term decreases, Increase the speed of parameter updates to quickly track scene changes; when the system is stable... and When the value is small, the exponent term approaches 1. Approaching Reduce the parameter update amplitude to enhance noise resistance;

[0094] The updated model weights are output and synchronized to the scene feature library to complete the parameter self-tuning closed loop.

[0095] The specific steps for constructing a cross-scenario dynamic coupling model of oxygen and carbon with process parameters are as follows:

[0096] Based on the scene feature library, the feature identifiers and process parameter ranges corresponding to the current operating condition are matched, and the pre-set oxygen-carbon reaction mechanism equations are invoked to construct a basic oxygen-carbon-process parameter dynamic coupling model framework. The pre-set oxygen-carbon reaction mechanism equations adopt a kinetic model that includes the Arrhenius equation, specifically in the following form:

[0097]

[0098] In the formula The rate of change of carbon monoxide concentration, Pre-exponential factor, The activation energy is given by R, the gas constant is given by R, which is 8.314 J / (mol·K), and the absolute temperature is given by T. Oxygen concentration, denoted as carbon monoxide concentration, and m and n as the reaction orders of oxygen and carbon monoxide, respectively.

[0099] Initialize model parameters by setting initial model weights and dynamic adjustment ranges based on the typical process parameter ranges and oxygen-carbon reaction mechanism thresholds in the scenario feature library.

[0100] An online parameter self-tuning algorithm is introduced, which inputs real-time collected data on oxygen and carbon concentration, temperature, pressure and material flow rate into the model, and dynamically updates the model weight parameters through recursive calculation;

[0101] A dynamic feedback structure for the coupled model is established, comparing the model output with the actual oxygen and carbon state of the physical system. The model structure parameters and response relationship are corrected in real time based on the deviation. The dynamic feedback structure adopts an online rolling correction framework for model predictive control. Specifically, using a fixed prediction time domain N and control time domain M as windows, within each control cycle, the deviation between the model output and the actual oxygen and carbon state is used as a feedback signal. This signal is substituted into the prediction error model to calculate the error compensation amount. The error compensation amount is then integrated into the model's state equations to correct the model output.

[0102]

[0103] In the formula The corrected model output. The predicted output for the original model. This is the error compensation amount;

[0104] An integrated scene adaptation layer performs scene-specific correction on the model output based on scene classifications and difference vectors in the feature library, generating a dynamic coupling model of oxygen and carbon-process parameters that can be used across different scenes. The scene adaptation layer adopts a linear offset correction method based on difference vectors. Specifically, it determines the correction coefficients through the feature difference vectors in the scene feature library and linearly adjusts the model output. The correction formula is as follows:

[0105]

[0106] In the formula This is the output of the model after scene adaptation, where K is the correction coefficient matrix. This is the feature difference vector between the current scene and the benchmark scene. This linear offset correction eliminates system biases in different scenes, enabling cross-scene adaptation of the model.

[0107] The digital twin mapping module is electrically connected to the scene adaptive modeling module, and is used to map the physical system state in real time, predict operating condition fluctuations, and provide real-time input to the coupled model.

[0108] The specific state of the real-time mapped physical system is as follows:

[0109] Real-time acquisition of oxygen and carbon concentration data, process parameter data, and equipment status data at various monitoring points in the physical system to form a multi-source heterogeneous real-time data stream;

[0110] Based on real-time data streams, a digital twin is constructed that is consistent with the topology of the physical system. The digital twin includes an oxygen and carbon distribution layer, a process parameter layer, and an equipment status layer, and dynamic relationships between the layers are established.

[0111] The time-series data synchronization engine maps the real-time data stream of the physical system to the corresponding layer of the digital twin. The time alignment of the multi-source heterogeneous real-time data stream adopts the linear interpolation completion method. For data sources with different sampling frequencies, the highest sampling frequency is used as the benchmark, and the low sampling frequency data is linearly interpolated in the time dimension to obtain the time series with equal time intervals. The spatial registration is based on the device topology coordinates of the physical system, and the location information of each monitoring point is mapped and matched with the layer coordinates of the digital twin to ensure the consistency of data in the spatial dimension.

[0112] The time-series data synchronization engine adopts a technical solution that combines Kafka message middleware with Flink stream processing platform. Kafka is responsible for high-throughput real-time data reception and caching, and achieves parallel data processing through partitioning strategy. Flink stream processing platform is responsible for real-time data computation and synchronous distribution. It uses its low-latency stream processing capability to ensure that the mapping latency of data from physical system to digital twin is controlled within 100ms. At the same time, it avoids data write conflicts through distributed lock mechanism to ensure data consistency.

[0113] A real-time carbon flow calculation unit is embedded in the digital twin. Based on oxygen and carbon distribution data and material flow topology, the carbon flow intensity of each node is calculated and updated in real time, generating a dynamic carbon flow map. The material flow topology map is constructed based on the equipment connection relationships and process flow diagram of the physical system. By analyzing the connection methods of process pipelines and the inlet and outlet positions of equipment, a topology structure containing nodes (equipment / monitoring points) and edges (material pipelines) is established, and the material flow direction is labeled for each edge. The specific calculation formula for carbon flow intensity is based on the stoichiometry of chemical reactions and the law of conservation of mass, namely:

[0114]

[0115] In the formula Carbon flux intensity, in kg / s. This represents the material flow rate into the node. This represents the carbon mass fraction in the incoming material. Let be the stoichiometric coefficient of carbon in the i-th carbon-containing reaction. Let be the reaction rate of the i-th carbon-containing reaction. The value is the molar mass of carbon, which is 12 g / mol.

[0116] The updated digital twin data is output to the scene adaptive modeling module in real time.

[0117] The method for predicting operating condition fluctuations and self-calibrating the model based on the digital twin mapping module is as follows:

[0118] A dynamic disturbance prediction unit is integrated into the digital twin. Based on historical operating data and the current real-time status, a time-series prediction model is used to predict the fluctuation trend of oxygen and carbon concentration and the range of process parameter changes within a preset future time window. The time-series prediction model adopts a long short-term memory network. This model controls the transmission and forgetting of information through gating units (input gate, forget gate, output gate), which can effectively capture the long-term and short-term dependencies in the time-series data and is suitable for predicting operating condition fluctuations in industrial scenarios. The model's input is a time-series sequence composed of historical oxygen and carbon concentration data, process parameter data, and equipment status data. The output is the predicted value and fluctuation range of oxygen and carbon concentration and process parameters within a preset future time window (such as 5 minutes or 10 minutes).

[0119] The dynamic disturbance prediction unit employs a reinforcement learning algorithm to generate a set of pre-control strategies for potential operating condition fluctuations based on prediction results and historical control effects. The state space of the reinforcement learning algorithm is defined as a vector composed of the current oxygen and carbon concentration, process parameters, equipment status, and predicted fluctuation trends. The action space is defined as a set of adjustment amounts for each control parameter, including material flow rate adjustment, temperature adjustment, and pressure adjustment. The reward function is defined as:

[0120]

[0121] In the formula, R is the reward value, and a, b, and c are weighting coefficients. This is due to the deviation in the oxygen-to-carbon ratio. For the increase in energy consumption, For equipment load fluctuations, an optimal set of pre-regulation strategies is generated by maximizing the cumulative reward value;

[0122] The set of pre-regulation strategies and the predicted fluctuation trend are sent to the multi-objective collaborative decision-making module as the basis for forward-looking regulation decisions.

[0123] A model self-calibration unit is set up in the digital twin to compare the measured data of the physical system with the output data of the digital twin in real time, and calculate the deviation index between the two; the deviation index is the root mean square error, and the calculation formula is as follows:

[0124]

[0125] In the formula, RMSE is the root mean square error, and n is the number of data samples. For measured data of physical systems, Output data for the digital twin; the threshold is determined based on the industry's allowable fluctuation range of process parameters and the oxygen and carbon concentration control accuracy requirements, combined with the statistical analysis results of historical data. For example, the RMSE threshold corresponding to oxygen and carbon concentration is set to 0.02 mol / L, and the RMSE threshold corresponding to process parameters is set according to the parameter type.

[0126] When the deviation index exceeds a set threshold, a parameter adaptive correction mechanism is triggered. This mechanism dynamically adjusts the state mapping relationship and prediction model parameters in the digital twin based on the deviation sequence, and synchronizes the corrected parameters to the scene adaptive modeling module. The specific algorithm steps of the parameter adaptive correction mechanism are as follows: First, trend analysis is performed on the deviation sequence to determine whether the deviation type is systematic or random. If it is a systematic deviation, the gradient descent algorithm is used to adjust the transformation coefficients in the state mapping relationship of the digital twin, i.e.:

[0127]

[0128] In the formula These are the corrected conversion factors. These are the conversion factors before correction. The value is the learning rate; if it is a random bias, the weight parameters of the time series prediction model are adjusted, and the neuron weights of the long short-term memory network are updated through the backpropagation algorithm until the bias index drops below the set threshold.

[0129] The multi-objective collaborative decision-making module is electrically connected to the digital twin mapping module and is used to output the optimal control command by combining the improved multi-objective optimization algorithm with real-time carbon flow data.

[0130] The specific method for generating the optimal control command is as follows:

[0131] Receive real-time oxygen and carbon status data, process parameter data, predicted operating condition fluctuation trends, and dynamic carbon flow intensity sequences from the digital twin mapping module;

[0132] A multi-objective optimization function is constructed, with the core optimization objective being the minimization of the oxygen-to-carbon ratio deviation. The constraints are set as an upper limit for carbon flow intensity, an industry benchmark value for energy consumption, and the fluctuation range of equipment operating load. The mathematical definition of minimizing the core optimization objective, the oxygen-to-carbon ratio deviation, is the sum of squares of the oxygen-to-carbon ratio deviations, i.e.:

[0133]

[0134] In the formula, J is the objective function value, N is the number of sampling points in the prediction time domain, and O / C(k) is the actual oxygen-to-carbon ratio at time k. The target oxygen-to-carbon ratio;

[0135] An improved multi-objective optimization algorithm is used to collaboratively solve the core objective and the carbon flow constraint to generate a preliminary optimized solution set. Specifically, the improved multi-objective optimization algorithm is an improved non-dominated sorting genetic algorithm III. The improvements mainly include the introduction of an elite retention strategy and a dynamic crowding calculation method. The elite retention strategy constructs an elite pool to retain non-dominated solutions in each generation of the population, avoiding the loss of excellent individuals. The dynamic crowding calculation method dynamically adjusts the crowding calculation weight according to the distribution density of individuals in the population, increasing the weight in sparsely distributed regions of the population to improve the uniformity of solution distribution.

[0136] The steps for outputting the optimal control command are as follows:

[0137] Based on the scene classification identifiers in the scene feature library, dynamic weights are assigned to different objectives in the multi-objective optimization function. The rules for assigning dynamic weights are determined based on the energy consumption sensitivity and carbon emission priority in the scene feature identifiers. Specifically, the energy consumption sensitivity coefficient and carbon emission priority coefficient of the current scene are queried from the scene feature library, and the dynamic weights of each objective are calculated using a weighted summation method.

[0138]

[0139]

[0140]

[0141] In the formula As the weight of energy consumption target, As the target weight for carbon flow, The target weight for the oxygen-to-carbon ratio deviation. and The normalization coefficient is... Energy sensitivity coefficient This is the carbon emission priority coefficient; when the scenario is a high-energy-consumption sensitive scenario... Increase The corresponding increase; when the scenario is a high-carbon emission constraint scenario. Increase Increase accordingly;

[0142] The digital twin constructed using the aforementioned digital twin mapping module is used to virtually execute and verify the multi-dimensional effects of the strategies in the initial optimized solution set. The specific process of virtual execution and effect verification involves real-time simulation within the digital twin, based on the timescale of the actual industrial scenario. The simulation speed is synchronized 1:1 with the physical system to ensure the timeliness of the simulation results. During the simulation, key indicators such as oxygen-to-carbon ratio deviation, carbon flux intensity, energy consumption, and equipment operating load are monitored in real time to verify the control effect of each optimization strategy under predicted operating condition fluctuations. The evaluation criterion for the best overall performance adopts a weighted total score method, that is, a weighted total score is calculated for each key indicator according to dynamic weights.

[0143]

[0144] In the formula, Score is the weighted total score. The score is given for the effectiveness of oxygen-carbon ratio control. To score the effectiveness of energy consumption control, The carbon flow control effect is scored, with a higher score indicating a better control effect. The strategy with the highest weighted total score is selected as the optimal control strategy.

[0145] Based on the verification results, the control strategy with the best overall performance is selected from the preliminary optimized solution set, and the final control command is generated and output.

[0146] The execution and feedback module is electrically connected to the multi-objective collaborative decision-making module and is used to execute control commands and feed back real-time parameters to the scene adaptive modeling module.

[0147] The method for executing control commands and feeding back real-time parameters to the scene adaptive modeling module is as follows:

[0148] Receive the final control instruction from the multi-objective collaborative decision-making module, and parse the specific actuator action type, adjustment range, and execution sequence contained in the instruction;

[0149] The parsed action commands are converted into control signals for the underlying physical actuators and sent synchronously to the corresponding regulating valves, frequency converters, heaters, or material conveying devices. The interface standard for converting high-level control commands into low-level control signals is the Modbus-RTU protocol. For digital control signals, switch signals are used for output; for analog control signals, they are converted into 4-20mA standard current signals, where different current values ​​correspond to different action amplitudes of the actuators.

[0150] During execution, the action status and response feedback of each actuator are monitored in real time, and dynamic coordination is performed to prevent execution conflicts or overshoot. The specific logic of the dynamic coordination mechanism to prevent execution conflicts adopts rule-based interlock judgment. An interlock rule library for actuator actions is established in advance, which includes the action priorities and mutual exclusion relationships between different actuators. Before the execution command is issued, the action commands of each actuator are interlocked. If there are conflicting actions, the execution sequence is adjusted according to the priority rules, and the higher priority command is executed first. For the possible overshoot problem, a segmented execution strategy is adopted, and the adjustment range is divided into multiple stages for gradual execution. The changing trend of the controlled parameter is monitored in real time. If overshoot signs appear, the adjustment range of the subsequent stages is adjusted immediately.

[0151] The actual system control quantity after execution is recorded and fed back to the digital twin mapping module for updating the state of the digital twin;

[0152] After the control command is executed, the updated oxygen and carbon concentration data, process parameter data, and equipment operating status data are collected in real time through the sensor network deployed in the physical system.

[0153] The collected feedback data is time-series aligned and noise-filtered to form a structured feedback data packet. The structured format of the feedback data packet includes fields such as timestamp, data source identifier, parameter name, value, unit, and confidence level. The timestamp is accurate to the millisecond level, the data source identifier corresponds to the specific sensor number, and the confidence level is calculated based on the sensor's measurement accuracy and data stability.

[0154] The feedback data packet is synchronously sent to the digital twin mapping module for updating the twin's state, and simultaneously sent to the scene adaptive modeling module;

[0155] The scenario adaptive modeling module, based on the received feedback data, triggers the online parameter self-tuning algorithm to update the weights and correct the parameters of the oxygen-carbon process parameter dynamic coupling model, completing closed-loop control. The conditions for triggering the online parameter self-tuning algorithm include periodic triggering and instantaneous triggering based on sudden changes in feedback data. Periodic triggering is executed at fixed time intervals (e.g., 1 second). The condition for instantaneous triggering is that the rate of change of the feedback data exceeds a set threshold. In the formula For the rate of change of the feedback data, The set mutation threshold is used to immediately trigger the online parameter self-tuning algorithm when the condition is met, thereby achieving rapid correction of the model parameters.

[0156] Reference Figure 2 As shown, the closed-loop control method based on the coupling model of oxygen, carbon, and process parameters includes the following steps:

[0157] S1. Collect process parameters and oxygen-carbon reaction data from different industrial scenarios, construct a multi-scenario feature library, and establish a cross-scenario applicable dynamic coupling model of oxygen-carbon and process parameters through parameter self-tuning algorithm.

[0158] S2. Based on real-time collected oxygen and carbon concentration, process parameters and equipment status data, construct a digital twin synchronized with the physical system to predict future operating condition fluctuations and calculate dynamic carbon flow intensity.

[0159] S3. Combining real-time oxygen and carbon data, predicted fluctuation trends and carbon flow intensity, a multi-objective optimization function is constructed with minimizing the oxygen-carbon ratio deviation as the core and carbon flow and energy consumption as constraints. The optimal control strategy is solved through optimization algorithms.

[0160] S4. Parse and execute control commands, collect system feedback data in real time, update the state of the digital twin and correct the parameters of the dynamic coupling model to form a closed-loop control.

[0161] The data flow and information triggering logic between the four steps in the entire closed-loop control method have clear dependencies and temporal relationships, forming a complete cyclical operation process.

[0162] In the S1 stage, a multi-scenario feature library is built by collecting data from multiple scenarios to provide basic data support for subsequent model building. The generated cross-scenario oxygen-carbon-process parameter dynamic coupling model provides the model basis for optimization decision-making in the S3 stage. This stage is the initialization and model building link of the entire closed loop. Its output model parameters and scenario feature library data continuously provide static basic data for subsequent stages.

[0163] Entering the S2 stage, the start of this stage depends on the model framework built in the S1 stage. By collecting physical system data in real time, on the one hand, a digital twin synchronized with the physical system is built to achieve real-time mapping of the physical state. On the other hand, based on the digital twin, operating condition fluctuation prediction and dynamic carbon flow calculation are performed. The output of this stage includes real-time oxygen and carbon data, process parameter data, predicted fluctuation trends, and dynamic carbon flow intensity sequences. These data serve as the core inputs of the S3 stage, triggering the optimization decision-making process of the S3 stage. The relationship between S2 and S3 is a real-time data-driven triggering relationship, and the data update cycle is consistent with the decision cycle of S3.

[0164] After receiving real-time input data from S2, the S3 stage constructs a multi-objective optimization function based on the coupled model built in S1 and solves for the optimal control strategy. The generated control command serves as the input to the S4 stage, triggering the execution process. This stage is the decision-making stage, and its output control command directly determines the execution action of the S4 stage.

[0165] After the S4 stage executes the control command, feedback data is collected through the sensor network. This feedback data is sent to the digital twin mapping module to update the state of the digital twin constructed in the S2 stage, so as to realize the synchronization between the digital twin and the physical system. On the other hand, it is sent to the scene adaptive modeling module to trigger the online parameter self-tuning algorithm in the S1 stage to correct the parameters of the coupled model and complete the dynamic update of the model.

[0166] The corrected model parameters and the updated digital twin provide more accurate support for the optimization decision-making in the S3 stage and the real-time mapping in the S2 stage, respectively. This forms a complete closed loop from data acquisition, model building, simulation prediction, optimization decision-making to execution feedback. The inputs and outputs of each link are closely interdependent, and the time sequence is cyclical, following the sequence of "data acquisition - model update - simulation prediction - decision execution - feedback correction", ensuring the real-time performance and accuracy of the entire control system.

[0167] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A closed-loop control system based on a coupled model of oxygen, carbon, and process parameters, characterized in that, include: The module includes a scene adaptive modeling module, a digital twin mapping module, a multi-objective collaborative decision-making module, and an execution and feedback module. The scenario adaptive modeling module is used to construct a cross-scenario oxygen-carbon process parameter dynamic coupling model based on a scenario feature library and an online parameter self-tuning algorithm. The digital twin mapping module is electrically connected to the scene adaptive modeling module, and is used to map the physical system state in real time, predict operating condition fluctuations, and provide real-time input to the coupled model. The multi-objective collaborative decision-making module is electrically connected to the digital twin mapping module and is used to output the optimal control command by combining the improved multi-objective optimization algorithm with real-time carbon flow data. The execution and feedback module is electrically connected to the multi-objective collaborative decision-making module and is used to execute control commands and feed back real-time parameters to the scene adaptive modeling module.

2. The closed-loop control system based on the oxygen-carbon and process parameter coupling model according to claim 1, characterized in that, The method for constructing the scene feature library is as follows: Based on oxygen and carbon reaction data from cross-industry industrial scenarios, typical process parameter sequences are obtained, and oxygen and carbon concentration, temperature, pressure, and material flow data are collected in real time. Outlier cleaning and normalization processing are performed on these data. Set a baseline scenario feature template and extract the process parameter range, oxygen-carbon reaction mechanism threshold, and typical interference factor types of the baseline scenario; The process parameter sequences of each scenario are matched and analyzed with the benchmark feature template to extract the feature difference and generate scenario feature identifiers. Calculate the distance metric between each scenario and the baseline scenario in terms of oxygen and carbon response characteristics, and perform feature clustering in combination with the mechanism knowledge base to form a set of scenario feature classifications; By integrating scene feature identifiers, feature difference vectors, and clustering results, a structured multi-scene feature library is constructed.

3. The closed-loop control system based on the oxygen-carbon and process parameter coupling model according to claim 2, characterized in that, The online parameter self-tuning algorithm is specifically as follows: Based on real-time collected data on oxygen and carbon concentration, temperature, pressure, and material flow rate, a dynamic input vector is constructed and standardized. Initialize the model weight parameters, set the initial value of the forgetting factor and the update step size; An improved forgetting factor recursive least squares method is used to calculate parameter correction based on the real-time input and output deviation and update the model weights. An adaptive forgetting factor adjustment mechanism is introduced to adjust the size of the forgetting factor based on the rate of change of the scene and the history of prediction error. The updated model weights are output and synchronized to the scene feature library to complete the parameter self-tuning closed loop.

4. The closed-loop control system based on the oxygen-carbon and process parameter coupling model according to claim 3, characterized in that, The specific steps for constructing a cross-scenario dynamic coupling model of oxygen and carbon with process parameters are as follows: Based on the scene feature library, the feature identifiers and process parameter ranges corresponding to the current working conditions are matched, and the pre-set oxygen and carbon reaction mechanism equations are called to construct a basic oxygen and carbon-process parameter dynamic coupling model framework. Initialize model parameters by setting initial model weights and dynamic adjustment ranges based on the typical process parameter ranges and oxygen-carbon reaction mechanism thresholds in the scenario feature library. An online parameter self-tuning algorithm is introduced, which inputs real-time collected data on oxygen and carbon concentration, temperature, pressure and material flow rate into the model, and dynamically updates the model weight parameters through recursive calculation; Establish a dynamic feedback structure for the coupled model, compare the model output with the actual oxygen and carbon state of the physical system, and correct the model structure parameters and response relationship in real time based on the deviation; An integrated scene adaptation layer is used to perform scene-specific correction on the model output based on scene classification and difference vectors in the feature library, generating a dynamic coupling model of oxygen and carbon-process parameters that can be used across scenes.

5. The closed-loop control system based on the oxygen-carbon and process parameter coupling model according to claim 4, characterized in that, The specific state of the real-time mapped physical system is as follows: Real-time acquisition of oxygen and carbon concentration data, process parameter data, and equipment status data at various monitoring points in the physical system to form a multi-source heterogeneous real-time data stream; Based on real-time data streams, a digital twin is constructed that is consistent with the topology of the physical system. The digital twin includes an oxygen and carbon distribution layer, a process parameter layer, and an equipment status layer, and dynamic relationships between the layers are established. The real-time data stream of the physical system is mapped to the corresponding layer of the digital twin through the time-series data synchronization engine. A real-time carbon flow calculation unit is embedded in the digital twin. Based on oxygen and carbon distribution data and material flow topology, the carbon flow intensity of each node is calculated and updated in real time to generate a dynamic carbon flow map. The updated digital twin data is output to the scene adaptive modeling module in real time.

6. The closed-loop control system based on the oxygen-carbon and process parameter coupling model according to claim 5, characterized in that, The method for predicting operating condition fluctuations and self-calibrating the model based on the digital twin mapping module is as follows: A dynamic disturbance prediction unit is integrated into the digital twin. Based on historical operating data and the current real-time status, a time-series prediction model is used to predict the fluctuation trend of oxygen and carbon concentration and the range of process parameter changes within a future preset time window. The dynamic disturbance prediction unit uses a reinforcement learning algorithm to generate a set of pre-control strategies for potential operating condition fluctuations based on the prediction results and historical control effects. The set of pre-regulation strategies and the predicted fluctuation trend are sent to the multi-objective collaborative decision-making module as the basis for forward-looking regulation decisions. A model self-calibration unit is set up in the digital twin to compare the measured data of the physical system with the output data of the digital twin in real time and calculate the deviation index between the two. When the deviation index exceeds the set threshold, the parameter adaptive correction mechanism is triggered. Based on the deviation sequence, the state mapping relationship and prediction model parameters in the digital twin are dynamically adjusted, and the corrected parameters are synchronized to the scene adaptive modeling module.

7. The closed-loop control system based on the oxygen-carbon and process parameter coupling model according to claim 6, characterized in that, The specific method for generating the optimal control command is as follows: Receive real-time oxygen and carbon status data, process parameter data, predicted operating condition fluctuation trends, and dynamic carbon flow intensity sequences from the digital twin mapping module; A multi-objective optimization function is constructed, with the core optimization objective set as minimizing the oxygen-carbon ratio deviation, and the constraints set as the upper limit of carbon flow intensity, the industry benchmark value of energy consumption, and the fluctuation range of equipment operating load. An improved multi-objective optimization algorithm is used to solve the core objective and the carbon flow constraint conditions in a coordinated manner to generate a preliminary optimized solution set.

8. The closed-loop control system based on the oxygen-carbon and process parameter coupling model according to claim 7, characterized in that, The steps for outputting the optimal control command are as follows: Based on the scene classification identifiers in the scene feature library, dynamic weights are assigned to different objectives in the multi-objective optimization function; The digital twin constructed using the aforementioned digital twin mapping module is used to virtually execute and verify the multi-dimensional effects of the strategies in the preliminary optimized solution set. Based on the verification results, the control strategy with the best overall performance is selected from the preliminary optimized solution set, and the final control command is generated and output.

9. The closed-loop control system based on the oxygen-carbon and process parameter coupling model according to claim 8, characterized in that, The method for executing control commands and feeding back real-time parameters to the scene adaptive modeling module is as follows: Receive the final control instruction from the multi-objective collaborative decision-making module, and parse the specific actuator action type, adjustment range, and execution sequence contained in the instruction; The parsed action commands are converted into control signals for the underlying physical actuators and sent synchronously to the corresponding regulating valves, frequency converters, heaters or material conveying devices. During execution, the action status and response feedback of each actuator are monitored in real time, and dynamic coordination is performed to prevent execution conflicts or overshoot; The actual system control quantity after execution is recorded and fed back to the digital twin mapping module for updating the state of the digital twin; After the control command is executed, the updated oxygen and carbon concentration data, process parameter data, and equipment operating status data are collected in real time through the sensor network deployed in the physical system. The collected feedback data is time-aligned and noise-filtered to form a structured feedback data packet; The feedback data packet is synchronously sent to the digital twin mapping module for updating the twin's state, and simultaneously sent to the scene adaptive modeling module; Based on the received feedback data, the scenario adaptive modeling module triggers the online parameter self-tuning algorithm to update the weights and correct the parameters of the oxygen-carbon process parameter dynamic coupling model, thereby completing closed-loop control.

10. A closed-loop control method based on a coupled model of oxygen, carbon, and process parameters, characterized in that, Includes the following steps: S1. Collect process parameters and oxygen-carbon reaction data from different industrial scenarios, construct a multi-scenario feature library, and establish a cross-scenario applicable dynamic coupling model of oxygen-carbon and process parameters through parameter self-tuning algorithm. S2. Based on real-time collected oxygen and carbon concentration, process parameters and equipment status data, construct a digital twin synchronized with the physical system to predict future operating condition fluctuations and calculate dynamic carbon flow intensity. S3. Combining real-time oxygen and carbon data, predicted fluctuation trends and carbon flow intensity, a multi-objective optimization function is constructed with minimizing the oxygen-carbon ratio deviation as the core and carbon flow and energy consumption as constraints. The optimal control strategy is solved through optimization algorithms. S4. Parse and execute control commands, collect system feedback data in real time, update the state of the digital twin and correct the parameters of the dynamic coupling model to form a closed-loop control.

Images