Roasting-carbon emission collaborative management and control method and system based on artificial intelligence

Abstract

The invention provides a roasting-carbon emission collaborative management and control method and system based on artificial intelligence. The method comprises the following steps: constructing a structured tensor; constructing a target variable vector according to the temperature values of the plurality of thermocouple measuring points at the current time point and the concentration at the current time point; inputting the structured tensor and the target variable vector into a pre-constructed time sequence prediction model to generate predicted values of a product quality index and a carbon emission index; according to the predicted values of the product quality index and the carbon emission index and the target variable vector, measuring the overall deviation from the current system predicted behavior to the optimal control state, and determining the suggested adjustment direction at the current stage; and based on the predicted values of the product quality index and the carbon emission index and the adjustable variable set, constructing a target control optimization function, generating a suggested value of a specific control parameter, and outputting the suggested value through an execution interface. The low-carbon intelligent roasting control requirement under the conditions of multiple furnace types and multiple working conditions can be met.

Description

Technical Field

[0001] This invention belongs to the technical field of artificial intelligence, and in particular relates to a method and system for coordinated management and control of roasting and carbon emissions based on artificial intelligence. Background Technology

[0002] The roasting process of carbon products is a crucial heat treatment step in the production process, which has a decisive impact on the product's bulk density, resistivity, mechanical strength, and electrical conductivity. In actual industrial production, the roasting process involves long-term high-temperature heating, complex gas flow and combustion reactions, resulting in a system with high thermal inertia, strong process nonlinearity, and complex parameter coupling. High-precision, dynamic, and coordinated control is particularly needed in areas such as temperature field distribution, gas flow rate, air ratio, and flue gas emissions. However, current roasting process control generally relies on manual experience and preset heating curves, lacking the ability to model and intelligently adjust the dynamic coupling relationship between furnace temperature changes, combustion efficiency, and carbon emissions. This leads to lag in process adjustment and significant fluctuations in product performance. Existing carbon emission monitoring in roasting processes mainly relies on backend data collection and post-process statistics, making real-time prediction and dynamic adjustment difficult, and failing to establish a direct link between emission fluctuations and changes in process parameters. This limits the initiative for low-carbon production and often makes it difficult to quickly trace the source of carbon emission anomalies, making it difficult to determine whether the problem originates from gas supply, damper regulation, or temperature control, thus resulting in long-term passive corrections. Although some studies have attempted to introduce data-driven modeling methods or expert knowledge systems, problems such as poor adaptability to complex working conditions, long response delays, and lack of interpretability of results still exist. A complete closed-loop control system capable of multi-objective prediction, variable response identification, and control command generation has not yet been established.

[0003] Therefore, there is an urgent need for an intelligent system that can integrate multi-source data sensing, temperature-carbon emission co-prediction, control variable identification, and process optimization adjustment to target the dynamic characteristics of the roasting process, so as to achieve proactive control and continuous optimization of carbon emissions while ensuring stable roasting quality. Summary of the Invention

[0004] The purpose of this invention is to propose a method and system for coordinated management of roasting and carbon emissions based on artificial intelligence, in order to solve the above-mentioned problems.

[0005] To achieve the above objectives, a method for coordinated management of roasting and carbon emissions based on artificial intelligence is provided in a first aspect of the present invention, comprising the following steps:

[0006] S1: Obtain the temperature values ​​of multiple thermocouple measuring points at the current time point, and the current time point... Concentration, gas flow rate signal, damper opening signal, and current process stage are used to construct a structured tensor;

[0007] S2: Based on the temperature values ​​of multiple thermocouple measuring points at the current time point and the current time point... Concentration is used to construct a target variable vector; the target variable vector includes product quality indicators and carbon emission indicators.

[0008] S3: Input the structured tensor and the target variable vector into the pre-built time series prediction model to generate predicted values ​​for product quality indicators and carbon emission indicators; wherein, the time series prediction model includes a regularization enhancement term constructed based on the derivatives of the control physical variables;

[0009] S4: Based on the predicted values ​​of the product quality indicators and carbon emission indicators, as well as the target variable vector, measure the overall deviation of the current system's predicted behavior from the optimal control state, determine the suggested adjustment direction for the current stage, and generate a set of adjustable variables; wherein, each variable in the set of adjustable variables will be labeled with a direction indicator;

[0010] S5: Based on the predicted values ​​of the product quality indicators and carbon emission indicators, and the set of adjustable variables, construct the target control optimization function, generate suggested values ​​for specific control parameters, and output them through the execution interface.

[0011] Furthermore, the current process stage includes a heating, holding, or cooling stage;

[0012] The product quality index is the absolute deviation between the mean value of multiple thermocouple measuring points and the target value, used to characterize whether the current temperature is close to the final temperature target set by the process; the carbon emission index is the carbon emission per unit time. The rate of change of concentration is used to reflect the emission stability of the system under the current control input.

[0013] Furthermore, the main structure of the input pre-built time series prediction model includes: a gated recurrent neural network with residual connections, a stage embedding layer, and a dual-branch output head structure;

[0014] The gated recurrent neural network with residual connections is used to extract time series features;

[0015] The first-stage embedding layer encodes the current process stage into an embedding vector and concatenates it with other inputs to represent the control mechanism of different stages; wherein, the current process stage is fed into the embedding layer after being encoded in 3D one-hot encoding, mapped into a continuous vector of dimension 4, and concatenated with the input of each time step;

[0016] The dual-branch output head structure is used to fit the predicted values ​​of product quality indicators and carbon emission indicators, respectively.

[0017] Furthermore, the pre-built time series prediction model is based on the mean squared error loss function, and a regularization enhancement term is added to the loss as a penalty factor for joint optimization. The training data comes from three consecutive months of data from the structured tensor. Each set of data contains the running curves of multiple furnaces, and missing values ​​have been cleaned and the time axis has been unified.

[0018] Furthermore, the regularization enhancement term constructed based on the derivatives of the controlled physical variables is generated based on the gas flow signal and the damper opening signal.

[0019] Furthermore, the measurement of the overall deviation of the current system's predicted behavior from the optimal control state is obtained based on the predicted values ​​of product quality indicators and carbon emission indicators, as well as a disturbance penalty term; the disturbance penalty term is used to penalize combinations of variables that are not suitable for drastic adjustment at the current stage.

[0020] The disturbance penalty term is generated by an exponential function based on the gas flow signal and the damper opening signal.

[0021] Furthermore, the determination of the recommended adjustment direction for the current stage is achieved through a numerical perturbation method, specifically as follows:

[0022] For the gas flow signal and damper opening signal, positive and negative disturbances are introduced respectively. The disturbed variables are then fed into the pre-constructed time-series prediction model to recalculate the predicted values ​​of the positive and negative target variable vectors, and the corresponding positive and negative deviations are calculated accordingly. The logic for determining the adjustment direction is as follows:

[0023] If the positive deviation is less than the overall deviation, it is recommended to adjust the variable upwards;

[0024] If the negative deviation is less than the overall deviation, it is recommended to adjust the variable downwards;

[0025] If neither of these measures improves the situation or the effect is minimal, then adjustment is not recommended.

[0026] Furthermore, the construction of the target control optimization function based on the predicted values ​​of the product quality indicators and carbon emission indicators, and the set of adjustable variables, specifically includes:

[0027] The set of adjustable variables is obtained based on structured tensors and the adjustment is performed to generate the adjustment amount and the adjusted target variable vector. These are then input into a pre-built time series prediction model to obtain the re-estimated prediction value.

[0028] Combining the aforementioned adjustment amount, the re-estimated predicted value, and the adjustment constraint function, a target control optimization function is constructed; wherein, the adjustment constraint function is used to incorporate the operational constraints of the roasting process into the determination logic of the adjustment direction, specifically:

[0029] During the heat preservation stage, adjusting the gas flow will trigger a high-weight penalty to prevent the thermal field stability from being disrupted; during the heating stage, adjusting the damper will be restricted to avoid interfering with combustion stability.

[0030] Furthermore, when solving for the optimization objective, the system determines the step size coefficient of each variable based on the current overall deviation magnitude, the direction of independent variable adjustment, and the historical operation amplitude.

[0031] Based on the step size coefficient and direction indicator, the suggested values ​​of each control variable are linearly corrected to generate suggested values ​​for specific control parameters.

[0032] A second aspect of the present invention provides an artificial intelligence-based roasting-carbon emission coordinated management system, the system comprising:

[0033] The data acquisition and processing module is used to acquire the temperature values ​​of multiple thermocouple measuring points at the current time point, and the current time point... Concentration, gas flow rate signal, damper opening signal, and current process stage are used to construct a structured tensor;

[0034] The collaborative target construction module is used to construct targets based on the temperature values ​​of multiple thermocouple measuring points at the current time point and the current time point. Concentration is used to construct a target variable vector; the target variable vector includes product quality indicators and carbon emission indicators.

[0035] The temperature-carbon trend joint prediction module is used to input the structured tensor and the target variable vector into a pre-built time series prediction model to generate predicted values ​​of product quality indicators and carbon emission indicators; wherein, the time series prediction model includes a regularization enhancement term constructed based on the derivative of the control physical variables;

[0036] The control variable response identification module is used to measure the overall deviation of the current system's predicted behavior from the optimal control state based on the predicted values ​​of the product quality indicators and carbon emission indicators and the target variable vector, and to determine the suggested adjustment direction for the current stage, generating a set of adjustable variables; wherein, each variable in the set of adjustable variables is labeled with a direction marker.

[0037] The control suggestion generation and output module is used to construct a target control optimization function based on the predicted values ​​of the product quality indicators and carbon emission indicators, and the set of adjustable variables, generate suggested values ​​for specific control parameters, and output them through the execution interface.

[0038] The beneficial technical effects of the present invention are at least as follows:

[0039] This system establishes a multi-source data structure including furnace temperature, gas flow rate, damper opening, and flue gas concentration, and constructs a collaborative control target system with calcination quality indicators and carbon emission trend indicators as the core. It also uses a dual-output time-series prediction model that embeds process stage information to achieve joint prediction of temperature and carbon trends in future calcination states.

[0040] Based on this, the system uses a variable response identification mechanism, disturbance sensitivity analysis and phased regularization term design to quantify the degree and direction of the influence of each control variable on the prediction results, thereby forming a set of control variables with physical constraints.

[0041] Subsequently, by combining prediction bias, control smoothness and process stage constraints, the system automatically generates optimal adjustment suggestions for key variables such as gas flow and damper opening under a multi-objective optimization framework, and outputs them through a visualization interface or PLC communication module to achieve closed-loop control from prediction to execution.

[0042] The innovation of this invention lies in proposing an overall scheme that integrates temperature-carbon collaborative modeling, responsive variable identification, and constraint optimization in the roasting process. It constructs an intelligent control structure with multi-stage adaptive capabilities, which can simultaneously ensure the dynamic balance between product quality and emission control. It features high practicality and engineering deployment, and can meet the low-carbon intelligent roasting control requirements under multiple furnace types and operating conditions. Attached Figure Description

[0043] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0044] Figure 1 This is a schematic diagram of the steps of the roasting-carbon emission synergistic control method based on artificial intelligence of the present invention. Detailed Implementation

[0045] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0046] In one or more embodiments, such as Figure 1 As shown, an artificial intelligence-based method for coordinated management of roasting and carbon emissions is disclosed, the method comprising the following steps:

[0047] S1: Obtain the temperature values ​​of multiple thermocouple measuring points at the current time point, and the current time point... Concentration, gas flow rate signal, damper opening signal, and current process stage are used to construct a structured tensor.

[0048] Among them, variables This indicates the temperature values ​​of multiple thermocouple measuring points at the current time. The data comes from the thermocouple array deployed in the roasting furnace, with a collection period of 30 seconds. The data is collected and uploaded via PLC analog signal acquisition. Indicates the current time point The concentration data comes from an infrared analyzer installed on the flue gas main, and is uploaded in real time via Modbus communication, with a sampling cycle of 5 minutes. This is a gas flow signal. The damper opening signal is transmitted back to the system by the corresponding field instruments or actuators. The current process stage is determined by the status output of the PLC program, indicating whether the current stage is heating, holding, or cooling.

[0049] S2: Based on the temperature values ​​of multiple thermocouple measuring points at the current time point and the current time point... Concentration is used to construct a target variable vector; the target variable vector includes product quality indicators and carbon emission indicators.

[0050] Specifically, this step involves constructing a collaborative control target variable for the roasting process to reflect the integrated control state of product quality and carbon emission levels. This target will be used for subsequent model prediction training and control adjustment suggestion generation, therefore requiring computability, interpretability, and physical consistency. In the roasting process, product quality is mainly achieved through furnace temperature control, while carbon emissions are mainly controlled through gas combustion and air supply regulation; these two aspects are coupled during both the heating and holding stages. Therefore, this step is based on the structured tensor obtained in the previous step. Construct a target vector consisting of two variables. , representing product quality indicators and carbon emission trend indicators, respectively.

[0051] Among them, product quality indicators To characterize whether the current temperature is close to the final temperature target set by the process, the absolute deviation between the average temperature of multiple measurement points and the target value is defined as:

[0052] ;

[0053] in, Indicates the first Normalized temperature values ​​at each measuring point The target temperature for normalization is... The total number of measuring points (e.g.) The range of values ​​for this indicator is: The closer the value is to 1, the closer the furnace temperature is to the set target, and the more stable the control is.

[0054] Carbon emission indicators Then it is defined as a unit of time. The rate of change of concentration is used to reflect the emission stability of the system under the current control input:

[0055] ;

[0056] in, For the current moment Concentration (normalized value) The concentration value at the previous moment. The interval between two sampling time points (in minutes, usually 1 minute). ).when This indicates a downward trend in carbon emissions. This indicates that emission intensity is increasing, reflecting incomplete combustion or system disturbances.

[0057] The final output target variable vector is , representing the current furnace temperature compliance level and carbon emission change rate, respectively. Subsequent steps will be based on this target vector to predict trends and design control strategies.

[0058] S3: Input the structured tensor and the target variable vector into the pre-built time series prediction model to generate predicted values ​​for product quality indicators and carbon emission indicators; wherein, the time series prediction model includes a regularization enhancement term constructed based on the derivatives of the control physical variables.

[0059] Specifically, this step involves constructing a trend prediction model for practical problems in the roasting process, used to jointly predict and coordinate control objectives. The evolution over a future period. Temperature data already collected in the previous step using the furnace thermocouple array. And the data collected by the flue gas infrared analyzer concentration Constructed (Roasting quality standards) and (Carbon emission trend indicators). These two indicators reflect whether the current roasting system is in an ideal state of "thermal field compliance" and "emission stability". To achieve the goal of early intervention, it is necessary to... and The short-term trend is predicted to serve as a direct basis for generating subsequent control and adjustment recommendations.

[0060] The input data used in this step comes from the previous structured tensor. These data are updated every 5 minutes and include: furnace temperature sequence. (Data collected by deploying a thermocouple array) concentration (Data collected via NDIR infrared analyzer) Gas flow rate (Measured by a vortex flow meter), damper opening (Obtained from feedback from the electric actuator), and the current process stage code. (Provided by the PLC state machine). These signals are normalized and then combined into a fixed-length input sequence using a time sliding window method.

[0061] Considering the following unique characteristics of the roasting process:

[0062] The operating conditions differ significantly between the stages (e.g., the direction of change of the control target is different in the heating and heat preservation stages).

[0063] Temperature control suffers from thermal inertia, nonlinearity, and local disturbances (such as heat loss from the furnace wall).

[0064] The relationship between flue gas emissions and the gas-air volume combination is nonlinear and lag-dependent;

[0065] When multiple furnaces operate in parallel, the operating status of each furnace will affect the local pressure and gas flow direction (this model is temporarily treated as a single furnace).

[0066] Therefore, standard neural network architectures cannot be used simply for time series forecasting. Instead, the following should be introduced into the structural design: stage condition modeling mechanism; collaborative structural design of target variables; and control terms to enhance the consistency of physical constraints.

[0067] This step employs a set of dual-output time series prediction models, based on the input sequence. Predicting future moments of and The main structure of the model consists of three parts:

[0068] A single gated recurrent neural network (GRU) with residual connections is used to extract time series features.

[0069] A stage embedding layer will embed the current process stage. The encoding is concatenated with other inputs to represent the control mechanisms at different stages;

[0070] A dual-branch output head structure, respectively fitted and However, through shared feature representation Establish collaborative relationships.

[0071] The output of the joint prediction model is calculated as follows:

[0072] ;

[0073] in, It is the feature vector output by the GRU module. It is a two-layer fully connected network that outputs two predicted values. It is a regularization enhancement term constructed based on the derivatives of controlled physical variables. It is an adjustable coefficient (usually in) ).

[0074] Regular terms The design of this step is one of the key innovations. It considers the actual rate of impact of changes in control input on the system response, avoiding the problem of stable model output but drastic fluctuations in control input. This item is defined as:

[0075] ;

[0076] in, This represents the rate of change of gas flow rate per unit time. The rate of change of the damper opening is represented by dividing the difference between data points at adjacent time points by the sampling period. For example, if , If the sampling period is 5 minutes, then this item is The introduction of regularization terms suppresses "prediction overfitting under jump adjustment," making the prediction results more valuable for practical implementation.

[0077] Current process stage Instead of simply using labels as input, the data is encoded in 3D one-hot encoding and then fed into the embedding layer, mapped to a continuous vector of dimension 4, which is then concatenated with the input at each time step. This approach allows the model to automatically learn different feature selection mechanisms at different stages. For example, during the heating stage, more attention is paid to the impact of airflow on carbon emissions, while during the heat preservation stage, more attention is paid to the impact of temperature deviation on mass.

[0078] The entire model training was based on the mean squared error loss function, with a regularization term added as a penalty factor for joint optimization. The training data came from three consecutive months of data from the factory's roasting workshop. Each data set contained operating curves from multiple furnace runs, with missing values ​​removed and the time axis standardized. The final output was... , indicating the predicted time in the future (Can be set to 5~15 minutes) Predicted values ​​of the quality compliance of the roasting system and the trend of carbon emission changes. This result will be directly used in the next step of the control variable identification and adjustment suggestion generation logic.

[0079] S4: Based on the predicted values ​​of the product quality indicators and carbon emission indicators, as well as the target variable vector, measure the overall deviation of the current system's predicted behavior from the optimal control state, determine the suggested adjustment direction for the current stage, and generate a set of adjustable variables; wherein, each variable in the set of adjustable variables will be labeled with a direction marker.

[0080] Specifically, the task of this step is to predict the results. Based on this, and combined with the current working conditions, input tensors The process involves identifying which controllable variables are most sensitive to changes in the prediction results and determining their adjustment direction. This process acts as a bridge in this invention—it connects the previous "temperature-carbon synergistic prediction" results with the next "process parameter optimization" logic, preventing the model from simply predicting without interpretation, and further preventing the generation of control recommendations from falling into a state of "blind tuning." This is particularly important in the roasting process scenario, where there are fewer control variables but their impact is complex, such as gas flow rate. Japanese-style door opening Simultaneously affects temperature distribution and Emissions control processes are highly nonlinear, so we cannot rely on linear sensitivity or static correlation indicators to judge the role of variables. We need to design a reproducible and responsive identification and direction determination method based on the dynamic structure of the prediction model.

[0081] This step takes two core variables as input: one is the predicted output. Second, the historical control sequence. In a physical sense, A value close to 1 indicates that the final temperature meets the standard and the thermal field is uniform. A value close to 0 indicates stable carbon emissions with no abnormal changes. Therefore, the goal is to maximize emissions reduction. and reduce To achieve this goal, a joint deviation evaluation function is defined in this step. To measure the overall deviation of the current system's predicted behavior from the optimal control state:

[0082] ;

[0083] in, These are the weighting coefficients for the predicted targets, which can be dynamically adjusted according to the process stage to meet the requirements. For example, setting during the heating phase Strengthen emission control and set up [measures] during the insulation stage. Enhanced temperature stability; It is the weight of the additional regularization factor; It is a perturbation penalty term designed for different stages of roasting, used to penalize combinations of variables that should not be drastically adjusted in the current stage. This term is an original design made in this step in combination with the process characteristics of roasting.

[0084] The format is as follows:

[0085] ;

[0086] in, This is an indicator function used to dynamically activate different adjustment and penalty mechanisms at different stages. For example, during the heat preservation stage (Stage=2), significant adjustments to the gas flow rate should be avoided, and the system should primarily focus on maintaining a stable furnace temperature. Therefore, the following adjustments should be made to the gas flow rate: Introducing penalties; conversely, during the heating phase, greater emphasis is placed on air supply stability. Enhanced control. This regularization term is engineering interpretable, embedding adjustment constraints in the calcination process, and providing a reasonable boundary for subsequent parameter optimization.

[0087] Variable response identification is achieved through numerical perturbation. For adjustable parameters... and Positive disturbances were introduced respectively. With negative disturbance (For example The perturbated variables are then fed back into the original prediction model to recalculate the predicted values ​​of the positive target variable vector. Predicted values ​​of the negative target variable vector Therefore, the corresponding positive deviation is calculated. With negative deviation Adjust direction The judgment logic is as follows:

[0088] like If so, it is recommended to increase the value of this variable;

[0089] like If so, a downward adjustment is recommended;

[0090] If neither of these measures improves the situation or the effect is minimal, then adjustment is not recommended.

[0091] This mechanism allows for the construction of a set of adjustment variables:

[0092] ;

[0093] Each variable They will all be marked with a direction sign. This indicates the recommended adjustment direction for the current stage.

[0094] Here's a real-world example: Data was collected during a certain insulation stage. , , , Initial deviation function .right The prediction after increasing by 2% After a 2% reduction ,but This means that a slight reduction in gas flow rate would be beneficial in reducing emissions. If changes in airflow direction are not significant, then... The current opening degree will remain unchanged.

[0095] S5: Based on the predicted values ​​of the product quality indicators and carbon emission indicators, and the set of adjustable variables, construct the target control optimization function, generate suggested values ​​for specific control parameters, and output them through the execution interface.

[0096] Specifically, this step aims to integrate the set of adjustable variables identified in the previous step. Combined with the corresponding collaborative prediction output This is transformed into a set of actionable control parameter recommendations. The data is then output via an interface for operator decision-making or sent to the control system for execution. This process not only considers prediction deviations but also fully integrates the physical constraints, safety boundaries, and actual adjustability of control variables during the roasting stage to ensure that the generated parameter recommendations achieve a dynamic balance between quality objectives, carbon emission constraints, and process stability. This step is the final link in the control chain of this invention, and its reproducibility and industrial adaptability are crucial to the successful implementation of the entire solution.

[0097] The input data comes from the industrial field control system and model output. Each variable in The adjustable controllable quantities of the current roasting system mainly include the gas flow rate. (Data collected via a flow meter on the main gas pipeline, in Nm³ / h) and damper opening. (Real-time angle signal fed back by the electric actuator, in percentage). These signals are acquired by the PLC and refreshed every minute; while the predicted output... and The dual-output model, constructed in the third step, generates values ​​representing the compliance rate of the calcination thermal field and the trend of carbon emission changes, respectively. Adjustment direction. The perturbation analysis results from the fourth step, with values ​​of -1 (downward adjustment), 0 (no adjustment), and +1 (upward adjustment), provide a qualitative description of the current trend of each variable's effect.

[0098] Furthermore, in order to integrate this information into specific recommendations, this step first constructs a multi-objective control optimization function. This function takes into account prediction bias, control smoothness, and stage adjustment constraints, and is defined as follows:

[0099] ;

[0100] in, and Indicates the adjustment amount The recalculated prediction value under the influence of the action is obtained by feeding the adjusted input tensor into the trained prediction model; and The weights for quality and carbon emission control targets are respectively (and can be dynamically adjusted according to user strategies, such as increasing them during peak emission periods). ); It adjusts the smoothness penalty coefficient to prevent excessive parameter fluctuations and improve system stability; It is a penalty factor for limitations in the process stage. The stage-related adjustment constraint function is defined as follows:

[0101] ;

[0102] This function incorporates operational constraints of the roasting process into the control suggestion generation logic. During the heat preservation stage (Stage=2), adjusting the gas flow rate will trigger a high-weight penalty to prevent disruption of thermal field stability; during the heating stage (Stage=1), adjusting the damper is restricted to avoid interfering with combustion stability.

[0103] Furthermore, when solving for the optimization objective, the system determines the step size coefficient for each variable based on the current prediction deviation, the direction of independent variable adjustment, and the historical operational magnitude. , usually set in Interval. For example, when A value significantly higher than 0.1 indicates an excessively strong carbon emission trend, and the system automatically relaxes the upper limit of the allowable adjustment range. This is combined with directional judgment. The system performs linear corrections to the suggested values ​​for each control variable:

[0104] like ,but ;

[0105] like ,but ;

[0106] like ,but .

[0107] in The absolute range that the variable is allowed to adjust is configured by the process system, for example... Nm³ / h. For example, when , , , The recommended value is Nm³ / h. This adjustment range ensures a balance between system response speed and operational safety.

[0108] All suggested results Encapsulated into a standardized output structure, including variable names, suggested values, adjustment directions, and descriptions of expected impacts, this information is uniformly transmitted to two types of interface modules: one is the operator interface in the control room, which displays information through charts and text descriptions, such as "It is recommended to reduce the gas flow rate to 89 Nm³ / h, with an expected carbon emission reduction of 8%"; the other is an optional PLC control interface, which writes the recommended parameters into the control buffer area via industrial Ethernet or Modbus RTU communication, and executes them after manual confirmation, or automatically writes them according to permission policies. Each suggestion record is stamped with a timestamp, furnace number, and process stage marker for subsequent tracking and auditing. This step fully demonstrates the feasibility and creativity of roasting-carbon emission coordinated control in its structure and details. Its core innovations are reflected in the following aspects: First, a unified objective function. This integrates prediction accuracy, stability, and process stage logic to form a multi-level target fusion mechanism; secondly, it includes stage penalty terms. It is a control strategy specifically designed for the safety boundary of the roasting process, which can significantly reduce system oscillations caused by improper adjustment; furthermore, the output interface design takes into account the communication and hierarchical permission mode of industrial systems, so that the suggestions have flexible adaptability from auxiliary decision-making to automatic execution.

[0109] In one or more embodiments, another embodiment of the present invention provides an artificial intelligence-based roasting-carbon emission coordinated management and control system, the system comprising:

[0110] The data acquisition and processing module is used to acquire the temperature values ​​of multiple thermocouple measuring points at the current time point, and the current time point... Concentration, gas flow rate signal, damper opening signal, and current process stage are used to construct a structured tensor;

[0111] The collaborative target construction module is used to construct targets based on the temperature values ​​of multiple thermocouple measuring points at the current time point and the current time point. Concentration is used to construct a target variable vector; the target variable vector includes product quality indicators and carbon emission indicators.

[0112] The temperature-carbon trend joint prediction module is used to input the structured tensor and the target variable vector into a pre-built time series prediction model to generate predicted values ​​of product quality indicators and carbon emission indicators; wherein, the time series prediction model includes a regularization enhancement term constructed based on the derivative of the control physical variables;

[0113] The control variable response identification module is used to measure the overall deviation of the current system's predicted behavior from the optimal control state based on the predicted values ​​of the product quality indicators and carbon emission indicators and the target variable vector, and to determine the suggested adjustment direction for the current stage, generating a set of adjustable variables; wherein, each variable in the set of adjustable variables is labeled with a direction marker.

[0114] The control suggestion generation and output module is used to construct a target control optimization function based on the predicted values ​​of the product quality indicators and carbon emission indicators, and the set of adjustable variables, generate suggested values ​​for specific control parameters, and output them through the execution interface.

[0115] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0116] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0117] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

[0118] The various illustrative logic blocks, modules, and circuits described in conjunction with the embodiments disclosed herein can be implemented or performed using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. The general-purpose processor may be a microprocessor, but in alternatives, it may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors cooperating with a DSP core, or any other such configuration.

[0119] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of both. The software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor such that the processor can read and write information to / from the storage medium. In an alternative, the storage medium may be integrated into the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and storage medium may reside as discrete components in the user terminal.

[0120] In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functionality may be stored or transmitted as one or more instructions or code on or through a computer-readable medium. A computer-readable medium includes both computer storage media and communication media, encompassing any medium that facilitates the transfer of a computer program from one location to another. A storage medium may be any available medium accessible to a computer. By way of example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and is accessible to a computer. Any connection is also legitimately referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of a medium. As used in this article, disk and disc include compact discs (CDs), laser discs, optical discs, digital multi-purpose discs (DVDs), floppy disks, and Blu-ray discs. Disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of these should also be included within the scope of computer-readable media.

[0121] The prior description of this disclosure is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not intended to be limited to the examples and designs described herein, but should be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for collaborative control of roasting and carbon emission based on artificial intelligence, characterized in that, The method comprises the following steps: S1: Obtain the temperature values ​​of multiple thermocouple measuring points at the current time point, and the current time point... Concentration, gas flow rate signal, damper opening signal, and current process stage are used to construct a structured tensor; S2: constructing a target variable vector according to the temperature values of the multiple thermocouple measuring points at the current time point and the current time point a concentration; the target variable vector comprises a product quality index and a carbon emission index; S3: inputting the structured tensor and the target variable vector into a pre-constructed time series prediction model to generate predicted values of the product quality index and the carbon emission index; wherein the time series prediction model comprises a regularization enhancement term constructed based on derivatives of control physical variables; S4: measuring overall deviation of a current system prediction behavior from an optimal control state based on the predicted values of the product quality index and the carbon emission index and the target variable vector, determining a recommended adjustment direction for a current stage, and generating an adjustable variable set; wherein each variable in the adjustable variable set is marked with a direction flag; S5: constructing a target control optimization function based on the predicted values of the product quality index and the carbon emission index and the adjustable variable set to generate recommended values of specific control parameters, and outputting the recommended values through an execution interface. 2.The AI-based roasting-carbon emission collaborative management method according to claim 1, characterized in that, The current process stage comprises a heating stage, a holding stage or a cooling stage; The product quality index is the absolute deviation between the mean value of multiple thermocouple measuring points and the target value, used to characterize whether the current temperature is close to the final temperature target set by the process; the carbon emission index is the carbon emission per unit time. The rate of change of concentration is used to reflect the emission stability of the system under the current control input. 3.The AI-based roasting-carbon emission collaborative management method according to claim 1, characterized in that, The main structure of the input pre-constructed time series prediction model comprises a one-layer gated recurrent neural network with residual connection, a stage embedding layer and a double-branch output head structure; The one-layer gated recurrent neural network with residual connection is used to extract time series features; The stage embedding layer encodes the current process stage into an embedding vector, which is spliced with other inputs to represent control mechanisms in different stages; wherein the current process stage is sent to the embedding layer after 3D one-hot encoding, mapped into a continuous vector with a dimension of 4, and spliced with the input of each time step; The double-branch output head structure is used to fit the predicted values of the product quality index and the carbon emission index, respectively. 4.The AI-based roasting-carbon emission collaborative management method of claim 1, wherein, The pre-constructed time series prediction model is based on a mean square error loss function, and a regularization enhancement term is added as a penalty factor in the loss for joint optimization. The training data comes from continuous 3-month data of the structured tensor, and each group of data contains the running curves of multiple heats, with missing values cleaned up and the time axis unified. 5.The AI-based roasting-carbon emission collaborative management method of claim 1, wherein, The regularization enhancement term constructed based on derivatives of control physical variables is generated based on the gas flow signal and the damper opening signal. 6.The AI-based roasting-carbon emission collaborative management method according to claim 1, wherein, The overall deviation of the current system prediction behavior from the optimal control state is measured based on the predicted values of the product quality index and the carbon emission index and a disturbance penalty term; the disturbance penalty term is used to punish variable combinations that are not suitable for drastic adjustment in the current stage; The disturbance penalty term is generated based on the gas flow signal and the damper opening signal through an exponential function.

7. The roasting-carbon emission collaborative management method based on artificial intelligence according to claim 6, characterized in that, The determination of the recommended adjustment direction for the current stage is realized by a numerical perturbation method, specifically: For the gas flow signal and the damper opening signal, positive and negative perturbations are introduced respectively, and the perturbed variables are sent to the pre-constructed time series prediction model to recalculate the predicted values of the positive and negative target variable vectors, and the corresponding positive and negative deviations are calculated; wherein the determination logic of the adjustment direction is as follows: If the positive deviation is less than the overall deviation, the variable is recommended to be adjusted upward; If the negative deviation is less than the overall deviation, the variable is recommended to be adjusted downward; If neither of them is improved or the effect is very weak, no adjustment is recommended.

8. The roasting-carbon emission collaborative management method based on artificial intelligence according to any one of claims 2 or 7, characterized in that, The target control optimization function is constructed based on the predicted values of the product quality indicators and the carbon emission indicators, the adjustable variable set, and specifically includes: Based on the structured tensor, the adjustable variable set is obtained and adjusted to generate an adjustment amount and an adjusted target variable vector, which are input into a pre-constructed time series prediction model to obtain re-estimated predicted values; The target control optimization function is constructed in combination with the adjustment amount, the re-estimated predicted values, and an adjustment constraint function; wherein the adjustment constraint function is used to integrate the operation limits of the roasting process into the judgment logic of the adjustment direction, and specifically includes: In the holding stage, adjusting the gas flow will trigger a high-weight penalty to avoid the destruction of the thermal field stability; in the heating stage, the air door is limited to avoid interfering with the combustion stability. 9.The AI-based roasting-carbon emission collaborative management method of claim 8, wherein, When solving the optimization target, the system determines the step length coefficient of each variable based on the current overall deviation size, the independent variable adjustment direction, and the historical operation amplitude; Based on the step length coefficient and the direction flag, the suggested value of each control variable is linearly corrected to generate the suggested value of the specific control parameter.

10. The roasting-carbon emission collaborative management system based on artificial intelligence, characterized in that, The system includes: The data acquisition and processing module is used to acquire the temperature values ​​of multiple thermocouple measuring points at the current time point, and the current time point... Concentration, gas flow rate signal, damper opening signal, and current process stage are used to construct a structured tensor; The collaborative target construction module is used to construct targets based on the temperature values ​​of multiple thermocouple measuring points at the current time point and the current time point. Concentration is used to construct a target variable vector; the target variable vector includes product quality indicators and carbon emission indicators. A temperature-carbon trend joint prediction module for inputting the structured tensor and the target variable vector into a pre-constructed time series prediction model to generate predicted values of the product quality indicators and the carbon emission indicators; wherein the time series prediction model includes a regular enhancement term constructed based on the derivative of the control physical variable; A control variable response identification module for measuring the overall deviation of the current system prediction behavior from the optimal control state based on the predicted values of the product quality indicators and the carbon emission indicators and the target variable vector, determining the suggested adjustment direction of the current stage, and generating an adjustable variable set; wherein each variable of the adjustable variable set is labeled with a direction flag; A control suggestion generation and output module for constructing a target control optimization function based on the predicted values of the product quality indicators and the carbon emission indicators, the adjustable variable set, generating the suggested value of the specific control parameter, and outputting through an execution interface.

Images