Optimization control method and system for residual oxygen of steel rolling heating furnace

Abstract

The invention relates to the field of heating furnace monitoring and control, in particular to a steel rolling heating furnace residual oxygen optimization control method and system.The method comprises the steps that historical variables and future control variables of operation of a heating furnace are collected, collected data are preprocessed, and an input sequence and an output sequence are constructed; an encoder-decoder structure is adopted, space attention and time attention are integrated, multivariable coupling and time variation are captured, the residual oxygen content in a future time domain heating furnace is predicted, and a sequence-to-sequence Seq2Seq prediction model of a space-time dual attention mechanism is constructed; the prediction model is combined with a model prediction control method, a residual oxygen prediction sequence is used as input, a multi-objective optimization problem is constructed, under the constraint conditions of the air-fuel ratio, the flow, the temperature and the like, the optimal control sequence is solved, only the first control action is executed, rolling iteration is conducted, and prospective and high-precision closed-loop control over the residual oxygen content in the heating furnace is achieved. Global optimization is achieved, the oxidation burn-out rate of the steel billet is reduced, energy consumption is stabilized, and energy is saved.

Description

Technical Field

[0001] This invention relates to the field of heating furnace monitoring and control, specifically to a method and system for optimizing the control of residual oxygen in a steel rolling heating furnace. Background Technology

[0002] The steel rolling heating furnace is a key piece of equipment in the steel production process, its main function being to heat steel billets to the required rolling temperature. During the heating process, if the residual oxygen content in the furnace is too high, it will cause severe oxidation and burning of the steel billets, reducing the yield and also causing thicker iron oxide scale, affecting the quality of subsequent rolling. If the residual oxygen content in the furnace is too low, incomplete fuel combustion will occur, resulting in energy waste and increased emissions of harmful gases such as carbon monoxide. Therefore, accurately controlling the residual oxygen content in the furnace within a reasonable range is of great significance for ensuring steel quality, reducing energy consumption, and minimizing environmental pollution.

[0003] Whether it's a conventional continuous heating furnace or a regenerative heating furnace, the industry currently uses manual adjustment or PID control based on operator experience to control residual oxygen in the furnace. Manual adjustment suffers from problems such as untimely response and large fluctuations in control accuracy; PID control, as a classic feedback control strategy, is difficult to optimize the combustion process for complex industrial objects like heating furnaces, which have large time lags, multivariate coupling, and nonlinearity.

[0004] To address these issues, researchers in the field have proposed various automated control schemes, such as patents CN201410769027.7 and CN201610785843.6. These schemes adjust the air-fuel ratio or excess air coefficient based on the measured values ​​of the residual oxygen meter, essentially a feedback control method. This method only begins adjusting after the residual oxygen content deviates, exhibiting significant lag in its control behavior. Patent 202310592749.9 introduces feedforward control based on furnace condition judgment into the feedback control based on measured residual oxygen. It determines the current heat load by analyzing the difference between the real-time furnace temperature and the set furnace temperature, along with the rate of change in gas flow, calculating the air compensation coefficient, and pre-compensating the excess air coefficient. However, this feedforward control relies on logical judgments based on preset rules and thresholds, depending on manual setting and adjustment. When production processes or equipment characteristics change, these rules may become invalid, resulting in limited adaptability of this method.

[0005] In summary, existing technologies have evolved from simple step-by-step feedback to composite control methods that combine furnace condition assessment, achieving significant progress in automation. However, their common drawback lies in the lack of accurate, long-term prediction capabilities for the system's future dynamics. The control logic is essentially still based on the correction of current or past errors. For complex systems like steel rolling furnaces, which exhibit strong nonlinearity, time-varying characteristics, large time lags, and multivariate coupling, existing technologies cannot yet achieve effective predictive optimal control. Summary of the Invention

[0006] In view of this, the present invention aims to overcome the problems of response lag, low control accuracy, inability to handle multivariate coupling and time-varying characteristics in existing residual oxygen control technologies for heating furnaces, and to provide a method that can accurately predict the trend of residual oxygen content change in the furnace and perform optimal control based on this prediction.

[0007] To achieve the above objectives, this invention provides the following technical solution: A method for optimizing residual oxygen control in a steel rolling furnace is proposed. This method integrates deep learning and model predictive control. By constructing a sequence-to-sequence (Seq2Seq) prediction model with a spatiotemporal attention mechanism, and combining it with the model predictive control (MPC) method, a forward-looking, multi-input, and optimized precise closed-loop control of the residual oxygen content in the furnace is achieved. The core steps are as follows: Step 1: Acquisition and Serialization Construction of Multidimensional Time Series Data 1. Data collection To build a predictive model, it is necessary to periodically collect operating data from the heating furnace. The collected data includes data based on past time domain data. t h Historical data and coverage of future forecast time domain t p The future control variable.

[0008] Future control variables are for future prediction in the time domain. t p The control sequence within the model is the air-fuel ratio setpoint for each heating section. During model training, this sequence represents the actual control values ​​from historical data; in online applications of model predictive control, this sequence represents the candidate control sequences generated by the optimizer.

[0009] 2. Process and serialize the collected data. (1) The collected data is cleaned to remove outliers, missing values ​​are interpolated, and downsampling, time alignment and normalization are performed.

[0010] (2) Divide the preprocessed data into multiple sample pairs, each sample pair containing an input sequence and an output sequence. The input sequence consists of the aforementioned historical time domain. t h Historical data sequences (including state variable sequences, control variable sequences, and disturbance variable sequences) and future prediction time domain t p It is composed of a sequence of control variables within the given time domain. The output sequence corresponds to the future prediction time domain. t p The true residual oxygen content sequence.

[0011] Furthermore, the historical data sequence includes one or more of the following variables: ①Historical state variables: Past time domain t h The residual oxygen content, furnace temperature, and furnace pressure in each section.

[0012] ② Historical control variables: the same past time domain t h The set values ​​for fuel flow rate, air flow rate, and air-fuel ratio for each heating section within the unit.

[0013] ③ Historical perturbation variables: in the past time domain t h The steel type, billet size, furnace loading time, billet position, charging temperature, target temperature, etc.

[0014] Step 2: Construction and Training of a Sequence-to-Sequence (Seq2Seq) Residual Oxygen Prediction Model Based on Spatiotemporal Attention Mechanism 1. Model Building One of the core aspects of this invention is the construction of a high-precision residual oxygen prediction model. This model employs a sequence-to-sequence (Seq2Seq) model with an encoder-decoder structure and integrates a spatiotemporal dual attention mechanism. Its overall architecture includes: ① Encoder: It adopts gated recurrent unit (GRU) or long short-term memory network (LSTM). When it is working, it reads and encodes the historical data sequence and compresses it into a set of context hidden state vectors that can represent the past dynamics of the system.

[0015] ② Decoder: Also employs a gated recurrent unit (GRU) or a long short-term memory (LSTM) network. During operation, it gradually generates a prediction of the future time domain based on the context information provided by the encoder and the future sequence of control variables. t h Predicted value of residual oxygen content.

[0016] ③ Spatial attention: This mechanism applies to the input data at each time step. It dynamically assigns attention weights to multiple input variables through a feedforward neural network, enabling the model to automatically identify the key variables that have the greatest impact on the current residual oxygen content at any given time.

[0017] ④ Temporal attention: This mechanism is used in the encoder's prediction generation process. In predicting the future... k When calculating the residual oxygen value at a given step, the temporal attention mechanism reviews and evaluates the hidden states of all historical time steps output by the encoder, assigning weights to them. This allows the model to dynamically focus on past events in predicting future time steps. k This is one of the most impactful historical moments.

[0018] 2. Model Training The input and output sequence samples constructed in step one are input into the Seq2Seq residual oxygen prediction model based on the spatiotemporal attention mechanism described above. A loss function is set, and iterative training is performed through the backpropagation algorithm and optimizer. The network parameters inside the model are continuously adjusted until the difference between the predicted residual oxygen sequence and the real residual oxygen sequence is minimized.

[0019] Step 3: Model Predictive Control (MPC)-based optimization control of residual oxygen rolling in the heating furnace 1. State prediction In each control cycle Dt The beginning will be the past time domain t h The data sequence and a set of candidate future control sequences provided by the optimizer are input into a pre-trained Seq2Seq model. The model outputs a time-domain prediction of the future. t p Predicted value of residual oxygen content.

[0020] 2. Construct the objective function The core of model predictive control is to solve constrained finite-time optimization problems online.

[0021] Furthermore, the objective function J as follows:

[0022] in, O 2,pred To predict residual oxygen concentration, O 2,ref To specify the residual oxygen concentration, t is the current time step, and k is the number of time steps from the current time t towards future prediction or control. u Air-fuel ratio, w y and w u These are adjustable weighting coefficients. P To predict the time domain, M To control the time domain, satisfy M≤P; The first term of the formula is the sum of squared errors between the predicted output and the specified output, which serves as the tracking error term and aims to ensure the predicted residual oxygen concentration... O 2,pred As close as possible to the desired specified value O 2,ref This is used to measure the accuracy of the model's predictions. The second term in the formula is the sum of squares of the changes in the control input, serving as the control increment term, designed to control the air-fuel ratio. u This ensures that input changes are smooth and avoids drastic fluctuations.

[0023] Furthermore, the finite-time optimization process must satisfy the actual physical and technological constraints; therefore, the constraints include one or a combination of the following: upper and lower limits of air-fuel ratio:

[0024] Limitation on the rate of change of air-fuel ratio:

[0025] Furnace temperature safety constraints:

[0026] Upper and lower limits of air and / or gas valve opening:

[0027] Air and / or gas valve rate of change limit:

[0028] Upper and lower limits of air and / or gas flow rates:

[0029] Limits on the rate of change of air and / or gas flow:

[0030] Where T is the furnace temperature, Pos is the opening degree of the air or gas valve, ΔPos is the rate of change of the air or gas valve, V is the air or gas flow rate, ΔV is the rate of change of the air or gas flow rate, t is the current time step, and k is the number of time steps from the current time t to the future prediction or control.

[0031] 3. Optimize and implement continuously. Based on the objective function and constraints, a numerical optimizer is used to solve for the objective function. J Minimize the optimal control sequence for the next M steps:

[0032] The first control action in the sequence u * (t) The data is sent to the furnace control actuator for execution. In the next control cycle, new measurement data is collected and the above "state prediction-optimization-implementation" process is repeated. This rolling optimization method enables the control system to continuously use the latest feedback information to correct prediction and control decisions, thereby providing strong resistance to disturbances and model errors.

[0033] Corresponding to the above method, the present invention also provides a residual oxygen optimization control system for steel rolling heating furnaces. The system includes: The data acquisition and preprocessing module is configured to collect historical and real-time data of the heating furnace operation and perform cleaning, interpolation, alignment and normalization processing to construct the model input sequence; The residual oxygen prediction module is configured to carry and run the sequence-to-sequence Seq2Seq residual oxygen prediction model based on the spatiotemporal attention mechanism, receive the input sequence from the data acquisition and preprocessing module, and output the predicted residual oxygen content sequence in the future time domain. The model predictive control optimization module is configured to minimize the difference between the predicted residual oxygen sequence and the target value as the optimization objective, and solve for the optimal control sequence online in combination with preset process constraints. The control execution and feedback module is configured to send the first control command in the optimal control sequence to the heating furnace actuator and collect a new round of feedback data to achieve closed-loop rolling optimization.

[0034] The various modules of the system work together to achieve the aforementioned method for optimizing the control of residual oxygen in a steel rolling heating furnace.

[0035] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by one or more processors, the processors are able to implement the residual oxygen optimization control method for steel rolling furnaces described in any of the preceding claims. The storage medium includes, but is not limited to, read-only memory (ROM), random access memory (RAM), magnetic disk, optical disk, flash memory, and other media capable of storing program code.

[0036] The beneficial effects of this invention are as follows: (1) The Sequence-to-Sequence Seq2Seq prediction model based on the spatiotemporal attention mechanism provided by this invention can effectively capture the temporal dynamics and nonlinear characteristics of the heating furnace system, realize accurate prediction of residual oxygen content, and provide predictability for control. The predictive control framework of this model can handle multi-input and multi-output systems, coordinate multiple variables such as fuel, air, and furnace pressure, realize global optimization, and obtain the optimal control strategy through optimization solution.

[0037] (2) The residual oxygen optimization control method for steel rolling heating furnace provided by the present invention can effectively overcome model mismatch and external disturbances through the rolling optimization correction mechanism, thereby enhancing the robustness of the control method.

[0038] (3) The residual oxygen optimization control method of the steel rolling heating furnace provided by the present invention can bring significant economic and environmental benefits. The control method controls the oxidation loss rate of steel billet within a relatively stable range through precise residual oxygen control, realizes efficient combustion of fuel, stabilizes product quality and product energy consumption, reduces emissions of harmful gases such as carbon monoxide, and improves the environmental performance of the heating furnace.

[0039] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0040] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a flowchart of the overall control method of the present invention.

[0041] Figure 2 This is a schematic diagram of the sequence-to-sequence prediction model structure that incorporates a spatiotemporal attention mechanism used in this invention.

[0042] Figure 3 This is a schematic diagram of the architecture of a residual oxygen optimization control system for a steel rolling heating furnace provided in an embodiment of the present invention. Detailed Implementation

[0043] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0044] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0045] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0046] To facilitate understanding, the present invention provides an implementation example. The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0047] like Figure 1 As shown, a method for optimizing residual oxygen control in a steel rolling furnace includes two parts: an offline training prediction model and an online rolling optimization control. This embodiment uses a regenerative heating furnace in a 2250mm hot rolling production line of a steel company as the controlled object. The furnace has a designed effective length of 44870mm and an inner width of 11700mm. It consists of four sections: a preheating section, a first heating section, a second heating section, and a soaking section. Each section is equipped with an infrared residual oxygen analyzer to measure the residual oxygen content. Thermocouples are installed on the furnace top and side walls to measure the furnace temperature. The gas and air flow valves and burner reversing valves in each section are controlled by a PLC (Power Supply Controller) for the heating furnace.

[0048] To achieve optimized control of residual oxygen content in each section of the furnace, the core steps in this example are as follows: Step 1: Time-series data acquisition and serialization 1. Data Acquisition: The furnace operation data is collected from the furnace instrument PLC via the OPC UA protocol. The billet production information is collected from the furnace L2 system and MES system using DBLink as the model training dataset. The data acquisition period is 6 months and the data sampling period is 1 second.

[0049] 2. Variable Selection: Select one or more of the following variables from the historical and future time domains for periodic data collection: ①Historical state variables: furnace temperature, residual oxygen content, and furnace pressure in each section.

[0050] ② Historical control variables: gas flow rate, air flow rate, and air-fuel ratio for each section.

[0051] ③Historical disturbance variables: length, width, thickness, steel grade, charging temperature, and target temperature of the billet entering the furnace.

[0052] ④ Future control variables: Air-fuel ratio setpoints for each segment.

[0053] 3. Data Preprocessing: The collected data is cleaned to remove outliers, and missing values ​​are interpolated. Downsampling, time alignment, and normalization are performed. Time series data are downsampled to 10-second intervals, and the past time domain is set to 10 minutes (60 sets of data) and the future prediction time domain is set to 2 minutes (12 sets of data). The preprocessed data is divided into multiple sample pairs, each containing an input sequence and an output sequence, to construct the training dataset.

[0054] Step 2: Build and train the prediction model 1. Model Construction: Construct a Seq2Seq model incorporating a spatiotemporal dual attention mechanism. (1) Both the encoder and decoder use a 2-layer stacked GRU network, with each layer containing 256 hidden units.

[0055] (2) Spatial attention mechanism is applied to the input of the encoder to weight different variables.

[0056] (3) The time attention mechanism is applied to the decoder to weight the historical time steps.

[0057] like Figure 2 The diagram shows the structure of the sequence-to-sequence prediction model constructed by this method, which incorporates a spatiotemporal attention mechanism.

[0058] 2. Model Training: The processed input-output sequence is split into training and validation sets, with a batch size of 128. The Adam optimizer is used for training, and the loss function is set to mean squared error. The model is trained for 100 epochs until the loss on the validation set no longer decreases significantly. The trained model weights are then saved for online deployment.

[0059] Step 3: Online Scrolling Optimization Control of MPC Controller 1. Objective function construction:

[0060] In the formula, the first term, the prediction time domain P=12, includes the next 2 minutes; the second term, the control time sequence M=12, includes the next 2 minutes; and the residual oxygen setpoint. O 2,ref Always set to 1.8%; weighting coefficient w y =1.0, w u =0.2, prioritizing precise control of residual oxygen while also considering the stability of control.

[0061] 2. Constraint settings: ①Gas room door constraints: opening range [5%, 95%]; opening change rate ≤ 2% / cycle; ② Air valve constraints: opening range [5%, 95%]; opening change rate ≤ 2% / cycle; ③ Gas flow rate constraint: Flow rate range [1000, 15000] Nm 3 / h; Flow rate change rate ≤500Nm 3 / h / cycle; ④ Airflow constraint: Flow range [2000, 35000] Nm 3 / h; Flow rate change rate ≤1000Nm 3 / h / cycle; ⑤ Furnace temperature safety constraints: The furnace temperature in each section shall be ≤1300℃.

[0062] 3. Online scrolling optimization control: (1) At the beginning of each control cycle (10 seconds), process data is collected from the PLC and L2 system; (2) Construct the input sequence for prediction; (3) Call the deployed Seq2Seq prediction model and combine it with the candidate future control sequence generated by the MPC optimizer to predict the residual oxygen content in the next 2 minutes; (4) Under the premise of satisfying all the above constraints, the MPC optimizer solves the objective function and calculates the optimal air-fuel ratio adjustment sequence for the next 12 steps (2 minutes); (5) Send the first air-fuel ratio setpoint of the optimal sequence to the PLC for execution via the OPC UA protocol; (6) Repeat the above process in the next control cycle.

[0063] Before using this method, the residual oxygen content in the furnace was controlled manually. When encountering large disturbances such as changes in rolling rhythm or billet specifications, the residual oxygen content in each section usually fluctuated within a wide range of 0 to 5%, with large overshoot and a long time required to recover stability.

[0064] By using this method, thanks to the model's accurate predictive capabilities and MPC's rolling optimization control, the system can anticipate the impact of disturbances on residual oxygen content and make stable optimal adjustments. Under similar operating conditions, the residual oxygen content in each segment is stably controlled within the target range of 1.6% to 2%, with fluctuations reduced by 70%.

[0065] Corresponding to the above method, the present invention also provides an implementation method for a residual oxygen optimization control system for a steel rolling heating furnace. Figure 3 The diagram shown is an architectural schematic of the system provided in an embodiment of the present invention. Those skilled in the art will understand that this system can be deployed in an industrial control computer, an edge computing server, or an industrial cloud platform, and its functions can be implemented through software, hardware, or a combination of both.

[0066] The system mainly includes the following modules: Data Acquisition and Preprocessing Module: This module serves as the foundation of the system, communicating with the heating furnace's underlying control system (such as a PLC), manufacturing execution system, and secondary process computer via an industrial network (e.g., OPC UA protocol). Its function is to periodically acquire historical state variables, historical control variables, historical disturbance variables, and future control variables listed in Step 1. This module incorporates data cleaning, outlier handling, interpolation, downsampling, timestamp alignment, and normalization logic, responsible for converting raw, heterogeneous time-series data into a standardized, reliable input sequence, providing a standardized data interface for subsequent prediction modules.

[0067] Residual Oxygen Prediction Module: This module is the intelligent core of the system, carrying and running the pre-trained Sequence-to-Sequence (Seq2Seq) prediction model based on a spatiotemporal attention mechanism. This module receives the standard input sequence processed by the data acquisition and preprocessing module, utilizes its internal encoder-decoder structure and spatiotemporal attention mechanism to perform forward computation and inference, and outputs the prediction result in the future time domain. t h The module predicts the residual oxygen content sequence within the system. It can run on a computing environment that supports deep learning frameworks such as TensorFlow and PyTorch to achieve efficient and low-latency online prediction.

[0068] Model Predictive Control (MPC) Optimization Module: This module is the system's decision center. It receives the residual oxygen prediction sequence from the residual oxygen prediction module and, based on a preset objective function and various process constraints, constructs a constrained optimization problem in the finite time domain. This module either integrates or calls a high-efficiency numerical optimization solver to solve the optimization problem online in real time, calculating the optimal air-fuel ratio control sequence in the future control time domain.

[0069] Control Execution and Rolling Optimization Module: This module serves as the execution interface between the system and the physical world. Its function is to receive the optimal air-fuel ratio control sequence calculated by the MPC optimization module and extract the first control action u from this sequence. * (t). Subsequently, the setpoint is safely and reliably sent to the actuators (such as the controllers of gas regulating valves and air regulating valves) at the bottom of the heating furnace through a control network (such as OPC UA). After one control cycle ends, the module triggers a new round of data acquisition, prediction, optimization and control execution process, thereby realizing rolling optimization based on the latest feedback information, enabling the system to continuously adapt to changes in operating conditions and possessing strong anti-interference and robustness.

[0070] The above modules work together to form a closed-loop intelligent control system that integrates perception, prediction, decision-making, and execution, and together achieve high-precision and forward-looking optimized control of the residual oxygen content in the steel rolling heating furnace.

[0071] The present invention also provides an embodiment of a computer-readable storage medium. The storage medium may be an internal storage unit included in the control system, such as a read-only memory (ROM), a random access memory (RAM), etc.; or it may be an external storage device, such as a plug-in hard disk, a smart memory card (SMC), a secure digital card (SD), a flash memory card, etc.

[0072] Furthermore, the storage medium may include both internal storage units and external storage devices of the system. The storage medium is used to store computer programs and other types of data required for the system to run, forming a storage repository.

[0073] The computer program contains program instructions for executing all steps of the residual oxygen optimization control method for the steel rolling furnace described in this invention. For example, the program includes instructions for data acquisition and preprocessing, instructions for calling the residual oxygen prediction model for inference, instructions for constructing and solving the MPC optimization problem, and instructions for issuing control instructions and scheduling the process. When the computer program is executed by the system's processor, the processor can read the program instructions from the storage medium and execute the operations defined by the instructions, thereby realizing the optimized control function for residual oxygen in the heating furnace.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for optimizing and controlling residual oxygen in a steel rolling heating furnace, characterized in that, Includes the following steps: Step S1: Construct the input sequence for the prediction model, the input sequence including: based on the past time domain τ h Historical data sequences collected and processed, covering future prediction time domains. τ p The future sequence of control variables; Step S2: Input the input sequence into a pre-trained sequence-to-sequence (Seq2Seq) prediction model based on a spatiotemporal attention mechanism, and the model outputs a future prediction time domain. τ p The corresponding predicted residual oxygen content sequence; Step S3: The primary optimization objective is to minimize the difference between the predicted residual oxygen content sequence and the target residual oxygen setpoint sequence, with the stability of the control action as the secondary objective. Constraints are imposed on the control action sequence to establish a model for predictive control MPC optimization problem. Step S4: Solve the optimization problem to obtain a control sequence with the optimal air-fuel ratio in the future control time domain, extract the first air-fuel ratio of the optimal control sequence, and send it to the heating furnace control actuator; Step S5: In the next control cycle, repeat steps S1 to S4 to achieve rolling optimization and feedback correction of residual oxygen control in the heating furnace.

2. The method according to claim 1, characterized in that, The historical data sequence in step S1 includes one or more of the following variables: (1) Historical state variables: residual oxygen content, furnace temperature, and furnace pressure in each section; (2) Historical control variables: fuel flow rate, air flow rate, and air-fuel ratio setpoints for each heating section; (3) Historical disturbance variables: steel grade, billet size, furnace loading time, billet position, charging temperature, target temperature.

3. The method according to claim 1, characterized in that, The sequence-to-sequence (Seq2Seq) prediction model based on the spatiotemporal attention mechanism in step S2 is trained through the following steps: Step S301: Collect historical data of the heating furnace, construct the historical data sequence in the past time domain and the control variable sequence in the future prediction time domain as input, and use the residual oxygen measurement value in the future prediction time domain as the training dataset; The training dataset is preprocessed, and the preprocessed data is analyzed using data analysis tools. Step S302: Construct a sequence-to-sequence (Seq2Seq) prediction model with an encoder-decoder structure, integrating a spatiotemporal dual attention mechanism. The encoder and decoder use gated recurrent units (GRU) or long short-term memory networks (LSTM). The temporal attention mechanism is applied to each time step, and the spatial attention mechanism is applied to the input variables at each time step. Step S303: Supervised training of the model is performed using historical multivariate time series as input to the encoder, future prediction time domain control variable sequence as input to the decoder, and future residual oxygen content sequence as output. Step S304: Set the loss function and optimize the model parameters using the gradient descent algorithm until the model converges.

4. The method according to claim 1, characterized in that, The objective function of the optimization problem of model predictive control in step S3 is... J for: in, O 2,pred To predict residual oxygen concentration, O 2,ref To specify the residual oxygen concentration, t is the current time step, and k is the number of time steps from the current time t towards future prediction or control. u Air-fuel ratio, w y and w u These are adjustable weighting coefficients. P To predict the time domain, M To control the time domain, satisfy M≤P; The first term of the formula is the sum of squared errors between the predicted output and the specified output, which serves as the tracking error term and aims to ensure the predicted residual oxygen concentration... O 2,pred As close as possible to the desired specified value O 2,ref The first term is used to measure the accuracy of the model's predictions; the second term in the formula is the sum of squares of the changes in the control input, which serves as the control increment term and aims to control the air-fuel ratio. u The amount of change is adjusted to ensure smooth input changes and avoid drastic fluctuations.

5. The method according to claim 1, characterized in that, The constraints in step S3 include one or a combination of the following: upper and lower limits of air-fuel ratio: Limitation on the rate of change of air-fuel ratio: Furnace temperature safety constraints: Upper and lower limits of air and / or gas valve opening: Air and / or gas valve rate of change limit: Upper and lower limits of air and / or gas flow rates: Limits on the rate of change of air and / or gas flow: Where T is the furnace temperature, Pos is the opening degree of the air or gas valve, ΔPos is the rate of change of the air or gas valve, V is the air or gas flow rate, ΔV is the rate of change of the air or gas flow rate, t is the current time step, and k is the number of time steps from the current time t to the future prediction or control.

6. The method according to claim 1, characterized in that, The control sequence for the optimal air-fuel ratio in the future control time domain in step S4 is as follows: Where u is the air-fuel ratio, t is the current time step, and M is the control time domain. * (t) represents the first control action in the optimal control sequence, i.e., the first air-fuel ratio setpoint.

7. A residual oxygen optimization control system for a steel rolling heating furnace, characterized in that, The system employs the method as described in any one of claims 1 to 6, comprising: The data acquisition and preprocessing module is used to acquire and process furnace operation data to construct the input sequence; The residual oxygen prediction module has a built-in Seq2Seq residual oxygen prediction model based on the spatiotemporal attention mechanism, which is used to generate a predicted residual oxygen content sequence based on the input sequence. The model predictive control optimization module is used to construct and solve a constrained optimization problem based on the predicted residual oxygen content sequence, and generate the optimal air-fuel ratio control sequence. The control execution module is used to send the first control quantity in the optimal sequence to the actuator and realize rolling optimization control.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the residual oxygen optimization control method for steel rolling heating furnace as described in any one of claims 1 to 6.

Images