Electric precipitation coal gas recovery temperature control method based on load prediction

Abstract

The application discloses a temperature control method for electric dust removal and coal gas recovery based on load prediction, obtains time sequence operation data of a converter gas cooling system, and judges a current recovery state based on the time sequence operation data; after recognizing the intelligent regulation and control period according to the recovery state, the time sequence operation data is combined to intercept a converter blowing and refining period to construct a flue gas quantity feature matrix, time sequence features are extracted, a temperature lag parameter is calculated, and a temperature feature matrix with fused lag features is output; the flue gas quantity feature matrix and the temperature feature matrix are respectively input into corresponding time sequence prediction models, and corresponding prediction values are output; a theoretical water spraying sequence is calculated according to the prediction values; the theoretical water spraying sequence is optimized into an optimal water spraying sequence; the optimal water spraying sequence is feedback corrected according to a historical error weight matrix to obtain a final water spraying sequence, and the final water spraying sequence is issued to a water spraying execution mechanism for control.

Description

Technical Field

[0001] This invention belongs to the field of control and intelligent manufacturing technology of electrostatic precipitator systems in converter steelmaking, specifically relating to a method for controlling the temperature of gas recovery from electrostatic precipitators based on load prediction. Background Technology

[0002] In the converter dry electrostatic precipitator and gas recovery process, the outlet temperature control of the gas cooler is a key link to ensure system safety and recovery quality.

[0003] Currently, traditional gas cooler control systems have significant technical drawbacks: they generally employ a constant water spray cooling mode, which prevents the water spray volume from being adjusted in real time according to dynamic changes in parameters such as flue gas flow and temperature. This not only causes large fluctuations in the gas cooler outlet temperature but also leads to continuous energy waste from the cooling water pump. For stable outlet temperature control, existing technologies typically rely on basic PID control or staged experience-based control strategies. For example, some solutions adjust valve and fan speeds by real-time monitoring of gas concentration, supplemented by manual control of the coal-cooled spray guns; however, this method relies heavily on human experience and suffers from significant lag. Other solutions propose staged temperature control (e.g., controlling the evaporative cooler outlet between 220~225℃ and the electrostatic precipitator inlet between 150~160℃), and adding a high-flow pre-cooling spray gun at the inlet (spray volume greater than 230m³). 3 ( / h) to prevent gas rejection due to temperature exceeding 70°C. However, this type of method is essentially still a passive feedback control, unable to address the intermittent operating conditions of drastic fluctuations in flue gas parameters during converter steelmaking. Furthermore, although some existing technologies have begun to incorporate neural networks (such as CNN models) for data analysis, these are mostly limited to macroscopic predictions of feasible gas recovery rates and have not delved into the microscopic feedforward control of cooling water flow in cooling equipment.

[0004] In summary, existing methods for controlling the outlet temperature of converter gas coolers lack predictive control techniques, ignore the lag in heat transfer within the cooling system itself, and rely solely on a single fixed parameter for adjustment. This results in low temperature control accuracy, poor water spray volume matching, and unstable system operation, making it difficult to meet the intelligent, refined, and energy-saving production needs of modern steel enterprises. Summary of the Invention

[0005] In view of this, the main objective of the present invention is to provide a method for controlling the temperature of gas recovery from electrostatic precipitators based on load prediction.

[0006] To achieve the above objectives, the technical solution of the present invention is implemented as follows: This invention provides a method for controlling the temperature of coal gas recovery from electrostatic precipitators based on load prediction, the method comprising: Acquire the timing operation data of the converter gas cooling system, and determine the current recovery status based on the timing operation data; After identifying the intelligent control period based on the recovery status, the converter blowing cycle is extracted to construct a flue gas volume feature matrix, and the time-series operation data is combined with the time-series operation data to extract the time-series features and calculate the temperature lag parameter, and output the temperature feature matrix that integrates the lag features. The flue gas volume feature matrix and the temperature feature matrix are respectively input into the corresponding time series prediction model, and the corresponding predicted values ​​are output. The dynamic comprehensive specific heat is defined based on the heat exchange formula, and the theoretical water spray sequence is calculated based on the predicted value. The theoretical water spray sequence was optimized into the optimal water spray sequence. The optimal water spray sequence is corrected by feedback based on the historical error weight matrix to obtain the final water spray sequence, which is then sent to the water spray actuator for control.

[0007] Preferably, the step of acquiring the time-series operation data of the converter gas cooling system and determining the current recovery status based on the time-series operation data specifically includes: The operating data of the converter gas cooling system is collected in real time by DCS or PLC according to the converter smelting cycle, forming a time-series dataset containing 13 dimensions. ,in, Represents the 13-dimensional data vector at time k. in, The flue gas volumetric flow rate, The inlet flue gas temperature of the coal cooler. The temperature of the flue gas at the coal-cooled outlet. This is the evaporative cold inlet temperature. For the fan outlet pressure, This is the location of the oxygen lance. U(t) represents the oxygen flow rate in the converter, and u(t) represents the water spray rate. CO concentration, The CO recovery concentration threshold, O2 concentration, For the opening degree of the recovery valve, This is a signal indicating the refining status; Define recycling status ; When the blowing state signal is detected At that time, the slope of the CO concentration change trend was calculated using the least squares method within the sliding window. and the slope of the trend of O2 concentration ; If the trend change condition is met simultaneously: Determined as =1, indicating the current recycling status is before recycling; When the recovery valve opening degree satisfy When (5-5), determine =2, confirming the current recycling status as recycling in progress; simultaneously, record the current timestamp t as the most recent valid recycling time. ; When the recovery valve opening degree The process transitions from fully open to closed, with an effective recovery time. When recording, determine the time interval. ,according to judge When =3, the current recycling status is either temporary interruption or... When =4, the current recycling status is "recycling stopped"; where, Defined as the threshold for the pause interval time during recycling; When there is no effective recovery time During recording, and the opening degree of the recovery valve. At that time, make a judgment When the value is 5, the current recycling status is "not recycled".

[0008] Preferably, the step of identifying the intelligent control period based on the current recycling status specifically includes: based on Determine when At that time, it is determined that the current period is in intelligent control, and settings are made. Initiate load forecasting and optimization control; when in other stages, determine it as a non-intelligent control period and set... Maintain a fixed low-frequency water spray.

[0009] Preferably, the step of combining the time-series operation data, extracting the converter blowing cycle to construct a flue gas volume feature matrix, extracting time-series features and calculating temperature lag parameters, and outputting a temperature feature matrix that integrates the lag features specifically includes: Multiple blowing cycle data segments identified as belonging to the intelligent control period are concatenated to form a cycle flow feature vector. ; Extracting flue gas flow-related features to construct the flue gas flow-related feature matrix ; Define the cross-correlation function between the coal-cooled inlet temperature and the evaporator-cooled inlet temperature. Where τ is the lag time, and These are the average temperatures; The optimal lag time is determined by finding the maximum value of R(τ): ; Based on τ, construct multi-order hysteresis feature vectors: ; Extract flue gas temperature-related features from the acquired time-series operational data to construct a temperature feature matrix of the inlet flue gas temperature. .

[0010] Preferably, the flue gas volume feature matrix and the temperature feature matrix are respectively input into the corresponding time-series prediction model to output the corresponding predicted values, specifically including: All the time-series prediction models employ a two-layer LSTM network. The flue gas volume feature matrix and the temperature feature matrix Each as input The inputs are fed into their respective first-layer LSTM networks. The calculation formula for the first-layer LSTM network is as follows: ,in, For the hidden state vector, Let's consider the cell state vector, with the initial state as follows: and It is a 128-dimensional all-zero vector, with a Dropout rate of 0.2. The hidden state vector of the first layer LSTM network As input, it is fed into the second-layer LSTM network, and the calculation formula for the second-layer LSTM network is: ,in, For the hidden state vector, This is a cell state vector with a dimension of 64; The hidden state vector of the second layer LSTM network The input is a fully connected layer with an output layer, and the calculation formula is as follows: , where is the ReLU activation function. This represents element-wise multiplication, where W and b are the weight matrix and bias vector, respectively.

[0011] Based on the fully connected output layer, output the future... Predicted flue gas flow rate corresponding to step ,future The corresponding predicted inlet flue gas temperature value .

[0012] Preferably, the step of defining the dynamic comprehensive specific heat based on the heat exchange formula and calculating the theoretical water spray sequence based on the predicted value specifically includes: determining the comprehensive specific heat capacity. Power function relationship with temperature drop ΔT of gas ;in, , For the sensible heat and specific heat capacity of flue gas, m 冷凝 Let m be the mass of the condensate, r be the latent heat of vaporization of water, and m be the mass of the condensate.气 For flue gas quality, ΔT 气 The temperature of the smoke drops; Based on the first law of thermodynamics, neglecting heat loss from the coal cooling equipment, an energy conservation equation is established. ; The cooling water volume is determined based on the power function relationship and the energy conservation equation. ; According to the flue gas flow rate Inlet flue gas temperature outlet flue gas temperature Cooling water volume Cooling water inlet temperature Cooling water outlet temperature Historical operating data determines the comprehensive specific heat capacity ; The power function relationship is used as the dynamic comprehensive specific heat, and the theoretical water demand is calculated by substituting it into the energy conservation equation. For each prediction step k=1,2,...,n, based on the prediction and Calculate theoretical water demand ;in, The target coal cooling outlet temperature value is set. The theoretical water demand corresponding to each prediction step is combined to obtain the theoretical spray sequence. .

[0013] Preferably, optimizing the theoretical water spray sequence into the optimal water spray sequence specifically includes: Define the prediction time domain Np and the control time domain Nc, and construct an optimization objective function that includes a temperature tracking error term and a water spray volume energy consumption term. ;in, It is a temperature tracking error; The constraints that the optimization problem must satisfy are set, including upper and lower limits of water spray volume, upper limit of temperature, and specific heat capacity model effectiveness constraints. The optimization objective function and the constraints together constitute the MPC optimization problem:

[0014] st ; The theoretical water spray sequence is used as the initial value through a sequential quadratic programming algorithm. Iteratively solve for the optimal solution that satisfies all the aforementioned constraints. ; If the numerical optimization of the sequential quadratic programming algorithm fails to meet the constraints, a particle swarm optimization algorithm is used for global optimization to output the optimal solution that satisfies the constraints. The optimal water spray sequence is obtained. .

[0015] Preferably, the step of performing feedback correction on the optimal water spray sequence based on the historical error weight matrix to obtain the final water spray sequence specifically includes: A working condition-based error weight matrix is ​​introduced for dynamic compensation. The working condition space is defined as a two-dimensional grid of inlet flow rate Qg and coal cooling inlet temperature Tin, which is divided into M×N sub-regions. The flow range of each grid (i,j) Temperature range ;in, ; For each grid (i,j), calculate the average historical control error under that operating condition. ; in, The set of historical moments belonging to grid (i,j) is updated online using a sliding window strategy. It is the number of elements in the set; The historical error weight matrix is ​​composed of the average historical errors of all grids. ; In each control cycle, the grid to which the control belongs is determined based on the prediction results, the corresponding error compensation value is extracted, and the required water spray volume is corrected. Combining the optimal water spray sequence, the final cooling water spray volume sequence is obtained as the final water spray sequence.

[0016] Preferably, the first two-layer LSTM network used to predict flue gas flow rate in the time-series prediction model uses mean absolute error as the loss function during its training process. .

[0017] Preferably, the second two-layer LSTM network used to predict the inlet flue gas temperature in the time-series prediction model uses root mean square error as the loss function during its training process. .

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention acquires time-series operational data and accurately determines the recovery status to initiate intelligent control, eliminating interference from ineffective operating conditions for subsequent control. During the intelligent control period, a flue gas volume feature matrix is ​​constructed by extracting the blowing cycle and integrating temperature lag characteristics. Combined with a time-series prediction model, accurate predictions of future steps are output, thereby enabling early prediction of converter intermittent operating condition fluctuations. Subsequently, this invention couples the prediction results with the dynamic comprehensive specific heat defined based on the heat exchange formula to calculate the theoretical water spray sequence, and further obtains the optimal water spray sequence through optimized control. Finally, a historical error weight matrix is ​​introduced to provide feedback correction for the optimal sequence before execution. The closed-loop logic of the above-mentioned operating condition identification-feature prediction-mechanism calculation-optimization and feedback correction completely breaks through the limitations of traditional control's strong lag and blind regulation, realizing a transformation from passive feedback to feedforward + feedback composite control. While ensuring that the gas cooler outlet temperature accurately, quickly, and stably meets the standards, it effectively takes into account energy-saving requirements and significantly reduces the system's energy consumption. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and, together with their descriptions, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 A flowchart of a method for controlling the temperature of gas recovery from an electrostatic precipitator based on load prediction, provided in an embodiment of the present invention; Figure 2 This is a flue gas flow prediction curve diagram in an embodiment of the present invention; Figure 3 This is a curve showing the predicted inlet flue gas temperature in an embodiment of the present invention. Figure 4 This is a regression fitting diagram of the comprehensive specific heat capacity in an embodiment of the present invention; Figure 5 The curves represent the working condition identification and intelligent control in this embodiment of the invention.

[0020] Detailed Implementation To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0021] This invention provides a method for controlling the temperature of coal gas recovery from electrostatic precipitators based on load prediction, such as... Figure 1 As shown, the method includes: Step 101: Obtain the timing operation data of the converter gas cooling system, and determine the current recovery status based on the timing operation data; Specifically, the operating data of the converter gas cooling system is collected in real time by DCS or PLC based on the converter smelting cycle, forming a time-series dataset containing 13 dimensions. ,in, Represents the 13-dimensional data vector at time k. in, The flue gas volumetric flow rate, The inlet flue gas temperature of the coal cooler. The temperature of the flue gas at the coal-cooled outlet. This is the evaporative cold inlet temperature. For the fan outlet pressure, This is the location of the oxygen lance. U(t) represents the oxygen flow rate in the converter, and u(t) represents the water spray rate. CO concentration, The CO recovery concentration threshold, O2 concentration, For the opening degree of the recovery valve, This is a signal indicating the refining status; Furthermore, the acquired time-series data is cleaned by handling outliers and missing values.

[0022] Define recycling status ; The pre-, during-, and post-converter gas recovery phases are defined as the intelligent control period, while the remaining production phases are defined as the non-intelligent control period. By jointly identifying the blowing state and the converter gas recovery state, the automatic determination of the equipment control strategy switching period is achieved.

[0023] When the blowing state signal is detected At that time, the slope of the CO concentration change trend was calculated using the least squares method within the sliding window. and the slope of the trend of O2 concentration ; For example, when a blowing signal is detected ( When in a sliding window Trend verification of CO and O2 concentrations was performed within a timeframe (typically 30-60 seconds), and a threshold for the CO rise slope was defined. and O2 descent slope threshold It is used to statistically determine whether the trends in CO and O2 changes are sustained.

[0024] The slope of the CO concentration change trend was calculated using the least squares method. ;in, ; The slope of the O2 concentration change trend was calculated using the same method. ; If the trend change condition is met simultaneously: Determined as =1, indicating the current recycling status is before recycling; When the recovery valve opening degree satisfy When (5-5), determine =2, confirming the current recycling status as recycling in progress; simultaneously, record the current timestamp t as the most recent valid recycling time. ; When the recovery valve opening degree The process transitions from fully open to closed, with an effective recovery time. When recording, determine the time interval. ,according to judge When =3, the current recycling status is either temporary interruption or... When =4, the current recycling status is "recycling stopped"; where, Defined as the threshold for the pause interval time during recycling; When there is no effective recovery time During recording, and the opening degree of the recovery valve. At that time, make a judgment When the value is 5, the current recycling status is "not recycled".

[0025] Step 102: After identifying the intelligent control period based on the recovery status, combine the time-series operation data to extract the converter blowing cycle to construct the flue gas volume feature matrix, extract the time-series features and calculate the temperature lag parameter, and output the temperature feature matrix that integrates the lag features. Specifically, according to Determine when At that time, it is determined that the current period is in intelligent control, and settings are made. Initiate load forecasting and optimization control; when in other stages, determine it as a non-intelligent control period and set... Maintain a fixed low-frequency water spray.

[0026] Each blowing cycle in converter steelmaking follows similar physicochemical processes, exhibiting periodic characteristics. A cycle concatenation strategy is employed to concatenate multiple blowing cycle data segments identified as the intelligent control period, forming a cycle flow characteristic vector. ; Extracting flue gas flow-related features to construct the flue gas flow-related feature matrix ; The gas cooler is located at the downstream end of the electrostatic precipitator system. Its temperature change lags behind that of the flue gas generation end. Cross-correlation analysis is used to quantify this lag time, and a cross-correlation function between the coal-cooled inlet temperature and the evaporator-cooled inlet temperature is defined. Where τ is the lag time, and These are the average temperatures; The optimal lag time is determined by finding the maximum value of R(τ): ; Based on τ, construct multi-order hysteresis feature vectors: ; Extract flue gas temperature-related features from the acquired time-series operational data to construct a temperature feature matrix of the inlet flue gas temperature. .

[0027] Step 103: Input the flue gas volume feature matrix and the temperature feature matrix into the corresponding time series prediction models respectively, and output the corresponding predicted values; Specifically, all the time series prediction models use a two-layer LSTM network; The flue gas volume feature matrix and the temperature feature matrix Each as input The inputs are fed into their respective first-layer LSTM networks. The calculation formula for the first-layer LSTM network is as follows: ,in, For the hidden state vector, Let's consider the cell state vector, with the initial state as follows: and It is a 128-dimensional all-zero vector, with a Dropout rate of 0.2. The hidden state vector of the first layer LSTM network As input, it is fed into the second-layer LSTM network, and the calculation formula for the second-layer LSTM network is: ,in, For the hidden state vector, This is a cell state vector with a dimension of 64; The hidden state vector of the second layer LSTM network The input is a fully connected layer with an output layer, and the calculation formula is as follows: , where is the ReLU activation function. This represents element-wise multiplication, where W and b are the weight matrix and bias vector, respectively.

[0028] The first-layer LSTM network and the second-layer LSTM network have the same internal computation mechanism, which controls the information flow through forget gates, input gates, and output gates. The computation mechanism formula is as follows: Forgotten Gate:

[0029] Input Gate:

[0030] Candidate values:

[0031] Cell status:

[0032] Output gate:

[0033] Hidden state: ; Fully connected output layer , where is the ReLU activation function. This represents element-wise multiplication, where W and b are the weight matrix and bias vector, respectively.

[0034] Based on the fully connected output layer, output the future... Predicted flue gas flow rate corresponding to step ,future The corresponding predicted inlet flue gas temperature value .

[0035] The first two-layer LSTM network used to predict flue gas flow rate in the time-series prediction model uses the mean absolute error as the loss function during its training process. .

[0036] The second two-layer LSTM network used to predict the inlet flue gas temperature in the time-series prediction model employs root mean square error as the loss function during its training process. .

[0037] Figure 2 This is a comparison between the actual and predicted values ​​of the flue gas flow rate in the coal cooling system in step 103. During the blowing cycle, the model accuracy of the flue gas flow rate prediction is 97.3% for most production periods, with some deviation only during a few sudden fluctuation periods (such as local peaks and sudden drops at the end). The overall fitting effect is good, proving that the prediction method can effectively capture the dynamic changes in flue gas flow rate.

[0038] Figure 3 This is a comparison between the actual and predicted values ​​of the inlet flue gas temperature in step 103, with a model accuracy of 98.1%. It can be seen that during the blowing and recovery period, the inlet flue gas temperature generally shows an upward trend, indicating that the temperature prediction model can effectively track the dynamic changes in flue gas temperature.

[0039] Step 104: Define the dynamic comprehensive specific heat based on the heat exchange formula, and calculate the theoretical water spray sequence based on the predicted value; Specifically, determine the overall specific heat capacity. Power function relationship with temperature drop ΔT of gas ;in, , For the sensible heat and specific heat capacity of flue gas, m 冷凝Let m be the mass of the condensate, r be the latent heat of vaporization of water, and m be the mass of the condensate. 气 For flue gas quality, ΔT 气 The temperature of the smoke drops; During flue gas cooling, the flue gas releases both sensible heat (temperature decrease) and latent heat due to water vapor condensation. Since the real-time flue gas composition and condensate volume are difficult to measure precisely, and the latent heat released by water vapor condensation varies at different initial temperatures, a comprehensive specific heat capacity is defined. As the equivalent heat capacity of the process, it characterizes the total heat release capacity of a unit mass of flue gas per unit temperature drop. ;in: : sensible heat specific heat capacity of flue gas [kJ / (kg·℃)]; mcondensate: mass of condensate [kg]; r: latent heat of vaporization of water [kJ / kg]; mgas: mass of flue gas. ΔTgas: Flue gas temperature drop [°C].

[0040] The total heat released by the flue gas can be expressed as: ; Based on the first law of thermodynamics, neglecting heat loss from the coal cooling equipment, an energy conservation equation is established. ; The cooling water volume is determined based on the power function relationship and the energy conservation equation. ; According to the flue gas flow rate Inlet flue gas temperature outlet flue gas temperature Cooling water volume Cooling water inlet temperature Cooling water outlet temperature Historical operating data determines the comprehensive specific heat capacity ; The power function relationship is used as the dynamic comprehensive specific heat, and substituted into the energy conservation equation to calculate the theoretical water demand. ; For each prediction step k=1,2,...,n, based on the prediction and Calculate theoretical water demand ;in, The target coal cooling outlet temperature value is set. The theoretical water demand corresponding to each prediction step is combined to obtain the theoretical spray sequence. .

[0041] Figure 4This is the distribution of the comprehensive specific heat capacity of the flue gas temperature drop and the power function fitting results from step 104. The distribution in the figure shows that the comprehensive specific heat capacity decreases monotonically with increasing flue gas temperature drop. After fitting with the power function, the goodness of fit reaches 0.9420, indicating that the power function model can describe the quantitative relationship between flue gas temperature drop and comprehensive specific heat capacity, providing a mathematical model for the thermal calculation of coal-cooled systems.

[0042] Step 105: Optimize the theoretical water spray sequence into the optimal water spray sequence; Specifically, we define the prediction time domain Np and the control time domain Nc, and construct an optimization objective function that includes a temperature tracking error term and a water spray volume energy consumption term. ;in, It is a temperature tracking error; The constraints that the optimization problem must satisfy are set, including upper and lower limits of water spray volume, upper limit of temperature, and specific heat capacity model effectiveness constraints. The optimization objective function and the constraints together constitute the MPC optimization problem:

[0043] st ; The theoretical water spray sequence is used as the initial value through a sequential quadratic programming algorithm. Iteratively solve for the optimal solution that satisfies all the aforementioned constraints. ; If the numerical optimization of the sequential quadratic programming algorithm fails to meet the constraints, a particle swarm optimization algorithm is used for global optimization to output the optimal solution that satisfies the constraints. The optimal water spray sequence is obtained. .

[0044] The particle swarm size is 50, the number of iterations (iter) is 30, and the inertia weight is: (Linearly decreasing); Learning factor: ; Step 106: Perform feedback correction on the optimal water spray sequence according to the historical error weight matrix to obtain the final water spray sequence, and send the final water spray sequence to the water spray actuator for control.

[0045] Specifically, to improve control accuracy and robustness, an error weight matrix based on operating conditions is introduced for dynamic compensation. The operating condition space is defined as a two-dimensional grid of inlet flow rate Qg and coal cooling inlet temperature Tin, which is divided into M×N sub-regions. The flow range of each grid (i,j) Temperature range ;in, ; For each grid (i,j), calculate the average historical control error under that operating condition. ; in, The set of historical moments belonging to grid (i,j) is updated online using a sliding window strategy. It is the number of elements in the set; The historical error weight matrix is ​​composed of the average historical errors of all grids. ; In each control cycle, based on the prediction and Determine the grid to which it belongs in the historical error weight matrix Extract the corresponding error compensation value Adjust the required water volume ; Combining the optimal spray sequence, the final cooling water spray volume sequence is obtained as the final spray sequence. .

[0046] like Figure 5 As shown in the figure, the intelligent control strategy can accurately match the working cycle of converter blowing, etc. While ensuring that the gas and oxygen recovery targets are met, it significantly reduces the cooling water flow consumption and achieves an intelligent power saving rate of 41.54%, proving that the control scheme can effectively improve the energy utilization efficiency and operating economy of the coal cooling system.

[0047] This invention significantly improves the control precision of the coal cooling process, greatly reducing the fluctuation range of the core parameter, coal cooling outlet temperature, and accurately matching the control requirements of the national standard (GB / T45754-2025) for production processes. The system demonstrates outstanding energy-saving and consumption-reducing effects, effectively reducing energy consumption and operating costs during production through effective identification and intelligent control of operating conditions. Production safety is further enhanced; precise control of the process avoids process risks caused by parameter deviations, ensuring the stable and reliable operation of equipment and the production system. The overall level of intelligence is significantly improved, reducing the frequency and difficulty of manual intervention and promoting the transformation of the production process towards a refined, automated, and intelligent control model.

[0048] Compared to traditional control and PID control methods, this solution has significant technical advantages, primarily reflected in innovative upgrades in control logic and model construction: In terms of control principle, it transforms from hysteresis feedback to feedforward + feedback composite control, enabling early prediction of process conditions and timely adjustment responses, effectively solving the hysteresis problem of traditional control; at the model dependency level, it breaks free from the limitations of single empirical parameters, adopting a hybrid modeling approach of mechanistic model + data-driven approach, preserving clear physical meaning while adapting to changes in operating conditions based on actual production data; in terms of control objectives, it achieves multi-objective optimization design, precisely controlling temperature parameters while also considering energy-saving requirements, achieving dual optimization of process control and energy consumption management; furthermore, the solution is adaptive, automatically adjusting the control strategy according to dynamic changes in production conditions, maintaining excellent control performance, and adapting to different production operation scenarios.

[0049] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention.

Claims

1. A method for controlling the temperature of gas recovery from electrostatic precipitators based on load prediction, characterized in that, The method includes: Acquire the timing operation data of the converter gas cooling system, and determine the current recovery status based on the timing operation data; After identifying the intelligent control period based on the recovery status, the converter blowing cycle is extracted to construct a flue gas volume feature matrix, and the time-series operation data is combined with the time-series operation data to extract the time-series features and calculate the temperature lag parameter, and output the temperature feature matrix that integrates the lag features. The flue gas volume feature matrix and the temperature feature matrix are respectively input into the corresponding time series prediction model, and the corresponding predicted values ​​are output. The dynamic comprehensive specific heat is defined based on the heat exchange formula, and the theoretical water spray sequence is calculated based on the predicted value. The theoretical water spray sequence was optimized into the optimal water spray sequence. The optimal water spray sequence is corrected by feedback based on the historical error weight matrix to obtain the final water spray sequence, which is then sent to the water spray actuator for control.

2. The method for controlling the temperature of electrostatic precipitator gas recovery based on load prediction according to claim 1, characterized in that, The acquisition of time-series operation data of the converter gas cooling system, and the determination of the current recovery status based on the time-series operation data, specifically includes: The operating data of the converter gas cooling system is collected in real time by DCS or PLC according to the converter smelting cycle, forming a time-series dataset containing 13 dimensions. ,in, Represents the 13-dimensional data vector at time k. in, The flue gas volumetric flow rate, The inlet flue gas temperature of the coal cooler. The temperature of the flue gas at the coal-cooled outlet. This is the evaporative cold inlet temperature. For the fan outlet pressure, This is the location of the oxygen lance. U(t) represents the oxygen flow rate in the converter, and u(t) represents the water spray rate. CO concentration, The CO recovery concentration threshold, O2 concentration, For the opening degree of the recovery valve, This is a signal indicating the refining status; Define recycling status ; When the blowing state signal is detected At that time, the slope of the CO concentration change trend was calculated using the least squares method within the sliding window. and the slope of the trend of O2 concentration ; If the trend change condition is met simultaneously: Determined as =1, indicating the current recycling status is before recycling; When the recovery valve opening degree satisfy When (5-5), determine =2, confirming the current recycling status as recycling in progress; simultaneously, record the current timestamp t as the most recent valid recycling time. ; When the recovery valve opening degree The process transitions from fully open to closed, with an effective recovery time. When recording, determine the time interval. ,according to judge When =3, the current recycling status is either temporary interruption or... When =4, the current recycling status is "recycling stopped"; where, Defined as the threshold for the pause interval time during recycling; When there is no effective recovery time During recording, and the opening degree of the recovery valve. At that time, make a judgment When the value is 5, the current recycling status is "not recycled".

3. The method for controlling the temperature of electrostatic precipitator gas recovery based on load prediction according to claim 1 or 2, characterized in that, The process of identifying the intelligent control period based on the current recycling status specifically includes: based on... Determine when At that time, it is determined that the current period is in intelligent control, and settings are made. Initiate load forecasting and optimization control; when in other stages, determine it as a non-intelligent control period and set... Maintain a fixed low-frequency water spray.

4. The method for controlling the temperature of electrostatic precipitator gas recovery based on load prediction according to claim 3, characterized in that, The process of combining the time-series operational data, extracting the converter blowing cycle to construct a flue gas volume feature matrix, extracting time-series features and calculating temperature lag parameters, and outputting a temperature feature matrix incorporating lag features, specifically includes: Multiple blowing cycle data segments identified as belonging to the intelligent control period are concatenated to form a cycle flow feature vector. ; Extracting flue gas flow-related features to construct the flue gas flow-related feature matrix ; Define the cross-correlation function between the coal-cooled inlet temperature and the evaporator-cooled inlet temperature. Where τ is the lag time, and These are the average temperatures; The optimal lag time is determined by finding the maximum value of R(τ): ; Based on τ, construct multi-order hysteresis feature vectors: ; Extract flue gas temperature-related features from the acquired time-series operational data to construct a temperature feature matrix of the inlet flue gas temperature. .

5. The method for controlling the temperature of electrostatic precipitator gas recovery based on load prediction according to claim 4, characterized in that, The flue gas volume feature matrix and the temperature feature matrix are respectively input into the corresponding time-series prediction model, and the corresponding predicted values ​​are output, specifically including: All the time-series prediction models employ a two-layer LSTM network. The flue gas volume feature matrix and the temperature feature matrix Each as input The inputs are fed into their respective first-layer LSTM networks. The calculation formula for the first-layer LSTM network is as follows: ,in, For the hidden state vector, Let's consider the cell state vector, with the initial state as follows: and It is a 128-dimensional all-zero vector, with a Dropout rate of 0.

2. The hidden state vector of the first layer LSTM network As input, it is fed into the second-layer LSTM network, and the calculation formula for the second-layer LSTM network is: ,in, For the hidden state vector, This is a cell state vector with a dimension of 64; The hidden state vector of the second layer LSTM network The input is a fully connected layer with an output layer, and the calculation formula is as follows: , where is the ReLU activation function. This represents element-wise multiplication, where W and b are the weight matrix and bias vector, respectively. Based on the fully connected output layer, output the future... Predicted flue gas flow rate corresponding to step ,future The corresponding predicted inlet flue gas temperature value .

6. The method for controlling the temperature of electrostatic precipitator gas recovery based on load prediction according to claim 5, characterized in that, The process of defining the dynamic comprehensive specific heat based on the heat exchange formula and calculating the theoretical water spray sequence based on the predicted value specifically includes: determining the comprehensive specific heat capacity. Power function relationship with temperature drop ΔT of gas ;in, , For the sensible heat and specific heat capacity of flue gas, m 冷凝 Let m be the mass of the condensate, r be the latent heat of vaporization of water, and m be the mass of the condensate. 气 For flue gas quality, ΔT 气 The temperature of the smoke drops; Based on the first law of thermodynamics, neglecting heat loss from the coal cooling equipment, an energy conservation equation is established. ; The cooling water volume is determined based on the power function relationship and the energy conservation equation. ; According to the flue gas flow rate Inlet flue gas temperature outlet flue gas temperature Cooling water volume Cooling water inlet temperature Cooling water outlet temperature Historical operating data determines the comprehensive specific heat capacity ; The power function relationship is used as the dynamic comprehensive specific heat, and the theoretical water demand is calculated by substituting it into the energy conservation equation. For each prediction step k=1,2,...,n, based on the prediction and Calculate theoretical water demand ;in, The target coal cooling outlet temperature value is set. The theoretical water demand corresponding to each prediction step is combined to obtain the theoretical spray sequence. .

7. The method for controlling the temperature of electrostatic precipitator gas recovery based on load prediction according to claim 6, characterized in that, The optimization of the theoretical water spray sequence into the optimal water spray sequence specifically includes: Define the prediction time domain Np and the control time domain Nc, and construct an optimization objective function that includes a temperature tracking error term and a water spray volume energy consumption term. ;in, It is a temperature tracking error; The constraints that the optimization problem must satisfy are set, including upper and lower limits of water spray volume, upper limit of temperature, and specific heat capacity model effectiveness constraints. The optimization objective function and the constraints together constitute the MPC optimization problem: st ； The theoretical water spray sequence is used as the initial value through a sequential quadratic programming algorithm. Iteratively solve for the optimal solution that satisfies all the aforementioned constraints. ; If the numerical optimization of the sequential quadratic programming algorithm fails to meet the constraints, a particle swarm optimization algorithm is used for global optimization to output the optimal solution that satisfies the constraints. The optimal water spray sequence is obtained. .

8. The method for controlling the temperature of electrostatic precipitator gas recovery based on load prediction according to claim 7, characterized in that, The step of performing feedback correction on the optimal water spray sequence based on the historical error weight matrix to obtain the final water spray sequence specifically includes: A working condition-based error weight matrix is ​​introduced for dynamic compensation. The working condition space is defined as a two-dimensional grid of inlet flow rate Qg and coal cooling inlet temperature Tin, which is divided into M×N sub-regions. The flow range of each grid (i,j) Temperature range ;in, ; For each grid (i,j), calculate the average historical control error under that operating condition. ; in, The set of historical moments belonging to grid (i,j) is updated online using a sliding window strategy. It is the number of elements in the set; The historical error weight matrix is ​​composed of the average historical errors of all grids. ; In each control cycle, the grid to which the control belongs is determined based on the prediction results, the corresponding error compensation value is extracted, and the required water spray volume is corrected. Combining the optimal water spray sequence, the final cooling water spray volume sequence is obtained as the final water spray sequence.

9. The method for controlling the temperature of electrostatic precipitator gas recovery based on load prediction according to claim 5, characterized in that, The first two-layer LSTM network used to predict flue gas flow rate in the time-series prediction model uses the mean absolute error as the loss function during its training process. .

10. The method for controlling the temperature of electrostatic precipitator gas recovery based on load prediction according to claim 5, characterized in that, The second two-layer LSTM network used to predict the inlet flue gas temperature in the time-series prediction model employs root mean square error as the loss function during its training process. .

Images