A blast furnace hot blast stove stove changing control system

Abstract

The present application relates to blast furnace hot blast stove technical field, more specifically, relate to a kind of blast furnace hot blast stove change furnace control system, for solving the problem that prior art cannot realize dynamic adaptation to different working conditions, cannot adjust change furnace strategy in time, increase change furnace cost, cannot guarantee the stable operation of blast furnace;The present application utilizes reinforcement learning algorithm by change furnace optimization control module, based on smelting state and blast pressure prediction, provides intelligent decision support, realizes dynamic adaptation to different working conditions by defining state space and change furnace action, and combining reward function to comprehensively consider efficiency, cost and stability, this module improves smelting efficiency, reduces change furnace cost, guarantees the stable operation of blast furnace, optimizes strategy using historical data, responds to environmental changes in real time, adjusts change furnace strategy in time, ensures production continuity and safety.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of blast furnace hot blast stove, more particularly, to a blast furnace hot blast stove change control system. BACKGROUND

[0002] The hot blast stove is one of the important equipment for blast furnace ironmaking, which functions to heat the cold air into high-temperature hot blast through fuel combustion and then deliver it into the blast furnace to improve the smelting efficiency of the blast furnace. However, during the change of the hot blast stove, the wind pressure fluctuates greatly, which often leads to the imbalance of the blast furnace condition and affects the stability and efficiency of the smelting process.

[0003] The patent application with the publication number CN118685580A discloses a hot blast stove change switching control device and method for multiple blast furnaces, which communicates each air supply main pipe connected with the blast furnace hot blast stove through the air switching valve, can supply air to the corresponding air supply main pipe through the air switching valve when the air blower fails, and timely start the standby air blower to supplement the air pressure, so as to ensure the stability of the air pressure of the hot blast stove, thereby making the system air pressure stable, ensuring the iron tapping quality and effectively improving the safety. According to the corresponding relationship between the air blower and the hot blast stove of the blast furnace, a constant air pressure PID control model for the hot blast stove change switching control of multiple blast furnaces is established, the static vane of the current air blower is controlled, and the constant air pressure control is ensured to run stably. The static vane control is performed by using the feedforward assignment, the static vane opening degree is controlled according to the gradient, and the large air pressure fluctuation affecting the system safety is avoided.

[0004] However, the above-mentioned reference patent sets up the standby air blower and the constant air pressure PID control model, supplements the air pressure when the main air blower fails, adjusts the static vane opening degree according to the distance and performance of the air blower and the blast furnace, ensures the safe and stable operation of the hot blast stove change, improves the iron tapping quality and efficiency, but cannot realize the accurate evaluation of the blast furnace operation state, cannot optimize the process parameters to improve the efficiency and quality, cannot accurately predict the air supply pressure, cannot help the production plan and strategy making, reduces the efficiency and product quality, and cannot realize the dynamic adaptation to different working conditions, cannot timely adjust the change strategy, increases the change cost, and cannot ensure the stable operation of the blast furnace.

[0005] Therefore, the present application proposes a blast furnace hot blast stove change control system aiming at the above-mentioned problems. SUMMARY

[0006] The present application aims to provide a blast furnace hot blast stove change control system, which solves the problems that the prior art cannot realize the accurate evaluation of the blast furnace operation state, cannot optimize the process parameters to improve the efficiency and quality, cannot accurately predict the air supply pressure, cannot help the production plan and strategy making, reduces the efficiency and product quality, and cannot realize the dynamic adaptation to different working conditions, cannot timely adjust the change strategy, increases the change cost, and cannot ensure the stable operation of the blast furnace.

[0007] The object of the present application is achieved by the following technical solutions:

[0008] A blast furnace hot blast stove furnace replacement control system applied to a furnace replacement control platform, comprising:

[0009] A data acquisition module for real-time acquisition of operating parameters of the blast furnace and the hot blast stove, and preprocessing of the acquired operating parameters;

[0010] A smelting state evaluation module for real-time monitoring of smelting evaluation parameters of the blast furnace and monitoring and evaluation of the smelting state of the blast furnace;

[0011] A blast pressure prediction module for collecting historical operating parameters of the blast furnace and the hot blast stove, constructing a blast pressure prediction model, and predicting the blast pressure in the future period of time according to the model;

[0012] A furnace replacement optimization control module for optimizing the furnace replacement strategy by using a reinforcement learning algorithm according to the smelting state evaluation result and the blast pressure prediction result.

[0013] As a preferred embodiment of the present application, the specific process of the smelting state evaluation module for monitoring and evaluating the smelting state of the blast furnace is as follows:

[0014] Obtaining smelting evaluation parameters of the blast furnace, the smelting evaluation parameters including the top temperature, the tuyere temperature, the blast pressure and the coke consumption, generating a monitoring period, and equally dividing the monitoring period into a plurality of monitoring time periods;

[0015] Obtaining the top temperature change rate of the blast furnace in the plurality of monitoring time periods, the top temperature change rate representing the ratio between the top temperature change amount and the corresponding time period length, and calculating the arithmetic mean of the plurality of obtained top temperature change rates, and recording the arithmetic mean of the plurality of top temperature change rates as the average top temperature change rate PLD;

[0016] The average tuyere temperature change rate PFW, the average blast pressure change rate PFY and the average coke consumption change rate PJX can be obtained by using the same method as that for obtaining the average top temperature change rate.

[0017] As a preferred embodiment of the present application, the average top temperature change rate PLD, the average tuyere temperature change rate PFW, the average blast pressure change rate PFY and the average coke consumption change rate PJX are obtained, and the smelting state evaluation coefficient YZP is calculated by the following formula: ;

[0018] Wherein b1, b2, b3 and b4 are all preset proportional factor coefficients, b4>b3>b2>b1>0, the smelting state evaluation coefficient YZP is compared with a preset first smelting state evaluation coefficient threshold and a preset second smelting state evaluation coefficient threshold, the preset first smelting state evaluation coefficient threshold is smaller than the preset second smelting state evaluation coefficient threshold;

[0019] If the smelting state evaluation coefficient YZP is smaller than the preset first smelting state evaluation coefficient threshold, it indicates that the smelting state of the blast furnace is excellent;

[0020] If the smelting state evaluation coefficient YZP is greater than or equal to the preset first smelting state evaluation coefficient threshold and smaller than the preset second smelting state evaluation coefficient threshold, it indicates that the smelting state of the blast furnace is normal;

[0021] If the smelting state evaluation coefficient YZP is greater than or equal to the preset second smelting state evaluation coefficient threshold, it indicates that the smelting state of the blast furnace is abnormal.

[0022] As a preferred embodiment of the present application, the specific process that the blast air pressure prediction module predicts the blast air pressure in the future period of time is as follows:

[0023] The historical operation parameters of the blast furnace and the hot blast furnace are acquired, the operation parameters include blast furnace air pressure, air volume, air temperature, hot blast furnace temperature, pressure, blower speed and pressure equalizing valve opening degree, a collection period is generated, the collection period is equally divided into m continuous sub-periods, and the midpoint time of each sub-period is marked to obtain m midpoint times;

[0024] Taking the m midpoint times as the base point, the s-1 expansion times are marked by extending forward and backward for the same length, and after the midpoint times and the s-1 expansion times are summarized, s detection times are obtained;

[0025] The blast furnace air pressure values at the s detection times are measured by instruments respectively to obtain s detection blast furnace air pressure values, and the s detection blast furnace air pressure values are added and averaged to obtain m sub-blast furnace air pressure values.

[0026] As a preferred embodiment of the present application, the expression of the sub-blast furnace air pressure value is: ;

[0027] In the formula, ZFYzm is the sub-blast furnace air pressure value of the mth sub-period, and ZFYjcmn is the nth detection blast furnace air pressure value of the mth sub-period;

[0028] The maximum value and the minimum value of the sub-blast furnace air pressure value are removed, the remaining m-2 sub-blast furnace air pressure values are added and averaged to obtain a blast furnace air pressure average value;

[0029] The expression of the blast furnace air pressure average value is: ;

[0030] In the formula, FYjz is the blast furnace air pressure average value, and ZFYzp is the blast furnace air pressure value of the pth sub-period;

[0031] The blast furnace air volume average value FLjz, the hot blast furnace air pressure average value RYjz, the blower speed average value GZjz, and the average pressure valve opening average value FKjz can be obtained by using the method for calculating the blast furnace air pressure average value.

[0032] As a preferred embodiment of the present application, the blast furnace air pressure average value FYjz, the blast furnace air volume average value FLjz, the hot blast furnace air pressure average value RYjz, the blower speed average value GZjz, and the average pressure valve opening average value FKjz are combined to construct a blast furnace air pressure prediction matrix SYJ, the blast furnace air pressure prediction matrix SYJ is taken as the input of a machine learning model, and the blast furnace air pressure in a future period of time corresponding to each group of blast furnace air pressure prediction matrix SYJ is taken as the output of the machine learning model, the blast furnace air pressure in the future period of time is taken as the prediction target, the sum of prediction errors of all training data is minimized as the training target, the machine learning model is trained until the sum of prediction errors converges, and a blast furnace air pressure prediction model is obtained.

[0033] As a preferred embodiment of the present application, the expression formula of the blast furnace air pressure prediction model is as follows: ;

[0034] Wherein, η1, η2, η3, η4, and η5 are regression coefficients, λ is a random error term, and SFY represents the blast furnace air pressure in the future period of time.

[0035] Real-time running parameters of the blast furnace and the hot blast furnace are obtained, which are converted into corresponding blast furnace air pressure prediction matrix SYJ and input into the blast furnace air pressure prediction model, and real-time blast furnace air pressure in the future period of time is obtained through the blast furnace air pressure prediction model.

[0036] As a preferred embodiment of the present application, the specific process of optimizing the furnace change strategy by using the reinforcement learning algorithm in the furnace change optimization control module is as follows:

[0037] The smelting state evaluation result and the blast furnace air pressure prediction result are obtained, the smelting state evaluation result is that the smelting state of the blast furnace is excellent, normal or abnormal, and the blast furnace air pressure prediction result is the real-time blast furnace air pressure in the future period of time;

[0038] The specific steps of optimizing the furnace change strategy by using the reinforcement learning algorithm are as follows:

[0039] First, a state space S is defined, the state space is composed of the smelting state and the blast furnace air pressure prediction, and the state space can be represented as a vector: S=[Ssmelt, Spressure].

[0040] Ssmelt is a vector representation of smelt state, represented by a numerical vector:

[0041] Excellent: [1, 0, 0], Normal: [0, 1, 0], Abnormal: [0, 0, 1];

[0042] Spressure is a continuous variable vector representing the predicted air supply pressure for the next n time steps:

[0043] Spressure = [p1, p2, …, pn];

[0044] Where pi is the predicted air supply pressure at the i-th time step;

[0045] Define action time A, where action represents the change of furnace strategy, discretized as:

[0046] A = 0: No change of furnace;

[0047] A = 1: Immediate change of furnace;

[0048] A = 2: Delayed change of furnace by t1 time;

[0049] A = 3: Delayed change of furnace by t2 time;

[0050] And so on, multiple delay times can be set according to actual conditions.

[0051] As a preferred embodiment of the present application, the reward function R is then defined, which takes into account multiple factors: Where w1, w2, and w3 are weight coefficients;

[0052] Rsmelt calculates the reward based on the smelt state evaluation results:

[0053] Excellent: Rsmelt = 1, Normal: Rsmelt = 0, Normal: Rsmelt = -1;

[0054] Rpressure calculates the reward based on the air supply pressure stability: Where ptarget is the target air supply pressure;

[0055] Rchange calculates the reward based on the change of furnace time: Where t is the actual change of furnace time, tbest is the optimal change of furnace time, and k is a constant.

[0056] As a preferred embodiment of the present application, the Q-learning algorithm is finally trained, and the training process is as follows:

[0057] T1: Initialize Q table Q(s, a), where s represents state and a represents action, and initialize all Q values to 0 or a small random value;

[0058] T2: Repeat the following steps until the termination condition is met:

[0059] T21: obtaining a current state s, selecting an action a according to the current state s, and selecting the action by the following strategy:

[0060] T211: randomly selecting an action with a probability of epsilon, and selecting an action that maximizes Q(s,a) with a probability of 1-epsilon, and epsilon gradually decreases during the training process;

[0061] T22: executing the selected action a, observing the reward r and the next state s';

[0062] T23: updating the Q table according to the Q-learning update formula: ;

[0063] wherein is a learning rate, controlling the step size of each update, and gamma is a discount factor, indicating the discount rate of future rewards; T3: after reaching the termination condition, the training is completed;

[0064] After the Q-learning algorithm is trained, the action a corresponding to maxaQ(s,a) is selected according to the current state s as the converter strategy.

[0065] Compared with the prior art, the advantages of the present application are:

[0066] (1) In the present application, the key parameters of the blast furnace are monitored in real time by the smelting state evaluation module, the change rate is calculated and the evaluation coefficient YZP is generated, the running state of the blast furnace is accurately evaluated, early abnormal warning is provided, the process parameters are optimized to improve the efficiency and quality, and the energy consumption is reduced, the model reliability is enhanced based on the proportion factor set according to the actual data, scientific decision is supported, and the flexibility and adaptability of the system are guaranteed by adjusting the proportion factor, thereby improving the safety, stability and efficiency of production;

[0067] (2) In the present application, the historical operation parameters are analyzed by the blast pressure prediction module, the extreme values are removed, and a machine learning model is constructed to accurately predict the blast pressure, which provides early warning, supports process optimization, ensures stable production, enhances the model generalization ability by detailed data processing, flexibly adjusts parameters to adapt to different working conditions, reliable prediction results help production planning and strategy making, improve efficiency and product quality, continuously update and optimize to ensure long-term accuracy, and promote the safety, stability and efficiency of production;

[0068] (3) In the present application, the change furnace optimization control module uses a reinforcement learning algorithm to provide intelligent decision support based on smelting state and blast pressure prediction. By defining the state space and change furnace action, and combining the reward function to comprehensively consider efficiency, cost and stability, dynamic adaptation to different working conditions is achieved. This module improves smelting efficiency, reduces change furnace cost, ensures stable blast furnace operation, optimizes strategies using historical data, responds to environmental changes in real time, adjusts change furnace strategies in a timely manner, and ensures production continuity and safety. BRIEF DESCRIPTION OF DRAWINGS

[0069] Figure 1 is a system block diagram of example one in the present application;

[0070] Figure 2 is a system block diagram of example two in the present application;

[0071] Figure 3 is a logic flow diagram of example one in the present application;

[0072] Figure 4 is a training flowchart of the Q-learning algorithm in example two of the present application;

[0073] Figure 5 is a gas regulation schematic diagram of example three of the present application;

[0074] Figure 6 is another gas regulation schematic diagram of example three of the present application. DETAILED DESCRIPTION

[0075] The technical solutions of the embodiments of the present application will be described below with reference to the accompanying drawings; it is obvious that the described embodiments are only part of the embodiments of the present application; rather than all embodiments; based on the embodiments of the present application; all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the scope of the present application.

[0076] Example one: as shown in Figure 1 and Figure 3 , the present application proposes a blast furnace hot blast stove change furnace control system, applied to a change furnace control platform, comprising:

[0077] A data acquisition module is used to acquire the operating parameters of the blast furnace and hot blast stove in real time, and to preprocess the acquired operating parameters, the preprocessing operation including but not limited to data cleaning, filtering processing and normalization processing;

[0078] The data acquisition module acquires the operation parameters of the blast furnace and the hot blast furnace in real time, and improves the data accuracy and stability through data cleaning, filtering and normalization processing, which not only eliminates errors and noises, but also adjusts the data of different magnitudes to the same scale, thereby optimizing the analysis efficiency and supporting real-time monitoring and decision-making.

[0079] The smelting state evaluation module is used for monitoring and evaluating the smelting state of the blast furnace.

[0080] The specific process of monitoring and evaluating the smelting state of the blast furnace by the smelting state evaluation module is as follows:

[0081] The smelting evaluation parameters of the blast furnace are obtained, including the top temperature, the tuyere temperature, the wind pressure and the coke consumption, a monitoring period is generated, and the monitoring period is set to 24h, and the monitoring period is equally divided into multiple monitoring time periods;

[0082] The top temperature change rate of the blast furnace in the multiple monitoring time periods is obtained, the top temperature change rate represents the ratio between the top temperature change amount and the corresponding time period length, and the arithmetic mean of the obtained multiple top temperature change rates is calculated, and the arithmetic mean of the multiple top temperature change rates is denoted as the average top temperature change rate PLD;

[0083] The average tuyere temperature change rate PFW, the average wind pressure change rate PFY and the average coke consumption change rate PJX can be obtained by using the method of obtaining the average top temperature change rate;

[0084] The average top temperature change rate PLD, the average tuyere temperature change rate PFW, the average wind pressure change rate PFY and the average coke consumption change rate PJX are obtained, and the smelting state evaluation coefficient YZP is calculated by the following formula: ;

[0085] Wherein b1, b2, b3 and b4 are all preset proportional factor coefficients, b4>b3>b2>b1>0, when the specific values of b1, b2, b3 and b4 are selected, the actual production experience and historical data need to be determined, and b1, b2, b3 and b4 reflect the influence degree of different parameters on the overall smelting state;

[0086] The smelting state evaluation coefficient YZP is compared with the preset first smelting state evaluation coefficient threshold and the preset second smelting state evaluation coefficient threshold, and the preset first smelting state evaluation coefficient threshold is less than the preset second smelting state evaluation coefficient threshold;

[0087] If the smelting state evaluation coefficient YZP is less than the preset first smelting state evaluation coefficient threshold, it indicates that the smelting state of the blast furnace is excellent;

[0088] If the smelting state evaluation coefficient YZP is greater than or equal to the preset first smelting state evaluation coefficient threshold, and the smelting state evaluation coefficient YZP is less than the preset second smelting state evaluation coefficient threshold, it indicates that the smelting state of the blast furnace is normal;

[0089] If the smelting state evaluation coefficient YZP is greater than or equal to the preset second smelting state evaluation coefficient threshold, it indicates that the smelting state of the blast furnace is abnormal;

[0090] The smelting state evaluation module monitors the key parameters (such as the furnace top temperature, the tuyere temperature, the wind pressure and the coke consumption) of the blast furnace in real time, calculates the change rate and generates the evaluation coefficient YZP, realizes the accurate evaluation of the running state of the blast furnace, provides early abnormal warning, optimizes the process parameters to improve the efficiency and quality, reduces the energy consumption, enhances the model reliability based on the actual data setting ratio factor, supports scientific decision-making, and adjusts the ratio factor to ensure the flexibility and adaptability of the system, thereby improving the safety, stability and efficiency of production.

[0091] The blast furnace pressure prediction module is used to collect the historical operation parameters of the blast furnace and the hot blast furnace, construct a blast furnace pressure prediction model, and predict the blast furnace pressure in the future period according to the model;

[0092] The specific process of the blast furnace pressure prediction module predicting the blast furnace pressure in the future period is as follows:

[0093] The historical operation parameters of the blast furnace and the hot blast furnace are obtained, the operation parameters include the blast furnace pressure, the blast volume, the blast temperature, the hot blast furnace temperature, the pressure, the blast fan speed and the pressure equalizing valve opening, a collection period is generated, the collection period is set to one month, the collection period is equally divided into m continuous sub-periods, and the midpoint time of each sub-period is marked, and m midpoint times are obtained;

[0094] Taking the m midpoint times as the base point, respectively extending forward and backward for the same length, marking s-1 expansion times, and after collecting the midpoint times and the s-1 expansion times, s detection times are obtained;

[0095] The blast furnace pressure values at the s detection times are measured by instruments respectively, s detection blast furnace pressure values are obtained, and after accumulating the s detection blast furnace pressure values, the average is obtained, m sub-blast furnace pressure values are obtained;

[0096] The expression of the sub-blast furnace pressure value is: ;

[0097] In the formula, ZFYzm is the sub-blast furnace pressure value of the mth sub-period, and ZFYjcmn is the nth detection blast furnace pressure value of the mth sub-period;

[0098] The maximum value and the minimum value of the blast furnace air pressure values of the sub blast furnace are removed, the remaining m-2 sub blast furnace air pressure values are accumulated and averaged to obtain the blast furnace air pressure average value;

[0099] The expression of the blast furnace air pressure average value is: ;

[0100] In the formula, FYjz is the blast furnace air pressure average value, and ZFYzp is the sub blast furnace air pressure value of the pth sub period;

[0101] The method for calculating the blast furnace air pressure average value can be used to obtain the blast furnace air volume average value FLjz, the hot blast furnace air pressure average value RYjz, the blower speed average value GZjz, and the average pressure valve opening average value FKjz.

[0102] The blast furnace air pressure average value FYjz, the blast furnace air volume average value FLjz, the hot blast furnace air pressure average value RYjz, the blower speed average value GZjz, and the average pressure valve opening average value FKjz are combined to construct a blast furnace pressure prediction matrix SYJ. The blast furnace pressure prediction matrix SYJ is used as the input of a machine learning model, and the blast furnace pressure in a future period of time corresponding to each group of blast furnace pressure prediction matrix SYJ is used as the output of the machine learning model. The blast furnace pressure in the future period of time is used as the prediction target, the sum of the prediction errors of all training data is minimized as the training target, the machine learning model is trained until the sum of the prediction errors converges, and the blast furnace pressure prediction model is obtained.

[0103] The expression formula of the blast furnace pressure prediction model is as follows: ;

[0104] Wherein, η1, η2, η3, η4 and η5 are regression coefficients, λ is a random error term, and SFY represents the blast furnace pressure in the future period of time.

[0105] Real-time running parameters of the blast furnace and the hot blast furnace are obtained, which are converted into corresponding blast furnace pressure prediction matrix SYJ and input into the blast furnace pressure prediction model. The real-time blast furnace pressure in the future period of time is obtained through the blast furnace pressure prediction model.

[0106] The historical running parameters are analyzed by the blast furnace pressure prediction module, the extreme values are removed, and the machine learning model is constructed to accurately predict the blast furnace pressure. It provides early warning, supports process optimization, ensures stable production, enhances the generalization ability of the model by detailed data processing, flexibly adjusts parameters to adapt to different working conditions, and reliable prediction results help production planning and strategy making, improve efficiency and product quality, continuously update and optimize to ensure long-term accuracy, and promote the safety, stability and efficiency of production.

[0107] The technical solution of the embodiment of the present application is different from that of the first embodiment.

[0108] As Figure 2 and Figure 4 shown, the furnace replacement optimization control module is used to optimize the furnace replacement strategy by using a reinforcement learning algorithm according to the smelting state evaluation result and the blast pressure prediction result;

[0109] The specific process of the furnace replacement optimization control module using a reinforcement learning algorithm to optimize the furnace replacement strategy is as follows:

[0110] The smelting state evaluation result and the blast pressure prediction result are obtained, the smelting state evaluation result is that the smelting state of the blast furnace is excellent, normal or abnormal, and the blast pressure prediction result is the real-time blast pressure in the future period of time;

[0111] The specific steps of using a reinforcement learning algorithm to optimize the furnace replacement strategy are as follows:

[0112] First, define a state space S, which represents the current state of the environment, and the state space is composed of the smelting state and the blast pressure prediction, and the state space can be represented as a vector: S=[Ssmelt, Spressure];

[0113] Ssmelt is a vector representation of the smelting state, represented by a numerical vector:

[0114] Excellent: [1, 0, 0], normal [0, 1, 0], and abnormal [0, 0, 1];

[0115] Spressure is a continuous variable vector representing the predicted blast pressure value for the next n time steps:

[0116] Spressure=[p1, p2, …, pn];

[0117] Where pi is the predicted blast pressure at the i-th time step, and the size of n depends on the time length of the blast pressure prediction model;

[0118] Then define the action time A, which represents the furnace replacement strategy, discretized as:

[0119] A=0: no furnace replacement;

[0120] A=1: immediate furnace replacement;

[0121] A=2: delay t1 time furnace replacement (for example, t1=5 minutes);

[0122] A=3: delay t2 time furnace replacement (for example, t1=10 minutes);

[0123] … and so on, multiple delay times can be set according to actual conditions;

[0124] Then define the reward function R, which considers multiple factors: where w1, w2 and w3 are weight coefficients, the weight coefficients w1, w2 and w3

[0125] By analyzing historical data, the influence degree of the weight coefficients w1, w2 and w3 on the converter strategy is understood, and the specific values of w1, w2 and w3 are determined according to the historical data, so as to ensure that the defined reward function can balance the smelting efficiency, converter cost and stability of the blast furnace operation;

[0126] Rsmelting is calculated according to the smelting state evaluation result:

[0127] Excellent: Rsmelting = 1, Normal: Rsmelting = 0, Normal: Rsmelting = -1;

[0128] Rpressure is calculated according to the blast pressure stability: where ptarget is the target blast pressure;

[0129] Rconverter is calculated according to the converter time, the more timely the converter is, the higher the reward is, and the converter is punished in advance: where t is the actual converter time, tbest is the optimal converter time, and k is a constant;

[0130] Finally, the Q-learning algorithm is trained, and the training process is as follows:

[0131] T1: Initialize the Q table Q(s, a), where s represents the state and a represents the action, and initialize all Q values to 0 or a small random value;

[0132] T2: Repeat the following steps until the termination condition is met (reach the maximum number of iterations or the Q value converges):

[0133] T21: Obtain the current state s, and select the action a according to the current state s, and select the action by the following strategy:

[0134] T211: Randomly select an action with a probability of ε, and select the action that maximizes Q(s, a) with a probability of 1-ε, and ε gradually decreases during the training process;

[0135] T22: Execute the selected action a, and observe the reward r and the next state s';

[0136] T23: Update the Q table according to the Q-learning update formula: ;

[0137] where is the learning rate ( ), which controls the step size of each update, and γ is the discount factor (0≤γ≤1), which represents the discount rate of future rewards;

[0138] T3: Training ends after the termination condition is met;

[0139] After the Q-learning algorithm is trained, the action 'a' corresponding to maxaQ(s,a) is selected as the furnace switching strategy based on the current state 's'. The furnace switching optimization control module uses reinforcement learning algorithm to provide intelligent decision support based on smelting state and blast pressure prediction. By defining the state space and furnace switching actions, and combining the reward function to comprehensively consider efficiency, cost, and stability, it achieves dynamic adaptation to different operating conditions. This module improves smelting efficiency, reduces furnace switching costs, ensures stable blast furnace operation, optimizes strategies using historical data, responds to environmental changes in real time, and adjusts the furnace switching strategy in a timely manner to ensure production continuity and safety.

[0140] Example 3: Please refer to Figure 5 and Figure 6 As shown, this embodiment discloses a blast furnace hot blast stove replacement control method, and provides a detailed explanation of how to regulate gas to avoid fluctuations in air supply pressure. A gas storage tank (such as...) is added to the original hot blast stove system. Figure 5 (as shown)

[0141] Coal gas enters the hot blast stove through the coal gun pipe, passing through the gas flow meter, gas shut-off valve 1, gas shut-off valve 2, and gas flow regulating valve. Combustion air enters the hot blast stove through the pipe, passing through the air flow meter, combustion air shut-off valve, and combustion air flow regulating valve, where it mixes with the coal gas and is burned. The exhaust gas from combustion passes through the flue valve and is discharged through the flue gas pipe. When the above valves (combustion air shut-off valve, gas shut-off valve 1, gas shut-off valve 2, and flue valve) are closed, the furnace operation stops, and excess coal gas in the pipe is discharged through the gas vent valve.

[0142] High-pressure nitrogen and oxygen enter the gas storage tank through the pressurization valve. The gas storage tank is equipped with an oxygen detector to ensure that the nitrogen and oxygen ratio is consistent with that of air. When the pressure in the gas storage tank reaches the target pressure, the oxygen and nitrogen pressurization valves are closed. After the hot blast stove has finished firing, when air needs to be sent to the blast furnace, the external gas source pressurization valve is opened. When the pressure in the hot blast stove is consistent with the pressure in the cold air pipeline, the cold air valve is opened, the external gas source pressurization valve is closed, and the nitrogen and oxygen lance pressurization valves are reopened to pressurize the gas storage tank.

[0143] This process can effectively avoid the problem of large fluctuations in the pressure of the cold air duct during the original process of balancing the pressure of the hot air furnace and the cold air duct through the pressure valve, and at the same time, it can significantly shorten the pressure time and avoid the hot air furnace temperature from dropping due to excessive time.

[0144] Cold air enters the hot blast stove through the cold air valve. After being heated, the air passes through the hot blast valve and enters the blast furnace through the hot blast duct. Another part of the cold air passes through the mixing air flow regulating valve and the mixing air cut-off valve, and mixes with the heated air to regulate the temperature of the hot blast entering the blast furnace, thus completing the blast furnace air supply.

[0145] After the air supply is completed, the cold air valve, the hot air valve, and the mixed air cut-off valve are closed, the pressure relief valve is opened to discharge the high-pressure air in the hot blast furnace, and when the pressure of the hot blast furnace is balanced with the pressure of the flue gas pipeline, the pressure relief valve is closed, the combustion-supporting air cut-off valve, the gas cut-off valve 1, the gas cut-off valve 2, and the flue valve are opened again, and a new cycle of furnace burning and air supply is performed;

[0146] The gas source of the gas tank can come from an air compressor, as shown in the figure. Figure 6 As shown, the flow rates of the gas and the combustion-supporting air are measured by a gas flow meter and an air flow meter respectively, and the ratio of the gas to the combustion-supporting air is adjusted by a gas flow regulating valve and a combustion-supporting air regulating valve to ensure the stability of the combustion process.

[0147] A temperature detector is arranged in the hot air pipeline to adjust the ratio of the hot air to the cold air by a mixed air ratio regulating valve so that the temperature meets the requirements of the blast furnace.

[0148] The above is only a preferred specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can make equivalent replacements or changes to the technical solution and the improved concept of the present application within the technical range disclosed by the present application, which should be covered by the protection scope of the present application.

Claims

1. A blast furnace hot blast stove changeover control system, applied to a changeover control platform, characterized in that, include: The data acquisition module is used to collect the operating parameters of the blast furnace and hot blast stove in real time and to perform preprocessing operations on the collected operating parameters; The smelting status assessment module is used to monitor the smelting assessment parameters of the blast furnace in real time and to monitor and assess the smelting status of the blast furnace. The air supply pressure prediction module is used to collect historical operating parameters of blast furnace and hot blast stove, build an air supply pressure prediction model, and predict the air supply pressure for a future period of time based on the model. The furnace switching optimization control module is used to optimize the furnace switching strategy based on the smelting status assessment results and the air supply pressure prediction results using reinforcement learning algorithms. The specific process by which the furnace replacement optimization control module optimizes the furnace replacement strategy using reinforcement learning algorithms is as follows: The smelting condition assessment results and the air supply pressure prediction results are obtained. The smelting condition assessment results indicate whether the blast furnace is in an excellent, normal or abnormal smelting condition, and the air supply pressure prediction results indicate the real-time air supply pressure for a period of time in the future. The specific steps for optimizing the furnace switching strategy using reinforcement learning algorithms are as follows: First, define a state space S, which consists of the smelting state and the predicted air supply pressure. The state space can be represented as a vector: ; It is a vector representation of the smelting state, expressed as a numerical vector: Excellent: [1,0,0], Normal: [0,1,0], Abnormal: [0,0,1]; It is a continuous variable vector representing the predicted air supply pressure values ​​for the next n time steps: ； in It is the predicted supply air pressure at the i-th time step; Let the action time A be redefined, and the action represent the furnace switching strategy, which can be discretized as follows: A=0: No furnace replacement; A=1: Replace the furnace immediately; A=2: Delay the furnace changeover by time t1; A=3: Delay furnace replacement by time t2; ...and so on, multiple delay times can be set according to the actual situation; Next, we define the reward function R, which takes into account multiple factors: , where w1, w2 and w3 are all weighting coefficients; Rewards are calculated based on the smelting condition assessment results: Excellent: ; Rewards are calculated based on the stability of the supply air pressure. ,in It is the target air supply pressure; Rewards are calculated based on furnace changeover time: Where t is the actual furnace replacement time. For the optimal furnace changeover time, k is a constant; Finally, the Q-learning algorithm is trained, and the training process is as follows: T1: Initialize the Q-table Q(s,a), where s represents the state and a represents the action. Initialize all Q values ​​to 0 or a small random value. T2: Repeat the following steps until the termination condition is met: T21: Obtain the current state s, and select action a based on the current state s, using the following strategy: T211: Randomly select an action with probability ε, and select the action that maximizes Q(s,a) with probability 1-ε, where ε gradually decreases during the training process; T22: Execute the selected action a, observe the reward r and the next state s'; T23: Update the Q table based on the Q-learning update formula: ； in The learning rate controls the step size for each update, and γ is the discount factor, representing the discount rate for future rewards. T3: Training ends after the termination condition is met; After the Q-learning algorithm is trained, select based on the current state s. The corresponding action 'a' is used as the furnace replacement strategy.

2. The blast furnace hot blast stove changeover control system according to claim 1, characterized in that, The specific process by which the smelting condition assessment module monitors and assesses the smelting condition of the blast furnace is as follows: The smelting evaluation parameters of the blast furnace are obtained, including furnace top temperature, tuyere temperature, air pressure and coke consumption. A monitoring cycle is generated and the monitoring cycle is divided into multiple monitoring periods. The furnace top temperature change rate is obtained in multiple monitoring periods. The furnace top temperature change rate represents the ratio between the change in furnace top temperature and the duration of the corresponding time period. The arithmetic mean of the multiple furnace top temperature change rates is calculated and recorded as the average furnace top temperature change rate PLD. Similarly, the average tuyere temperature change rate PFW, average air pressure change rate PFY, and average coke consumption change rate PJX can be obtained using the same method as for calculating the average furnace top temperature change rate.

3. The blast furnace hot blast stove changeover control system according to claim 2, characterized in that, The average furnace top temperature change rate PLD, average tuyere temperature change rate PFW, average blast pressure change rate PFY, and average coke consumption change rate PJX are obtained. The smelting condition assessment coefficient YZP is calculated using the following formula: ； Where b1, b2, b3 and b4 are all preset proportional factor coefficients, b4>b3>b2>b1>0. The smelting state evaluation coefficient YZP is compared with the preset first smelting state evaluation coefficient threshold and the preset second smelting state evaluation coefficient threshold. The preset first smelting state evaluation coefficient threshold is less than the preset second smelting state evaluation coefficient threshold. If the smelting condition evaluation coefficient YZP is less than the preset first smelting condition evaluation coefficient threshold, it indicates that the blast furnace has an excellent smelting condition. If the smelting condition evaluation coefficient YZP is greater than or equal to the preset first smelting condition evaluation coefficient threshold, and the smelting condition evaluation coefficient YZP is less than the preset second smelting condition evaluation coefficient threshold, it indicates that the smelting condition of the blast furnace is normal. If the smelting condition evaluation coefficient YZP is greater than or equal to the preset second smelting condition evaluation coefficient threshold, it indicates that the smelting condition of the blast furnace is abnormal.

4. The blast furnace hot blast stove changeover control system according to claim 1, characterized in that, The specific process by which the air supply pressure prediction module predicts the air supply pressure over a future period is as follows: The historical operating parameters of the blast furnace and hot blast stove are obtained, including blast furnace air pressure, air volume, air temperature, hot blast stove temperature, pressure, blower speed and pressure equalization valve opening. A collection period is generated, and the collection period is divided into m consecutive sub-periods. The midpoint time of each sub-period is marked to obtain m midpoint times. Using m midpoint times as base points, extend forward and backward by equal time intervals to mark s-1 extended times. Then, combine the midpoint times and s-1 extended times to obtain s detection times. The blast furnace air pressure values ​​are measured at s detection times using instruments to obtain s detected blast furnace air pressure values. The average of the s detected blast furnace air pressure values ​​is then calculated to obtain m sub-blast furnace air pressure values.

5. A blast furnace hot blast stove changeover control system according to claim 4, characterized in that, The expression for the blast furnace air pressure value is: ; In the formula, The sub-blast furnace air pressure value for the m-th sub-cycle. This refers to the nth detected blast furnace air pressure value in the mth sub-cycle; Remove the maximum and minimum values ​​of the sub-blast furnace air pressure, and then sum the remaining m-2 sub-blast furnace air pressure values ​​and calculate the average value to obtain the average blast furnace air pressure. The expression for the average blast furnace air pressure is: ; In the formula, This represents the average blast furnace air pressure. Let be the sub-blast furnace air pressure value for the p-th sub-cycle; The average blast furnace air volume can be obtained by analogy using the same method as calculating the average blast furnace air pressure. Average air pressure of hot blast stove Average blower speed Average opening of equalizing valve .

6. The blast furnace hot blast stove changeover control system according to claim 5, characterized in that, Average blast furnace air pressure Average blast furnace air volume Average air pressure of hot blast stove Average blower speed and the average opening value of the equalizing valve A combined air supply pressure prediction matrix SYJ is constructed. The air supply pressure prediction matrix SYJ is used as the input of the machine learning model, and the air supply pressure corresponding to a certain period of the future for each set of air supply pressure prediction matrices SYJ is used as the output of the machine learning model. The air supply pressure in the future period is used as the prediction target, and the sum of prediction errors of all training data is used as the training objective. The machine learning model is trained until the sum of prediction errors converges, and then the training stops, thus obtaining the air supply pressure prediction model.

7. A blast furnace hot blast stove changeover control system according to claim 6, characterized in that, The formula for the air supply pressure prediction model is as follows: ； Where η1, η2, η3, η4 and η5 are regression coefficients, λ is the random error term, and SFY represents the air supply pressure in the future period; The real-time operating parameters of the blast furnace and hot blast stove are obtained and converted into the corresponding air supply pressure prediction matrix SYJ. This matrix is ​​then input into the air supply pressure prediction model, which is used to obtain the real-time air supply pressure for a future period of time.

Images