Intelligent injection rate control method for blast furnace coal injection system

Abstract

The invention discloses an intelligent injection rate control method for a blast furnace coal injection system, which comprises the following steps: collecting blast furnace coal injection historical production data, and filtering the data to obtain a sample database; an improved CNN-Multihead-Attention model is adopted to create a tank pressure prediction model and a gas carrying capacity prediction model, and a genetic algorithm is adopted to optimize model hyper-parameters; the injection rate is adjusted in a combined manner of a feedforward adjustment based on data driving and a feedback adjustment based on a deviation between an injection rate set value and an actual value. Feed-forward adjustment of the injection rate is achieved through adjustment of the tank pressure, the adjustment effect is further optimized based on feedback adjustment of injection errors as an auxiliary means, high stability and high accuracy of the injection rate are enhanced, the labor cost is reduced, and real-time accurate control over the target value of the injection rate is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of automatic control of blast furnace coal injection systems, in particular to an injection rate intelligent control method for blast furnace coal injection systems. Background Art

[0002] Blast furnace coal injection (BFI) is a crucial component of the current ironmaking process and an effective means of conserving coke, reducing consumption, and regulating furnace conditions. Therefore, BFs place high demands on the continuity and stability of the BFI process, making it crucial to achieve a high level of automated control during this process.

[0003] At present, most steel mills' coal injection systems are operated by traditional manual experience. Job workers rely on experience to control tank pressure, fluidizing gas flow, carrier gas volume, etc. to stabilize the injection rate. However, there are many factors that affect the injection rate, such as coal powder quality, humidity, temperature, fluidity, coal powder regulating valve opening, etc. These parameters often change in actual production. Manual adjustment alone cannot respond in real time, and often loses sight of one thing while focusing on another. It is difficult to arrive at the right time, consumes a lot of manpower, and the control effect is poor, affecting the quality and efficiency of blast furnace production.

[0004] In the existing technology, there is also a method of tracking and controlling the injection rate through PID adjustment (Chinese invention patent application publication number CN105368996A), but there are still certain defects. For example, the control accuracy using fuzzy control theory is poor, and the fuzzy control rules cannot be adjusted in real time according to changes in working conditions, and the generalization ability is weak; simply adjusting based on the difference between the actual injection rate and the set injection rate has a large lag. Summary of the Invention

[0005] In order to overcome the above problems existing in the prior art, the present invention proposes an injection rate intelligent control method for a blast furnace coal injection system.

[0006] The technical solution adopted by the present invention to solve the technical problem is: an injection rate intelligent control method for a blast furnace coal injection system, comprising the following steps: Step 1, data collection: collect historical production data of blast furnace coal injection, and filter the data to obtain a sample database; Step 2, model creation: Use the improved CNN-Multihead-Attention model to create tank pressure prediction models and gas volume prediction models, and use genetic algorithms to optimize the model hyperparameters; Step 3, injection rate control: This includes feedforward regulation and feedback regulation. The feedforward regulation is specifically as follows: based on the tank pressure prediction model and the carrier gas volume prediction model obtained in step 2, predictions are made on the sample database obtained in step 1 respectively. The tank pressure prediction model outputs a target pressure value for the injection tank, which is achieved by adjusting the controllers of the air supply regulating valve and the small relief valve. The carrier gas volume prediction model outputs an appropriate carrier gas volume and transmits it to the controller for control; The feedback regulation is specifically as follows: real-time monitoring of the difference between the actual value and the set value of the injection rate, calculation of the average error ME in the past three seconds, and adjustment of the opening of the fluidization control valve and the pulverized coal control valve according to the size of the average error ME.

[0007] The above-mentioned method for intelligent control of the injection rate of a blast furnace coal injection system, the step 1 specifically comprises: collecting historical production data of blast furnace coal injection, including injection rate, carrier gas volume, fluidizing gas flow, coal loading volume, coal powder humidity, coal powder temperature, tank pressure, and coal powder regulating valve opening, using KepServer configuration software to collect data from the PLC once per second and store them in the database; based on the collected original sample data, the data is mean filtered with a time period of 5 seconds, and the processed data is used to construct a sample database, and the sample data that has completed injection is divided into a training data set and a verification data set in a ratio of 8:2, and the sample data that has not completed injection is used as a test data set.

[0008] In the above-mentioned intelligent control method for the injection rate of a blast furnace coal injection system, the improved CNN-Multihead-Attention model in step 2 specifically comprises: using the BatchNormalization function after each convolutional layer; using the Adam optimization algorithm to optimize the convergence speed of the objective function; inputting the features extracted by the CNN into the Multi-head Attention mechanism, calculating the attention weights of each time step and each variable, and weighting the results.

[0009] The above-mentioned method for intelligent control of injection rate of blast furnace coal injection system, in step 2, uses genetic algorithm to optimize model hyperparameters, specifically: CNN model parameters and the number of heads of Multi-head Attention are encoded as chromosomes, the search space is determined according to specific working conditions or experience, and N groups of hyperparameter combinations are randomly generated as initial solutions; the evaluation indicator uses mean square error as the fitness value, and the model training and evaluation are completed under a fixed number of training times; new populations are generated through the three genetic operator operations of selection, crossover, and mutation until the termination condition is reached, and the optimal hyperparameter combination is determined.

[0010] In the above-mentioned intelligent control method for injection rate of coal injection system of blast furnace, the average error ME is calculated as follows: in, Indicates the actual value, represents the target value, and n represents the number of samples.

[0011] In the above-mentioned intelligent control method for injection rate of a blast furnace coal injection system, the feedback adjustment in step 3 is specifically as follows: when the average error ME is ≥ 0.5, the opening of the fluidization control valve and the pulverized coal control valve is reduced by 1% every 10 seconds; When the average error ME ≥ 0 and ME < 0.5, the openings of the fluidization control valve and the pulverized coal control valve remain unchanged; When the average error ME is less than 0, the opening of the fluidization control valve and the pulverized coal control valve increases by 1% every 10 seconds.

[0012] The beneficial effects of the present invention are: 1) The present invention uses an improved convolutional neural network model to train high-quality samples, deeply explores the intrinsic relationship between tank pressure and injection rate, and between carrier gas volume and injection rate under different working conditions, and forms knowledge, so that when the working conditions change, the tank pressure and carrier gas volume are adjusted to appropriate values in real time, realizing feedforward regulation and ensuring the stability and accuracy of the injection rate at all times.

[0013] 2) The adjustment of tank pressure realizes feedforward regulation of the injection rate. Feedback regulation based on injection error is used as an auxiliary means to further optimize the regulation effect and enhance the high stability and accuracy of the injection rate.

[0014] 3) It not only combines the operating experience of on-site personnel, but also integrates the concept of optimization settings into the control, which not only reduces labor costs but also achieves real-time and accurate control of the injection rate target value. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic flow chart of the present invention. DETAILED DESCRIPTION

[0016] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] like Figure 1As shown, this embodiment discloses an intelligent control method for the injection rate of a blast furnace coal injection system, which mainly realizes the stability and uniformity of the injection rate by combining data-driven feedforward regulation with feedback regulation based on the deviation between the injection rate set value and the actual value. This solution constructs a sample library by collecting historical production data, and uses deep learning methods to train the optimal tank pressure prediction model and the carrier gas volume prediction model. When the injection rate target value or other influencing factors change, the model outputs the appropriate tank pressure value and the appropriate carrier gas volume value in real time to realize feedforward regulation. However, the number of factors affecting the injection rate is large and the complexity is large, and some variables cannot be collected or quantified. If the injection rate error exceeds a certain threshold after feedforward regulation, feedback regulation of the fluidization control valve and the pulverized coal control valve is started as a compensation measure to make the injection rate more stable and uniform. Specifically, it includes the following steps: S1. Data Collection Collect historical production data of blast furnace coal injection, mainly including injection rate, carrier gas volume, fluidizing gas flow, coal charge, coal powder humidity, coal powder temperature, tank pressure, coal powder regulating valve opening, etc. Use KepServer configuration software to collect data from PLC once per second and store it in the database.

[0018] S2. Sample database construction Based on the collected original sample data, in order to improve the authenticity and validity of the data, the data was mean filtered with a time period of 5 seconds, and then the processed data was used to construct a sample database. The sample data that had completed the injection was divided into a training data set and a verification data set in an 8:2 ratio, and the sample data that had not completed the injection was used as the test data set.

[0019] S3, optimal model creation Because the blast furnace coal injection production process has strong nonlinear and tightly coupled characteristics, the tank pressure prediction model and the carrier gas volume prediction model are both created and trained using an improved CNN-Multihead-Attention model. A genetic algorithm (GA) is used to optimize the model hyperparameters. The model is evaluated on a validation set, and the optimal parameters are selected for retraining to obtain the optimal model, which is then saved for later use.

[0020] The input variables of the tank pressure prediction model are carrier gas volume, fluidizing gas flow, coal loading, pulverized coal humidity, pulverized coal temperature, pulverized coal regulating valve opening, and actual value of injection rate, and the output variable is the injection tank pressure; the input variable of the carrier gas volume prediction model is the actual value of injection rate, and the output variable is the carrier gas volume.

[0021] In view of the shortcomings of traditional CNN, this embodiment makes the following improvements to the CNN model: (1) The BatchNormalization function is used after each convolutional layer, which not only simplifies the parameter adjustment process, but also greatly improves the training speed, greatly accelerates the convergence process, and effectively prevents overfitting.

[0022] (2) The Adam optimization algorithm is used to optimize the convergence speed of the objective function, improve the model training speed, and prevent the occurrence of local optimal points.

[0023] (3) Introducing a multi-head attention mechanism into CNN to enhance attention to important features. The features extracted by CNN are input into the multi-head attention mechanism, which calculates the attention weights for each time step and each variable, and weights the results to highlight important information.

[0024] The CNN-Multihead-Attentionn model hyperparameters are optimized based on the GA genetic algorithm. The key parameters to be optimized are generally selected: such as the number of convolutional layers, convolution kernel size, batch size, learning rate, number of multi-heads, etc. The specific steps are as follows: 1. Hyperparameter encoding and population initialization Encode CNN model parameters (such as convolution kernel size, number of convolution layers, pooling layer size, learning rate, batch size, etc.) and the number of heads for Multi-head Attention as chromosomes. Either binary or real number encoding can be used. Set the parameter range, and determine the search space based on specific working conditions or experience. For example, consider 1-5 convolution layers, 1-5 convolution kernel sizes, 16-128 batch sizes, 0.001-0.1 learning rates, and 4-16 heads for Multi-head Attention. Randomly generate N sets of hyperparameter combinations as initial solutions, such as 50-100 individuals.

[0025] 2. Fitness function design The evaluation indicator uses mean-squared error (MSE) as the fitness value, and completes model training and evaluation under a fixed number of training times.

[0026] The mean square error (MSE) calculation formula is: in, represents the predicted value, represents the actual value, and n represents the number of samples.

[0027] 3. Genetic operator operation New populations are generated through the three genetic operators of selection, crossover, and mutation until the termination condition is reached and the optimal hyperparameter combination is determined.

[0028] According to the GA-CNN-Multihead-Attention model, the optimal tank pressure prediction model and carrier gas volume prediction model are obtained and saved.

[0029] S4, spray rate control It is divided into two steps: feedforward regulation and feedback regulation: 1. Data-driven feedforward regulation The tank pressure prediction model and carrier gas volume prediction model generated by S3 are used to predict the test data set in S2. The tank pressure prediction model outputs an appropriate target pressure for the injection tank, which is automatically achieved by adjusting the PID controllers of the air supply control valve and the small relief valve. To ensure the safe and efficient operation of the blast furnace coal injection system, the tank pressure target value is set within a range based on operating conditions, such as 0.8-1.5MPa. If the model output value exceeds the upper limit of the range, the upper limit is set as the target value; if it is below the lower limit, the lower limit is set as the target value. The carrier gas volume prediction model outputs an appropriate carrier gas volume and transmits it to the PLC for control.

[0030] 2. Results-based feedback regulation The system monitors the difference between the actual value and the set value of the injection rate in real time, calculates the average error ME of the past three seconds, and adjusts the opening of the fluidization control valve and the pulverized coal control valve according to the size of the average error ME.

[0031] When the average error ME is ≥ 0.5, the opening of the fluidization control valve and the pulverized coal control valve is reduced by 1% every 10 seconds; When the average error ME ≥ 0 and ME < 0.5, the openings of the fluidization control valve and the pulverized coal control valve remain unchanged; When the average error ME is less than 0, the opening of the fluidization control valve and the pulverized coal control valve increases by 1% every 10 seconds; The calculation formula of mean error ME is: in, Indicates the actual value, Represents the target value, n represents the number of samples, which is 3 here.

[0032] This embodiment collects historical production data and uses intelligent algorithms to summarize previous experience in injection rate control to form knowledge, so that it can intelligently learn, process, and solve the injection rate fluctuation problems caused by changes in tank pressure, fluidizing gas flow rate, and carrier gas volume, thereby reducing labor costs and achieving real-time and precise control of the injection rate target value.

[0033] The above embodiments are merely exemplary embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art may make various modifications or equivalent substitutions to the present invention within the spirit and scope of protection of the present invention, and such modifications or equivalent substitutions shall also be deemed to fall within the scope of protection of the present invention.

Claims

1. An intelligent control method for the injection rate of a blast furnace coal injection system, characterized in that: The steps include: Step 1, data collection: collect historical production data of blast furnace coal injection, and filter the data to obtain a sample database; Step 2, model creation: Use the improved CNN-Multihead-Attention model to create tank pressure prediction models and gas volume prediction models, and use genetic algorithms to optimize the model hyperparameters; Step 3, injection rate control: This includes feedforward regulation and feedback regulation. The feedforward regulation is specifically as follows: based on the tank pressure prediction model and the carrier gas volume prediction model obtained in step 2, predictions are made on the sample database obtained in step 1 respectively. The tank pressure prediction model outputs a target pressure value for the injection tank, which is achieved by adjusting the controllers of the air supply regulating valve and the small relief valve. The carrier gas volume prediction model outputs an appropriate carrier gas volume and transmits it to the controller for control; The feedback regulation is specifically as follows: real-time monitoring of the difference between the actual value and the set value of the injection rate, calculation of the average error ME in the past three seconds, and adjustment of the opening of the fluidization control valve and the pulverized coal control valve according to the size of the average error ME.

2. The method for intelligently controlling the injection rate of a blast furnace coal injection system according to claim 1, characterized in that: The step 1 is specifically as follows: collecting historical production data of blast furnace coal injection, including injection rate, carrier gas volume, fluidizing gas flow, coal loading, coal powder humidity, coal powder temperature, tank pressure, and coal powder regulating valve opening, using KepServer configuration software to collect data from the PLC once per second and store them in the database; based on the collected original sample data, performing mean filtering on the data with a time period of 5 seconds, and then using the processed data to construct a sample database, the sample data that has completed injection is divided into a training data set and a verification data set in a ratio of 8:2, and the sample data that has not completed injection is used as a test data set.

3. The method for intelligently controlling the injection rate of a blast furnace coal injection system according to claim 1, wherein: The improved CNN-Multihead-Attention model in step 2 is specifically as follows: using the BatchNormalization function after each convolutional layer; using the Adam optimization algorithm to optimize the convergence speed of the objective function; inputting the features extracted by CNN into the Multi-head Attention mechanism, calculating the attention weights of each time step and each variable, and weighting the results.

4. The method for intelligently controlling the injection rate of a blast furnace coal injection system according to claim 1, wherein: In step 2, the genetic algorithm is used to optimize the model hyperparameters as follows: the CNN model parameters and the number of heads of Multi-head Attention are encoded as chromosomes, the search space is determined according to specific working conditions or experience, and N groups of hyperparameter combinations are randomly generated as initial solutions; the mean square error is used as the fitness value as the evaluation indicator, and the model training and evaluation are completed under a fixed number of training times; new populations are generated through the three genetic operators of selection, crossover, and mutation until the termination condition is reached to determine the optimal hyperparameter combination.

5. The method for intelligently controlling the injection rate of a blast furnace coal injection system according to claim 1, wherein: The mean error ME is calculated as follows: in, Indicates the actual value, represents the target value, and n represents the number of samples.

6. The method for intelligently controlling the injection rate of a blast furnace coal injection system according to claim 5, characterized in that: The feedback adjustment in step 3 is specifically as follows: when the average error ME is greater than or equal to 0.5, the opening of the fluidization control valve and the pulverized coal control valve is reduced by 1% every 10 seconds; When the average error ME ≥ 0 and ME < 0.5, the openings of the fluidization control valve and the pulverized coal control valve remain unchanged; When the average error ME is less than 0, the opening of the fluidization control valve and the pulverized coal control valve increases by 1% every 10 seconds.

Images