Dynamic intelligent self-adaptive blast furnace pushing control method
Through the dynamic intelligent adaptive control method, the GAN and IDP algorithms and the neural network controller are used to solve the problems of instability and inefficiency in the traditional blast furnace thrust method, and the precise and efficient control of blast furnace thrust and safe production are achieved.
Patent Information
- Application Number
- CN202410997405.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-12
AI Technical Summary
The traditional blast furnace transfer method relies on manual operation and simple mechanical control, resulting in instability and inefficiency, difficulty in achieving precise control, and safety risks.
Using dynamic intelligent adaptive control method, the virtual lapse data is generated by generating adversarial networks (GANs), combined with intelligent adaptive dynamic optimization algorithm (IDP) and a robust adaptive controller based on neural networks, the blast furnace lapse parameters can be monitored and adjusted in real time to achieve Accurate and efficient control.
It realizes accurate and efficient control of the blast furnace pushing process, ensures stable and reliable pushing, optimizes energy consumption, reduces costs, and improves blast furnace recovery and production efficiency.
Smart Images

Figure CN118938674A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of industrial automation and intelligent control, and in particular to a dynamic intelligent adaptive blast furnace movement control method. Background Art
[0002] In the steel smelting industry, the blast furnace is one of the key equipment, and its operating efficiency and safety directly affect the operating conditions of the entire production line. With the continuous advancement of science and technology and the ever-changing market demand, the expansion, transformation or overhaul of blast furnaces has become a necessary production link to ensure that they continue to meet production requirements. In these important processes, the pushing operation of the blast furnace has become a vital task; however, traditional blast furnace pushing methods often rely on manual operation and simple mechanical control. Manual operation is easily affected by human factors, resulting in increased instability and uncertainty in the pushing process; and simple mechanical control cannot be accurately adjusted according to the real-time status of the blast furnace, and it is difficult to meet the complex and changeable pushing needs; this makes it impossible to achieve precise control of the blast furnace push, and there are problems such as high safety risks and low efficiency. Summary of the invention
[0003] The technical problem to be solved by the present invention is to provide a dynamic intelligent adaptive blast furnace movement control method, which is suitable for the expansion, transformation, maintenance or other scenarios where the blast furnace needs to be moved in the steel smelting industry.
[0004] The technical solution adopted by the present invention to solve its technical problem is:
[0005] A dynamic intelligent adaptive blast furnace movement control method comprises the following steps:
[0006] S1. Collect the recorded data of the actual blast furnace movement and the operating data of the sensors on the blast furnace;
[0007] S2, training a generative adversarial network (GAN) with recorded data and running data to generate virtual shift data;
[0008] S3, inputting the actual recorded data and the virtual shift data in step S1 into the intelligent adaptive dynamic optimization algorithm (IDP), training the blast furnace shift control model, and establishing the optimization control strategy for the blast furnace shift;
[0009] S4. Combined with a robust adaptive controller based on a neural network, real-time identification and adjustment of parameters are performed to achieve blast furnace movement control.
[0010] In step S2, the generative adversarial network (GAN) can generate virtual transition data by learning the distribution of actual data.
[0011] In step S3, the intelligent adaptive dynamic optimization algorithm (IDP) ensures that the relevant control quantities of the blast furnace moving direction and speed remain bounded through the constraints of the controller.
[0012] In step S3, the formula of the intelligent adaptive dynamic optimization algorithm (IDP) is as follows:
[0013]
[0014] In formula (1), s0(x) is the initial control law; Y0 is the initial performance index function; h(x) is the error for any initial state;
[0015] R(h(x)) and F(s(x)) are both positive definite functions. (R m is an m-dimensional real vector space), the calculation formula is as follows:
[0016]
[0017] In formula (II), h T is the transpose of the initial state error h(x); set up is the saturation bound of the controller constraint, s(x) is the controller constraint input; Q and R are diagonal positive definite matrices; dv represents the integral of v; and is a monotone bounded increasing odd function that satisfies Its first-order derivative is bounded by a constant;
[0018] The performance indicator function is further updated as follows:
[0019] Y1(h(x))=R(h(x))+F(s0(x))+Y0(h(x+1)) (3)
[0020] In formula (3), h(x+1)=f(h(x)+η(x))+g(h(x)+η(x))(s0(x)+uε(x))-η(x+1);
[0021] Then iterate according to the following formula:
[0022] s i (x) = argmin s(x) {R(h(x))+F(s(x))+Y i (h(x+1))}
[0023] Y i+1 (h(x))=R(h(x))+F(s0(x))+Y i(h(x+1))
[0024] h(x+1)=f(h(x)+η(x))+g(h(x)+η(x))(s i (x)+u h (x))-η(x+1)
[0025] In the above formula, s i (x) is the control input after the i-th iteration; η(x) is the reference trajectory; Y i is the performance index function after the i-th iteration; u h (x) is steady-state input control; is a Lipschitz continuous function; R and F are positive definite functions in formula 2.
[0026] In step S4, the neural network robust adaptive controller is a BP neural network based model to simulate the unknown nonlinear characteristics of the blast furnace movement and convert it into an input-output mode as follows:
[0027]
[0028] In formula (IV), Z is the output (the predicted output value of the displacement related to the blast furnace), u(x) represents the input at time x (the control input value of the displacement related to the blast furnace), d is the relative order of the system, and c(x) is the system state vector; g d-1 and f d-1 It is a nonlinear function after a multi-layer neural network; w is the weight connecting the input layer and the hidden layer; s is the weight connecting the hidden layer and the output layer;
[0029] Said and Determined by the neural network structure and the number of neurons, assuming that the number of hidden neurons in the network is a and b respectively, the expression is as follows:
[0030]
[0031] In formula (5), w i and i are the weights of the i-th neuron; w ij and ij are the weights between the ith neuron and the jth input; the difference is that w is the weight associated with the input vector c(x), and s is the weight associated with the control input u(x); P(·) is a continuous, bounded, and non-constant monotonically increasing function, Its control law is as follows:
[0032]
[0033] In formula (6), r is the reference output (indicating the expected system output, target value); u(x) is the control input at time x, which is used to adjust the operation of the system; c(x) is the input vector at time x, containing information about the current state of the system; w(x) and s(x) are both weight vectors of the neural network at time x; r(x) is the reference signal;
[0034] The control quantity u(x) acts on the displacement system object and the neural network model; the weight of the neural network can be adjusted in real time to achieve optimization based on the error between the actual output of the system and the predicted output of the model; in order to quantify the difference, we introduce an error indicator, which is expressed as follows:
[0035] z(x+1)=f d-1 [c(x-d+1)]+g d-1 [c(x-d+1)]u(x-d+1)
[0036]
[0037] At the same time, the estimated output and system error of the model are as follows:
[0038]
[0039] h * (x+1)=z * (x+1)-z(x+1)
[0040] In the above formula, z is the estimated output of the model after neural network adaptive control; c is the real-time input vector; h is the system error; s is the weight related to the control input; w is the weight related to the input vector; and u is the control input used to adjust the system.
[0041] The sensors include position sensors, speed sensors and installation force sensors, which are used to monitor various parameters in the blast furnace moving process in real time.
[0042] The beneficial effects of the present invention are:
[0043] 1. By using GAN, IDP algorithm and robust adaptive control based on neural network, it is possible to monitor the blast furnace movement data received by each sensor in real time, dynamically and intelligently calculate the optimal control parameters, and drive the optimal movement instructions to adapt to the changing conditions during the blast furnace movement process. Accurate and efficient control of the blast furnace movement process is achieved, ensuring that the blast furnace is stably and reliably moved to the designated position, enabling efficiency improvement for the safe implementation of the blast furnace movement project and the smooth resumption of blast furnace production.
[0044] 2. Through real-time monitoring of key node data of the blast furnace pushing process, dynamic calculation of the optimal control parameters of the blast furnace pushing model, optimization of pushing instructions, adaptation to different working conditions during the blast furnace pushing process, optimization of energy consumption, and reduction of input costs, the safety and stability of the blast furnace pushing process can be ensured, thereby enabling and improving the efficiency of the safe implementation of the blast furnace pushing project and the smooth resumption of blast furnace production.
[0045] 3. Generative adversarial networks (GANs) can effectively solve the problem that it may be difficult to obtain enough displacement data in actual production. By learning the distribution of actual data through GAN, a large amount of virtual displacement data can be generated. These data can be used to enhance the diversity and coverage of training samples and improve the generalization ability of the algorithm. At the same time, the actual collected blast furnace operation data can simulate various displacement scenarios in the virtual data generated by GAN, so that the control algorithm can be trained under different working conditions, thereby improving its ability to cope with different displacement phenomena.
[0046] 4. By inputting actual and virtual displacement data into the IDP algorithm, the blast furnace displacement control model is trained and the optimal control strategy for blast furnace displacement is established; the real-time operating parameters of the blast furnace of the displacement system (hydraulic temperature, displacement and speed, etc.) are effectively improved, and the optimal control strategy is quickly obtained, while limiting the fluctuation range of the blast furnace displacement control quantity. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION
[0048] The invention is further described below with reference to the accompanying drawings and embodiments.
[0049] Implementation 1
[0050] A dynamic intelligent adaptive blast furnace movement control method comprises the following steps:
[0051] S1. Collect the recorded data of the actual blast furnace movement in the past and the operating data of the sensors at the key parts of the blast furnace;
[0052] S2, training a generative adversarial network (GAN) with recorded data and running data to generate virtual shift data;
[0053] S3, inputting the actual recorded data and the virtual shift data in step S1 into the intelligent adaptive dynamic optimization algorithm (IDP), training the blast furnace shift control model, and establishing the optimization control strategy for the blast furnace shift;
[0054] S4. Combined with a robust adaptive controller based on a neural network, parameters can be identified and adjusted in real time to improve the adaptability and robustness of the system and achieve stable and efficient blast furnace movement control.
[0055] In step S2, the generative adversarial network (GAN) can generate a large amount of virtual transition data by learning the distribution of actual data.
[0056] In step S3, the intelligent adaptive dynamic optimization algorithm (IDP) ensures that the relevant control variables such as the blast furnace moving direction and speed remain bounded through the constraints of the controller.
[0057] In step S3, the formula of the intelligent adaptive dynamic optimization algorithm (IDP) is as follows:
[0058]
[0059] In formula (1), s0(x) is the initial control law; Y0 is the initial performance index function; h(x) is the error for any initial state;
[0060] R(h(x)) and F(s(x)) are both positive definite functions. (R m is an m-dimensional real vector space), the calculation formula is as follows:
[0061]
[0062] In formula (II), h T is the transpose of the initial state error h(x); set up is the saturation bound of the controller constraint, s(x) is the controller constraint input; Q and R are diagonal positive definite matrices; dv represents the integral of v; and is a monotone bounded increasing odd function that satisfies Its first-order derivative is bounded by a constant;
[0063] The performance indicator function is further updated as follows:
[0064] Y1(h(x))=R(h(x))+F(s0(x))+Y0(h(x+1)) (3)
[0065] In formula (3), h(x+1)=f(h(x)+η(x))+g(h(x)+η(x))(s0(x)+u ε (x))-ηx+1);
[0066] Then iterate according to the following formula:
[0067] s i (x) = argmin s(x) {R(h(x))+F(s(x))+Y i(h(x+1))}
[0068] Y i+1 (h(x))=R(h(x))+F(s0(x))+Y i (h(x+1))
[0069] h(x+1)=f(h(x)+η(x))+g(h(x)+η(x))(s i (x)+u h (x))-η(x+1)
[0070] In the above formula, s i (x) is the control input after the i-th iteration; η(x) is the reference trajectory; Y i is the performance index function after the i-th iteration; u h (x) is steady-state input control; is a Lipschitz continuous function; R and F are positive definite functions in formula 2.
[0071] In step S4, the neural network robust adaptive controller is a BP neural network based model to simulate the unknown nonlinear characteristics of the blast furnace movement and convert it into an input-output mode as follows:
[0072]
[0073] In formula (IV), Z is the output (the predicted output value of the displacement related to the blast furnace), u(x) represents the input at time x (the control input value of the displacement related to the blast furnace), d is the relative order of the system, and c(x) is the system state vector; g d-1 and f d-1 It is a nonlinear function after a multi-layer neural network; w is the weight connecting the input layer and the hidden layer; s is the weight connecting the hidden layer and the output layer;
[0074] Said and Determined by the neural network structure and the number of neurons, assuming that the number of hidden neurons in the network is a and b respectively, the expression is as follows:
[0075]
[0076] In formula (5), w i and i are the weights of the i-th neuron; w ij and ij are the weights between the ith neuron and the jth input; the difference is that w is the weight associated with the input vector c(x), and s is the weight associated with the control input u(x); P(·) is a continuous, bounded, and non-constant monotonically increasing function, Its control law is as follows:
[0077]
[0078] In formula (6), r is the reference output (indicating the expected system output, target value); u(x) is the control input at time x, which is used to adjust the operation of the system; c(x) is the input vector at time x, containing information about the current state of the system; w(x) and s(x) are both weight vectors of the neural network at time x; r(x) is the reference signal;
[0079] The control quantity u(x) acts on the displacement system object and the neural network model; the weight of the neural network can be adjusted in real time to achieve optimization based on the error between the actual output of the system and the predicted output of the model; in order to quantify the difference, we introduce an error indicator, which is expressed as follows:
[0080] z(x+1)=f d-1 [c(x-d+1)]+g d-1 [c(x-d+1)]u(x-d+1)
[0081]
[0082] At the same time, the estimated output and system error of the model are as follows:
[0083]
[0084] h * (x+1)=z * (x+1)-z(x+1)
[0085] In the above formula, z is the estimated output of the model after neural network adaptive control; c is the real-time input vector; h is the system error; s is the weight related to the control input; w is the weight related to the input vector; and u is the control input used to adjust the system.
[0086] The sensors include position sensors, speed sensors and installation force sensors, which are used to monitor various parameters in the blast furnace moving process in real time.
[0087] By using GAN, IDP algorithm and robust adaptive control based on neural network, it is possible to monitor the blast furnace movement data received by each sensor in real time, dynamically and intelligently calculate the optimal control parameters, and drive the optimal movement instructions to adapt to the changing conditions during the blast furnace movement process. Accurate and efficient control of the blast furnace movement process is achieved, ensuring that the blast furnace is stably and reliably moved to the designated position, enabling and improving the efficiency of the safe implementation of the blast furnace movement project and the smooth resumption of blast furnace production.
[0088] By real-time monitoring of key node data in the blast furnace moving process, dynamically calculating the optimal control parameters of the blast furnace moving model, optimizing the moving instructions, adapting to different working conditions during the blast furnace moving process, optimizing energy consumption, reducing input costs, and ensuring the safety and stability of the blast furnace moving process, it enables and improves efficiency for the safe implementation of the blast furnace moving project and the smooth resumption of blast furnace production.
[0089] Generative adversarial networks (GANs) can effectively solve the problem that it may be difficult to obtain enough displacement data in actual production. By learning the distribution of actual data through GAN, a large amount of virtual displacement data can be generated, which can be used to enhance the diversity and coverage of training samples and improve the generalization ability of the algorithm. At the same time, the actual collected blast furnace operation data can simulate various displacement scenarios in the virtual data generated by GAN, so that the control algorithm can be trained under different working conditions, thereby improving its ability to cope with different displacement phenomena.
[0090] By inputting actual and virtual shifting data into the IDP algorithm, the blast furnace shifting control model is trained and the optimal control strategy for blast furnace shifting is established. This effectively improves the real-time operating parameters of the blast furnace (hydraulic temperature, displacement and speed, etc.) of the shifting system, and quickly obtains the optimal control strategy, while limiting the fluctuation range of the blast furnace shifting control quantity.
[0091] Example 2
[0092] First, the blast furnace moving coordinate system is established with the body installation positioning system as the coordinate origin and the track direction as the axis, and the state equation of the system is automatically established with displacement x and velocity As a state variable, The input quantity Y = CX, and the state equation is as follows:
[0093]
[0094] In the above formula, A is determined by the system itself, B is determined by the hydraulic push rod actuator, and D is determined by the actual resistance of the reverse thrust track.
[0095] Install position sensors and speed sensors on the push platform, install force sensors on the hydraulic push system, ensure that the sensors are correctly installed on the blast furnace push support and are stably connected to the controller; at the same time, ensure that the controller establishes a communication connection with the execution component to control the transmission and execution of instructions; start the blast furnace push control system to enter the monitoring mode; the sensor monitors the temperature, pressure and other data of the blast furnace in real time, as well as the displacement and speed information during the push process, and inputs the data into GAN for training. GAN generates a large amount of virtual push data to simulate various push scenarios. Combine the actual data with the virtual data and input them into the intelligent adaptive dynamic optimization (IDP) algorithm to train the blast furnace push model, establish the optimization control strategy for blast furnace push, and dynamically calculate and control the controller in real time.
[0096] The implementation of the IDP algorithm includes two parts: the evaluation network and the execution network. The evaluation network is used to approximate the performance index function. The weight matrix between the input layer and the hidden layer is Sc, and the weight matrix between the hidden layer and the output layer is Fc. The input is h(x), and the output is expressed as follows:
[0097]
[0098] The target expression is:
[0099]
[0100] Define the error expression of the judgment network:
[0101]
[0102] The role of the judgment network is to minimize the following function:
[0103]
[0104] Based on the gradient descent method, the weight update rule of the evaluation network is expressed as:
[0105]
[0106] In the above formula, β c To judge the learning rate of the network.
[0107] Among them, the control law is obtained by executing the network, the weight matrix between the input layer and the hidden layer is Sd, the weight matrix between the hidden layer and the output layer is Fd, the input is h(x), and the output is expressed as follows:
[0108]
[0109] The target expression is:
[0110]
[0111] Define the error expression for executing the network:
[0112]
[0113] Similarly, executing the network minimizes the following function:
[0114]
[0115] The weight update rule of the execution network is similar to that of the judgement; according to the gradient descent method, we can get:
[0116]
[0117] In the above formula, β d The learning rate of the execution network. The IDP algorithm effectively limits the fluctuation range of the blast furnace movement; at the same time, the BP neural network is used and the dead zone identification algorithm is introduced to adaptively adjust the weight parameters of the neural network to cope with the changes of various operating parameters in the blast furnace movement system over time.
[0118] The entire blast furnace movement control system continuously monitors the blast furnace movement status and makes real-time adjustments and optimizations based on real-time monitoring data; once abnormal conditions or environmental changes are detected, the controller will quickly adjust the control parameters to ensure the safety and stability of the movement process.
[0119] Through the visual monitoring and operation functions of the intelligent control system, the administrator can select and input the target position where the blast furnace needs to be pushed on the interface; when the blast furnace is pushed to the specified position, the movement is completed; in addition, the operator can monitor the movement of the blast furnace in real time through the visual interface, and perform necessary operations and adjustments to meet the movement requirements in different scenarios.
[0120] The above description is only a preferred embodiment of the invention and is not intended to limit the invention. For those skilled in the art, the present invention may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A dynamic intelligent adaptive blast furnace movement control method, Features: The following steps are involved: S1. Collect the recorded data of the actual blast furnace movement and the operating data of the sensors on the blast furnace; S2, training a generative adversarial network (GAN) with recorded data and running data to generate virtual shift data; S3, inputting the actual recorded data and the virtual shift data in step S1 into the intelligent adaptive dynamic optimization algorithm (IDP), training the blast furnace shift control model, and establishing the optimization control strategy for the blast furnace shift; S4. Combined with a robust adaptive controller based on a neural network, real-time identification and adjustment of parameters are performed to achieve blast furnace movement control.
2. A dynamic intelligent adaptive blast furnace movement control method as claimed in claim 1, Features: In step S2, the generative adversarial network (GAN) can generate virtual transition data by learning the distribution of actual data.
3. A dynamic intelligent adaptive blast furnace movement control method as claimed in claim 1, Features: In step S3, the intelligent adaptive dynamic optimization algorithm (IDP) ensures that the relevant control quantities of the blast furnace moving direction and speed remain bounded through the constraints of the controller.
4. A dynamic intelligent adaptive blast furnace movement control method as claimed in claim 1, Features: In step S3, the formula of the intelligent adaptive dynamic optimization algorithm (IDP) is as follows: In formula (1), s 0 (x) is the initial control law; Y 0 is the initial performance index function; h(x) is the error for any initial state; R(h(x)) and F(s(x)) are both positive definite functions. (R m is an m-dimensional real vector space), the calculation formula is as follows: In formula (2), h T is the transpose of the initial state error h(x); set up is the saturation bound of the controller constraint, s(x) is the controller constraint input; Q and R are diagonal positive definite matrices; dv represents the integral of v; and is a monotone bounded increasing odd function that satisfies Its first-order derivative is bounded by a constant; The performance indicator function is further updated as follows: Y 1 (h(x)) = R(h(x)) + F(s 0 (x)) + Y 0 (h(x + 1)) (III) In Equation (III), h(x + 1) = f(h(x) + η(x)) + g(h(x) + η(x))(s 0 (x) + uε(x)) - η(x + 1); Then iterate according to the following formula: s i (x)=arg min s(x) {R(h(x))+F(s(x))+Y i (h(x+1))} Y i+1 (h(x))=R(h(x))+F(s 0 (x))+Y i (h(x+1)) h(x+1)=f(h(x)+η(x))+g(h(x))(s i (x)+u h (x))-(x+1) In the above formula, s i (x) is the control input after the i-th iteration; η(x) is the reference trajectory; Y i is the performance index function after the i-th iteration; u h (x) is steady-state input control; is a Lipschitz continuous function; R and F are positive definite functions in formula 2.
5. A dynamic intelligent adaptive blast furnace movement control method as claimed in claim 1, Features: In step S4, the neural network robust adaptive controller is a BP neural network based model to simulate the unknown nonlinear characteristics of the blast furnace movement and convert it into an input-output mode as follows: In formula (IV), Z is the output (the predicted output value of the displacement related to the blast furnace), u(x) represents the input at time x (the control input value of the displacement related to the blast furnace), d is the relative order of the system, and c(x) is the system state vector; g d-1 and f d-1 It is a nonlinear function after passing through a multi-layer neural network; w is the weight connecting the input layer and the hidden layer; S is the weight connecting the hidden layer and the output layer; Said and Determined by the neural network structure and the number of neurons, assuming that the number of hidden neurons in the network is a and b respectively, the expression is as follows: In formula (5), w i and i are the weights of the i-th neuron; w ij and ij are the weights between the ith neuron and the jth input; the difference is that w is the weight associated with the input vector c(x), and s is the weight associated with the control input u(x); P(·) is a continuous, bounded, and non-constant monotonically increasing function, Its control law is as follows: In formula (6), r is the reference output (indicating the expected system output, target value); u(x) is the control input at time x, which is used to adjust the operation of the system; c(x) is the input vector at time x, containing information about the current state of the system; w(x) and s(x) are both weight vectors of the neural network at time x; r(x) is the reference signal; The control quantity u(x) acts on the displacement system object and the neural network model; the weight of the neural network can be adjusted in real time to achieve optimization based on the error between the actual output of the system and the predicted output of the model; in order to quantify the difference, we introduce an error indicator, which is expressed as follows: z(x+1)=f d-1 [c(x-d+1)]+g d-1 [c(x-d+1)]u(x-d+1) At the same time, the estimated output and system error of the model are as follows: h * (x+1)=z * (x+1)-z(x+1) In the above formula, z is the estimated output of the model after neural network adaptive control; c is the real-time input vector; h is the system error; s is the weight related to the control input; w is the weight related to the input vector; and u is the control input used to adjust the system.
6. A dynamic intelligent adaptive blast furnace movement control method as claimed in claim 1, Features: The sensors include position sensors, speed sensors and installation force sensors, which are used to monitor various parameters in the blast furnace moving process in real time.
