Intelligent equipment control method of electrical smelting furnace for magnesia based on neural network self-adaptive PID (proportion integration differentiation)

Abstract

The invention discloses a method for controlling intelligent equipment of an electric smelting furnace for magnesia based on neural network self-adaptive PID, and relates to the field of industrial control of equipment with great energy consumption. According to the method, modeling is carried out on the electric smelting magnesium furnace, the current working condition and the production boundary condition are analyzed to obtain an ideal current set value, and a corresponding PID controller is designed with the purpose of eliminating the error between the current three-phase electrode current and the set value. And the three-phase electrode current can rapidly track the changing set current value, so that the safe and stable operation of the production process is ensured. Designing a self-adaptive control method based on a radial basis function neural network; designing a self-adaptive law aiming at meeting Lyapunov stability conditions, and approaching key variables of the self-adaptive law by the radial basis function neural network; and finally, in combination with PID parameters and an adaptive law, obtaining the dynamic PID controller of the electrical smelting furnace for magnesia, which is embedded with the neural network and has self-tuning and self-adaptive gains. Compared with the prior art, the neural network adaptive PID controller has the advantages that the control effect is obviously improved, and the transient and steady-state performance of the system can be effectively improved.

Description

Technical Field

[0001] The invention relates to the field of industrial control of major energy-consuming equipment, and in particular to an intelligent equipment control method for a fused magnesium furnace based on a neural network adaptive PID. Background Art

[0002] Large energy-consuming equipment is very common in factories, manufacturing, data centers and other industries that require high energy consumption. The electric magnesium furnace is a typical large energy-consuming equipment, and the electricity cost accounts for more than 60% of the total production cost. Due to its large power consumption and the complexity and uncertainty of the system itself, what advanced control method to use to improve its working efficiency has always been a key issue of concern in the industry.

[0003] Fused magnesia is an important magnesium oxide, usually made from magnesite ore. Fused magnesia has the characteristics of high purity, high melting point, oxidation resistance, complete structure, strong insulation, etc. It is mainly used to produce various magnesium refractory materials with excellent performance. Although my country has huge reserves of magnesite ore, the ore generally has the characteristics of low quality, large fluctuations in composition, and complex mineral composition. It needs to be smelted in my country's unique three-phase AC fused magnesium furnace. The performance indicator used to reflect the energy consumption of the product in the smelting process of the fused magnesium furnace is the energy consumption per ton of the product, which is directly related to the value and fluctuation range of the smelting current.

[0004] PID control is a feedback control algorithm widely used in automatic control systems. The PID controller adjusts the system output according to the system deviation (i.e. the difference between the set value and the actual output value) so that the output can gradually approach the set value, thereby achieving precise control.

[0005] Adaptive PID is an improved PID control method that aims to address the problem of insufficient performance of traditional PID controllers in complex, nonlinear or time-varying systems. The adaptive PID controller adjusts the parameters of the PID controller in real time to enable it to adapt to changes in the dynamic characteristics of the system, thereby improving the control effect.

[0006] Neural network adaptive PID control is an advanced control method that combines neural networks and PID controllers. It uses the learning ability and adaptability of neural networks to adjust the parameters of PID controllers in real time to adapt to the dynamic characteristics of complex, nonlinear, and time-varying systems. This can maintain good control performance under a wider range of working conditions. It has broad application prospects in modern industrial control, robot control, intelligent systems and other fields.

[0007] The control of the molten magnesium furnace is a tracking control problem of a complex system. Due to the complexity and uncertainty of the dynamic model, and considering the coupling factors and non-zero interference, any model-based control is expensive and almost impractical. Therefore, the low-complexity proportional integral differential control based on the model-free method has its advantages here. Unlike the traditional fixed-gain PID controller, the controller described in this method is a molten magnesium furnace PID controller with self-tuning adaptive gain embedded in a neural network. Its tuning strategy is based on the stability and performance indicators of the system. The gain is continuously and adaptively adjusted through the neural network learning mechanism to ensure the transient and steady-state performance of the system. Summary of the invention

[0008] In order to solve the above technical problems, the present invention proposes an intelligent equipment control method for an electric molten magnesium furnace based on a neural network adaptive PID.

[0009] The technical problem to be solved by the present invention is achieved by the following technical solutions: a method for controlling an intelligent device of an electric molten magnesium furnace based on a neural network adaptive PID. The method models the electric molten magnesium furnace, analyzes the current working conditions and production boundary conditions to obtain an ideal current setting value, and designs a corresponding PID controller for the purpose of eliminating the error between the current three-phase electrode current and the setting value, so that the three-phase electrode current can quickly track the changing setting current value to ensure the safe and stable operation of the production process. Subsequently, an adaptive control method based on a radial basis function neural network is designed: an adaptive law is designed for the purpose of meeting the Lyapunov stability condition, and its key variables are approximated by a radial basis function neural network. Finally, a dynamic PID controller for an electric molten magnesium furnace with a self-tuning adaptive gain embedded in a neural network is obtained by combining PID parameters and adaptive laws.

[0010] As a further improvement of the present invention, the electric fused magnesium furnace model includes: an operation layer model with single ton energy consumption as output and three-phase electrode current as input, and a control layer model with three-phase electrode current as output and the speed and direction of the three-phase traction motor as input.

[0011] As a further improvement of the present invention, in the smelting process of the electric fused magnesium furnace, the performance index for reflecting the energy consumption of the product is the energy consumption per ton of the product, that is, the electric energy consumed for producing one ton of qualified products.

[0012] As a further improvement of the present invention, the operating layer output, i.e., the energy consumption per ton, needs to be controlled within the target value range as much as possible and as small as possible. According to the relationship between the product energy consumption per ton index and the electrode current setting value, current loop tracking control is performed.

[0013] As a further improvement of the present invention, the current tracking system adjusts the up and down movement of the electrode by controlling the rotation time and direction of the motor according to the tracking error between the three-phase current value and the current set value under the heating and melting conditions, so that the electrode and the raw material are in a suitable position to form an arc and release heat.

[0014] As a further improvement of the present invention, firstly, the tracking control of the three-phase electrode current is realized by PID control, so as to ensure that the current tracks the appropriate set value of the underlying three-phase electrode current control loop.

[0015] The traditional PID controller is:

[0016]

[0017] where k p1 , k I1 , k d1 are the proportional coefficient, integral coefficient and differential coefficient of the PID controller of the electric magnesium furnace, u is the output of the controller, and e is the three-phase electrode current error signal.

[0018] As a further improvement of the present invention, in order to improve the current loop tracking accuracy, adaptive control is adopted for the molten magnesium furnace and an adaptive law is designed according to the Lyapunov stability condition.

[0019] The adaptive law used in this method is:

[0020]

[0021] In the formula, is the adaptive law of the electric fused magnesium furnace control system, is the derivative of the adaptive law with respect to time, δ1 and c1 are the free design parameters of the adaptive law, i is the sequence number of the currently selected channel of the electric fused magnesium furnace control system, and m is the total number of channels of the control system; φ i is the maximum 2-norm of the basic function vector of the radial basis neural network selected for the electric fused magnesium furnace control system, is the generalized error set of three-phase electrode current.

[0022] As a further improvement of the present invention, the radial basis neural network has the characteristics of simple structure and strong approximation ability, and is used to approximate the key variable Φ in the adaptive rate. i .

[0023] According to the universal approximation theory, there exists an ideal neural network with an appropriate network structure and a sufficient number of neurons that can approximate any given continuous nonlinear function f(Z) in a compact set and can be written as:

[0024] f(Z)=P *T ψ(Z)+ζ(Z)

[0025] Where f(Z) is any continuous nonlinear function, specifically the variable Φ in the adaptive rate i , Z is the neural network input, i.e., the three-phase electrode current and its error, ψ(Z) is the variable Φ describing how the radial basis function neural network approximates the adaptive law of the fused magnesium furnace i The basic function vector of ζ(Z) is the approximate error that must exist in the approximation process, P * is the optimal constant weight that minimizes the approximation error ζ(Z).

[0026] A classic choice for ψ(Z) and also the choice for this method is:

[0027]

[0028] Among them, μ n ,φ n They are respectively the center and width of ψ(Z) selected according to the control system of the electric molten magnesium furnace.

[0029] As a further improvement of the present invention, the intelligent equipment control method of the electric fused magnesium furnace based on neural network adaptive PID combines the PID parameters selected according to the electric fused magnesium furnace system and the adaptive law embedded in the radial basis function neural network to obtain a dynamic PID controller of the electric fused magnesium furnace with self-tuning adaptive gain embedded in the radial basis function neural network. The gain is continuously and adaptively adjusted through the neural network learning mechanism, which can effectively realize the tracking control of the three-phase electrode current of the electric fused magnesium furnace.

[0030] The present invention provides a method for controlling an intelligent device of a fused magnesium furnace based on a neural network adaptive PID. Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. Unlike the existing research on fixed structure systems, this method can be applied to the fused magnesium furnace system with variable structure. The fused magnesium furnace model has the characteristics of strong nonlinearity, multivariable coupling, unclear smelting mechanism, and difficulty in establishing a mathematical model. Its dynamic characteristics are affected by boundary conditions such as raw material particle length and impurity composition and changes in smelting conditions.

[0032] 2. Unlike the traditional PID controller with fixed structural gain, the present invention embeds a self-tuning adaptive gain based on a neural network, and continuously and adaptively adjusts and updates its gain through the learning mechanism of the neural network. Such an adjustment process can not only achieve stable tracking, but also ensure the predetermined transient and steady-state performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Flow chart of this method.

[0034] Figure 2 The effect diagram of the electric magnesium furnace under the traditional PID control method. (a) is the traditional PID current signal; (b) is the traditional PID control signal; (c) is the traditional PID gain change.

[0035] Figure 3 The effect diagram of the electric magnesium furnace under the neural network adaptive PID control method. (a) Neural network adaptive PID current signal; (b) Neural network adaptive PID control signal; (c) Neural network adaptive PID gain change. DETAILED DESCRIPTION

[0036] Step (a) Adjust the current setting value: According to the energy consumption index per ton of fused magnesia products, consider the current setting value under the current working conditions and production boundary conditions.

[0037] Step (ii) Model loading: Load the LSTM nonlinear model and linear model trained using historical data such as three-phase current, motor motion, three-phase voltage, and material disturbance.

[0038] Step (iii) Integration and controller limiting: Set the PID controller integration upper limit according to the size of the historical three-phase electrode current error to prevent integration saturation.

[0039] Step (iv) PID design: The PID controller takes the error between the actual value and the set value of the three-phase electrode current as input, and adjusts the three parameters of the PID controller of the electric magnesium furnace, namely the proportional coefficient, integral coefficient and differential coefficient, in the simulation.

[0040] Step (V) Error transformation: Since the preset performance can be guaranteed when the following conditions are met:

[0041]

[0042] where, for any time t ≥ 0, α i and are respectively the positive upper and lower limit constants of the three-phase electrode current error, ρ i (t) is the performance function describing the three-phase electrode current error, e i (t) is the original tracking error of the three-phase electrode current.

[0043] The following error transformation is introduced to convert the original tracking error equivalently:

[0044] e i (t) = ρ i (t)q i (ε i )

[0045] Among them, ε i is the three-phase electrode current conversion error, qi (ε i ) is smooth and strictly increasing with respect to ε i A reversible function that satisfies:

[0046] 1.

[0047] 2.

[0048] 3.q i (0) = 0

[0049] From this, we can inversely solve for ε i , and introduce the three-phase electrode current filtering error S i

[0050]

[0051] Among them, λ i These are parameters selected according to the electric fused magnesium furnace system.

[0052] Step (six) Generalized error design: Design the generalized error of the three-phase electrode current for easy control.

[0053]

[0054] Where E is the generalized error of the three-phase electrode current, σ is the positive design parameter obtained from the electric fused magnesium furnace system, and S(.) is the filtering error mentioned above.

[0055] Step (VII) Adaptive law design: To satisfy the Lyapunov stability condition, the adaptive law is designed as follows.

[0056]

[0057] In the formula, is the adaptive law of the electric fused magnesium furnace control system, is the derivative of the adaptive law with respect to time, 61 and c1 are the free design parameters of the adaptive law, i is the sequence number of the currently selected channel of the electric fused magnesium furnace control system, m is the total number of channels of the control system; φi is the maximum 2-norm of the basic function vector of the radial basis neural network selected by the electric fused magnesium furnace control system, is the generalized error set of three-phase electrode current.

[0058] Step 8: Radial Basis Neural Network: Use radial basis neural network to approximate the key variable Φ in the adaptive law i .

[0059] Step (IX) Dynamic Differentiation: The dynamic differential coefficients of the fused magnesium furnace control system are formed from the parameters obtained in the previous steps.

[0060]

[0061] In the formula, Δk d is the dynamic differential coefficient of the electric magnesium furnace control system.

[0062] Step (ten) Dynamic PI: Based on the dynamic differential coefficient, the dynamic proportional coefficient and the dynamic integral coefficient are obtained, and then the pre-tuned PID parameters are added with the dynamic PID parameters to form a new PID controller.

[0063] Δk p =2σΔk d , Δk I =σ 2 Δk d

[0064]

[0065] In the formula, Δk d , Δk p , Δk I They are respectively the dynamic differential coefficient, dynamic proportional coefficient and dynamic integral coefficient of the electric magnesium furnace control system, k p , k I , k d They are respectively the proportional coefficient, integral coefficient and differential coefficient of the PID controller of the electric magnesium furnace.

[0066] Step (XI) Prediction results: Using the prediction capability of the LSTM model, the prediction results of the control layer output, i.e., the three-phase electrode current, are obtained.

[0067] Step (XII) ends: The control of the fused magnesium furnace based on the neural network adaptive PID is completed and compared with the classic PID control results. The effect diagrams of the traditional PID control method and the neural network adaptive PID control method are shown in Figure 11. Figure 2 and Figure 3 The evaluation results are shown in Table 1. Figure 2 , Figure 3 It can be seen from Table 1 that by adopting the control method based on neural network adaptive PID, the number of current root error exceeding the allowable error range is significantly reduced, which effectively improves the tracking error convergence accuracy and convergence speed of the loop control and ensures the transient and steady-state performance of the system.

[0068] Table 1

[0069]

[0070] The above simulation values ​​prove the correctness and effectiveness of the neural network-based adaptive PID control method, which has been successfully applied in the control of the electric molten magnesium furnace and achieved significant energy-saving and emission reduction effects.

Claims

1. A method for controlling intelligent equipment of a fused magnesium furnace based on a neural network adaptive PID, characterized in that: The electric fused magnesium furnace is modeled, and the current operating conditions and production boundary conditions are analyzed to obtain the ideal current set value. The corresponding PID controller is designed to eliminate the error between the current three-phase electrode current and the set value, so that the three-phase electrode current can quickly track the changing set current value to ensure the safe and stable operation of the production process; then an adaptive control method based on radial basis function neural network (RBFNN) is designed: an adaptive law is designed to meet the Lyapunov stability condition, and its key variables are approximated by radial basis function neural network; finally, the PID parameters are combined with the adaptive law to obtain a dynamic PID controller for the electric fused magnesium furnace with self-tuning adaptive gain embedded in a neural network.

2. The method for controlling an intelligent device of a fused magnesium furnace based on a neural network adaptive PID according to claim 1 is characterized in that: The model of the electric fused magnesium furnace modeling includes: an operation layer model with single ton energy consumption as output and three-phase electrode current as input, and a control layer model with three-phase electrode current as output and the speed and direction of the three-phase traction motor as input.

3. The method for controlling an intelligent device of a fused magnesium furnace based on a neural network adaptive PID according to claim 1 is characterized in that: In the smelting process of the electric fused magnesium furnace, the performance index used to reflect the energy consumption of the product is the energy consumption per ton of the product, that is, the electric energy consumed for producing one ton of qualified products.

4. The method for controlling an intelligent device of a fused magnesium furnace based on a neural network adaptive PID according to claim 1 is characterized in that: With the goal of controlling the energy consumption per ton within the target value range and as small as possible, current loop tracking control is carried out according to the relationship between the product's energy consumption per ton index and the electrode current setting value.

5. The method for controlling an intelligent device of a fused magnesium furnace based on a neural network adaptive PID according to claim 1 is characterized in that: The current tracking system adjusts the up and down movement of the electrode by controlling the rotation time and direction of the motor according to the tracking error between the three-phase electrode current value and the current setting value under the heating and melting condition, so that the electrode and the raw material are in a suitable position to form an arc to release heat.

6. The method for controlling an intelligent device of a fused magnesium furnace based on a neural network adaptive PID according to claim 5 is characterized in that: Firstly, the tracking control of the three-phase electrode current is realized by PID control to ensure that it tracks the set value of the underlying three-phase electrode current control loop; The PID controller is: where k p1 , k I1 , k d1 are the proportional coefficient, integral coefficient and differential coefficient of the PID controller of the electric magnesium furnace, u is the output of the controller, and e is the three-phase electrode current error signal.

7. The method for controlling an intelligent device of a fused magnesium furnace based on a neural network adaptive PID according to claim 6 is characterized in that: In order to improve the current loop tracking accuracy, adaptive control is adopted for the fused magnesium furnace and the adaptive law is designed according to the Lyapunov stability condition. The adaptive law used is: In the formula, is the adaptive law of the electric fused magnesium furnace control system, is the derivative of the adaptive law with respect to time, δ1 and c1 are the free design parameters of the adaptive law; i is the sequence number of the currently selected channel of the electric fused magnesium furnace control system, and m is the total number of channels of the control system; φ i is the maximum 2-norm of the basic function vector of the radial basis neural network selected for the electric fused magnesium furnace control system, is the generalized error set of three-phase electrode current.

8. The method for controlling an intelligent device of a fused magnesium furnace based on a neural network adaptive PID according to claim 7 is characterized in that: The radial basis function neural network selected by the control system of the fused magnesium furnace is used to approximate the key variable Φ in the adaptive rate i ; According to the universal approximation theory, there exists an ideal neural network with an appropriate network structure and a sufficient number of neurons that can approximate any given continuous nonlinear function f(Z) in a compact set. The selected radial basis neural network is: f(Z)=P* T ψ(Z)+ζ(Z) Where f(Z) is any continuous nonlinear function, specifically the variable Φ in the adaptive rate i , Z is the neural network input, i.e., the three-phase electrode current and its error, ψ(Z) is the variable Φ describing how the radial basis function neural network approximates the adaptive law of the fused magnesium furnace i The basic function vector of , ζ(Z) is the approximate error that must exist in the approximation process, and P* is the optimal constant weight that minimizes the approximate error ζ(Z); The choice of ψ(Z) is: Among them, μ n ,φ n They are respectively the center and width of ψ(Z) selected according to the control system of the electric molten magnesium furnace.

9. The method for controlling an intelligent device of a fused magnesium furnace based on a neural network adaptive PID according to claim 8 is characterized in that: Combining the PID parameters selected according to the electric fused magnesium furnace system and the adaptive law embedded in the radial basis function neural network, a dynamic PID controller for the electric fused magnesium furnace with self-tuning adaptive gain embedded in the radial basis function neural network is obtained. Its gain is continuously and adaptively adjusted through the neural network learning mechanism, which can effectively realize the tracking control of the three-phase electrode current of the electric fused magnesium furnace.

Images