Submerged arc furnace advanced control system based on electrode insertion depth soft measurement

Abstract

The invention provides a submerged arc furnace advanced control system based on electrode insertion depth soft measurement, and the system comprises a GPC unit which receives a deviation value between a smelting power set value and a real-time power value, the electrode voltage / current / active power of a submerged arc furnace, an electrode insertion depth soft measurement value, and an interference variable as input variables, the GPC unit outputs electrode insertion depth bias and gear bias to the electrode lifting / gear controller as output variables; the PCA unit is used for carrying out dimension reduction processing on the received electrical parameters and the working condition parameters and sending low-dimension data to the BPN unit; and the BPN unit receives the low-dimensional data, performs soft measurement calculation on the electrode insertion depth based on the low-dimensional data, generates an electrode insertion depth soft measurement value and sends the electrode insertion depth soft measurement value to the GPC unit. Soft measurement of the electrode insertion depth can be achieved, automatic adjustment of the submerged arc furnace electrode is achieved through a generalized predictive control algorithm, production is stabilized, economic benefits are improved, and the overall intelligent level of the submerged arc furnace is improved.

Description

Technical Field

[0001] The invention relates to the technical field of smelting ore-fired furnaces, and in particular to an advanced control system for ore-fired furnaces based on soft measurement of electrode insertion depth. Background Art

[0002] At present, the design and production capacity of smelting ore-fired furnaces is huge, and some have a capacity close to 100MVA or more. Generally, three AC single-phase transformers are used for power supply. The primary high-voltage side is connected to the high-voltage power supply system by star connection or delta connection, and the secondary low-voltage side is connected to the three electrodes by delta connection. In the actual production process, the three electrodes are often operated separately. The process operator selects a certain electrode for lifting and lowering operations according to the furnace conditions. When the deviation from the production target is large, the adjustment is completed by upshifting and downshifting the single-phase transformer. On the whole, the transformer is used for coarse adjustment and the electrode is lifted and lowered for fine adjustment. In the process of adjustment, the three-phase electrical parameters are balanced. However, due to the strong coupling of the three-phase power supply system, the electrical parameters of the three-phase electrodes are difficult to balance in the actual production process, and the fluctuations are large. The power consumption varies depending on the different operating methods of the process operators. In the case of extreme imbalance, it will also affect the furnace structure, damage the masonry refractory and the furnace bottom, seriously affecting the life of the ore-fired furnace and causing serious economic losses.

[0003] The production furnace condition is closely related to the electrode insertion depth. Controlling the electrode insertion depth is the main means to ensure the stability of the furnace condition. The electrode is inserted into the charge for submerged arc smelting, and the electrode tip is constantly consumed. The actual insertion depth is difficult to measure. The process operator needs to roughly estimate the electrode insertion depth based on many parameters such as electrode current, voltage, active power, furnace temperature, flue gas temperature, etc. The empirical method is often used to operate, which often causes three-phase imbalance and is difficult to adjust. Therefore, in order to obtain the actual insertion depth, the production personnel need to shut down the power every 1 to 2 days to measure the actual insertion depth. Some calculate it by inserting a steel chisel according to the angle, and some directly pull out all the electrodes and then reinsert them into the furnace after the measurement is completed. It is still difficult to measure accurately during the production process. In summary, a fast and accurate soft measurement system needs to be established to provide real-time electrode insertion depth information, realize stable control of the submerged arc furnace, and ensure product quality. Summary of the invention

[0004] In view of this, the purpose of the present invention is to propose an advanced control system for an electric arc furnace based on soft measurement of electrode insertion depth, so as to solve the technical problem that the existing three-phase AC electric arc furnace cannot obtain the electrode insertion depth.

[0005] The technical means adopted by the present invention are as follows:

[0006] An advanced control system for a submerged arc furnace based on soft measurement of electrode insertion depth, comprising:

[0007] A GPC unit, wherein the GPC unit receives a deviation value between a smelting power setting value and a real-time power value, an ore-arc furnace electrode voltage / current / active power, an electrode insertion depth soft measurement value, and an interference variable as input variables, and the GPC unit outputs an electrode insertion depth bias and a gear position bias to an electrode lifting / gear position controller as output variables;

[0008] An electrode lifting / shift controller, which receives an electrode insertion depth bias and a shift offset, and generates a corresponding control signal based on the electrode insertion depth bias and the shift offset and sends it to the ore-bearing furnace;

[0009] The submerged arc furnace receives the control signal and performs control, the submerged arc furnace sends the electrical parameters and operating parameters of the submerged arc furnace to the PCA unit, and the submerged arc furnace sends the electrode voltage / current / active power of the submerged arc furnace to the GPC unit; the submerged arc furnace outputs a real-time power value;

[0010] Subtract the real-time power value from the smelting power setting value to generate a deviation value between the smelting power setting value and the real-time power value and send it to the GPC unit;

[0011] A PCA unit, wherein the PCA unit performs dimensionality reduction processing on the received electrical parameters and operating condition parameters, and sends the low-dimensional data to the BPN unit;

[0012] The BPN unit receives the low-dimensional data and performs soft measurement calculation on the electrode insertion depth based on the low-dimensional data, generates a soft measurement value of the electrode insertion depth and sends it to the GPC unit.

[0013] Furthermore, the interference variables include furnace bottom temperature, flue gas temperature, O2 content and H2 content.

[0014] Further, the calculation method of the GPC unit is as follows:

[0015] S1. For objects subject to random interference, the controlled autoregressive integrated moving average model used in minimum variance control is adopted:

[0016] A(q -1 )y(k)=B(q -1 )u(k-1)+C(q -1 )ξ(k) / Δ

[0017] Where: A(q -1 )=1+a1q -1 +…+a n q -n , B(q -1 )=b0+b1q -1 +…+b m q -m , C(q -1) takes the value of 1;

[0018] Among them, q -1 is the backward shift operator, which indicates the corresponding amount of moving back one sampling period; Δ = 1-q -1 is a differential operator; ξ(k) is an uncorrelated random noise sequence, which represents the influence of a type of random noise on the system; y(k) is the output variable, u(k) is the input variable, and d(k) is the interference variable; coarse adjustment is achieved by adjusting the gear of the single-phase transformer and fine adjustment is achieved by raising and lowering the electrode position;

[0019] S2, control constraints are imposed on the output variable y(k), input variable u(k) and control increment Δu(k), where Δu(k) is the difference between the input variable of the current sampling period and the previous sampling period, y min (k)≤y(k)≤y max (k),u min (k)≤u(k)≤u max (k), Δu min (k)≤Δu(k)≤Δu max (k);

[0020] S3. Assuming the prediction length is j, the output prediction model of the system at time k+j is:

[0021] A(q -1 )y(k+j)=B(q -1 )u(k+j-1)+ξ(k+j) / Δ

[0022] The Diophantine equation is used to obtain the predicted system output value y at time k+j M (k+j), the Diophantine equation is as follows:

[0023] 1=A(q -1 )ΔE j (q -1 )+q -j F j (q -1 )

[0024] In the formula, E j (q -1 ), F j (q -1 ) is composed of A(q -1 ) and the predicted length j are uniquely determined by the polynomial, taking E1 = 1, As E j , F j The initial value of , then the formula can be derived:

[0025]

[0026] E j+1 =E j +e j+1,j q -1 =E j +f j,0 q -1

[0027]

[0028] in,

[0029] S4, let the controlled variable set value at time k be SP, and the actual tracking value is the softened set value w k , the object output expected value adopts the following reference trajectory:

[0030] w(k)=y(k)

[0031] w(k+j)=αw(k+j-1)+(1-α)SP

[0032] Among them, 0<α<1;

[0033] S5. Make the predicted output at time k+j as close as possible to the reference trajectory w(k+j), then the optimization target at time k is:

[0034]

[0035] Where, E is the mathematical expectation; w is the expected value of the output; N1, N2 are the initial and final values ​​of the optimization time domain; NU is the control time domain; λ(j) is the control weighting coefficient, which is generally a constant λ;

[0036] S6. Minimize the quadratic objective function J(k), let The control law is obtained as follows:

[0037] Δu=(G T G+λI) -1 G T (WH)

[0038] Where I is the identity matrix;

[0039]

[0040] Among them, g j,i =g i+1 (i<j) is the step response coefficient; W=[w(k+1)…w(k+N)] T ;

[0041] H=[h1…h N ] T ,in:

[0042] h1(t)=[G1(q -1 )-g 1,0 ]Δu(t)+F1y(t)

[0043] h2(t)=q[G2(q -1 )-q -1 g 2,1 -g 2,0 ]Δu(t)+F2y(t)

[0044]

[0045] h N (t) = q N-1 [G N (q -1 )-q -(N-1) g N,N-1 -…-g N,0 ]Δu(t)+F N y(t)

[0046] S7, set g T For (G T G+λI) -1 G T The first line of , then the instantaneous control increment at time k is:

[0047] Δu(k)=g T (WH)

[0048] S8. Apply the control quantity u(k)=u(k-1)+Δu(k) to the system to complete the control of this step.

[0049] Furthermore, S2 specifically includes:

[0050] The 3% and 97% position points of the hydraulic cylinder stroke of each electrode lifting drive are set as the lower and upper limits of the electrode position; based on the average value of the three electrode positions, the upper limit of the imbalance of the three electrodes is set to 20%. When the three electrode position values ​​are completely equal, the imbalance value is 0, and no lower limit is set; the upper limit of the single adjustment of the three electrode positions is set to 10mm, and no lower limit is set; the lower and upper limits of the single-phase transformer gear are set according to the gear configuration of the secondary side of the single-phase transformer. When the gear is the lowest gear and the highest gear, it is no longer adjusted to the boundary and can only be adjusted within the gear protection limit; the maximum gear difference of the three single-phase transformers is set to 2 gears. In the actual adjustment process, the strategy of the same gear increase and decrease of the three single-phase transformers is preferred. Only when the furnace condition is unstable, the adjustment strategy with gear difference is selected; the upper limit of the single adjustment of the three single-phase transformers is set to 1 gear, and no lower limit is set; based on the rated current of the secondary side of the single-phase transformer The multiple value is used as the upper limit of the electrode current, and 30% of the rated current is set as the lower limit of the electrode current. If the electrode is judged to be broken, the program control will exit the MPC electrode automatic adjustment and switch to manual process operation; based on the average current value of the three electrodes, the upper limit of the current imbalance of the three electrodes is set to 2%. When the current values ​​of the three electrodes are completely equal, the imbalance value is 0, and no lower limit is set; based on the rated voltage of the secondary side of the single-phase transformer The multiple value is used as the upper limit of the electrode voltage, and 70% of the rated voltage is set as the lower limit of the electrode voltage. If an abnormal situation occurs in the production process of the electric arc furnace, the program control exits the MPC electrode automatic adjustment and switches to manual process operation; based on the voltage average value of the three electrodes, the upper limit of the voltage imbalance of the three electrodes is set to 5%. When the voltage values ​​of the three electrodes are completely equal, the imbalance value is 0, and no lower limit is set; 120% multiple value of the rated power of the secondary side of the single-phase transformer is used as the upper limit of the electrode power, and no lower limit is set; based on the power average value of the three electrodes, the upper limit of the power imbalance of the three electrodes is set to 7%, and when the power values ​​of the three electrodes are completely equal, the imbalance value is 0, and no lower limit is set; the upper limits of the furnace bottom temperature and the flue gas temperature are set to 500℃ and 800℃ respectively, and no lower limit is set; the upper limits of the oxygen content in the flue gas and the hydrogen content in the flue gas are set to 0.5% and 8% respectively, and no lower limit is set.

[0051] Furthermore, the program execution flow of the GPC unit is as follows:

[0052] T1, initialization procedure;

[0053] T2, according to the latest input and output data, use the recursive least squares method to estimate the model parameters and obtain A(q -1 ) and B(q -1 );

[0054] T3, according to the obtained A(q -1 ), calculate E(q -1 ) and F(q -1 );

[0055] T4, according to B(q -1 )、E(q -1 ) and F(q -1 )Calculate G(q -1 ) element g i , and calculate H;

[0056] T5. Calculate g T , and calculate the control action deviation Δu(k);

[0057] T6. Apply the control quantity u(k)=u(k-1)+Δu(k) to the control system. The output adjustment of this cycle is completed and jump to T2 for loop execution.

[0058] Furthermore, the dimensionality reduction processing steps of the PCA unit are as follows:

[0059] Obtain electrical parameters and operating condition parameters to form an original matrix;

[0060] The electrical parameters and operating condition parameters are processed by zero-meaning the feature dimension data, that is, the mean of each feature dimension is subtracted from the data of the feature to obtain the sampling feature;

[0061] Obtain the covariance matrix of the sampled features;

[0062] Calculate the eigenvalues ​​of the covariance matrix and its corresponding unit eigenvector;

[0063] Sort the eigenvectors by eigenvalue from largest to smallest;

[0064] Select the eigenvectors corresponding to the first k eigenvalues ​​that contribute the most to the overall features;

[0065] The first k principal components are selected according to their cumulative variance contribution rate > 0.85; these k eigenvectors are then used as row vectors to form a dimensionality reduction matrix;

[0066] Multiply the reduced dimension matrix by the original matrix on the left to complete the dimension reduction.

[0067] Furthermore, in the BPN unit, a soft-sensing model is established for the principal variable using the BP back-propagation neural network;

[0068] The BP back propagation neural network includes an input layer A, a hidden layer B, and an output layer C. The transfer function of the input layer of the BP back propagation neural network adopts a linear function, and the activation functions of the hidden layer and the output layer adopt a Sigmoid function. The generalized delta algorithm is used for training, the error adopts the standard error, and the initial distribution of the network weights adopts a normal distribution.

[0069] Furthermore, the soft measurement calculation method of the BPN unit is as follows:

[0070] P0, initialization program;

[0071] P1, give weight v between -1 and +1 ij and w ij Random assignment, in the generalized 6-rule, the thresholds of all hidden layers and output layers must be 1, that is, T(2, j) = T(3, k) = 1;

[0072] P2, input vector I i Send it to the neural network and calculate the output of layer 1 according to the following formula:

[0073] x i =I i -T1I =I i -0=I i

[0074]

[0075] P3. Given the output of layer 1, use the following formula to calculate the output of layer 2:

[0076]

[0077] Where, f() is the Sigmoid function;

[0078] P4. Given the output of layer 2, calculate the output layer result according to the following formula:

[0079]

[0080] P5. Continue P1 to P4 for the M training patterns sent to the output layer, and calculate the total square error E according to the following formula:

[0081]

[0082] Where M is the number of training patterns sent to the output layer, n is the number of processing units in the output layer, is the expected output value of the mth training pattern on the kth processing unit;

[0083]

[0084] Among them, f() is the Sigmoid function:

[0085]

[0086] The partial derivative of the Sigmoid function is:

[0087]

[0088] x k is the weighted sum of the inputs of the kth node in the output layer, that is, for the mth training process:

[0089]

[0090] For the mth training pattern, is the output value of the jth element in the hidden layer, is the threshold of the output layer;

[0091] P6, given the mth mode, use the following formula to calculate That is, the gradient descent term of the jth processing unit in the front hidden layer:

[0092]

[0093]

[0094] The partial derivative of the Sigmoid function is:

[0095]

[0096] P7, known hidden layer and the output layer The weight change is calculated using the following formula:

[0097]

[0098]

[0099] Among them, η is the learning rate, is the momentum coefficient; α is in the range of 0<α<1; therefore, the momentum term and is the fractional value of the weight change from the previous iteration;

[0100] P8. Knowing the weight change, calculate the weight according to the following formula:

[0101]

[0102]

[0103] in, is the connection weight between the first i element of the input layer and the jth element of the hidden layer for the mth iteration training; is the connection weight between the jth element of the hidden layer and the kth element of the output layer for the mth iteration training. For all training modes, repeat steps P2 to P8 until the square error is 0 or less than the specified value.

[0104] Compared with the prior art, the present invention has the following advantages:

[0105] The present invention is based on a controlled autoregressive integral sliding model in minimum variance control, but has low requirements on the model;

[0106] The present invention adopts a rolling optimization strategy, performs rolling calculations within a limited time domain, and replaces global optimization with local optimization;

[0107] The present invention uses measured information to estimate model parameters online for feedback correction, thereby enhancing the robustness of control;

[0108] The application of the soft instrument technology of the present invention solves the problem of electrode insertion depth detection, does not require careful maintenance, has rapid dynamic response, is insensitive to interference other than process output or unmeasurable disturbances, and can continuously give indicated values. BRIEF DESCRIPTION OF THE DRAWINGS

[0109] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0110] Figure 1 This is a system structure diagram of the present invention.

[0111] Figure 2 This is a diagram of the MPC control system of the present invention.

[0112] Figure 3 This is a diagram of the steps for executing the GPC process of the present invention.

[0113] Figure 4 This is a diagram of the PCA data dimensionality reduction steps of the present invention.

[0114] Figure 5 This is a three-layer BP network diagram of the present invention.

[0115] Figure 6 This is a step diagram of the BP learning algorithm of the present invention. DETAILED DESCRIPTION

[0116] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0117] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0118] In the field of process control, model predictive control (MPC) is a typical representative of advanced process control (APC). MPC is the most successful control method and technology after PID. It can effectively handle constraints, multivariables, coupling and pure lag. It is widely used in industries such as petroleum, chemical industry, cement, etc., but it is rarely successfully applied in ore-fired furnace smelting. The MPC multivariable control function uses a model to predict the future behavior of process output under input changes, improves the dynamic performance of process control, can minimize the fluctuation of the control system, and promote the production process to the control target of quality (or economic) constraints that play a key role in the production process, and reduce the variance of the control system. What's more, MPC can smoothly operate the system on certain boundaries of multiple constraints, so that the control system can run at the edge, which is the key to MPC's economic benefits. At the same time, MPC can quickly suppress the influence of interference and can also handle the strong coupling of complex multivariable systems. Due to the strong coupling between the three electrodes caused by the unique structure of the ore-fired furnace power supply system, MPC is very suitable for the automatic control of the electrical parameter balance of the three-phase electrode. Compared with the model algorithm control (MAC) and dynamic matrix control (DMC) in the MPC strategy, the generalized predictive control (GPC) algorithm is proposed for random discrete systems, which does not require high accuracy of the mathematical model and maintains the rolling optimization strategy in the predictive control algorithm. After the MPC was put into use, the electrodes of the submerged arc furnace were automatically controlled and adjusted, the operating intensity of the process personnel was greatly reduced, the automation rate was significantly improved, the stability and safety of the device operation were guaranteed, the product quality was guaranteed, the target product yield was improved, the operating cost was reduced, and the economic benefits were significantly improved.

[0119] Therefore, we developed a submerged arc furnace control system based on the soft measurement technology of electrode insertion depth, which can realize soft measurement of electrode insertion depth, use generalized predictive control algorithm to realize automatic adjustment of submerged arc furnace electrodes, stabilize production, improve economic benefits, and improve the overall intelligence level of submerged arc furnace.

[0120] MPC (Model Predictive Control) is the core technology of APC (Advanced Process Control), which is mainly used in process control engineering. MPC uses model control algorithms to predict the future behavior of process outputs when inputs change. The submerged arc furnace control system we developed based on the soft measurement technology of electrode insertion depth can realize soft measurement of electrode insertion depth and use generalized predictive control algorithms to realize automatic adjustment of submerged arc furnace electrodes. The structure of the control system is as follows: Figure 2 As shown, the control system consists of a sensor unit, a data acquisition unit, a central processing unit, a human-computer interaction unit, a model prediction control unit, a numerical control output unit, and an actuator unit.

[0121] Sensor unit: It senses and detects various physical data parameters such as the actual position of the three electrodes, the actual gear position of the single-phase transformer, the electrode voltage, the electrode current, the electrode power, the furnace bottom temperature, the flue gas temperature, the oxygen content in the flue gas, the hydrogen content in the flue gas, etc. during the production process of the submerged arc furnace, and converts the sensed information into electrical signals according to certain rules for output. The output signal can be used by the data acquisition unit;

[0122] Data acquisition unit: collects the electrical signals output by the sensor unit according to the millisecond scanning cycle, processes the collected signals using anti-shake, filtering, passivation and other methods to improve the availability of data, and then outputs the collected data to the central processing unit;

[0123] Central processing unit: The signals collected from the data acquisition unit are processed by logic control in a cycle of seconds, and then the processed real-time information is transmitted to the human-machine interaction unit for the process operator to view, and the control instructions sent by the human-machine interaction unit are received to enter the logic control of the central processing unit, completing the two-way data communication with the human-machine interaction unit, so as to achieve true human-machine friendly interaction; the central processing unit sends a large amount of production process data of the submerged arc furnace to the model prediction control unit for model prediction control data calculation, and then the model prediction control unit returns the calculation result information to the central processing unit; the central processing unit processes the received calculation result information and adjusts the production process control strategy, and transmits the control signal to the CNC output unit through the CNC output unit The actuator unit completes the final real action of the on-site equipment, and then the process data parameters of the production site change accordingly, and are sensed by the sensor unit, repeating over and over again, ultimately ensuring the reliable, normal and optimal operation of the entire production system; the central processing unit also protects the submerged arc furnace equipment based on the upper and lower limit protection values ​​of various physical data parameters such as the actual position of the three electrodes of the submerged arc furnace, the actual gear of the single-phase transformer, the electrode voltage, electrode current, electrode power, furnace bottom temperature, flue gas temperature, oxygen content in the flue gas, and hydrogen content in the flue gas manually set in the human-computer interaction unit. When the limit is exceeded, a reverse adjustment command is output, or when the limit is seriously exceeded, the single-phase transformer power supply main circuit command is cut off and output to the CNC output unit, and the protection action is finally realized through the actuator unit.

[0124] Human-machine interaction unit: realizes friendly communication with process operators, displays the information received from the central processing unit in real time, and sends the control instructions of the process operators to the central processing unit;

[0125] Model prediction control unit: the core component for model prediction control calculation of the ore-fired furnace. It obtains a large amount of ore-fired furnace generation process data in the central processing unit, and then executes the model prediction control data calculation function developed in this project. After a single model prediction control calculation task is completed, the prediction calculation result information is transmitted back to the central processing unit in a minute-level cycle.

[0126] CNC output unit: receives control instructions from the central processing unit in real time, refreshes output change control information at millisecond level, adjusts output changes in real time, and converts information into various physical signals and transmits them to the actuator unit;

[0127] Actuator unit: Receives various physical signals from the CNC output unit and completes various equipment actions in the production process of the submerged arc furnace. This development mainly involves the actuators for the lifting and lowering adjustment of the gear positions of three single-phase transformers, which are on-load voltage regulators driven by the forward and reverse motors, the actuators for the lifting and lowering adjustment of the positions of three electrodes, which are switch solenoid valves with hydraulic driving force, and the primary high-voltage vacuum circuit breaker opening and closing device of the single-phase transformer for serious over-limit protection of process parameters.

[0128] The model prediction control unit is the core software component for model prediction control of the submerged arc furnace. It is generally set in the edge server Edge-PC hardware for submerged arc furnace production management. The server can be deployed as a local server. The network is configured in the upper-level terminal bus of the central processing controller hardware and is independent of the submerged arc furnace production process control factory network. The model prediction control unit is used for model prediction control calculations for automatic adjustment of submerged arc furnace electrodes. The result information data interacts in real time with the central processing unit and the human-computer interaction unit on the terminal bus to achieve the purpose of intelligent prediction control. Data edge computing in the edge server Edge-PC is an open system composed of hardware and software, which can flexibly execute packaged applications based on high-level languages; intuitive integrated factory functions are used to automatically collect, process and exchange data, and realize intelligent use of data on a factory basis. The edge server Edge-PC is based on open industrial Ethernet communication and is compatible with multiple Ethernet communication protocols. It supports OPC, TCP, UDP, ISO-On-TCP, I-Device, and S7-Routing communication protocols. It also supports IT communication services such as SNMP, Web, FTP, and NTP. It can easily transplant the data processing and algorithm unit functions of this model predictive control system directly to other similar industrial intelligent control systems.

[0129] MPC can smoothly operate the system on certain boundaries of multiple constraints, so that the control system runs on the upper or lower limit, which is the key to MPC's economic benefits; MPC can quickly suppress the impact of interference and can handle the strong coupling of complex multivariable systems. MPC uses the model and the measured values ​​of the current process to calculate the future actions of the manipulated variables, and ensures that all inputs and outputs meet the constraints. Then MPC downloads the calculated multiple control input values ​​to the central processing unit regulator control loop, and the regulator completes the basic control function. At the same time, the feedback mechanism in MPC can compensate for the prediction error between the model and the process. MPC is a model-based finite time domain (or infinite time domain) open-loop optimal control algorithm. Its essence is to use the process model to predict the future state or output. At each sampling moment, the controller uses the current system state as the initial state, and obtains the control action by minimizing the difference between the future output and the reference trajectory (open-loop optimization problem), and optimizes to generate a control input sequence, in which the first component will be executed and the other components will be discarded.

[0130] The specific complete steps of the MPC system are as follows:

[0131] 1. The sensor unit, which consists of temperature sensors, electrical parameter detectors, gas analyzers, etc. at the production site of the submerged arc furnace, senses and detects various physical data parameters such as the actual position of the three electrodes, the actual gear position of the single-phase transformer, the electrode voltage, the electrode current, the electrode power, the furnace bottom temperature, the flue gas temperature, the oxygen content in the flue gas, the hydrogen content in the flue gas, etc. during the production process, and converts the sensed information into electrical signals according to certain rules, and transmits them to the data acquisition unit via data cables or communication cables;

[0132] 2. The data acquisition unit collects the output signal of the sensor unit according to the millisecond scanning cycle, and then transmits the collected data to the central processing unit;

[0133] 3. The central processing unit performs logic control processing on the signals collected from the data acquisition unit in a cycle of seconds, and then transmits the data to the human-computer interaction unit and the model prediction control unit;

[0134] 4. The human-computer interaction unit displays important information in the central processing unit in real time, and receives control instructions from process operators and transmits them back to the central processing unit;

[0135] 5. The model prediction control unit uses a large amount of submerged arc furnace generation process data obtained in the central processing unit to Figure 2 The implementation method of the flowchart shown in the figure performs model predictive control of the production process of the ore-fired furnace. After a single model predictive control calculation task is completed, the prediction calculation result information is transmitted back to the central processing unit in a minute-level cycle.

[0136] 6. When the actual values ​​of various physical data parameters such as the actual position of the three electrodes of the submerged arc furnace, the actual gear position of the single-phase transformer, the electrode voltage, electrode current, electrode power, furnace bottom temperature, flue gas temperature, oxygen content in flue gas, and hydrogen content in flue gas exceed the upper and lower limit protection values, the central processing unit automatically switches the reverse adjustment command output of the production process control strategy of the submerged arc furnace, or accepts the adjustment strategy command from the process operator of the human-computer interaction unit to protect the submerged arc furnace equipment. In the case of serious over-limit, the command to cut off the main power supply circuit of the single-phase transformer is immediately output;

[0137] 7. The central processing unit outputs control instructions and data to the numerical control output unit, and the numerical control output unit continuously refreshes the output change control information from the central processing unit at the millisecond level;

[0138] 8. The control signal is transmitted to the actuator unit composed of solenoid valves, motors, etc. through the numerical control output unit. The on-load voltage regulator driven by the actuator for the forward and reverse rotation of the motor completes the lifting and lowering adjustment of the gear position of the three single-phase transformers. The switch solenoid valve driven by the actuator for hydraulic force completes the lifting and lowering adjustment of the position of the three electrodes. When the process parameters are seriously exceeded, the opening and closing action of the primary high-voltage vacuum circuit breaker of the single-phase transformer is completed;

[0139] 9. Finally, the process data parameters of the electric arc furnace production site change accordingly and are sensed by the sensor unit. This process is repeated continuously to ensure that the entire production system always operates reliably, stably and optimally.

[0140] The present invention provides an advanced control system for a submerged arc furnace based on soft measurement of electrode insertion depth, i.e., the central processing unit in the MPC system. The generalized predictive control (GPC) strategy is used to realize the automatic adjustment control of the submerged arc furnace electrodes, and the PCA-BPN is used to realize the soft measurement of the electrode insertion depth, and the measured value is used as an important input variable of the GPC control system.

[0141] The control scheme of the submerged arc furnace control system, such as Figure 1 As shown, it is composed of a GPC unit, an electrode lifting / shift controller, an electric arc furnace, a PCA unit, and a BPN unit.

[0142] GPC unit: It is adjusted according to the power setting, with the deviation between the smelting power setting value and the real-time power, the voltage / current / active power of the submerged arc furnace electrode, and the soft measurement value of the electrode insertion depth as input, and the furnace bottom temperature, flue gas temperature, O2 content, and H2 content as interference variables, and outputs the electrode insertion depth bias and gear offset;

[0143] Electrode lifting / shift controller: receives the offset output from the GPC unit and outputs the control signal to the actuator of the submerged arc furnace;

[0144] Submerged arc furnace: The control object of this control system is controlled by the electrode lifting / shift controller. The electrical parameters and working condition parameters of the main body are collected and used by other systems.

[0145] PCA unit: reduces the dimension of the electrical parameters and operating parameters of the submerged arc furnace to provide low-dimensional and highly reliable data for the BPN unit;

[0146] BPN unit: Based on the data provided by the PCA unit, soft measurement calculation of the electrode insertion depth is performed and the output signal is transmitted to the GPC unit.

[0147] The generalized predictive control (GPC) MPC strategy is selected, and the controlled autoregressive integrated moving average model (CARIMA) used in minimum variance control is used to describe the object subject to random interference. Compared with the model algorithm control (MAC) and dynamic matrix control (DMC) in the MPC strategy, the generalized predictive control (GPC) algorithm is proposed for random discrete systems, does not require high accuracy of the mathematical model, and maintains the rolling optimization strategy in the predictive control algorithm. The specific calculation method is as follows:

[0148] ① According to MPC control theory, for objects subject to random interference, the controlled autoregressive integral moving average model used in minimum variance control is adopted:

[0149] A(q -1 )y(k)=B(q -1 )u(k-1)+C(q -1 )ξ(k) / Δ

[0150] Where: A(q -1 )=1+a1q -1 +…+a n q -n , B(q -1 )=b0+b1q -1 +…+b m q -m , C(q -1 ) takes a value of 1.

[0151] Among them, q -1 is the backward shift operator, which indicates the corresponding amount of moving back one sampling period; Δ = 1-q -1is a differential operator; ξ(k) is an unrelated random noise sequence, which represents the impact of a type of random noise on the system. The control system developed in this study based on the generalized predictive model control of the electrode insertion depth soft measurement selects the electrode position and the single-phase transformer gear as the output variable y(k) for the three electrodes, selects the electrode insertion depth (subsequent soft measurement algorithm provides data), electrode voltage, electrode current, and electrode active power as the control input variable u(k), and selects the furnace bottom temperature, flue gas temperature, oxygen content in flue gas, and hydrogen content in flue gas as the measurable external interference variable d(k). Rough adjustment is achieved by adjusting the gear of the single-phase transformer and fine adjustment is achieved by raising and lowering the electrode position. In the actual stable production process, the automatic control adjustment of the electrode is mostly carried out by fine-tuning the electrode position. In the adjustment process, the three-phase electrical parameters are balanced.

[0152] ② Control constraints are imposed on the output variable y(k), the control variable u(k) and the control increment Δu(k), y min (k)≤y(k)≤y max (k),u min (k)≤u(k)≤u max (k), Δu min (k)≤Δu(k)≤Δu max (k). The 3% and 97% position points of the hydraulic cylinder driving the lifting of each electrode are set as the lower and upper limits of the electrode position; based on the average value of the three electrode positions, the upper limit of the imbalance of the three electrodes is set to 20%. When the three electrode position values ​​are completely equal, the imbalance value is 0, and no lower limit is set; the upper limit of the single adjustment of the three electrode positions is set to 10mm, and no lower limit is set; the lower and upper limits of the single-phase transformer gear are set according to the gear configuration of the secondary side of the single-phase transformer. When the gear is at the lowest gear and the highest gear, it can no longer be adjusted to the boundary and can only be adjusted within the gear protection limit; the maximum gear difference of the three single-phase transformers is set to 2 gears. In the actual adjustment process, the strategy of the three single-phase transformer gears rising and falling at the same time is preferred. Only when the furnace condition is unstable, the adjustment strategy with gear difference is selected; the upper limit of the single adjustment of the three single-phase transformers is set to 1 gear, and no lower limit is set; based on the rated current of the secondary side of the single-phase transformer The multiple value is used as the upper limit of the electrode current, and 30% of the rated current is set as the lower limit of the electrode current. If the electrode is judged to be broken, the program control will exit the MPC electrode automatic adjustment and switch to manual process operation; based on the average current value of the three electrodes, the upper limit of the current imbalance of the three electrodes is set to 2%. When the current values ​​of the three electrodes are completely equal, the imbalance value is 0, and no lower limit is set; based on the rated voltage of the secondary side of the single-phase transformer The multiple value is used as the upper limit of the electrode voltage, and 70% of the rated voltage is set as the lower limit of the electrode voltage. If an abnormal situation occurs in the production process of the electric arc furnace, the program control exits the MPC electrode automatic adjustment and switches to manual process operation; based on the voltage average value of the three electrodes, the upper limit of the voltage imbalance of the three electrodes is set to 5%. When the voltage values ​​of the three electrodes are completely equal, the imbalance value is 0, and no lower limit is set; 120% multiple value of the rated power of the secondary side of the single-phase transformer is used as the upper limit of the electrode power, and no lower limit is set; based on the power average value of the three electrodes, the upper limit of the power imbalance of the three electrodes is set to 7%, and when the power values ​​of the three electrodes are completely equal, the imbalance value is 0, and no lower limit is set; the upper limits of the furnace bottom temperature and the flue gas temperature are set to 500℃ and 800℃ respectively, and no lower limit is set; the upper limits of the oxygen content in the flue gas and the hydrogen content in the flue gas are set to 0.5% and 8% respectively, and no lower limit is set.

[0153] ③ Assuming the prediction length is j, the output prediction model of the system at time k+j is

[0154] A(q -1 )y(k+j)=B(q -1 )u(k+j-1)+ξ(k+j) / Δ

[0155] In order to obtain the predicted system output value y at time k+j M (k+j), the following Diophantine equation needs to be used

[0156] 1=A(q -1 )ΔE j (q -1 )+q -j F j (q -1 )

[0157] In the formula, E j (q -1 ), F j (q -1 ) is composed of A(q -1 ) and the predicted length j are uniquely determined by the polynomial, which will be omitted for convenience in writing. -1 . Take E1=1, As E j , F j The initial value of , then the formula can be derived to obtain:

[0158]

[0159] E j+1 =E j +e j+1,j q -1 =E j+f j,0 q -1

[0160]

[0161] in,

[0162] ④ Assume that the controlled variable set value at time k is SP, and the actual tracking value is the softened set value w k , the object output expected value adopts the following reference trajectory:

[0163] w(k)=y(k)

[0164] w(k+j)=αw(k+j-1)+(1-α)SP

[0165] Where 0<α<1.

[0166] ⑤ The predicted output at time k+j should be as close as possible to the reference trajectory w(k+j), so the optimization target at time k is

[0167]

[0168] Among them, E is the mathematical expectation; w is the expected value of the output; N1, N2 are the initial value and the final value of the optimization time domain; NU is the control time domain; λ(j) is the control weighting coefficient, which is generally taken as a constant λ.

[0169] ⑥ Minimize the quadratic objective function J(k), let Get the control law

[0170] Δu=(G T G+λI) -1 G T (WH)

[0171] Where I is the identity matrix;

[0172]

[0173] where g j,i =g i+1 (i<j) is the step response coefficient;

[0174] W=[w(k+1)…w(k+N)] T ;

[0175] H=[h1…h N ] T ,

[0176] in:

[0177] h1(t)=[G1(q -1 )-g1,0 ]Δu(t)+F1y(t)

[0178] h2(t)=q[G2(q -1 )-q -1 g 2,1 -g 2,0 ]Δu(t)+F2y(t)

[0179]

[0180] h N (t) = q N-1 [G N (q -1 )-q -(N-1) g N,N-1 -…-g N,0 ]Δu(t)+F N y(t)

[0181] ⑦Let g T For (G T G+λI) -1 G T The first line of , then the instantaneous control increment at time k is

[0182] Δu(k)=g T (WH)

[0183] ⑧Finally, the control quantity u(k)=u(k-1)+Δu(k) is applied to the system to complete the control of this step.

[0184] In the actual application process, the process of implementing GPC in the model prediction control unit is as follows: Figure 3 As shown, the process execution steps are as follows:

[0185] T1, initialization procedure;

[0186] T2, according to the latest input and output data, use the recursive least squares method to estimate the model parameters and obtain A(q -1 ) and B(q -1 );

[0187] T3, according to the obtained A(q -1 ), calculate E(q -1 ) and F(q -1 );

[0188] T4, according to B(q -1 )、E(q -1 ) and F(q -1 )Calculate G(q -1 ) element g i , and calculate H;

[0189] T5. Calculate g T , and calculate the control action deviation Δu(k);

[0190] T6, the control quantity u(k)=u(k-1)+Δu(k) is applied to the control system, the output adjustment of this cycle is completed, and the process jumps to T2 for loop execution;

[0191] The control input variable electrode insertion depth of the GPC generalized predictive control in the MPC strategy is an extremely important parameter for stabilizing furnace conditions, and plays a decisive role in stabilizing furnace conditions and ensuring product quality. In order to obtain the actual insertion depth, production personnel need to shut down the power every 1 to 2 days to measure the actual insertion depth. The detected data is offline data, which has little real-time guidance significance for production operations. The soft sensor model is different from the general mathematical model. It emphasizes obtaining the best estimate of the target dominant variable through auxiliary variables. A section of data before power outage for each measurement of the actual electrode insertion depth is intercepted as historical data and online correction data, and electrical parameters and operating parameters such as electrode voltage, electrode current, electrode power, electrode bearing chamber pressure of electrode lifting hydraulic cylinder, furnace bottom temperature, flue gas temperature, oxygen content in flue gas, and hydrogen in flue gas are selected as input variables. The actually measured electrode insertion depth is selected as the network output variable, and the PCA-BPN network model method is used for soft measurement of electrode insertion depth. Principal component analysis PCA is used to preprocess high-dimensional input variables, and low-dimensional principal component variables reflecting process information are constructed. BP back propagation neural network is then used to establish a soft measurement model for the principal component variables. This method not only simplifies the structure of the neural network model, but also can use the principal component analysis method to detect data errors caused by instrument and process failures, and avoid erroneous output of the model, which is of great significance to model-based optimization control. Artificial Neural Network (ANN) has large-scale parallelism, redundancy, fault tolerance, inherent nonlinearity, self-organization, and self-learning adaptability. Using artificial neural networks for information processing can reduce data analysis and modeling work, and complete complex nonlinear mapping in a non-programmed, adaptive processing method, making it have good application prospects in the field of automatic control. The application of ANN does not require prior knowledge of the object, but directly models the object based on its input and output data; its unique non-traditional expression method and inherent learning ability make it have great potential in solving highly nonlinear and seriously uncertain system control. The BP (Back-Propagation) neural network in ANN adopts tutor learning. By organizing a batch of correct input and output data pairs, adding the input data to the network input, comparing the actual response output of the network with the correct (expected) output to obtain the error, and then modifying each connection weight according to the error, so that the network can continuously change in the direction of correct response until the difference between the actual response output and the expected output is within the allowable range. The BP network has a simple structure, stable working state, and is easy to implement in hardware. When the parameters are appropriate, it can converge to a smaller mean square error. The idea of ​​principal component analysis (PCA) is to project the high-dimensional data space composed of variables such as process data and quality data into a low-dimensional feature space. The principal element variables in the feature space retain the characteristic information of the original variables and eliminate redundant information.PCA can solve data correlation problems well. It obtains the main variable that reflects the data change information from PCA, ignores the influence of secondary variables, and is conducive to improving the model's operation speed in modeling. Obviously, for black box models such as neural networks, using PCA to preprocess the input variables, eliminate irrelevant variables that do not affect the output, and use a small number of irrelevant comprehensive variables to extract the structural relationship between the input variables, reflecting most of the information contained in the original high-dimensional variables, can not only simplify the structure and scale of the network model, but also improve the efficiency and accuracy of the model.

[0192] like Figure 4 As shown in the figure, the specific steps of data dimensionality reduction of principal component analysis (PCA) are as follows: first, obtain electrical parameters and operating parameters to form the original matrix; then perform zero-mean processing on the feature dimension data of electrical parameters and operating parameters, that is, subtract the mean of each feature dimension from the data of the feature; then obtain the covariance matrix of the sampling features; then calculate the eigenvalues ​​of the covariance matrix and their corresponding unit eigenvectors; sort the eigenvectors from large to small according to the eigenvalues; select the eigenvectors corresponding to the first k eigenvalues ​​with the largest contribution rate to the overall feature, and intercept them according to the cumulative variance contribution rate of the first k principal components>0.85; then use these k eigenvectors as row vectors to form a dimensionality reduction matrix, so that unimportant features are discarded and new features with higher principal component contribution rates are retained; then the dimensionality reduction matrix is ​​multiplied by the original matrix on the left to complete the dimensionality reduction, so that the main information of the research object is described, and the amount of calculation for subsequent data processing is greatly reduced, completing the dimensionality reduction of the original data and removing noise.

[0193] The transfer function of the input layer of this BP network adopts a linear function, and the activation function of the hidden layer and the output layer adopts a Sigmoid function. The generalized delta algorithm is used for training, the error adopts the standard error, and the initial distribution of the network weight adopts the normal distribution. In essence, the BP algorithm takes the square of the network error as the objective function and uses the gradient descent method to calculate the minimum value of the objective function. The generalized delta rule algorithm is an iterative gradient descent method that minimizes the square error. It uses a method called momentum to accelerate training. Momentum is an additional weight added to the adjusted weight factor, which increases the training speed by accelerating the change of the weight factor. The generalized delta rule uses a bias function instead of an internal threshold, and adds a bias function when performing node summation.

[0194] The Sigmoid function, also called the Logistic function, is the most basic and standard activation function used in BP neural networks. It is used for neuron output, and the output value range is (0, 1). It can map a real number to the interval (0, 1). It is equivalent to compressing the output of the neuron into the interval (0, 1), which can also be understood as normalizing the output of each neuron.

[0195] Using a three-layer BP network, such as Figure 5 As shown, it consists of input layer A, hidden layer B, and output layer C. The input signal I L From input node a i , through the hidden node b j , and finally reaches the output node c k . There is no connection between nodes in the same layer of BPN, and the output of each layer of nodes only serves as the input of the next layer of nodes. The input mode is the "mode forward propagation" process from the middle layer to the output layer. The error signal of the difference between the expected output of the network and the actual output of the network is the "error back propagation" process of the output layer correcting the connection weights layer by layer through the middle layer. The network "memory training" process is repeated by alternating "mode forward propagation" and "error back propagation". The network tends to converge, that is, the global error of the network tends to the minimum value of the "learning convergence" process. In summary, it is the process of "mode forward propagation" - "error back propagation", "memory training" - "learning convergence". Figure 6 As shown, the specific process execution steps are as follows:

[0196] P0, initialization program;

[0197] P1, give weight v between -1 and +1 ij and w ij Random assignment, in the generalized delta rule, the thresholds of all hidden layers and output layers must be 1, that is, T(2, j) = T(3, k) = 1;

[0198] P2, input vector I i Send it to the neural network and calculate the output of layer 1 according to the following formula

[0199] x i =I i -T 1I =I i -0=I i

[0200]

[0201] P3. Given the output of layer 1, use the following formula to calculate the output of layer 2

[0202]

[0203] Where f() is the Sigmoid function.

[0204] P4. Given the output of layer 2, calculate the output layer result according to the following formula

[0205]

[0206] P5. Continue steps P1 to P4 for the M training patterns sent to the output layer, and calculate the total square error E according to the following formula:

[0207]

[0208] Where M is the number of training patterns sent to the output layer, n is the number of processing units in the output layer, is the expected output value of the mth training pattern on the kth processing unit.

[0209]

[0210] Where f( ) is the Sigmoid function

[0211]

[0212] The partial derivative of the Sigmoid function is

[0213]

[0214] x k is the weighted sum of the inputs of the kth node in the output layer, that is, for the mth training process

[0215]

[0216] For the mth training pattern, is the output value of the jth element in the hidden layer, is the threshold of the output layer.

[0217] P6, given the mth mode, use the following formula to calculate That is, the gradient descent term of the jth processing unit in the first hidden layer (layer 2)

[0218]

[0219]

[0220] The partial derivative of the Sigmoid function is

[0221]

[0222] P7, known hidden layer and the output layer The weight change is calculated using the following formula

[0223]

[0224]

[0225] Where η is the learning rate, is the momentum coefficient, and α is in the range of 0<α<1. Therefore, the momentum term and is the fractional value of the weight change from the previous iteration.

[0226] P8. Knowing the weight change, calculate the weight according to the following formula

[0227]

[0228]

[0229] in is the connection weight between the first i element of the input layer and the jth element of the hidden layer for the mth iteration training; is the connection weight between the jth element of the hidden layer and the kth element of the output layer for the mth iteration training. For all training modes, repeat steps P2 to P8 until the square error is 0 or sufficiently small.

[0230] From the above steps, it can be seen that the generalized delta rule first calculates the output error, and then generates a new weight value based on the gradient of the Sigmoid function and the output error.

[0231] Although the nonlinearity of the production process of the electric arc furnace is absolute, most large-scale production processes are continuous and stable. Therefore, the steady-state and dynamic characteristics of the process in a certain interval near the operating point show a relatively consistent linear relationship. The corresponding model can more accurately characterize the input-output relationship of the process, so that the steady-state target calculation has a better application effect. As the most effective multivariable control method, the role of MPC is to keep the multivariable control system at the optimal working point, that is, to make the system quickly return to the optimal working point when interference occurs. When operating near the optimal working point, the characteristics reflected by the controlled system are linear. The purpose of effectively controlling the system can be achieved by using this linear multivariable system model. The optimization in MPC does not adopt an unchanging global optimization index, but adopts a rolling optimization strategy in a finite time domain. It is a closed-loop optimization based on feedback correction. This rolling implementation of optimization can cope with uncertainties caused by model mismatch, time variation, interference, etc., timely correct the model and compensate for it, and always establish the new optimization on the actual basis to keep the control actually optimal, that is, to determine the current optimal control strategy by rolling optimization, so that the deviation between the controlled variable and the expected value in the future period of time is minimized.

[0232] The key technology of the advanced control system of the electric arc furnace based on soft measurement of electrode insertion depth of the present invention can be partially applied, expanded and transplanted to the control systems of DC electric arc furnaces or other types of industrial furnaces. This part of the technology is also within the protection scope of this scheme and is subject to the constraints of this technical scheme.

[0233] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An advanced control system for a submerged arc furnace based on soft measurement of electrode insertion depth, characterized in that: include: A GPC unit, wherein the GPC unit receives a deviation value between a smelting power setting value and a real-time power value, an ore-arc furnace electrode voltage / current / active power, an electrode insertion depth soft measurement value, and an interference variable as input variables, and the GPC unit outputs an electrode insertion depth bias and a gear position bias to an electrode lifting / gear position controller as output variables; An electrode lifting / shift controller, which receives an electrode insertion depth bias and a shift offset, and generates a corresponding control signal based on the electrode insertion depth bias and the shift offset and sends it to the ore-bearing furnace; The submerged arc furnace receives the control signal and performs control, the submerged arc furnace sends the electrical parameters and operating parameters of the submerged arc furnace to the PCA unit, and the submerged arc furnace sends the electrode voltage / current / active power of the submerged arc furnace to the GPC unit; the submerged arc furnace outputs a real-time power value; Subtract the real-time power value from the smelting power setting value to generate a deviation value between the smelting power setting value and the real-time power value and send it to the GPC unit; A PCA unit, wherein the PCA unit performs dimensionality reduction processing on the received electrical parameters and operating condition parameters, and sends the low-dimensional data to the BPN unit; The BPN unit receives the low-dimensional data and performs soft measurement calculation on the electrode insertion depth based on the low-dimensional data, generates a soft measurement value of the electrode insertion depth and sends it to the GPC unit.

2. The advanced control system for a submerged arc furnace based on soft measurement of electrode insertion depth according to claim 1 is characterized in that: The disturbance variables include furnace bottom temperature, flue gas temperature, O2 content and H2 content.

3. The advanced control system for a submerged arc furnace based on soft measurement of electrode insertion depth according to claim 1 is characterized in that: The calculation method of the GPC unit is as follows: S1. For objects subject to random interference, the controlled autoregressive integrated moving average model used in minimum variance control is adopted: A(q -1 )y(k)=B(q -1 )u(k-1)+C(q -1 )ξ(k) / Δ Where: A(q -1 )=1+a1q -1 +…+a n q -n , B(q -1 )=b0+b1q -1 +…+b m q -m , C(q -1 ) takes the value of 1; Among them, q -1 is the shift operator, which indicates the corresponding amount of one sampling period backward; Δ = 1-q -1 is a differential operator; ξ(k) is an uncorrelated random noise sequence, which represents the influence of a type of random noise on the system; y(k) is the output variable, u(k) is the input variable, and d(k) is the interference variable; coarse adjustment is achieved by adjusting the gear of the single-phase transformer and fine adjustment is achieved by raising and lowering the electrode position; S2. Control constraints are imposed on the output variable y(k), the input variable u(k) and the control increment Δu(k). Δu(k) is the difference between the input variable of the current sampling period and the previous sampling period. min (k)≤y(k)≤y max (k),u min (k)≤u(k)≤u max (k), Δu min (k)≤Δu(k)≤Δu max (k); S3. Assuming the prediction length is j, the output prediction model of the system at time k+j is: Tongue -1 )y(k+j)=B(q -1 )u(k+j-1)+ξ(k+j) / Δ The Diophantine equation is used to obtain the predicted system output value y at time k+j M (k+j), the Diophantine equation is as follows: 1=A(q -1 )ΔE j (q -1 )+q -j F j (q -1 ) In the formula, E j (q -1 ), F j (q -1 ) is composed of A(q -1 ) and the predicted length j are uniquely determined by the polynomial, taking E1 = 1, As E j , F j The initial value of , then the formula can be derived: AND j+1 =And j +e j+1,j q -1 =And j +f j,0 q -1 in, S4, let the controlled variable set value at time k be SP, and the actual tracking value is the softened set value w k , the object output expected value adopts the following reference trajectory: w(k)=y(k) w(k+j)=αw(k+j-1)+(1-α)SP Among them, 0<α<1; S5. Make the predicted output at time k+j as close as possible to the reference trajectory w(k+j), then the optimization target at time k is: Where, E is the mathematical expectation; w is the expected value of the output; N1, N2 are the initial and final values ​​of the optimization time domain; NU is the control time domain; λ(j) is the control weighting coefficient, which is generally a constant λ; S6. Minimize the quadratic objective function J(k), let The control law is obtained as follows: Δu=(G T G+λI) -1 G T (W-H) Where I is the identity matrix; Among them, g j,i =g i+1 (i<j) is the step response coefficient; W=[w(k+1)…w(k+N)] T ; H=[h1…h N ] T ,in: h1(t)=[G1(q -1 )-g 1,0 ]Δu(t)+F1y(t) h2(t)=q[G2(q -1 )-q -1 g 2,1 -g 2,0 ]Δu(t)+F2y(t) h N (t)=q N-1 [G N (q -1 )-q -(N-1) g N,N-1 -…-g N,0 ]Δu(t)+F N y(t) S7, set g T For (G T G+λI) -1 G T The first line of , then the instantaneous control increment at time k is: Δu(k)=g T (WH) S8. Apply the control quantity u(k)=u(k-1)+Δu(k) to the system to complete the control of this step.

4. The advanced control system for a submerged arc furnace based on soft measurement of electrode insertion depth according to claim 3 is characterized in that: S2 specifically includes: The 3% and 97% position points of the hydraulic cylinder stroke of each electrode lifting drive are set as the lower and upper limits of the electrode position; based on the average value of the three electrode positions, the upper limit of the imbalance of the three electrodes is set to 20%. When the three electrode position values ​​are completely equal, the imbalance value is 0, and no lower limit is set; the upper limit of the single adjustment of the three electrode positions is set to 10mm, and no lower limit is set; the lower and upper limits of the single-phase transformer gear are set according to the gear configuration of the secondary side of the single-phase transformer. When the gear is the lowest gear and the highest gear, it is no longer adjusted to the boundary and can only be adjusted within the gear protection limit; the maximum gear difference of the three single-phase transformers is set to 2 gears. In the actual adjustment process, the strategy of the same gear increase and decrease of the three single-phase transformers is preferred. Only when the furnace condition is unstable, the adjustment strategy with gear difference is selected; the upper limit of the single adjustment of the three single-phase transformers is set to 1 gear, and no lower limit is set; based on the rated current of the secondary side of the single-phase transformer The multiple value is used as the upper limit of the electrode current, and 30% of the rated current is set as the lower limit of the electrode current. If the electrode is judged to be broken, the program control will exit the MPC electrode automatic adjustment and switch to manual process operation; based on the average current value of the three electrodes, the upper limit of the current imbalance of the three electrodes is set to 2%. When the current values ​​of the three electrodes are completely equal, the imbalance value is 0, and no lower limit is set; based on the rated voltage of the secondary side of the single-phase transformer The multiple value is used as the upper limit of the electrode voltage, and 70% of the rated voltage is set as the lower limit of the electrode voltage. If an abnormal situation occurs in the production process of the electric arc furnace, the program control exits the MPC electrode automatic adjustment and switches to manual process operation; based on the voltage average value of the three electrodes, the upper limit of the voltage imbalance of the three electrodes is set to 5%. When the voltage values ​​of the three electrodes are completely equal, the imbalance value is 0, and no lower limit is set; 120% multiple value of the rated power of the secondary side of the single-phase transformer is used as the upper limit of the electrode power, and no lower limit is set; based on the power average value of the three electrodes, the upper limit of the power imbalance of the three electrodes is set to 7%, and when the power values ​​of the three electrodes are completely equal, the imbalance value is 0, and no lower limit is set; the upper limits of the furnace bottom temperature and the flue gas temperature are set to 500℃ and 800℃ respectively, and no lower limit is set; the upper limits of the oxygen content in the flue gas and the hydrogen content in the flue gas are set to 0.5% and 8% respectively, and no lower limit is set.

5. The advanced control system for a submerged arc furnace based on soft measurement of electrode insertion depth according to claim 1 is characterized in that: The program execution flow of the GPC unit is as follows: T1, initialization procedure; T2, according to the latest input and output data, use the recursive least squares method to estimate the model parameters and obtain A(q -1 ) and B(q -1 ); T3, according to the obtained A(q -1 ), calculate E(q -1 ) and F(q -1 ); T4, according to B(q -1 )、E(q -1 ) and F(q -1 )Calculate G(q -1 ) element g i , and calculate H; T5. Calculate g T , and calculate the control action deviation Δu(k); T6. Apply the control quantity u(k)=u(k-1)+Δu(k) to the control system. The output adjustment of this cycle is completed and jump to T2 for loop execution.

6. The advanced control system for a submerged arc furnace based on soft measurement of electrode insertion depth according to claim 1 is characterized in that: The dimensionality reduction processing steps of the PCA unit are as follows: Obtain electrical parameters and operating condition parameters to form an original matrix; The electrical parameters and operating condition parameters are processed by zero-meaning the feature dimension data, that is, the mean of each feature dimension is subtracted from the data of the feature to obtain the sampling feature; Obtain the covariance matrix of the sampled features; Calculate the eigenvalues ​​of the covariance matrix and its corresponding unit eigenvector; Sort the eigenvectors by eigenvalue from largest to smallest; Select the eigenvectors corresponding to the first k eigenvalues ​​that contribute the most to the overall features; The first k principal components are selected according to their cumulative variance contribution rate > 0.85; these k eigenvectors are then used as row vectors to form a dimensionality reduction matrix; Multiply the reduced dimension matrix by the original matrix on the left to complete the dimension reduction.

7. The advanced control system for a submerged arc furnace based on soft measurement of electrode insertion depth according to claim 1 is characterized in that: In the BPN unit, the BP back propagation neural network is used to establish a soft measurement model for the principal variable; The BP back propagation neural network includes an input layer A, a hidden layer B, and an output layer C. The transfer function of the input layer of the BP back propagation neural network adopts a linear function, and the activation functions of the hidden layer and the output layer adopt a Sigmoid function. The generalized delta algorithm is used for training, the error adopts the standard error, and the initial distribution of the network weights adopts a normal distribution.

8. The advanced control system for a submerged arc furnace based on soft measurement of electrode insertion depth according to claim 7 is characterized in that: The soft measurement calculation method of the BPN unit is as follows: P0, initialization program; P1, give weight v between -1 and +1 ij and w ij Random assignment, in the generalized delta rule, the thresholds of all hidden layers and output layers must be 1, that is, T(2, j) = T(3, k) = 1; P2, input vector I i Send it to the neural network and calculate the output of layer 1 according to the following formula: x i =I i -T 1I =I i -0=I i P3. Given the output of layer 1, use the following formula to calculate the output of layer 2: Where f( ) is the Sigmoid function; P4. Given the output of layer 2, calculate the output layer result according to the following formula: P5. Continue P1 to P4 for the M training patterns sent to the output layer, and calculate the total square error E according to the following formula: Where M is the number of training patterns sent to the output layer, n is the number of processing units in the output layer, is the expected output value of the mth training pattern on the kth processing unit; Among them, f() is the Sigmoid function: The partial derivative of the Sigmoid function is: x k is the weighted sum of the inputs of the kth node in the output layer, that is, for the mth training process: For the mth training pattern, is the output value of the jth element in the hidden layer, is the threshold of the output layer; P6, given the mth mode, use the following formula to calculate That is, the gradient descent term of the jth processing unit in the front hidden layer: The partial derivative of the Sigmoid function is: P7, known hidden layer and the output layer The weight change is calculated using the following formula: Among them, η is the learning rate, is the momentum coefficient; α is in the range of 0<α<1; therefore, the momentum term and is the fractional value of the weight change from the previous iteration; P8. Knowing the weight change, calculate the weight according to the following formula: in, is the connection weight between the first i element of the input layer and the jth element of the hidden layer for the mth iteration training; is the connection weight between the jth element of the hidden layer and the kth element of the output layer for the mth iteration training. For all training modes, repeat steps P2 to P8 until the square error is 0 or less than the specified value.

Images