Calcium carbide furnace electrode current regulating method and system

Abstract

This application provides a method and system for regulating the electrode current of a calcium carbide furnace, relating to the technical field of electrode current regulation in calcium carbide furnaces. The method includes: collecting and estimating unmeasurable state parameters of the electrode system online using an extended Kalman filter algorithm based on three-phase current data and the lifting / lowering speed data of the three-phase electrodes; obtaining a three-phase current setpoint; and, based on the deviation between the three-phase current setpoint and the collected three-phase current data, the online identified three-phase coupling coefficient matrix, and the unmeasurable state parameters, performing rolling optimization in the prediction time domain using coupled model predictive control, considering three-phase coupling constraints and system time lag, generating a predictive control quantity for the lifting / lowering speed of the three-phase electrodes; and outputting the predictive control quantity for the lifting / lowering speed of the three-phase electrodes to the electrode lifting / lowering actuator to adjust the electrode position. By employing the above method, the technical effect of improving the accuracy of three-phase current control is achieved.

Description

Technical Field

[0001] This application relates to the technical field of electrode current regulation in calcium carbide furnaces, and in particular to a method and system for regulating electrode current in calcium carbide furnaces. Background Technology

[0002] The calcium carbide furnace is the core equipment in calcium carbide production. It uses the thermal energy of an electric arc to heat raw materials such as lime and coke to high temperatures for a chemical reaction. It typically uses a three-phase AC power supply, with each phase equipped with a graphite electrode. The arc length is controlled by adjusting the position of the electrodes, which in turn regulates the current in each phase to maintain thermal balance within the furnace. Precise adjustment of the electrode current is crucial for the stable operation of the calcium carbide furnace. However, in actual production, there is a strong electromagnetic coupling effect between the three-phase electrodes. Adjusting one phase electrode will cause fluctuations in the current of the other two phases.

[0003] Chinese invention patent CN108444280B, published on September 24, 2019, provides a precise control method for electrode raising and lowering in a calcium carbide furnace. This method classifies each furnace run as a cycle, decomposing the cycle into three working stages: the furnace exit period, the adjustment period, and the stabilization period. Different control strategies are implemented for each of these three working stages, with each stage controlling the three-phase electrodes individually. Electrode correlation is established based on the interrelationships between the three-phase electrodes, and the current and voltage of the original electrode are controlled by adjusting the position of the associated electrode holder. The advantage of this method is that it recognizes the mutual influence between the three-phase electrodes and attempts to compensate for it through associated electrodes. However, the "electrode correlation" in this method is a static logic control achieved through a preset rule table. These correlation rules are pre-set and fixed, and cannot be adjusted in real time according to dynamic changes in raw material composition, furnace temperature distribution, electrode state, and other operating conditions. Therefore, when furnace conditions change, a deviation occurs between the preset correlation rules and the actual coupling characteristics, leading to a decrease in decoupling effect and further reducing the accuracy of three-phase current control. Summary of the Invention

[0004] To improve the accuracy of three-phase current control, this application provides a method and system for adjusting the electrode current of an calcium carbide furnace.

[0005] In a first aspect, this application provides a method for adjusting the electrode current of a calcium carbide furnace, employing the following technical solution: A method for adjusting the electrode current of a calcium carbide furnace includes the following steps: Collect three-phase current data and three-phase electrode lifting and lowering speed data of the calcium carbide furnace; Based on the three-phase current data and the lifting and lowering speed data of the three-phase electrodes, the unmeasurable state parameters of the electrode system are estimated online using the extended Kalman filter algorithm. The unmeasurable state parameters include the furnace entry depth of the three-phase electrodes and the equivalent resistance of the three-phase material layer. The three-phase current setpoint is obtained. Based on the deviation between the three-phase current setpoint and the collected three-phase current data, the three-phase coupling coefficient matrix obtained online, and the unmeasurable state parameters, and considering the three-phase coupling constraints and system time lag, the three-phase electrode rise and fall speed is generated by rolling optimization in the prediction time domain through coupling model predictive control. The predicted control value of the three-phase electrode lifting speed is output to the electrode lifting actuator to adjust the electrode position.

[0006] By adopting the above technical solutions, the extended Kalman filter uses measurable data to estimate the electrode entry depth and equivalent resistance of the material layer online, providing dynamic state information for control. Coupled model predictive control, based on current deviation, the online identified coupling coefficient matrix, and state estimation results, comprehensively considers coupling constraints and hysteresis characteristics in the prediction time domain, generating the optimal control sequence through rolling optimization. This application updates the coupled model with operating conditions through online identification, grasps real-time operating conditions through state estimation, and compensates for hysteresis through predictive control, thus helping to improve the dynamic adaptability of three-phase current control under unstable conditions.

[0007] Optionally, the method further includes: obtaining the three-phase coupling coefficient matrix online by recursive least squares method with forgetting factor; The recursive least squares method with a forgetting factor performs the following update steps in each sampling period: Collect the changes in the positions of the three-phase electrodes and the changes in the three-phase current at the current moment; The prediction error at the current moment is calculated based on the changes in the positions of the three-phase electrodes, the changes in the three-phase current, and the coupling coefficient matrix at the previous moment. Calculate the gain vector at the current moment based on the covariance matrix of the previous moment, the change in the position of the three-phase electrodes, and the forgetting factor; The coupling coefficient matrix at the current time is updated based on the coupling coefficient matrix at the previous time step, the gain vector, and the prediction error. The covariance matrix at the current moment is updated based on the gain vector, the change in the position of the three-phase electrodes, and the covariance matrix at the previous moment.

[0008] By adopting the above technical solution, the online real-time identification of the three-phase coupling coefficient matrix is ​​realized. The coupling coefficient is dynamically corrected using the latest data, which helps to enhance the tracking ability of furnace condition changes and provides a real-time updated coupling model for decoupling control.

[0009] Optionally, the method further includes: calculating a decoupling compensation matrix based on the three-phase coupling coefficient matrix obtained through online identification; Calculate the current deviation between the three-phase current setpoint and the collected three-phase current data; Based on the decoupling compensation matrix and the current deviation, the three-phase electrode position adjustment amount is calculated and superimposed on the predicted control amount of the three-phase electrode lifting and lowering speed as a decoupling compensation amount.

[0010] By adopting the above technical solution, the current deviation is processed using the decoupling compensation matrix, and the decoupling compensation amount is generated and superimposed on the control command to counteract the electromagnetic coupling effect between the three-phase electrodes. This helps to improve the independent regulation capability of the three-phase current and reduce the current imbalance.

[0011] Optionally, the method further includes: performing nonlinear dynamic compensation through an RBF neural network inverse model to obtain the neural network compensation amount for the rise and fall speeds of the three-phase electrodes; The input of the RBF neural network inverse model is the three-phase current deviation and the current three-phase electrode position, and the output is the neural network compensation amount of the three-phase electrode rise and fall speed. The neural network compensation quantity is superimposed with the predicted control quantity of the three-phase electrode lifting and lowering speed and then output.

[0012] By adopting the above technical solution, the nonlinear characteristics of the system are compensated by using the inverse model of the RBF neural network. The current deviation and electrode position are used as inputs to map the required speed compensation amount, which is used to offset the influence of nonlinear links in advance, and helps to improve the tracking accuracy when the operating conditions change over a wide range.

[0013] Optionally, the construction of the RBF neural network inverse model includes: Collect historical operating data, which includes three-phase current, three-phase electrode position and corresponding electrode lifting speed; An RBF neural network is constructed using three-phase current deviation and three-phase electrode position as inputs and electrode lifting and lowering speed as outputs. The center vectors of neurons in the hidden layer of the RBF neural network are determined using the K-means clustering algorithm. The width parameter of the hidden layer neurons is determined based on the maximum distance between the center vectors of the hidden layer neurons and the number of hidden layer neurons. The weight matrix of the output layer of the RBF neural network is trained using the least squares method.

[0014] By adopting the above technical solution, a method for constructing RBF neural networks is provided, enabling neurons to cover different working conditions and providing an inverse model basis for nonlinear compensation.

[0015] Optionally, the method further includes: updating the weight matrix of the RBF neural network online through incremental learning during operation; The incremental learning process performs the following steps in each sampling period: Calculate the RBF neural network output at the current time step based on the input vector at the current time step and the weight matrix at the previous time step. The output error is calculated based on the expected speed corresponding to the actual control effect and the output of the RBF neural network. The weight matrix is ​​updated based on the output error, the hidden layer output vector at the current time, and the learning rate.

[0016] By adopting the above technical solution, online adaptive updating of the neural network inverse model is realized. The weights are corrected online using real-time feedback error, enabling the model to maintain compensation accuracy as the operating conditions drift, which helps to enhance the long-term stability of the system.

[0017] Optionally, the online estimation of the unmeasurable state parameters of the electrode system using an extended Kalman filter includes: A state-space model is established with the three-phase electrode entry depth and the equivalent resistance of the three-phase material layer as state vectors, the three-phase electrode lifting and lowering speed as control input, and the three-phase current as observation output. In each sampling period, the prediction step of the extended Kalman filter is executed, and the state prediction value and covariance prediction value at the current time are calculated based on the state estimate value, control input and state transition function of the previous time step. Collect the current three-phase current data as the observation value; Calculate the Kalman gain based on the predicted covariance, the Jacobian matrix of the observation function, and the observation noise covariance matrix. The state estimate at the current moment is updated based on the predicted state value, the Kalman gain, the observed value, and the observation function.

[0018] By adopting the above technical solution, the specific implementation of extended Kalman filtering is disclosed. By fusing model information and real-time measurements through a prediction-update recursive structure, state estimation is given in a nonlinear noise environment, providing state information for predictive control.

[0019] Optionally, the rolling optimization in the prediction time domain through coupled model predictive control includes: Define the prediction time domain and the control time domain; Based on the current state estimate and the three-phase coupling coefficient matrix, a prediction model is established to predict the system state and current at future times. Define a reference trajectory for a smooth transition from the current measured value to the three-phase current setpoint; Construct an optimization objective function, which includes the sum of squared current tracking errors in the prediction time domain and the sum of squared control increments in the control time domain; At each sampling time, the optimization objective function is solved, and the optimal control sequence in the control time domain is obtained under the condition of satisfying the preset constraints. The first control variable in the optimal control sequence is output as the predicted control variable for the acceleration and deceleration of the three-phase electrodes.

[0020] By adopting the above technical solution, the specific application of coupled model predictive control is illustrated. By predicting system behavior through rolling optimization, the tracking accuracy and control stability are balanced, which can overcome the effects of large lag and respond to changes in operating conditions in a timely manner.

[0021] Optionally, the preset constraints include three-phase coupling constraints, electrode velocity constraints, and current safety constraints; The three-phase coupling constraint is that, within the prediction time domain, the relationship between the predicted current change and the predicted electrode position change is determined by the three-phase coupling coefficient matrix. The electrode speed constraint is that the output three-phase electrode rising and falling speeds are within a preset minimum and maximum range; The current safety constraint is that the predicted current value is within a preset minimum and maximum range.

[0022] By adopting the above technical solution, key process limitations are incorporated into the optimization problem to ensure the consistency between the prediction model and the real system, prevent the actuator from exceeding the limit, and ensure the safe operation of the calcium carbide furnace.

[0023] Secondly, this application provides a calcium carbide furnace electrode current adjustment system, which adopts the following technical solution: An electrode current regulating system for a calcium carbide furnace, comprising: The data acquisition module is used to collect the three-phase current data and the lifting and lowering speed data of the three-phase electrodes of the calcium carbide furnace. The state estimation module is used to estimate the unmeasurable state parameters of the electrode system online using an extended Kalman filter algorithm based on the three-phase current data and the rising and falling speed data of the three-phase electrodes. The unmeasurable state parameters include the furnace entry depth of the three-phase electrodes and the equivalent resistance of the three-phase material layer. The predictive control module is used to acquire the three-phase current setpoint. Based on the deviation between the three-phase current setpoint and the acquired three-phase current data, the three-phase coupling coefficient matrix obtained online, and the unmeasurable state parameters, and considering the three-phase coupling constraints and system time lag, the module performs rolling optimization in the prediction time domain through coupled model predictive control to generate predictive control quantities for the rise and fall speeds of the three-phase electrodes. The control output module is used to output the predicted control quantity of the lifting speed of the three-phase electrodes to the electrode lifting actuator to adjust the electrode position.

[0024] In summary, this application includes at least one of the following beneficial technical effects: 1. By using a recursive least squares method with a forgetting factor to identify the three-phase coupling coefficient matrix online, the coupling model can be updated in real time according to changes in furnace conditions, overcoming the limitations of existing technologies that rely on static preset rule tables for decoupling compensation. When the raw material composition, furnace temperature distribution, or electrode state changes, the coupling coefficient matrix can be automatically adjusted to match the decoupling compensation amount with the current actual coupling characteristics, which helps to reduce the imbalance of the three-phase current and improve the independent regulation capability of the three-phase current.

[0025] 2. By employing an extended Kalman filter algorithm and utilizing measurable three-phase current and electrode rise / fall speed data, the electrode entry depth and equivalent resistance of the material layer, which cannot be directly measured, are estimated online. These state parameters are fundamental information for current control, and their online estimation provides real-time state input for coupled model predictive control, enabling the controller to obtain current furnace condition information and avoid control deviations caused by unknown key parameters or the use of fixed empirical values.

[0026] 3. Coupled model predictive control is employed for rolling optimization within the prediction time domain, taking into account the time lag between electrode action and current change when generating control commands. By establishing a predictive model to anticipate the system response at future moments and re-optimizing the solution in each sampling period, the control quantity can act in advance to offset the lag effect, which helps to suppress overshoot and oscillation and improve the dynamic response characteristics of current regulation. Attached Figure Description

[0027] Figure 1 This is a flowchart of Embodiment 1 of this application. Detailed Implementation

[0028] The following combination Figure 1 This application will be described in further detail.

[0029] Example 1: This example discloses a method for adjusting the electrode current of a calcium carbide furnace. The method includes the following steps: S1 data acquisition, S2 state estimation, S3 predictive control, and S4 control output. The execution process of each step in this example is as follows: S1 Data Acquisition: Acquires three-phase current data and three-phase electrode lifting and lowering speed data of the calcium carbide furnace.

[0030] Specifically, the instantaneous values ​​of the three-phase current are acquired in real time by current sensors installed on the electrodes of each phase of the calcium carbide furnace, and denoted as... , , Where the subscript k represents the current sampling time. Simultaneously, the lifting speed of the three-phase electrodes is obtained through the encoder or displacement sensor of the electrode lifting mechanism, denoted as... , , This speed is the actual execution speed output from the previous control cycle, used for state estimation at this moment.

[0031] S2 State Estimation: Based on the three-phase current data and the rise / fall speed data of the three-phase electrodes, the unmeasurable state parameters of the electrode system are estimated online using an extended Kalman filter algorithm. The unmeasurable state parameters include the furnace entry depth of the three-phase electrodes and the equivalent resistance of the three-phase material layer. The specific implementation includes the following sub-steps: S2.1 Establishing the State-Space Model. To apply Kalman filtering, the actual physical system needs to be described using a mathematical model. The state vector is defined as follows:

[0032] in, For state vectors, These are the furnace insertion depths of the three phase electrodes A, B, and C, respectively, in meters; A, B, and C are the equivalent resistances of the three-phase material layers, in ohms. The transpose symbol indicates that a row vector is transposed into a column vector.

[0033] The control input to the state-space model is the rise and fall speed of the three-phase electrodes. The unit is meters per second. The observed output of the state-space model is the three-phase current. The unit is ampere.

[0034] The state transition equation is established based on the relationship between the rate of change of electrode depth in the furnace and the lifting / lowering speed, as well as a random walk model of the resistance. The rate of change of electrode depth in the furnace is determined by the electrode lifting / lowering speed, i.e. (The negative sign indicates that the depth of the electrode entering the furnace increases as it descends.) in, The depth of the i-th phase electrode in the furnace is indicated by its position. Indicates time, This indicates the rising and falling speed of the i-th phase electrode; "-" indicates that the depth of entry into the furnace increases when the electrode descends (the speed is positive). The change in the equivalent resistance of the material layer can be modeled as a random walk process, i.e. ,in, This is process noise, used to represent the randomness of resistance changes; Let be the equivalent resistance of the i-th phase material layer. For time.

[0035] In this embodiment, the state transition equation is discretized, and the discretized state transition equation is a nonlinear function:

[0036] in, Let k be the state vector at time k. Let be the state vector at time k+1. The sampling period is the time interval between two calculations. It is a nonlinear state transition function, the specific form of which is determined by physical relationships, namely by the relationship between the change of electrode depth into the furnace and the lifting and lowering speed, as well as the random walk model of the resistance. This is the control input at time k, i.e., the rising and falling speeds of the three electrodes. The process noise has the following covariance matrix: 1, 1 represents the uncertainty of the model.

[0037] The observation equations are established based on circuit theory. The three-phase currents are related to the phase voltage, equivalent resistance, and equivalent reactance, and can be expressed as nonlinear functions:

[0038] in, The observed output at time k is the three current measurements. It is a nonlinear observation function, specifically it can be derived from... approximate, Let i be the current of phase i. Let i be the voltage of phase i. , The equivalent reactance of the i-th phase; The covariance matrix of the observation noise is: , This represents the measurement error of the sensor.

[0039] S2.2 Performs the prediction step of the extended Kalman filter. The prediction step uses the optimal estimate from the previous time step to extrapolate the current state and error covariance. Based on the state estimate from the previous time step... and control input The predicted state value at the current moment is calculated using the state transition function:

[0040] in, The state prediction value at time k is the estimate of the state at time k based on the information at time k-1. This is the state estimate at time k-1; The sampling period; It is a nonlinear state transition function; This is the control input at time k-1.

[0041] Simultaneously calculate the predicted value of the error covariance:

[0042] in, Let be the predicted error covariance at time k, representing the uncertainty of the predicted state value; Let be the Jacobian matrix of the state transition function. The Jacobian matrix is ​​obtained by taking the partial derivative of the nonlinear function. Its function is to linearize the nonlinear system near the current estimation point, thereby enabling the Kalman filter framework to handle nonlinear problems. This is the estimated value of the error covariance at time k-1. Let T be the process noise covariance matrix, and T be the transpose sign, which will be used in subsequent update steps.

[0043] S2.3 executes the update step of the extended Kalman filter. The update step uses the actual measured values ​​at the current moment to correct the predicted values, obtaining a more accurate state estimate. First, the three-phase current data at the current moment are collected as observations. Then, based on the covariance prediction value... Jacobian matrix of observation function Calculate the Kalman gain:

[0044] in, Here is the Kalman gain matrix. This is the predicted value of the error covariance. Let T be the Jacobian matrix of the observation function, and T be the transpose sign. Write the RV matrix to observe the noise.

[0045] Kalman gain This determines the weighting of predicted and measured values ​​when correcting the state. A larger gain indicates greater trust in the measured values, while a smaller gain indicates greater trust in the predicted values. Finally, the state estimate is updated using the observation residuals.

[0046] in, The state estimate at time k is the final estimate after fusing the predicted and measured values. This is the predicted state value; The Kalman gain matrix; These are the actual observed values, and the observed output at time k, i.e., the three current measurement values; These are theoretical observations calculated based on state predictions; The observation residual represents the difference between the actual measurement and the theoretical prediction.

[0047] And update the covariance estimate:

[0048] in, Let be the error covariance estimate at time k, used to represent the uncertainty of the state estimate; It is the identity matrix. For Kalman gain, Let Jacobian matrix be the observation function. This represents the predicted value of the error covariance.

[0049] After the above recursion, the state estimate at the current time is obtained. This includes the estimated electrode entry depth into the furnace. Estimated equivalent resistance of material layer These estimates cannot be measured directly, but can be obtained online from the measurable current using an EKF (Extended Kalman Filter), providing crucial state information for precise control.

[0050] S3 Predictive Control: The three-phase current setpoints are acquired. Based on the deviation between the setpoints and the collected three-phase current data, the online-identified three-phase coupling coefficient matrix, and the unmeasurable state parameters, and considering three-phase coupling constraints and system time lag, rolling optimization is performed in the prediction time domain using coupled model predictive control to generate predictive control quantities for the three-phase electrode rise and fall speeds. This step specifically includes the following sub-steps: S3.1 Online Identification of Three-Phase Coupling Coefficient Matrix. Strong electromagnetic coupling exists between the three-phase electrodes of an calcium carbide furnace; that is, adjusting one phase electrode will cause changes in the currents of the other two phases. To describe this coupling relationship, this invention employs a recursive least squares method with a forgetting factor to identify the three-phase coupling coefficient matrix online. The three-phase coupling model describes the relationship between electrode position changes and current changes:

[0051] in, This represents the change in three-phase current. These represent the changes in phase currents A, B, and C, respectively. This represents the change in the position of the three-phase electrodes. These represent the changes in the positions of phases A, B, and C, respectively. It is a 3×3 coupling coefficient matrix, whose elements This represents the effect of the change in the position of the j-th phase electrode on the current of the i-th phase. (Diagonal elements) These are the phase self-coupling coefficients, and the off-diagonal elements. (i≠j) is the mutual coupling coefficient.

[0052] The following update steps are performed for each sampling period: Collect the changes in the positions of the three-phase electrodes at the current moment. and changes in three-phase current The change in electrode position can be obtained by integrating the electrode's lifting and lowering speed, or by direct measurement using a displacement sensor.

[0053] Taking phase A as an example, define the parameter vector to be identified. Regression vector ;in, The parameter vector to be identified contains the self-coupling coefficients of phase A. and the mutual coupling coefficients of B and C relative to phase A ; The regression vector at time k is composed of the changes in the positions of the three-phase electrodes at the current time.

[0054] Calculate the prediction error:

[0055] in, Let A be the prediction error at time k. This represents the actual change in the phase A current at time k; These are the estimated values ​​of the A-phase parameters at time k-1. Let be the regression vector at time k. This is the predicted value of the current change based on the parameters at the previous moment.

[0056] Calculate the gain vector:

[0057] in, Let be the gain vector at time k. Let be the covariance matrix at time k-1. Let be the regression vector at time k. Forgetting factor (0 < ≤1), used to adjust the rate at which the algorithm forgets historical data: The closer it is to 1, the slower the algorithm forgets historical data and the smoother the estimation result; The smaller the value, the more sensitive the algorithm is to recent data, and the faster the tracking speed. The value is typically between 0.95 and 0.99.

[0058] Update parameter estimates:

[0059] in, The estimated value of phase A parameters at time k; The estimated value of phase A parameters at time k-1; It is the gain vector; This represents the prediction error.

[0060] Update the covariance matrix:

[0061] in, Let be the covariance matrix at time k. For previous factors, Let be the covariance matrix at time k-1. For the gain vector, This is the transpose of the regression vector.

[0062] Repeat the above process for phases B and C to obtain the complete estimated values ​​of the coupling coefficient matrix. This matrix is ​​updated online as the furnace conditions change, providing real-time and accurate coupling information for decoupled control and prediction models.

[0063] S3.2 Calculate the decoupling compensation amount. Based on the coupling coefficient matrix obtained from online identification, the decoupling compensation matrix can be calculated. The purpose of the decoupling compensation matrix is ​​to introduce a reverse compensation in the control loop, making the transfer function matrix from the control input to the current output approximately diagonalized, thereby eliminating the mutual influence between the three phases.

[0064] Obtain three-phase current setpoint Calculate the current deviation ,in, This is the three-phase current setpoint vector. These are the set values ​​for the currents of phases A, B, and C, respectively. The measured three-phase currents at the current time (time k) are... Let be the current deviation vector at time k. Then calculate the electrode position adjustment after decoupling:

[0065] in, Let be the decoupling compensation position adjustment vector at time k, representing the additional position adjustment required for each phase electrode to eliminate current deviation, considering the coupling effect. For decoupling compensation matrix, This is the current deviation vector.

[0066] S3.3 Calculation of Nonlinear Compensation. The calcium carbide furnace system exhibits strong nonlinear characteristics; the relationship between current and electrode position is not a simple linear one. To compensate for this nonlinearity, this invention uses an RBF neural network inverse model for nonlinear dynamic compensation, obtaining the neural network compensation amount for the three-phase electrode rise and fall speeds. .

[0067] The structure of the RBF neural network inverse model is as follows: the input layer contains 6 neurons, each corresponding to a three-phase current deviation. and current three-phase electrode positions The hidden layer contains N neurons (N=20 in this embodiment), using a Gaussian radial basis function as the activation function; the output layer contains 3 neurons, outputting the neural network compensation amount for the rise and fall speeds of the three-phase electrodes. The output of the hidden layer neurons is:

[0068] in, This represents the output value of the i-th hidden layer neuron. For the input vector, Let be the center vector of the i-th hidden layer neuron. Let be the width parameter of the i-th hidden layer neuron. Let Euclidean distance be the Euclidean distance between the input vector and the center vector. It is an exponential function. The role of radial basis functions is to map the input space to a high-dimensional feature space, making a problem that was originally nonlinearly separable linearly separable in the high-dimensional space. The network output is:

[0069] in, This is the output vector of the neural network, which represents the neural network compensation amount for the three electrode lifting and lowering velocities. The output layer weight matrix (dimension 1) ), , which is the output vector of the hidden layer.

[0070] The RBF neural network inverse model is constructed through a combination of offline training and online updating: Offline training: Training is performed using historical operational data (including three-phase current, three-phase electrode positions, and corresponding electrode rise and fall rates). First, the K-means clustering algorithm is used to determine the center vectors of the hidden layer neurons. This ensures that the center can cover the main region of the input space. Then, based on the maximum distance between the center vectors... The width parameter is determined by the number of neurons N. To ensure appropriate overlap of the effective ranges of each center, the least squares method is finally used to train the output layer weight matrix. That is, to solve ,in, This is the hidden layer output matrix, where each row corresponds to the hidden layer output vector of a single sample. This is the desired output matrix, where each row corresponds to the desired output (electrode lifting and lowering speed) of a sample.

[0071] Online Update: During operation, the network weights are updated in each sampling period using an incremental learning algorithm to adapt to dynamic changes in the furnace conditions. Specifically, based on the current input vector... and the weight matrix of the previous time step Calculate the current network output Based on the expected speed corresponding to the actual control effect. (This can be estimated by a feedback controller or human experience) and the output error is calculated using the network output. Update the weight matrix: ,in, The updated (time k) weight matrix, This is the weight matrix from the previous time step. The hidden layer output vector. This is the transpose of the output error. The learning rate (ranging from 0.001 to 0.01) allows the network to continuously adapt to new operating conditions and improve compensation accuracy through incremental learning.

[0072] S3.4 Coupled Model Predictive Control (CPDC) Rolling Optimization. Coupled Model Predictive Control is an advanced model-based control strategy that determines the optimal control action by solving an optimization problem in a finite time domain online. It effectively handles multivariable coupling, constrained, and large time-delay problems. This step sets the prediction time domain. and control time domain Considering a system time lag of approximately 30 to 60 seconds, the sampling period is set to 1-5 seconds, with the specific sampling period determined based on the performance of the computing device. The prediction time domain is set to 40 steps, and the control time domain to 8 steps.

[0073] Based on the current state estimate (From S2) and the identified coupling coefficient matrix (From S3.1), establish a predictive model. Use the state-space model to recursively predict the system state at future times:

[0074] in, Let k be the state vector predicted at time k+i+1. Let k be the state vector predicted at time k+i. The sampling period is It is a nonlinear state transition function. This is the control input at time k+i, which is suitable for the k-th programming.

[0075] The corresponding predicted current value is given by the observation equation:

[0076] in, Let be the current vector predicted at time k+i from time k. For nonlinear observation functions, Let be the state vector predicted at time k+i at time k.

[0077] At the same time, the current change and the electrode position change must satisfy the identified coupling relationship:

[0078] in, Let be the predicted change in current at time k+i, calculated at time k. The coupling coefficient matrix identified at time k. This represents the change in electrode position between adjacent time points. Let be the electrode position predicted at time k+i.

[0079] Define a reference trajectory to smoothly transition the current from the current measurement value to the set value according to an exponential law, in order to minimize the need for overly abrupt control actions.

[0080] in, Let be the reference value of the current at time k+i. For current setpoint vector, Let be the measured current value at time k. The sampling period is The reference trajectory time constant (which can be 10 to 20 seconds) determines the speed of the transition. Natural constant.

[0081] Construct an optimization objective function, including the sum of squared current tracking errors in the prediction time domain and the sum of squared control increments in the control time domain:

[0082] in, To optimize the objective function, To predict the length of the time domain, Let be the reference value of the current at time k+i. Let k be the current value predicted at time k+i. The weighted square norm is calculated by squaring each element of a vector and multiplying it by the weight matrix. ; To control the length of the time domain, To control the increment, and This is the weight matrix. Typically, a diagonal matrix is ​​used, and the diagonal elements determine the degree of importance attached to the current tracking error of each phase. We also use a diagonal matrix, where the diagonal elements determine the smoothness of the control actions. This embodiment uses... (Identity matrix) To balance tracking performance and control stability.

[0083] Consider the following constraints to ensure that the control actions are safe and feasible: Three-phase coupling constraint: The predicted current change and the predicted electrode position change must satisfy... This constraint directly embeds the coupled model into the prediction, so that the optimization results automatically satisfy the coupling relationship.

[0084] Electrode speed constraint: The output electrode lifting speed must be within the range allowed by the actuator, i.e. For example, the maximum lifting speed of the electrode is typically 5 to 10 mm / s.

[0085] Current safety constraint: The predicted current value must be within the safe range to prevent overcurrent tripping, i.e. Typically, the upper limit of the current is 110% of the rated current.

[0086] At each sampling time, the constrained quadratic programming optimization problem described above is solved to obtain the optimal control sequence in the control time domain. Based on the principle of rolling optimization, only the first control variable is used. As the output of the coupled model predictive control At the next sampling time, optimization is performed again to achieve feedback correction.

[0087] S3.5 Control Quantity Synthesis. The decoupling compensation quantity, nonlinear compensation quantity, and predictive control quantity are superimposed to obtain the final control quantity for the current three-phase electrode acceleration and deceleration rates:

[0088] in, The output at time k represents the control quantity for the acceleration and deceleration of the three-phase electrodes. For coupled model predictive control output, For neural network compensation quantity, To decouple and compensate for position adjustment, The sampling period is This involves converting the decoupling compensation quantity into a velocity quantity. In this way, the control quantity is responsible for the core optimization and adjustment, the neural network compensation quantity is responsible for nonlinear dynamic compensation, and the decoupling compensation quantity is responsible for eliminating three-phase coupling. The three work together to achieve high-performance control.

[0089] S4 control output: Predictive control quantity for the acceleration and deceleration speed of the three-phase electrodes. The output is sent to the electrode lifting actuator. The actuator (such as a hydraulic proportional valve or an electric push rod) drives the electrode to lift or lower according to the command, changing the electrode's depth in the furnace, thereby adjusting the three-phase current to quickly and smoothly track the set value.

[0090] Example 2: The difference between this example and Example 1 is that the extended Kalman filter parameters in the state estimation of step S2 are set differently and dynamically adjusted.

[0091] Process noise covariance matrix (Include Differentiation settings: Based on the differences in the physical characteristics of electrode position changes and resistance changes, Designed as a diagonal matrix:

[0092] in The corresponding electrode insertion depth reflects the measurement error of the electrode lifting speed and the uncertainty of the position model. Its value is relatively small, so we take... ; The equivalent resistance of the corresponding material bed reflects the impact of random fluctuations in furnace conditions on the resistance. Its value is relatively large, so we take... This differentiated setting allows the process noise covariance matrix to be matched with the dynamic characteristics of different state variables.

[0093] Observation noise covariance matrix (Include Dynamic adjustment of ) Observation noise covariance matrix Online dynamic adjustments are made based on the basic settings. First, the basic observation noise variance is set according to the accuracy of the current sensor. (Corresponding to 0.5%FS, rated current 75kA). During operation, a length is selected centered on the current moment. Using a sliding window, calculate the sample variance of the three-phase current measurements within the window. ( ), and adjust the observation noise variance at the current time according to the following rules:

[0094] in For adjustment coefficients, , These represent the upper and lower bounds of the variance. The observation noise covariance matrix at the current time is: This enables dynamic matching with real-time measured noise levels.

[0095] Example 3: This example differs from Example 1 in that the forgetting factor in step S3.1 coupling identification is adaptively adjusted. In Example 1, the forgetting factor... The value is fixed, but in this embodiment, the forgetting factor is dynamically adjusted according to the rate of change of the furnace condition.

[0096] The specific method is as follows: In each sampling period, calculate the prediction error at the current moment. (Taking phase A as an example) the absolute value, and compared with the preset error threshold. Comparison. If This indicates that the furnace condition may have changed drastically, requiring a faster tracking speed. Therefore, the forgetting factor should be reduced to a lower value (e.g., );like If the value remains below the threshold, it indicates that the furnace condition is relatively stable. Therefore, the forgetting factor should be gradually increased to a higher value (e.g., ...). This is done to improve the stationarity of parameter estimation. The adjustment rule for the forgetting factor is:

[0097] in, Let k be the forgetting factor at time k. The forgetting factor at time k-1 The step size (can be from 0.001 to 0.01). The minimum value of the same factor. This represents the maximum value of the forgetting factor. To set the prediction error threshold, This represents the absolute value of the prediction error. By adaptively adjusting the forgetting factor, the recursive least squares algorithm can respond quickly to changes in furnace conditions while maintaining parameter smoothness during steady-state operation, thus improving the robustness of decoupled control.

[0098] Example 4: This example differs from Example 1 in that the inverse model of the RBF neural network in step S3.3 nonlinear compensation is further refined. In Example 1, the number of hidden layer neurons in the RBF neural network is fixed, while in this example, the number of hidden layer neurons is dynamically determined according to the complexity of the input space to improve the generalization ability of the model.

[0099] The specific method is as follows: During the offline training phase, an incremental clustering method is used to gradually increase the number of neurons. First, the initial number of neurons is set. Then, the distance between each sample and the existing center vector is calculated. If the distance between a sample and the nearest center is greater than a preset threshold... If the sample is selected correctly, it becomes the new center, and a neuron is added; otherwise, the parameters of the nearest center are updated using the sample (e.g., by adjusting the center position through a moving average). In this way, the number of neurons in the hidden layer can be adaptively determined according to the data distribution.

[0100] Furthermore, during the online update phase, when a new operating condition is detected (e.g., the state estimate exceeds the historical range), new neurons can be dynamically added, with the center of the new neuron set to the current input vector and the width parameter set to the initial value. The weights are initialized to zero. Simultaneously, to prevent the number of neurons from growing indefinitely, neurons with low usage frequency are periodically pruned, removing those that have not been activated for a long time. By dynamically adjusting the network structure, the RBF neural network inverse model can better adapt to complex and changing furnace conditions.

[0101] Example 5: This example differs from Example 1 in that it provides a calcium carbide furnace electrode current regulation system, including: The data acquisition module is used to collect the three-phase current data and the lifting and lowering speed data of the three-phase electrodes of the calcium carbide furnace. The state estimation module is used to estimate the unmeasurable state parameters of the electrode system online using an extended Kalman filter algorithm based on the three-phase current data and the rising and falling speed data of the three-phase electrodes. The unmeasurable state parameters include the furnace entry depth of the three-phase electrodes and the equivalent resistance of the three-phase material layer. The predictive control module is used to acquire the three-phase current setpoint. Based on the deviation between the three-phase current setpoint and the acquired three-phase current data, the three-phase coupling coefficient matrix obtained online, and the unmeasurable state parameters, and considering the three-phase coupling constraints and system time lag, the module performs rolling optimization in the prediction time domain through coupled model predictive control to generate predictive control quantities for the rise and fall speeds of the three-phase electrodes. The control output module is used to output the predicted control quantity of the lifting speed of the three-phase electrodes to the electrode lifting actuator to adjust the electrode position.

[0102] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for adjusting the electrode current of a calcium carbide furnace, characterized in that, include: Collect three-phase current data and three-phase electrode lifting and lowering speed data of the calcium carbide furnace; Based on the three-phase current data and the lifting and lowering speed data of the three-phase electrodes, the unmeasurable state parameters of the electrode system are estimated online using the extended Kalman filter algorithm. The unmeasurable state parameters include the furnace entry depth of the three-phase electrodes and the equivalent resistance of the three-phase material layer. The three-phase current setpoint is obtained. Based on the deviation between the three-phase current setpoint and the collected three-phase current data, the three-phase coupling coefficient matrix obtained online, and the unmeasurable state parameters, and considering the three-phase coupling constraints and system time lag, the three-phase electrode rise and fall speed is generated by rolling optimization in the prediction time domain through coupling model predictive control. The predicted control value of the three-phase electrode lifting speed is output to the electrode lifting actuator to adjust the electrode position.

2. The method for adjusting the electrode current of a calcium carbide furnace according to claim 1, characterized in that, The method further includes: obtaining the three-phase coupling coefficient matrix online by recursive least squares method with forgetting factor; The recursive least squares method with a forgetting factor performs the following update steps in each sampling period: Collect the changes in the positions of the three-phase electrodes and the changes in the three-phase current at the current moment; The prediction error at the current moment is calculated based on the changes in the positions of the three-phase electrodes, the changes in the three-phase current, and the coupling coefficient matrix at the previous moment. Calculate the gain vector at the current moment based on the covariance matrix of the previous moment, the change in the position of the three-phase electrodes, and the forgetting factor; The coupling coefficient matrix at the current time is updated based on the coupling coefficient matrix at the previous time step, the gain vector, and the prediction error. The covariance matrix at the current moment is updated based on the gain vector, the change in the position of the three-phase electrodes, and the covariance matrix at the previous moment.

3. The method for adjusting the electrode current of a calcium carbide furnace according to claim 2, characterized in that, The method further includes: calculating a decoupling compensation matrix based on the three-phase coupling coefficient matrix obtained through online identification; Calculate the current deviation between the three-phase current setpoint and the collected three-phase current data; Based on the decoupling compensation matrix and the current deviation, the three-phase electrode position adjustment amount is calculated, and the three-phase electrode position adjustment amount is superimposed as the decoupling compensation amount into the predictive control amount of the three-phase electrode rise and fall speed.

4. The method for adjusting the electrode current of a calcium carbide furnace according to claim 1, characterized in that, The method further includes: performing nonlinear dynamic compensation through the inverse model of the RBF neural network to obtain the neural network compensation amount for the rise and fall speed of the three-phase electrodes; The input of the RBF neural network inverse model is the three-phase current deviation and the current three-phase electrode position, and the output is the neural network compensation amount of the three-phase electrode rise and fall speed. The neural network compensation quantity is superimposed with the predicted control quantity of the three-phase electrode lifting and lowering speed and then output.

5. The method for adjusting the electrode current of a calcium carbide furnace according to claim 4, characterized in that, The construction of the RBF neural network inverse model includes: Collect historical operating data, which includes three-phase current, three-phase electrode position and corresponding electrode lifting speed; Using the three-phase current deviation and the three-phase electrode position as inputs, and the electrode lifting and lowering speed as outputs, an inverse RBF neural network model is constructed. The center vectors of neurons in the hidden layer of the RBF neural network inverse model are determined using the K-means clustering algorithm. The width parameter of the hidden layer neurons is determined based on the maximum distance between the center vectors of the hidden layer neurons and the number of hidden layer neurons. The weight matrix of the output layer of the inverse model of the RBF neural network is trained using the least squares method.

6. The method for adjusting the electrode current of a calcium carbide furnace according to claim 5, characterized in that, The method further includes: during operation, updating the weight matrix of the RBF neural network inverse model online through incremental learning; The incremental learning process performs the following steps in each sampling period: Calculate the output of the inverse RBF neural network model at the current time step based on the input vector at the current time step and the weight matrix at the previous time step. The output error is calculated based on the expected speed corresponding to the actual control effect and the output of the inverse model of the RBF neural network. The weight matrix is ​​updated based on the output error, the hidden layer output vector at the current time, and the learning rate.

7. The method for adjusting the electrode current of a calcium carbide furnace according to claim 1, characterized in that, The online estimation of unmeasurable state parameters of the electrode system using the extended Kalman filter algorithm includes: A state-space model is established with the three-phase electrode entry depth and the equivalent resistance of the three-phase material layer as state vectors, the three-phase electrode lifting and lowering speed as control input, and the three-phase current as observation output. In each sampling period, the prediction step of the extended Kalman filter is executed. Based on the state estimate of the unmeasurable state parameters of the previous time step, the control input, and the state transition function, the state prediction value and covariance prediction value of the current time step are calculated. Collect the current three-phase current data as the observation value; Calculate the Kalman gain based on the predicted covariance, the Jacobian matrix of the observation function, and the observation noise covariance matrix. Based on the predicted state value, the Kalman gain, the observed value, and the observation function, the estimated state value of the unmeasurable state parameters at the current moment is updated.

8. The method for adjusting the electrode current of a calcium carbide furnace according to claim 1, characterized in that, The method of performing rolling optimization in the prediction time domain through coupled model predictive control includes: Define the prediction time domain and the control time domain; Based on the current state estimate and the three-phase coupling coefficient matrix, a prediction model is established to predict the system state and current at future times. Define a reference trajectory for a smooth transition from the current measured value to the three-phase current setpoint; Construct an optimization objective function, which includes the sum of squared current tracking errors in the prediction time domain and the sum of squared control increments in the control time domain; At each sampling time, the optimization objective function is solved, and the optimal control sequence in the control time domain is obtained under the condition of satisfying the preset constraints. The first control variable in the optimal control sequence is output as the predicted control variable for the acceleration and deceleration of the three-phase electrodes.

9. The method for adjusting the electrode current of a calcium carbide furnace according to claim 8, characterized in that, The preset constraints include three-phase coupling constraints, electrode velocity constraints, and current safety constraints. The three-phase coupling constraint is that, within the prediction time domain, the relationship between the predicted current change and the predicted electrode position change is determined by the three-phase coupling coefficient matrix. The electrode speed constraint is that the output three-phase electrode rising and falling speeds are within a preset minimum and maximum range; The current safety constraint is that the predicted current value is within a preset minimum and maximum range.

10. A calcium carbide furnace electrode current adjustment system, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory is provided with a computer-readable storage medium, and a computer program is stored on the computer-readable storage medium. When the processor processes a computer program stored on the computer-readable storage medium, it implements the method as described in any one of claims 1-9.

Images