A model-free adaptive learning optimal control method and system for zinc electrolysis process

Abstract

The application relates to the technical field of industrial automation control, and discloses a model-free adaptive learning type optimal control method and system for a zinc electrolysis process, which comprises the following steps: obtaining a rectified current feedback value and calculating a discrete tracking error signal of a target rectified current set value; determining an adaptive correction step operator according to the change rate of the discrete tracking error signal, and decreasing the value of the adaptive correction step operator when the value is greater than a preset breaking threshold; updating a pseudo partial derivative estimation parameter by using the adaptive correction step operator; and solving a final control law command based on the updated pseudo partial derivative estimation parameter and the discrete tracking error signal and outputting the final control law command. When a transient impact working condition is encountered, the parameter correction rate is reduced to block the conduction damage of noise to parameter estimation, the phase lag and the numerical fault are eliminated, and the steady-state constraint of the control loop is maintained.

Description

Technical Field

[0001] This invention relates to a model-free adaptive learning-based optimization control method and system for zinc electrolysis processes, belonging to the field of industrial automation control technology. Background Technology

[0002] Current data-driven adaptive control methods are widely used in the nonparametric dynamic linearization design of controlled processes with large inertia, time-varying characteristics, and large time delays. The system utilizes input and output time-series data to estimate the dynamic pseudo-parameter matrix online, eliminating the dependence of the control loop on an exact algebraic model of the controlled object. This enables rapid convergence of tracking errors and steady-state control of the time-varying nonlinear system under smooth and continuous operating conditions. The stable operation of this data-driven linearization mechanism relies on the assumption of smooth data flow variation, meaning that the nonlinear time-varying disturbance terms of the controlled object's time-series data within the continuous discrete sampling time window satisfy a first-order differentiable or zero-mean bounded distribution. The dynamic parameter estimation matrix is ​​based on this characteristic. By smoothly correcting the control gain through time-series differential residual projection, a monotonic deterministic control correlation is established between the input change and the output response. However, when the controlled system encounters a large-scale discontinuous transient step disturbance, the potential transients caused by instantaneous local dendrite short circuits on the plates or sudden instantaneous voltage jumps in the slots can lead to non-Gaussian numerical discontinuities in the time-series data. The control loop is prone to misjudging the transient distortion discrete residual energy as time-varying drift of conventional parameters, forcing the pseudo-parameter update operator to follow high-frequency broadband noise. This results in phase lag and numerical discontinuities in the estimation matrix, ultimately causing saturation of the final control command output and spontaneous physical divergence of the power loop.

[0003] To address such sudden nonlinear fault impacts, linear attenuation methods such as artificially expanding the dead zone of the control loop or globally proportionally reducing the updated gain are commonly used. However, this reduces the tracking response speed under normal smooth operating conditions and cannot eliminate the system divergence problem caused by estimation mismatch under continuous large-scale step stress. Therefore, in view of the shortcomings of the existing technology, how to capture the sudden discontinuous jump trend through the multi-order algebraic relationship of the discrete residuals of input and output, perform adaptive gain yield suppression when encountering transient impact conditions to block the transmission and destruction of noise on the parameter estimation matrix, and coordinate to offset the phase delay of the return gap of the physical mechanism, has become the technical problem to be solved by the present invention. Summary of the Invention

[0004] To address the problems in the background art, the technical solution of the present invention is as follows: A model-free adaptive learning-based optimization control method for zinc electrolysis process, comprising the following steps: Step S101: Obtain the rectified current feedback value of the current conversion controlled unit, and calculate the discrete tracking error signal between the rectified current feedback value and the target rectified current set value; Step S102: Determine the adaptive correction step size operator based on the rate of change of the discrete tracking error signal; when the rate of change of the discrete tracking error signal is greater than the preset breaking threshold, use the attenuation relationship of multiplying the square inverse of the rate of change of the discrete tracking error signal with the basic control step size to adjust the adaptive correction step size operator to produce a numerical decrease in order to reduce the parameter correction rate. Step S103: Using the adaptive correction step size operator, correct the estimation weights of the pseudo-partial derivative estimation parameters at the current sampling time, and update the pseudo-partial derivative estimation parameters at the current sampling time. Step S104: Based on the updated pseudo-partial derivative estimation parameters and discrete tracking error signal, the objective function containing the tracking error deviation term and the penalty term for the change of control input independent variable is minimized as the optimization rule, and the transmission gap time delay compensation caused by the transmission resistance torque is superimposed to solve the final control law command at the current sampling time. The final control law command is output to the current conversion controlled unit to change the rectified current feedback value.

[0005] Preferably, between steps S101 and S102, the following anti-divergence stability control sub-steps are included by utilizing the dynamic response of the discrete tracking error signal: Step S1011, the sum of squares of the discrete tracking error signal within a set period is continuously calculated using a sliding window accumulator to obtain the time-domain error accumulation parameter; Step S1012, when the rate of change of the time-domain error accumulation parameter within three consecutive sampling periods is greater than the set drift threshold, a gain attenuation factor is added to the control loop to limit the rate of change of the amplitude of the final control law command.

[0006] Preferably, step S102 specifically includes the following sub-steps: Step S1021, calculate the algebraic difference between the discrete tracking error signal at the current sampling time and the historical sampling time to obtain the first-order discrete difference component and the second-order discrete difference component of the discrete tracking error signal; Step S1022, calculate the ratio of the absolute values ​​of the first-order discrete difference component and the second-order discrete difference component to obtain the rate of change of the discrete tracking error signal; Step S1023, when the rate of change of the discrete tracking error signal is greater than a preset breaking threshold, calculate the square inverse of the rate of change of the discrete tracking error signal, and multiply the square inverse by a preset basic control step size to generate an adaptive correction step size operator.

[0007] Preferably, the calculation of the transmission gap delay compensation and the solution of the final control law command in step S104 includes the following sub-steps: Step S1041, based on the algebraic difference component of the final control law command at the previous sampling time and the online measured system response delay parameter, calculate the transmission gap deviation estimate characterizing the transmission loss; Step S1042, use the sign function to reverse the sign of the transmission gap deviation estimate to generate the transmission gap delay compensation; Step S1043, superimpose the feedback control adjustment quantity solved by the objective function onto the control quantity reference at the previous sampling time, and superimpose the transmission gap delay compensation to generate the final control law command at the current sampling time.

[0008] Preferably, after step S104, the following parallel main steps are also included to implement amplitude limiting protection for the final control law command: step S105, the final control law command is introduced into the amplitude limiting limiter; step S106, when the final control law command exceeds the set industrial safety voltage boundary range, the final control law command is truncated by the amplitude limiting limiter, and the amplitude limiting rectification control command within the safe amplitude range is output.

[0009] Preferably, in step S1043, the feedback control adjustment quantity obtained by solving the objective function increases monotonically with the product term of the pseudo-partial derivative estimation parameter at the current sampling time and the discrete tracking error signal at the current sampling time, and decreases monotonically with the increase of the penalty term for the change of the control input independent variable.

[0010] Preferably, after step S106, the following parallel main steps are included to implement long-term operating state evolution control: Step S107, accumulating the rectified current feedback value and system operating temperature data of the controlled unit in the historical operating cycle; Step S108, calculating the system drift index characterizing the performance degradation of the rectifier equipment based on the accumulated rectified current feedback value and system operating temperature data; Step S109, when the system drift index is greater than the set degradation threshold, outputting an equipment abnormality warning signal, and increasing the penalty term for the change of the control input independent variable in the objective function to limit the adjustment range of the final control law command.

[0011] Preferably, the sampling period for obtaining the rectified current feedback value in step S101 is 1ms to 10ms, and the range of the preset fracture threshold in step S102 is 1.5 to 3.0.

[0012] Preferably, the system response delay parameter in step S1041 is obtained by inputting a pseudo-random binary disturbance signal and detecting the dynamic phase response of the rectified current feedback value when the rectified current feedback value is stable within a preset range.

[0013] A system for implementing a model-free adaptive learning-based optimization control method for zinc electrolysis processes, comprising: The signal acquisition unit is used to acquire the rectified current feedback value of the current-converting controlled unit and calculate the discrete tracking error signal between the rectified current feedback value and the target rectified current set value. The signal acquisition unit also includes a sliding window accumulator, which is used to continuously calculate the sum of squares of the discrete tracking error signal within a set period to obtain the time-domain error accumulation parameter. When the rate of change of the time-domain error accumulation parameter is greater than the set drift threshold within three consecutive sampling periods, a gain attenuation factor is added to the control loop to limit the rate of change of the amplitude of the final control law command. The step size determination unit is used to determine the adaptive correction step size operator based on the rate of change of the discrete tracking error signal. When the rate of change of the discrete tracking error signal is greater than the preset breakage threshold, the adaptive correction step size operator is adjusted to produce a numerical decrease by multiplying the square inverse of the rate of change of the discrete tracking error signal by the preset basic control step size to reduce the parameter correction rate. The parameter update unit is used to use the adaptive correction step size operator to correct the estimation weights of the pseudo-partial derivative estimation parameters at the current sampling time and update the pseudo-partial derivative estimation parameters at the current sampling time. The control solving unit is used to estimate parameters based on updated pseudo-partial derivatives and discrete tracking error signals, using the objective function that minimizes the tracking error deviation term and the penalty term for changes in the control input independent variable as the optimization rule, and superimposing the transmission backlash time delay compensation caused by the transmission resistance torque, to solve for the final control law command at the current sampling moment, and output the final control law command to the current-converting controlled unit to change the rectified current feedback value; it is also used to input the final control law command into the amplitude limiting limiter, and cut off the final control law command when it exceeds the set industrial safety voltage boundary range, outputting the limited rectification control command within the safe amplitude range; it is also used to accumulate the rectified current feedback value and system operating temperature data of the current-converting controlled unit in the historical operating cycle, calculate the system drift index characterizing the performance degradation of the rectifier equipment based on the accumulated rectified current feedback value and system operating temperature data, and output the equipment abnormality warning signal and increase the penalty term for changes in the control input independent variable in the objective function when the system drift index is greater than the set degradation threshold.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. In the model-free adaptive learning optimization control of zinc electrolysis, by tracking the multi-order discrete difference algebraic proportional relationship of the error signal, the nonlinear jump trend caused by potential changes and mass transfer resistance step changes inside the controlled process is dynamically captured. When the data stream is discontinuous, the square inverse mapping is used to suppress the gain, so that the variable step size update operator produces a step-like instantaneous drop, and the online learning rate is actively compressed to the resting state. In this way, the inherent physical inertia of the controlled object is relied upon to overcome the disturbance impact period, cut off the transmission path of noise to the pseudo-partial derivative estimation matrix, eliminate the phase lag and numerical discontinuity phenomenon in conventional feedback correction, and maintain the steady-state constraint of the control loop.

[0015] 2. By continuously monitoring the sum of squared tracking errors of the cumulative sampling period using the error pseudo-energy integrator, when the algebraic slope of the integrator exceeds the set critical threshold in multiple consecutive sampling periods, it is determined that the system has a divergence trend. Subsequently, the control gain yield adjustment is triggered. By introducing an exponential decay suppression factor into the output gain, the output saturation of the control command is forcibly suppressed, thus gaining a physical response window for the re-convergence of the estimated parameters and avoiding the risk of physical divergence and physical burnout of the control loop due to sudden transient shocks.

[0016] 3. To address the inherent electromagnetic response delay and transmission gap of the power actuator, the logic rack bounce, which characterizes physical losses, is calculated using the differential component of the control command and the delay parameters obtained from online measurement. During the optimal control law generation stage, the logic rack bounce is used as a feedforward compensation term and superimposed into the final control law command through a sign function to offset the phase disruption caused by transmission gap and physical lag, and to suppress the low-frequency oscillations accompanying the closed-loop control circuit. Attached Figure Description

[0017] Figure 1 This is a flowchart of the control steps of a model-free adaptive learning-based optimization control method for zinc electrolysis process according to the present invention. Figure 2 This is a state diagram of a model-free adaptive learning optimization control system for a zinc electrolysis process according to the present invention.

[0018] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0020] This application provides a model-free adaptive learning-based optimization control method and system for zinc electrolysis, including the following steps: Step S101: Obtain the rectified current feedback value of the current conversion controlled unit, and calculate the discrete tracking error signal between the rectified current feedback value and the target rectified current set value; Step S102: Determine the adaptive correction step size operator based on the rate of change of the discrete tracking error signal; when the rate of change of the discrete tracking error signal is greater than the preset breaking threshold, use the attenuation relationship of multiplying the square inverse of the rate of change of the discrete tracking error signal with the basic control step size to adjust the adaptive correction step size operator to produce a numerical decrease in order to reduce the parameter correction rate. Step S103: Using the adaptive correction step size operator, correct the estimation weights of the pseudo-partial derivative estimation parameters at the current sampling time, and update the pseudo-partial derivative estimation parameters at the current sampling time. Step S104: Based on the updated pseudo-partial derivative estimation parameters and discrete tracking error signal, the objective function containing the tracking error deviation term and the penalty term for the change of control input independent variable is minimized as the optimization rule, and the transmission gap time delay compensation caused by the transmission resistance torque is superimposed to solve the final control law command at the current sampling time. The final control law command is output to the current conversion controlled unit to change the rectified current feedback value.

[0021] Preferably, between steps S101 and S102, the following anti-divergence stability control sub-steps are included by utilizing the dynamic response of the discrete tracking error signal: Step S1011, the sum of squares of the discrete tracking error signal within a set period is continuously calculated using a sliding window accumulator to obtain the time-domain error accumulation parameter; Step S1012, when the rate of change of the time-domain error accumulation parameter within three consecutive sampling periods is greater than the set drift threshold, a gain attenuation factor is added to the control loop to limit the rate of change of the amplitude of the final control law command.

[0022] Preferably, step S102 specifically includes the following sub-steps: Step S1021, calculate the algebraic difference between the discrete tracking error signal at the current sampling time and the historical sampling time to obtain the first-order discrete difference component and the second-order discrete difference component of the discrete tracking error signal; Step S1022, calculate the ratio of the absolute values ​​of the first-order discrete difference component and the second-order discrete difference component to obtain the rate of change of the discrete tracking error signal; Step S1023, when the rate of change of the discrete tracking error signal is greater than a preset breaking threshold, calculate the square inverse of the rate of change of the discrete tracking error signal, and multiply the square inverse by a preset basic control step size to generate an adaptive correction step size operator.

[0023] Preferably, the calculation of the transmission gap delay compensation and the solution of the final control law command in step S104 includes the following sub-steps: Step S1041, based on the algebraic difference component of the final control law command at the previous sampling time and the online measured system response delay parameter, calculate the transmission gap deviation estimate characterizing the transmission loss; Step S1042, use the sign function to reverse the sign of the transmission gap deviation estimate to generate the transmission gap delay compensation; Step S1043, superimpose the feedback control adjustment quantity solved by the objective function onto the control quantity reference at the previous sampling time, and superimpose the transmission gap delay compensation to generate the final control law command at the current sampling time.

[0024] Preferably, after step S104, the following parallel main steps are also included to implement amplitude limiting protection for the final control law command: step S105, the final control law command is introduced into the amplitude limiting limiter; step S106, when the final control law command exceeds the set industrial safety voltage boundary range, the final control law command is truncated by the amplitude limiting limiter, and the amplitude limiting rectification control command within the safe amplitude range is output.

[0025] Preferably, in step S1043, the feedback control adjustment quantity obtained by solving the objective function increases monotonically with the product term of the pseudo-partial derivative estimation parameter at the current sampling time and the discrete tracking error signal at the current sampling time, and decreases monotonically with the increase of the penalty term for the change of the control input independent variable.

[0026] Preferably, after step S106, the following parallel main steps are included to implement long-term operating state evolution control: Step S107, accumulating the rectified current feedback value and system operating temperature data of the controlled unit in the historical operating cycle; Step S108, calculating the system drift index characterizing the performance degradation of the rectifier equipment based on the accumulated rectified current feedback value and system operating temperature data; Step S109, when the system drift index is greater than the set degradation threshold, outputting an equipment abnormality warning signal, and increasing the penalty term for the change of the control input independent variable in the objective function to limit the adjustment range of the final control law command.

[0027] Preferably, the sampling period for obtaining the rectified current feedback value in step S101 is 1ms to 10ms, and the range of the preset fracture threshold in step S102 is 1.5 to 3.0.

[0028] Preferably, the system response delay parameter in step S1041 is obtained by inputting a pseudo-random binary disturbance signal and detecting the dynamic phase response of the rectified current feedback value when the rectified current feedback value is stable within a preset range.

[0029] A system for implementing a model-free adaptive learning-based optimization control method for zinc electrolysis processes, comprising: The signal acquisition unit is used to acquire the rectified current feedback value of the current-converting controlled unit and calculate the discrete tracking error signal between the rectified current feedback value and the target rectified current set value. The signal acquisition unit also includes a sliding window accumulator, which is used to continuously calculate the sum of squares of the discrete tracking error signal within a set period to obtain the time-domain error accumulation parameter. When the rate of change of the time-domain error accumulation parameter is greater than the set drift threshold within three consecutive sampling periods, a gain attenuation factor is added to the control loop to limit the rate of change of the amplitude of the final control law command. The step size determination unit is used to determine the adaptive correction step size operator based on the rate of change of the discrete tracking error signal. When the rate of change of the discrete tracking error signal is greater than the preset breakage threshold, the adaptive correction step size operator is adjusted to produce a numerical decrease by multiplying the square inverse of the rate of change of the discrete tracking error signal by the preset basic control step size to reduce the parameter correction rate. The parameter update unit is used to use the adaptive correction step size operator to correct the estimation weights of the pseudo-partial derivative estimation parameters at the current sampling time and update the pseudo-partial derivative estimation parameters at the current sampling time. The control solving unit is used to estimate parameters based on updated pseudo-partial derivatives and discrete tracking error signals, using the objective function that minimizes the tracking error deviation term and the penalty term for changes in the control input independent variable as the optimization rule, and superimposing the transmission backlash time delay compensation caused by the transmission resistance torque, to solve for the final control law command at the current sampling moment, and output the final control law command to the current-converting controlled unit to change the rectified current feedback value; it is also used to input the final control law command into the amplitude limiting limiter, and cut off the final control law command when it exceeds the set industrial safety voltage boundary range, outputting the limited rectification control command within the safe amplitude range; it is also used to accumulate the rectified current feedback value and system operating temperature data of the current-converting controlled unit in the historical operating cycle, calculate the system drift index characterizing the performance degradation of the rectifier equipment based on the accumulated rectified current feedback value and system operating temperature data, and output the equipment abnormality warning signal and increase the penalty term for changes in the control input independent variable in the objective function when the system drift index is greater than the set degradation threshold.

[0030] Example 1: A model-free adaptive learning optimization control method for zinc electrolysis operates in a zinc electrolysis cell rectifier power supply system. The system includes a rectifier unit, a current sensor, and a control processing module. The control processing module collects the DC current feedback value flowing through the electrolysis cell with a sampling period of 10ms, and subtracts it from the target rectified current setpoint to generate the tracking error signal at the current discrete moment. The system performs a differential operation on the tracking error signal, calculates the numerical algebraic difference between the tracking error signal at the current discrete moment and the tracking error signal at the previous sampling moment, and obtains the first-order discrete difference component. Further, a second differential operation is performed on the tracking error signal to obtain the second-order discrete difference component.

[0031] The system calculates the ratio of the absolute value of the second-order discrete difference component to the absolute value of the first-order discrete difference component to obtain the dynamic rate of change of the tracking error signal at the current moment. When the dynamic rate of change is less than or equal to a preset fracture threshold of 2.0, the system maintains the basic control step size and controls the update rate of the pseudo-partial derivatives by the basic learning rate coefficient. When the dynamic rate of change is greater than the preset fracture threshold of 2.0, the system constructs a variable step size update operator by multiplying the nonlinear adjustment factor by the inverse of the square of the dynamic rate of change. This operator decays inversely with the square of the increase in the dynamic rate of change, causing the online correction rate of the system parameters to slip instantly, suppressing the destruction of the pseudo-partial derivative estimation matrix by transient impact noise caused by short circuit of the electrode dendrite.

[0032] Based on the variable step-size update operator, the system continuously corrects the pseudo-partial derivative estimation parameters under the full-format dynamic linearization framework. The updated pseudo-partial derivative estimation parameters and the tracking error signal are then substituted into the control input criterion objective function for optimization. This objective function includes a tracking error deviation term and a penalty term for changes in the control input independent variable. Specifically, the objective function is constructed using a weighted summation, where the tracking error deviation term is the square of the predicted tracking error signal at the next time step, and the penalty term for changes in the control input independent variable is the square of the difference between the control command at the current time step and the control command at the previous time step multiplied by a preset non-negative weight coefficient. The optimization rule is to calculate the partial derivative of the objective function with respect to the control command at the current time step and set the derivative to zero. At each sampling time step, the previous sampling time step is used... The changes in control input, system output feedback, and the current adaptive correction step size operator at each sampling time are used to iteratively correct the current pseudo-partial derivative estimation parameters using a projection algorithm. Specifically, the pseudo-partial derivative estimation value from the previous time step is added to the projection component of the product of the step size operator and the input change in the residual direction. The updated pseudo-partial derivative estimation parameters, the current discrete tracking error signal, and the non-negative weighting coefficients are then substituted into the algebraic control law calculated from the derivative equation to directly calculate the feedback control adjustment. This adjustment is equal to the product of the pseudo-partial derivative estimation parameters, the step size factor, and the current error signal, divided by the sum of the square of the pseudo-partial derivative estimation parameters and the non-negative weighting coefficients. This process yields the control quantity at the current sampling time. In addition to the baseline, the system calculates the logic rack bounce as a feedforward compensation term. This logic rack bounce is obtained by multiplying the 12ms system response delay parameter measured online by the system and the algebraic change of the current control command relative to the previous moment. The sign is then changed using a sign function and inversely superimposed onto the final control law command to offset the phase delay caused by the transmission gap of the physical actuator. In actual industrial settings, the aforementioned transmission resistance torque, transmission gap, and logic rack bounce do not represent the physical entities of the overall mechanical transmission components, but rather serve as equivalent simulation models to characterize the large inductive hysteresis effect of the thyristor-controlled rectifier bridge during trigger angle adjustment, the electromagnetic dead time during bidirectional switching, and the rectifier transformer under high load impact. The magnetic circuit saturation nonlinear transition, specifically, due to the large inductance and inertia of the rectifier circuit, when the trigger angle undergoes a step adjustment, the response of the control current exhibits a phase lag blind zone equivalent to the mechanical return gap. This invention abstracts the time delay and dead zone characteristics of this electromagnetic response as a transmission gap caused by the transmission resistance torque, and uses a logic rack bounce model to calculate the nonlinear phase deviation of the trigger pulse caused by magnetic circuit saturation. This equivalent deviation is then reversed using a sign function and superimposed onto the control law command. Essentially, this implements electromagnetic dead zone offset compensation with a leading phase in the calculation of the thyristor trigger pulse conduction angle, thereby eliminating low-frequency electrical oscillations caused by hysteresis and dead zone within the pure power electronic control architecture.The control processing module outputs the final control law command to the rectifier power actuator to achieve current control. When this command exceeds the industrial safety physical boundary range of 1.2V to 4.8V, the system triggers a rigid cutoff, locking the output at the boundary extreme value to ensure production safety.

[0033] Example 2: This experiment aims to verify the dynamic tracking performance and anti-disturbance capability of a model-free adaptive learning optimization control method for zinc electrolysis under different electrolysis conditions. The experimental platform consists of a series unit containing 12 electrolytic cells of the same specifications. The rectifier unit adopts a thyristor-controlled rectifier bridge, and the current feedback sampling frequency is set to 100Hz. The control processing module runs on a real-time control platform to verify the parameter convergence characteristics of the variable step size update operator in an industrial noise environment. The sample group of this invention adopts the model-free adaptive learning optimization control method, while the control group adopts the traditional fixed gain proportional-integral-derivative control algorithm. The first stage of the experiment verifies the system's suppression effectiveness against nonlinear fault impacts caused by sudden changes in operating conditions. During normal system operation, a step-like change in electrode connection state is created by manually switching the electrode connection state. In the short-circuit disturbance test, the test records show that when the rectified current output is affected by the transient impact caused by the dendrite short circuit, the discrete tracking error signal monitored by the control processing module generates an instantaneous oscillation with an amplitude of 250A. According to the dynamic change rate calculation logic of the tracking error signal, after the system detects that the dynamic change rate exceeds the 2.0 threshold, it immediately starts the variable step size update operator adjustment logic, reducing the parameter correction gain from 0.5 to 0.15. The control group uses the traditional fixed gain algorithm, which cannot adjust the update rate under transient impact, causing the pseudo-partial derivative estimation matrix to diverge, and the current feedback control loop experiences an oscillation process lasting 1.2s. The sample of this invention, due to the immediate triggering of gain yield damping, limits the oscillation amplitude of the pseudo-partial derivative estimation to within 15A, and the tracking error signal recovers to stability within 0.2s.

[0034] The second phase of the experiment verified the system's compensation effectiveness for transmission lag in the mechanical actuator. The system tested the response characteristics of the control loop to a step change in the setpoint by changing the rectified voltage regulation command. The original rectified voltage command switched between 280V and 320V. The test data showed that without introducing the logic rack bounce as a feedforward compensation term, the current output had a phase lag relative to the command due to the influence of transmission gap and physical delay of magnetic circuit response. The measured lag time was 35ms. The system started logic rack bounce feedforward compensation. Based on the 12ms system response delay parameter observed online, the rectified control law command was calculated and corrected in advance. The phase lag time of the current output was reduced to 6ms. This data proves that the logic rack bounce compensation path effectively offsets the physical gap delay of the transmission system by introducing a reverse correction operator to pre-shift the phase of the control law command in the adjustment direction.

[0035] The third stage of the experiment defined the applicable performance boundaries and conducted gradient tests on the sensitivity of the preset fracture threshold. The preset fracture threshold was set to three levels: 0.5, 2.0, and 4.0. When the threshold was set to 0.5, the system frequently triggered parameter degradation under normal small fluctuation conditions, resulting in a decrease in current steady-state accuracy and an average tracking error of 5A. When the threshold was set to 4.0, the system triggered a response delay when encountering a short-circuit impact, and could not effectively suppress pseudo-partial derivative oscillations, leading to instability in the control loop. Within the value range of 2.0, the system maintained an average tracking error of less than 1.5A under normal operating conditions while achieving a millisecond-level response to short-circuit impacts. The gradient test data proved that the value setting of 2.0 is the optimal engineering window for balancing the suppression of normal fluctuations and the damping of abnormal sudden changes in the electrolysis process. Exceeding this window limit will lead to the deterioration of steady-state control performance or the failure of the ability to suppress sudden disturbances. The test results show that under disturbance input, the control method eliminates numerical discontinuities in the control command through the synergistic effect of the variable step size update operator and the feedforward compensation operator, and maintains the steady-state constraint of the electrolysis current loop.

[0036] Example 3: This example combines Figures 1 to 2 This paper describes a model-free adaptive learning-based optimal control method and system for a zinc electrolysis process, such as... Figure 1 As shown, the control sequence of a model-free adaptive learning optimization control method for zinc electrolysis consists of four cascaded execution links. Initially, step S101 obtains the rectified current feedback value of the current-converting controlled unit and calculates the discrete tracking error signal between the rectified current feedback value and the target rectified current setpoint. The process proceeds sequentially to step S102, where, based on the rate of change of the discrete tracking error signal, an adaptive correction step size operator is determined. When the rate of change exceeds a preset breaking threshold, the attenuation relationship between the reciprocal of the square of the rate of change and the basic control step size is used to adjust the control... The operator value is decreased to reduce the parameter correction rate. After the step size adjustment is completed, the logic proceeds to step S103. The adaptive correction step size operator is used to correct the estimated weights of the pseudo-partial derivative estimation parameters at the current sampling time, and the pseudo-partial derivative estimation parameters at the current sampling time are updated. Finally, the parameters are collected into the pseudo-partial derivative estimation parameters and discrete tracking error signal updated in step S104. The objective function containing the tracking error deviation term and the penalty term is minimized as the optimization rule. The transmission gap time delay compensation is superimposed to solve the final control law command, and the output is used to change the rectified current feedback value.

[0037] like Figure 2As shown, the control flow logic of this invention uses "maintaining the basic control step size for smooth correction" as the normal switching benchmark, and triggers corresponding adjustment branches according to different real-time monitoring indicators. When the rate of change of the monitoring indicator is greater than the preset breaking threshold, the system unidirectionally flows to the node where the adaptive correction step size operator value decreases. When the system state evaluation reaches a rate of change less than or equal to the preset breaking threshold, it returns from this operating state to the node where the basic control step size is maintained for smooth correction. At the same time, if the rate of change of the time domain error accumulation parameter is greater than the set drift threshold, the system switches to the amplitude limiting control state, and a gain attenuation factor is introduced into the control loop to limit the rate of change of the amplitude of the final control law command. After the adjustment is completed... After the estimated parameters converge and the loop stabilizes, the closed loop switches back to the node for maintaining the basic control step size for smooth correction. Similarly, when the system drift index is determined to be greater than the set attenuation threshold, the system triggers an output warning and increases the penalty state for changes in independent variables. In addition, for external and hardware loss factors, once periodic triggering and single-slot impedance deviation are detected, the system jumps to the trigger self-reconfiguration logic to correct the impedance reference module until the incremental translation correction and reference update are completed, and then merges into the node for maintaining the basic control step size for smooth correction. If the electrolytic plate is replaced, the system directly restores to the basic operating state of maintaining the basic control step size for smooth correction after executing the initial calibration procedure.

[0038] Example 4: In the zinc electrolysis cell production workshop, when the rectifier power supply system is connected to the power grid via a transformer, fluctuations in the power grid voltage can cause high-frequency electromagnetic interference to superimpose on the rectified current feedback signal, directly affecting the convergence of the pseudo-partial derivative estimation matrix. This can trigger large-amplitude oscillations in the current regulation loop. The system's preset offline data preprocessing procedure uses a digital filtering algorithm by the control processing module to reduce noise in the original current feedback value. The sampling frequency is set to 100Hz. The system introduces a moving average filter with 5 sampling points to smooth out high-frequency electromagnetic interference noise. At the same time, the system sets a noise discrimination threshold based on the absolute value of the second-order discrete difference component. When the transient noise amplitude is detected to exceed 3% of the preset rated current feedback value, the system automatically performs data removal and interpolation smoothing to prevent distorted data points from entering the iterative calculation process of the pseudo-partial derivative estimation matrix.

[0039] During the industrial deployment phase, the control processing module invokes a preset parameter initial calibration procedure to perform initial benchmark calibration on the discrete linearized model, which includes the dynamic characteristics of the rectifier unit and the impedance characteristics of the electrolytic cell. The control processing module adjusts the rectified voltage setpoint with discrete step size and records the rectified current feedback value and voltage command response gait under 50%, 75%, and 100% load conditions, respectively. The system uses the recorded response gait data to calculate the system open-loop gain coefficient and response delay parameters. After completing the parameter measurement at each load point, the system establishes a mapping function between different load conditions and pseudo-partial derivative estimation parameters through linear interpolation logic and stores it in the storage space of the control processing module as a necessary initialization operation. This calibration process is performed before the rectifier power supply system is put into operation for the first time or after the electrolytic plates are replaced to ensure that the initial value of the pseudo-partial derivative obtained by the control processing module has physical homogeneity with the actual operating conditions of the controlled object, providing a stable and accurate starting point for subsequent model-free adaptive learning control.

[0040] Example 5: In a continuously operating zinc electrolysis production line, the system faces the problem of uneven current density distribution caused by electrode plate wear. In order to achieve long-term steady-state operation, the control processing module calls the preset electrode state calibration procedure. In the initial stage after the electrolysis electrode plate is replaced, the system records multiple sets of voltage and current static characteristic curves under different loads by measuring the correspondence between the voltage values ​​of each shunt circuit output by the rectifier unit and the current feedback values. The system uses least squares fitting logic to determine the unit impedance reference value of each electrolytic cell and writes this reference value into the storage unit of the control processing module as the original reference data for subsequent evaluation of the current density distribution of each electrolytic cell.

[0041] During continuous operation of the electrolysis process, the system automatically triggers a periodic reference drift correction process every 24 hours. The control processing module reads the real-time current shunt signal of each electrolytic cell and calculates it against the unit impedance reference value to determine whether there is an abnormal state where the impedance of a single cell deviates by more than 5%. In order to quantitatively assess the overall attenuation of the rectifier equipment, the control processing module simultaneously calculates the system drift index in this process. Specifically, the system continuously accumulates and reads the rectified current feedback value and system operating temperature data of the aforementioned current conversion controlled units in the historical operating cycle, and performs an arithmetic mean calculation on the current discrete points and temperature discrete points collected in the past 24 hours over the time dimension to obtain... The average current and average temperature values ​​of the current cycle serve as the basic input data for subsequent index calculations. When a deviation in the impedance of a certain tank is detected, and the degree of deviation increases linearly over time, it is determined to be long-term irreversible loss of the electrode plate. The system automatically triggers the parameter self-reconfiguration logic. Specifically, the parameter self-reconfiguration logic is implemented by calling the impedance correction thread through a software interrupt. In incremental translation correction, the control processing module calculates the absolute difference between the current measured average impedance and the originally set unit impedance reference value. This absolute difference is used as the translation correction increment. The original unit impedance reference value is added to this translation correction increment, thereby completing the overall step-like translation and raising of the reference value through algebraic addition. This ensures that the underlying optimization reference system of the control algorithm matches the physical impedance drift caused by electrode losses. Based on the current impedance deviation, the control processing module incrementally shifts and corrects the unit impedance reference value in the storage unit, forming a dynamically updated reference suitable for the current electrode loss state. This updated reference replaces the original reference in subsequent iterative optimization calculations of the pseudo-partial derivative matrix. By introducing this self-repairing logic, the system limits the target current distribution deviation caused by electrode losses to within 0.5%, ensuring that the zinc electrolysis process parameters remain within the target deviation range throughout the entire lifespan of the electrode. To eliminate the conflict between the different physical dimensions of current and temperature and establish a deterministic mapping relationship, in In the quantification and fusion process of the aforementioned system drift index, the control processing module adopts a dimensionless transformation interface; the average current value is divided by the rated maximum output current, which is mapped to a dimensionless current degradation coefficient between 0 and 1, and the average temperature value is divided by the rectifier bridge safety limit temperature, which is mapped to a dimensionless temperature overheating coefficient between 0 and 1; the system sets a linear weight allocation rule, multiplying the dimensionless current degradation coefficient by a weight of 0.6 and the dimensionless temperature overheating coefficient by a weight of 0.4, and then summing the two products algebraically, thereby transforming the multidimensional heterogeneous characteristics into a single-dimensional system drift index scalar value, which is used to perform subsequent comparison and judgment with the scalar value of the set attenuation threshold.

[0042] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0043] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A model-free adaptive learning-based optimization control method for zinc electrolysis, characterized in that, Includes the following steps: Step S101: Obtain the rectified current feedback value of the current conversion controlled unit, and calculate the discrete tracking error signal between the rectified current feedback value and the target rectified current set value; Step S102: Determine the adaptive correction step size operator based on the rate of change of the discrete tracking error signal; When the rate of change of the discrete tracking error signal is greater than the preset fracture threshold, the attenuation relationship between the square inverse of the rate of change of the discrete tracking error signal and the basic control step size is used to adjust the adaptive correction step size operator to produce a numerical decrease in order to reduce the parameter correction rate. Step S103: Using the adaptive correction step size operator, correct the estimation weights of the pseudo-partial derivative estimation parameters at the current sampling time, and update the pseudo-partial derivative estimation parameters at the current sampling time. Step S104: Based on the updated pseudo-partial derivative estimation parameters and discrete tracking error signal, the objective function containing the tracking error deviation term and the penalty term for the change of control input independent variable is minimized as the optimization rule, and the transmission gap time delay compensation caused by the transmission resistance torque is superimposed to solve the final control law command at the current sampling time. The final control law command is output to the current conversion controlled unit to change the rectified current feedback value.

2. The model-free adaptive learning-based optimization control method for zinc electrolysis process according to claim 1, characterized in that, Between steps S101 and S102, the following anti-divergence stability control sub-steps are included, based on the dynamic response of the discrete tracking error signal: Step S1011, the sum of squares of the discrete tracking error signal within a set period is continuously calculated using a sliding window accumulator to obtain the time-domain error accumulation parameter; Step S1012, when the rate of change of the time-domain error accumulation parameter within three consecutive sampling periods is greater than the set drift threshold, a gain attenuation factor is added to the control loop to limit the rate of change of the amplitude of the final control law command.

3. The model-free adaptive learning-based optimization control method for zinc electrolysis process according to claim 1, characterized in that, Step S102 specifically includes the following sub-steps: Step S1021, calculate the algebraic difference between the discrete tracking error signal at the current sampling time and the historical sampling time to obtain the first-order discrete difference component and the second-order discrete difference component of the discrete tracking error signal; Step S1022, calculate the absolute value ratio of the first-order discrete difference component and the second-order discrete difference component to obtain the rate of change of the discrete tracking error signal; Step S1023, when the rate of change of the discrete tracking error signal is greater than the preset breakage threshold, calculate the square reciprocal of the rate of change of the discrete tracking error signal, and multiply the square reciprocal by the preset basic control step size to generate an adaptive correction step size operator.

4. The model-free adaptive learning-based optimization control method for zinc electrolysis process according to claim 1, characterized in that, Step S104, which calculates the transmission backlash delay compensation and solves for the final control law command, includes the following sub-steps: Step S1041, based on the algebraic difference component of the final control law command at the previous sampling time and the online measured system response delay parameter, calculate the transmission backlash deviation estimate characterizing the transmission loss; Step S1042, use a sign function to reverse the sign of the transmission backlash deviation estimate to generate the transmission backlash delay compensation; Step S1043, superimpose the feedback control adjustment quantity solved by the objective function onto the control quantity reference at the previous sampling time, and superimpose the transmission backlash delay compensation to generate the final control law command at the current sampling time.

5. The model-free adaptive learning-based optimization control method for zinc electrolysis process according to claim 1, characterized in that, Following step S104, the following parallel main steps are also included to implement amplitude limiting protection for the final control law command: Step S105, the final control law command is fed into the amplitude limiting limiter; Step S106, when the final control law command exceeds the set industrial safety voltage boundary range, the final control law command is truncated by the amplitude limiting limiter, and the amplitude limiting rectification control command within the safe amplitude range is output.

6. The model-free adaptive learning-based optimization control method for zinc electrolysis process according to claim 4, characterized in that, In step S1043, the feedback control adjustment quantity obtained by solving the objective function increases monotonically with the product term of the pseudo-partial derivative estimation parameter at the current sampling time and the discrete tracking error signal at the current sampling time, and decreases monotonically with the increase of the penalty term for the change of the control input independent variable.

7. The model-free adaptive learning-based optimization control method for zinc electrolysis process according to claim 1, characterized in that, Following step S106, the following parallel main steps are also included to implement long-term operating state evolution control: Step S107, accumulating the rectified current feedback value and system operating temperature data of the controlled unit in the historical operating cycle; Step S108, calculating the system drift index characterizing the performance degradation of the rectifier equipment based on the accumulated rectified current feedback value and system operating temperature data; Step S109, when the system drift index is greater than the set degradation threshold, outputting an equipment abnormality warning signal and increasing the penalty term for the change of the control input independent variable in the objective function to limit the adjustment range of the final control law command.

8. The model-free adaptive learning-based optimization control method for zinc electrolysis process according to claim 1, characterized in that, In step S101, the sampling period for obtaining the rectified current feedback value is 1ms to 10ms, and in step S102, the preset fracture threshold value ranges from 1.5 to 3.

0.

9. The model-free adaptive learning-based optimization control method for zinc electrolysis process according to claim 4, characterized in that, In step S1041, the system response delay parameter is obtained by inputting a pseudo-random binary disturbance signal and detecting the dynamic phase response of the rectified current feedback value when the rectified current feedback value is stable within a preset range.

10. A system for implementing the model-free adaptive learning-based optimization control method for the zinc electrolysis process as described in claim 1, characterized in that, include: The signal acquisition unit is used to acquire the rectified current feedback value of the current conversion controlled unit and calculate the discrete tracking error signal between the rectified current feedback value and the target rectified current set value. The signal acquisition unit also includes a sliding window accumulator, which is used to continuously calculate the sum of squares of discrete tracking error signals within a set period to obtain the time-domain error accumulation parameter. When the rate of change of the time-domain error accumulation parameter is greater than the set drift threshold within three consecutive sampling periods, a gain attenuation factor is added to the control loop to limit the rate of change of the amplitude of the final control law command. The step size determination unit is used to determine the adaptive correction step size operator based on the rate of change of the discrete tracking error signal; When the rate of change of the discrete tracking error signal is greater than the preset fracture threshold, the attenuation relationship between the square inverse of the rate of change of the discrete tracking error signal and the preset basic control step size is used to adjust the adaptive correction step size operator to generate a numerical decrease in order to reduce the parameter correction rate. The parameter update unit is used to correct the estimation weights of the pseudo-partial derivative estimation parameters at the current sampling time using the adaptive correction step size operator, and update the pseudo-partial derivative estimation parameters at the current sampling time. The control solving unit is used to estimate parameters based on updated pseudo-partial derivatives and discrete tracking error signals, using the objective function that minimizes the tracking error deviation term and the penalty term for changes in the control input independent variable as the optimization rule, and superimposing the transmission backlash time delay compensation caused by the transmission resistance torque, to solve for the final control law command at the current sampling moment, and output the final control law command to the current-converting controlled unit to change the rectified current feedback value; it is also used to input the final control law command into the amplitude limiting limiter, and cut off the final control law command when it exceeds the set industrial safety voltage boundary range, outputting the limited rectification control command within the safe amplitude range; it is also used to accumulate the rectified current feedback value and system operating temperature data of the current-converting controlled unit in the historical operating cycle, calculate the system drift index characterizing the performance degradation of the rectifier equipment based on the accumulated rectified current feedback value and system operating temperature data, and output the equipment abnormality warning signal and increase the penalty term for changes in the control input independent variable in the objective function when the system drift index is greater than the set degradation threshold.

Images