A method and system for synergistic adaptive control of molding pressure and temperature for a coke molding

Abstract

The present application relates to the technical field of adaptive control, and more particularly, to a kind of collaborative adaptive control method and system of forming pressure and temperature of type coke, comprising: obtaining the set value and measured value of the pressure and temperature of type coke forming, constructing the performance index consisting of integral absolute error, overshoot and establishment time, when performance index breaks through trigger boundary, start online setting;System identification is carried out using sequential relay excitation, and limit loop signal is generated using bias relay in pressure control loop.The present application constructs feedforward compensation decoupling matrix based on identification result, reduces the cross action between pressure control and temperature control, then, calculate proportional-integral-derivative control parameters based on decoupled model, and realize smooth and undisturbed switching update of new and old parameters through parameter fusion strategy, which helps to keep type coke forming process stable operation, improve control system reliability and type coke product forming quality.

Description

Technical Field

[0001] This invention belongs to the field of adaptive control technology, and in particular relates to a method and system for coordinated adaptive control of coke forming pressure and temperature. Background Technology

[0002] There is a bidirectional coupling between molding pressure and temperature: pressure changes alter the internal friction state and heat transfer conditions of the material, thus affecting temperature distribution; temperature fluctuations, in turn, change the plasticity and thermal expansion characteristics of the coal, having a reverse effect on molding pressure. This strong coupling, combined with fluctuations in raw material properties and external environmental disturbances, makes it difficult for the control system to maintain stability, inducing oscillations in temperature and pressure parameters, leading to uneven product quality, increased scrap rates, and increased energy consumption. Existing methods typically combine decoupling control strategies with proportional-integral-derivative (PID) control techniques, and use feedforward compensation algorithms to reduce the mutual interference between temperature and pressure. Simultaneously, relay feedback excitation tuning technology is used to identify the critical parameters of the controlled object during operation and adjust the controller parameters accordingly.

[0003] In related technologies, for example, Chinese patent document CN117784723B discloses an online multi-band light-sensing decoupled micro-optical component forming process monitoring method. By selecting process parameters with desired height, the method collects multi-band illuminance curves of light during the processing from multiple directions online and extracts feature values ​​through decoupling. Combining the correlation between light-sensing feature values ​​and process parameters in past experience data, the method predicts processing parameters in real time, including hot-pressing temperature, hot-pressing pressure, and holding time. It also combines the hot-melt compression theory to characterize the surface stress of micro-forming and finally corrects the equipment status and controls the micro-forming scale through feedback control algorithm.

[0004] However, existing methods lack a multi-dimensional performance evaluation mechanism, making it difficult to construct performance boundaries and trigger online tuning based on indicators such as integral absolute error, overshoot, and settling time. This can easily lead to unreasonable tuning timing or response lag. Furthermore, when relay excitation identification is used in strongly correlated systems, the excitation test of one loop can interfere with another. Without a signal cleanup and compensation mechanism based on a disturbance observer, the extracted critical gain and critical period can easily deviate from the true main channel characteristics. In addition, existing technologies often directly replace old parameters with new ones, which can easily cause instantaneous changes in control output, making it difficult to achieve disturbance-free parameter switching and affecting the stable operation of the coke continuous forming system. Summary of the Invention

[0005] In view of the problem that existing technologies often use a direct overwrite method to replace old parameters after obtaining new parameters, which can easily cause instantaneous changes in control output, make it difficult to achieve disturbance-free parameter switching, and affect the stable operation of the coke continuous forming system, this invention provides a collaborative adaptive control method and system for coke forming pressure and temperature.

[0006] In a first aspect, this invention provides a collaborative adaptive control method for coke forming pressure and temperature, comprising: S1: acquiring setpoints and measured values ​​of coke forming pressure and temperature, constructing a performance index composed of integral absolute error, overshoot, and settling time, and initiating online tuning when the performance index exceeds the trigger boundary; S2: using sequential relay excitation for system identification, generating a limit loop signal in the pressure control loop using a bias relay, filtering to extract the critical gain and critical period of the main pressure channel, synchronously identifying the first coupling transfer function of pressure to temperature, performing excitation identification in the temperature control loop, constructing a disturbance observer based on the first coupling transfer function and the pressure control output, compensating for disturbances to obtain a clean signal, and analyzing the clean signal to obtain... S3: Based on the critical gain and critical period of the main temperature channel, the second coupling transfer function of temperature to pressure is identified simultaneously; S4: Based on the critical gain and critical period of the main pressure channel and the main temperature channel, a transfer function is constructed, which together with the first coupling transfer function and the second coupling transfer function to construct a feedforward compensation decoupling matrix. The feedforward compensation decoupling matrix is ​​combined with the controlled object, and the critical gain and critical period of the main pressure channel and the main temperature channel are converted into the transformed critical gain and transformed critical period of the decoupled dual-loop model. The proportional integral differential parameters of pressure and temperature are calculated based on the transformed critical gain and transformed critical period. The parameter fusion strategy is used to switch and update the newly calculated parameters with the current parameters without disturbance.

[0007] By constructing multi-dimensional performance indicators based on integral absolute error, overshoot, and settling time, online tuning is triggered when control performance truly degrades, avoiding the blindness and lag inherent in existing fixed-cycle tuning or manual experience-based tuning. Simultaneously, a sequential relay excitation method is used to identify the pressure main channel, temperature main channel, and bidirectional coupling channel. During temperature identification, a disturbance observer constructed using a pressure-to-temperature coupling model is used to purify the feedback signal, reducing the contamination of the identification results of the other loop by the excitation of a single loop in a strongly coupled system. Furthermore, a feedforward compensation decoupling matrix is ​​constructed based on the identified main channel and coupling channel models, transforming the originally mutually constraining pressure and temperature dual-variable control problem into a decoupled dual-loop tuning problem. A parameter fusion strategy is used to achieve a smooth transition between old and new PID parameters. Compared to existing control methods that directly tune or replace parameters, this invention can suppress cross-disturbances, shorten pressure response settling time, reduce temperature fluctuations and integral errors, and improve the reliability of the control system and the consistency of coke product forming without interrupting coke forming production.

[0008] Furthermore, when a performance indicator exceeds the trigger boundary, online tuning is initiated, including: at the end of the response process after a step change in the setpoint or disturbance recovery, the error sequence of the entire response process is collected; the integral absolute error, maximum overshoot, and settling time of the entire response process are calculated, and the baseline expected value and standard deviation of each indicator are established using weighted moving average filtering; the baseline expected value is added to the standard deviation by a preset multiple as the trigger boundary of the corresponding performance indicator; when the calculated value of any performance indicator in multiple consecutive response processes exceeds its corresponding trigger boundary, it is determined as performance degradation, a tuning trigger command is generated, the integral state of the original controller is frozen and the steady-state control output before identification is maintained, and the system identification is prepared.

[0009] This method collects a complete error sequence after the setpoint step or disturbance recovery response ends, and establishes a dynamic benchmark and standard deviation trigger boundary based on the integral absolute error, maximum overshoot, and settling time. This allows the triggering conditions to adaptively adjust with equipment status, raw material batches, and normal operating conditions. By requiring multiple consecutive out-of-bounds responses before determining performance degradation, it can effectively filter out false triggers caused by occasional noise, single disturbances, or instantaneous anomalies. At the same time, it freezes the integral state during triggering and maintains the steady-state control output before identification, providing stable initial conditions for subsequent closed-loop identification, thereby improving the accuracy of online tuning start-up timing and the continuous stability of the molding process.

[0010] Further, the critical gain and critical period of the main pressure channel are extracted through filtering, including: after entering the online identification mode, freezing the integral state of the pressure proportional-integral-derivative controller, keeping the pressure control output near the steady-state control output before identification, and superimposing a bias relay excitation signal with asymmetric dead zone, output limiting, and bias on the steady-state control output; acquiring the limit loop signal generated by the pressure system after the bias relay, and smoothing it through a state variable filter to separate the fundamental component and high-frequency noise; using an orthogonal demodulation algorithm to extract the in-phase and quadrature components of the fundamental component, and calculating the actual amplitude and oscillation angular frequency of the limit loop signal; and using the describing function method, using the conversion gain of the bias relay and the actual amplitude and oscillation angular frequency of the extracted limit loop signal to solve for the critical gain and critical period of the main pressure channel.

[0011] This method superimposes an excitation from a bias relay with an asymmetric dead zone, output limiting, and bias near the steady-state pressure control output, causing the pressure loop to generate a limit loop response with controlled amplitude that more closely approximates the actual operating point. Then, a state variable filter separates the fundamental frequency from high-frequency noise, and orthogonal demodulation extracts the actual amplitude and oscillation angular frequency. Finally, the critical gain and critical period of the pressure main channel are calculated using the describing function method. Compared to directly reading from noisy waveforms or using ordinary relays for tuning, this invention can obtain more stable and accurate dynamic characteristics of the pressure main channel in complex industrial environments, providing a reliable basis for subsequent decoupling compensation and PID parameter calculation.

[0012] Furthermore, the first coupling transfer function of pressure to temperature is identified simultaneously, including: during the relay excitation of the pressure control loop to generate limit loop oscillation, simultaneously acquiring the control output sequence of the pressure control loop and the measurement feedback sequence of the temperature control loop; using the control output sequence as the excitation input and the measurement feedback sequence as the forced response output; using the recursive least squares method with a forgetting factor to perform online parameter estimation of the input and output sequences, and identifying a first-order inertial plus pure delay model composed of gain coefficient, time constant and pure time delay, as the first coupling transfer function of pressure to temperature.

[0013] This method simultaneously acquires the pressure control output sequence and temperature measurement feedback sequence during the limit cycle oscillation generated by the pressure relay excitation. It then employs a recursive least squares method with a forgetting factor to identify the first-order inertial plus pure delay model of pressure versus temperature online. This allows for the acquisition of the coupling gain, time constant, and pure lag time of pressure changes on the mold temperature without additional shutdown tests or independent disturbance tests. The advantage of this technique lies in its ability to quantitatively characterize the thermal inertia and hysteresis characteristics of temperature fluctuations caused by pressure pulsations. This provides a realistic and usable coupling model for subsequent disturbance observer compensation and feedforward decoupling, reducing erroneous temperature control caused by the pressure regulation process.

[0014] Furthermore, the critical gain and critical period of the temperature main channel are obtained by analyzing the purification signal, including: discretizing and convolving the first coupling transfer function obtained by identification with the incremental control output of the current pressure control loop relative to the steady-state control output before identification, and estimating the coupling interference of pressure control on the temperature object online; synchronously subtracting the original measurement feedback signal of the temperature control loop from the coupling interference to compensate for the main superposition effect of pressure fluctuation on the temperature response, and obtaining the purification signal that mainly characterizes the temperature main channel; performing zero-crossing detection and peak extraction on the purification signal, calculating the amplitude and period of the limit loop of the temperature control loop, and obtaining the critical gain and critical period of the temperature main channel.

[0015] This method utilizes the identified pressure-to-temperature first coupling transfer function and the pressure increment control output for discrete convolution to estimate the coupling interference generated by pressure control on the temperature object online. This interference component is then simultaneously subtracted from the original temperature feedback signal to obtain a purified signal that primarily characterizes the temperature's own dynamics. Zero-crossing detection and peak extraction are then performed on the purified signal to calculate the temperature critical gain and critical period. Compared to existing methods that directly analyze the original temperature oscillation waveform, this invention significantly reduces the contamination of temperature identification results by pressure loop excitation, improving the relevance and stability of temperature PID parameter calculation.

[0016] Furthermore, a feedforward compensation decoupling matrix is ​​constructed, including: constructing the transfer functions of the pressure main channel and the temperature main channel based on the extracted critical gain and critical period of the pressure main channel, and constructing the transfer function matrix of the dual-input dual-output system with the first coupling transfer function of pressure to temperature and the second coupling transfer function of temperature to pressure; based on the transfer function matrix, using the diagonal matrix decoupling principle, with the transfer function of each main channel as the denominator and the negative value of its corresponding coupling transfer function as the numerator, and using the filter and delay compensation factor to satisfy the realizability condition, calculating the feedforward compensator network parameters from pressure to temperature and from temperature to pressure respectively; and combining the feedforward compensator network parameters to obtain the feedforward compensation decoupling matrix.

[0017] This method constructs a dual-input, dual-output system model based on the pressure and temperature main channel transfer functions and the pressure-to-temperature and temperature-to-pressure bidirectional coupling transfer functions. It then uses the diagonal matrix decoupling principle to calculate the feedforward compensator network parameters for pressure-to-temperature and temperature-to-pressure, while introducing filters and delay compensation factors to meet physical realizability requirements. Therefore, this invention can form a deployable feedforward compensation decoupling matrix in engineering control systems, effectively counteracting the cross-effects between pressure and temperature control inputs. This prevents subsequent control parameter tuning from being repeatedly constrained by strong coupling relationships, improving the controllability and robustness of the dual-closed-loop control for coke forming.

[0018] Furthermore, the feedforward compensation decoupling matrix is ​​combined with the controlled object, including: multiplying the transfer function matrix of the controlled object with the feedforward compensation decoupling matrix to obtain the overall equivalent decoupling object matrix, and extracting the two fully decoupled single-loop subsystems from it, wherein the two single-loop subsystems are the pressure equivalent single-loop subsystem and the temperature equivalent single-loop subsystem, respectively.

[0019] This method multiplies the transfer function matrix of the controlled object with the feedforward compensation decoupling matrix to obtain the overall equivalent decoupling object matrix. From this matrix, it extracts the pressure equivalent single-loop subsystem and the temperature equivalent single-loop subsystem, transforming the originally mutually influential bivariate control problem into two relatively independent single-loop control problems. In practical applications, this technique allows for optimization of the pressure loop with a fast response objective and the temperature loop with a stable, low overshoot objective, reducing repeated trial-and-error adjustments during joint debugging, lowering the difficulty of on-site commissioning, and improving the predictability of control performance.

[0020] Furthermore, the proportional-integral-derivative (PID) parameters of pressure and temperature are calculated based on the switching critical gain and switching critical period. This includes: applying the improved gain margin and phase margin constraints to the decoupled equivalent single-loop subsystems of pressure and temperature, respectively; calculating the proportional coefficient, integral time, and derivative time that satisfy the gain margin requirement of the pressure control loop, based on the frequency domain characteristics of the equivalent decoupled object matrix at the critical oscillation point, combined with the obtained switching critical gain and switching critical period of the pressure control loop, using the improved formula of the frequency domain tuning rule; and calculating the proportional coefficient, integral time, and derivative time that satisfy the gain margin and response index of the temperature control loop, combined with the obtained switching critical gain and switching critical period of the temperature control loop, thus completing the solution of the dual-loop PID parameters of the decoupled model.

[0021] Furthermore, a parameter fusion strategy is adopted to seamlessly switch and update the newly calculated parameters with the current parameters. This includes: setting a transition time window and a fusion weight coefficient that increases over time, with the fusion weight coefficient smoothly increasing from 0 to 1 within the transition time window; in each control cycle within the transition time window, multiplying the currently running old proportional-integral-derivative (PID) parameters and the newly calculated PID parameters with the remaining weight and the fusion weight coefficient respectively, and then summing the results to obtain the actual PID parameters executed in the current control cycle; when the transition time window ends, the old PID parameters are discarded, completing the online seamless update of the controller parameters.

[0022] In a second aspect, the present invention also provides a coking pressure and temperature adaptive control system, comprising a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to implement the coking pressure and temperature adaptive control method described above.

[0023] Beneficial effects: First, by constructing performance indicators composed of integral absolute error, overshoot, and settling time, continuous monitoring of the control performance of the coke forming process is achieved, and online tuning is triggered in a timely manner when performance degrades. In the system identification stage, sequential relay excitation combined with disturbance observer technology is used to reduce the impact of coupling interference between pressure and temperature on the identification results, thereby extracting the critical parameters of the main channel and the bidirectional coupling transfer function. Second, based on the above identification results, a feedforward compensation decoupling matrix is ​​constructed to reduce the cross-effect between pressure control and temperature control. Subsequently, proportional-integral-derivative control parameters are calculated based on the decoupled model, and a parameter fusion strategy is used to achieve smooth and disturbance-free switching and updating of old and new parameters, which helps to maintain the stable operation of the coke forming process, improve the reliability of the control system, and enhance the forming quality of the coke product. Attached Figure Description

[0024] Figure 1 This is a flowchart of the coordinated adaptive control method for medium-sized coke forming pressure and temperature according to the present invention;

[0025] Figure 2 This is the boundary map for triggering the absolute integral error in this invention; Figure 3 This is a comparison diagram of temperature disturbances in this invention. Detailed Implementation

[0026] In this embodiment, a method for coordinated adaptive control of coke forming pressure and temperature is described, such as... Figure 1 As shown, it includes: S1 acquires data to construct performance metrics and triggers online tuning startup.

[0027] Specifically, the setpoints and measured values ​​of the pressure and temperature for coke forming are obtained, and a performance index consisting of integral absolute error, overshoot, and settling time is constructed. When the performance index exceeds the trigger boundary, online tuning is initiated.

[0028] Based on a programmable logic controller, the analog input module acquires the milliampere-level current signal and millivolt-level voltage signal transmitted in real time from the pressure sensor and thermocouple, converts them into pressure and temperature measurement values ​​in engineering quantity form, and then reads the pressure and temperature setpoints issued by the human-machine interface, and calculates the pressure deviation and temperature deviation using the difference command.

[0029] The integral absolute error within a certain time window is calculated using a trapezoidal integral numerical accumulation algorithm. The maximum measured value during the control process is detected, and the percentage difference between the measured value and the setpoint is calculated to obtain the overshoot. An internal system timer is used to count the setpoint step time until the measured value remains stable within the error band. The integral absolute error, overshoot, and settling time are compared with preset performance benchmark thresholds. When any performance index exceeds the corresponding trigger boundary after three consecutive setpoint step response processes or disturbance recovery response processes, the main program is triggered to call the online tuning procedure.

[0030] In some implementations, the setpoints and measured values ​​of the pressure and temperature for coke forming are obtained, and a performance index consisting of integral absolute error, overshoot, and settling time is constructed. When the performance index exceeds the trigger boundary, online tuning is initiated, including: At the end of the response process after a step change in the setpoint or a disturbance recovery, the error sequence of the entire response process is collected; the integral absolute error, maximum overshoot, and settling time of the entire response process are calculated, and the baseline expected value and standard deviation of each index are established using weighted moving average filtering; the baseline expected value is added to the standard deviation by a preset multiple as the trigger boundary of the corresponding performance index; when the calculated value of any performance index in multiple consecutive response processes exceeds its corresponding trigger boundary, it is determined as performance degradation, a tuning trigger command is generated, the integral state of the original controller is frozen and the steady-state control output before identification is maintained, and the system identification is prepared.

[0031] The step size of the set value is usually set to 5% to 10% of the current steady-state value. For example, the pressure set value for coke forming jumps from 20MPa to 22MPa. The condition for the end of the response process can be set to the time when the error continuously enters and remains in the steady-state zone for more than the preset establishment time threshold, such as ±2% of the set value, with an establishment time threshold of 30s.

[0032] During the response period, an error sequence e(k) of length N is acquired and constructed at a fixed sampling frequency, preferably 10Hz. For this error sequence e(k), the integral absolute error IAE is calculated as the sum of the absolute errors at each sampling point multiplied by the sampling period. The maximum overshoot Mp is the percentage of the difference between the maximum peak value and the set value divided by the set value. The settling time is the time required to enter and stabilize in the steady-state band.

[0033] After the performance indicators are calculated, a weighted moving average filter is used to calculate the baseline expected value at the current time. The baseline expected value at the current time is equal to the weighted sum of the expected value at the previous time and the current measured value. The weights are determined by the attenuation factor, which is preferably between 0.85 and 0.95, for example, an attenuation factor of 0.9. The variance is calculated recursively in the same way to obtain the baseline expected value and standard deviation of each performance indicator.

[0034] The trigger boundary is set based on the calculated expected value and standard deviation. The trigger boundary value equals the baseline expected value of each performance indicator, plus the product of a preset multiple and the corresponding standard deviation. The preferred range for the preset multiple is 2 to 3, for example, 2.5. If the baseline expected value of the integral absolute error is 150 and the standard deviation is 15, then the trigger boundary is calculated to be 187.5. Figure 2 As shown, the horizontal line with a baseline expected value of 150 and the dashed line with a trigger boundary of 187.5 divide the normal performance range and the performance degradation range.

[0035] Real-time monitoring of the system's performance after each disturbance or sudden change; once continuous [disruptions / changes] are detected... During the second step response, if any index, such as the integral absolute error greater than 187.5, the maximum overshoot, or the settling time exceeds its corresponding trigger boundary, it is determined that the control performance has degraded. The optimal value is 3. At this point, the logic controller will trigger the Boolean tuning instruction variable to be set to 1, freeze the integral state of the original proportional-integral-derivative PID controller, maintain the steady-state control output before identification, smoothly transition and prepare to carry out closed-loop system identification.

[0036] S2, a sequential excitation identification system, obtains critical parameters and coupling functions.

[0037] Specifically, sequential relay excitation is used for system identification. In the pressure control loop, a limit loop signal is generated using a bias relay. After filtering, the critical gain and critical period of the main pressure channel are extracted. The first coupling transfer function of pressure to temperature is identified simultaneously. Excitation identification is performed in the temperature control loop. A disturbance observer is constructed based on the first coupling transfer function and the pressure control output. The disturbance is compensated to obtain a clean signal. The critical gain and critical period of the main temperature channel are obtained by analyzing the clean signal. The second coupling transfer function of temperature to pressure is identified simultaneously.

[0038] After the online tuning procedure is triggered, the integral state of the conventional proportional-integral-derivative controller is frozen, and the control output is maintained near the steady-state control output before identification. A bias asymmetric relay excitation signal with hysteresis characteristics is superimposed on this steady-state control output, causing the pressure control loop to generate a limit loop response with controlled amplitude. The forward and reverse amplitudes of the bias relay, as well as the hysteresis width parameters, are set to produce a continuous periodic oscillation with controlled amplitude in the controlled pressure, which is the limit loop signal. Spectral analysis is performed on the acquired limit loop signal to filter out high-frequency noise and extract the fundamental component. The amplitude and oscillation period of the fundamental component are measured, and the oscillation period is used as the critical period of the main pressure channel. The equivalent gain of the relay describing function is calculated based on the output amplitude of the bias relay, the amplitude of the limit loop response, and the dead zone correction term. This equivalent gain is then used as the critical gain of the main pressure channel. .

[0039] Simultaneously, during the oscillation of the pressure control loop, the temperature response data is recorded. The first coupling transfer function of pressure input to temperature output is fitted using the least squares method. The relay excitation of the pressure control loop is disconnected, ensuring the pressure control loop maintains closed-loop stable operation. The incremental control output signal generated by the pressure controller to maintain pressure steady state is acquired in real time. The coupling disturbance caused by pressure to temperature is calculated using the first coupling transfer function. A feedforward disturbance observer is constructed to subtract the coupling disturbance from the actual measured temperature oscillation signal, obtaining the temperature clean signal after compensating for the main coupling interference of pressure. Similarly, the temperature clean signal is analyzed to extract the temperature fundamental wave, and the critical gain of the main temperature channel is calculated using the same gain period calculation formula. With critical period The temperature control loop's control output increment sequence is used as the excitation input, and the pressure control loop's measurement feedback increment sequence is used as the forced response output. The second coupling transfer function of temperature to pressure is identified by using the same recursive least squares method with forgetting factor as the first coupling transfer function.

[0040] In some implementations, a limit loop signal is generated using a bias relay in the pressure control loop, and the critical gain of the main pressure channel is extracted after filtering. With critical period ,include: After entering online identification mode, the integral state of the pressure proportional-integral-derivative controller is frozen, and the pressure control output is maintained near the steady-state control output before identification. A bias relay excitation signal with asymmetric dead zone, output limiting, and bias is superimposed on the steady-state control output. The limit loop signal generated by the pressure system after the bias relay is acquired and smoothed using a state variable filter to separate the fundamental component and high-frequency noise. An orthogonal demodulation algorithm is used to extract the in-phase and quadrature components of the fundamental component, and the actual amplitude and oscillation angular frequency of the limit loop signal are calculated. Based on the describing function method, the critical gain of the main pressure channel is obtained using the switching gain of the bias relay and the extracted actual amplitude and oscillation angular frequency. With critical period .

[0041] After entering online identification mode, the steady-state control output before identification by the pressure PID controller is used as a reference. A bias relay excitation signal is superimposed on this steady-state control output. Key parameters of the bias relay include the asymmetric dead zone *d*, the output limit *h*, and the bias amount *b*. The preferred range for the output limit *h* is 5% to 15% of the current steady-state control output. For example, if the current opening of the pressure control valve is 40%, then the output limit *h* can be 4%. The asymmetric dead zone *d* is set to 2 to 4 times the standard deviation of the pressure sensor's ambient noise. For example, when the standard deviation of the pressure noise is 0.05 MPa, the asymmetric dead zone *d* can be 0.1 MPa to 0.2 MPa to block high-frequency chatter. The bias amount *b* is set to 1% to 2% to compensate for asymmetric nonlinearity caused by system mechanical friction, etc.

[0042] Using this bias relay, the closed-loop system generates constant-amplitude oscillations. The pressure output response signal, affected by industrial noise, is recorded using a data acquisition card and then smoothed using a second-order Butterworth state-variable low-pass filter. The filter cutoff frequency is preferably set to five times the expected oscillation frequency, for example, 1Hz. This operation attenuates random noise above the cutoff frequency and extracts the fundamental component of the limiting loop signal used for amplitude and period estimation.

[0043] After obtaining the fundamental frequency component, a discrete digital processor employs an orthogonal demodulation algorithm to multiply the fundamental frequency component in real time with the reference sine wave sin(ωt) and cosine wave cos(ωt), respectively. Discrete numerical integration is then performed through a sliding time window with a length of 2000 sampling points to extract the in-phase and quadrature components of the fundamental frequency signal. After normalizing the in-phase and quadrature components according to the number of sampling points in the sliding window, the actual oscillation amplitude of the limiting loop signal is the square root of the sum of the squares of the in-phase and quadrature components. For example, if the calculated oscillation amplitude A is 1.5 MPa, the half-cycle is obtained by continuously monitoring the time difference Δt between adjacent opposite zero-crossing points, and the oscillation period is twice the time difference between adjacent opposite zero-crossing points, with the oscillation angular frequency being pi divided by this time difference. If the measured Δt is 2.5 s, then the critical period is... The duration is 5.0s and the oscillation angular frequency ω is 1.256 rad / s.

[0044] After removing the DC bias component from the bias relay output and converting the positive and negative output amplitudes to an equivalent symmetrical limiting height, the calculation method for the equivalent conversion gain of the dead-zone relay is set as follows: divide four times the equivalent symmetrical limiting height by the product of pi and the actual oscillation amplitude of the limit loop signal, then multiply by 1 and subtract the square root of the difference obtained by subtracting the square of the ratio of the dead zone width to the oscillation amplitude. For example, when the extracted oscillation amplitude is equal to 1.5, the set dead zone width is equal to 0.2, and the equivalent symmetrical limiting height is equal to 4, the equivalent conversion gain is approximately 3.36 according to the above logic. This calculated equivalent conversion gain is directly used as the critical gain of the pressure main channel, that is, the critical gain is considered to be approximately 3.36, and the critical period of the pressure main channel corresponding to this oscillation process is recorded as 5.0s.

[0045] In some implementations, the first coupling transfer function of pressure to temperature is identified simultaneously, including: During the relay excitation of the pressure control loop to generate limit loop oscillation, the control output sequence of the pressure control loop and the measurement feedback sequence of the temperature control loop are simultaneously acquired. The control output sequence is used as the excitation input, and the measurement feedback sequence is used as the forced response output. The recursive least squares method with forgetting factor is used to estimate the parameters of the input and output sequences online, and a first-order inertial plus pure delay model composed of gain coefficient, time constant and pure time delay is identified as the first coupling transfer function of pressure to temperature.

[0046] Within 15 to 20 consecutive oscillation cycles of applying relay excitation to the pressure control loop and maintaining the limit loop oscillation, the total test time is approximately 75 to 100 seconds with a critical period of 5 seconds. The distributed control system or data acquisition card is used to perform dual-channel synchronous data recording according to a fixed sampling period, wherein the sampling period is preferably 0.5 seconds.

[0047] The collected pressure control loop control commands or valve openings representing the actions of the molding machine actuators are used as the model excitation input sequence. Simultaneously, the measured temperature fluctuations inside the mold are used as the forced response output sequence of the system. Before identification, the steady-state baseline values ​​of the two sequences before oscillation are calculated and mean-free processing is performed to extract the incremental fluctuation of the model excitation input sequence, which represents the changing characteristics. Incremental fluctuation of the system's forced response output sequence This eliminates the interference of steady-state operating point offset on identification accuracy.

[0048] Within the controller, the initial coupling process from pressure to temperature is constructed as a first-order discrete model with inertia and pure delay. The parameters of this model are solved using a recursive least squares algorithm with a forgetting factor. The preferred range for the forgetting factor is 0.96 to 0.99, specifically set to 0.98. A regression data vector containing historical inputs and outputs is then constructed. ,in, Indicates the first The incremental fluctuation of the system's forced response output sequence at the sampling time; Indicates taking into account the discrete delay order After that, the The incremental fluctuation of the model excitation input sequence at the sampling time; Let be the discrete delay order of the pressure-to-temperature coupling channel. It can be determined by the peak value of input-output cross-correlation, candidate delay ergonomics, or the minimum residual criterion; and by... = The coupled pure time delay is calculated. During initialization, the covariance matrix is ​​set. ,in The identity matrix and parameter vector .

[0049] The gain matrix, covariance matrix, and parameter estimates are updated recursively for each sampling period until the rate of change of each parameter estimate in the parameter vector is less than 0.1% over 100 consecutive sampling steps, at which point the parameters are considered converged. After convergence, the discrete-domain parameters are inversely calculated into the parameters of the continuous-domain first-order inertial plus pure lag model based on the zero-order preserved discretization relation, and the first coupled transfer function, such as the standard FOPDT continuous transfer function, is calculated.

[0050] Such numerical calculations can yield specific results, such as the crossover process gain coefficient. It is 0.42℃ / %, temperature-time constant. The time is 45.5s, and the coupling pure time delay is... The model parameters are 12.0s. This model represents the hysteresis conduction and thermal inertia characteristics of mold temperature change caused by molding pressure pulsation.

[0051] In some implementations, a disturbance observer is constructed based on the first coupling transfer function and the pressure control output. The disturbance is compensated to obtain the clean signal, and the critical gain of the temperature main channel is obtained by analyzing the clean signal. With critical period ,include: The first coupling transfer function obtained from identification is discretized and convolved with the incremental control output of the current pressure control loop relative to the steady-state control output before identification to estimate the coupling interference of pressure control on the temperature object online. The original measurement feedback signal of the temperature control loop is synchronously subtracted from the coupling interference to compensate for the main superposition effect of pressure fluctuations on the temperature response, obtaining the purified signal that mainly characterizes the characteristics of the main temperature channel. Zero-crossing detection and peak extraction are performed on the purified signal, and the amplitude and period of the limit loop of the temperature control loop are calculated to obtain the critical gain of the main temperature channel. With critical period .

[0052] The numerical calculation of the perturbation observer is based on identifying the first coupling transfer function that converges. Assuming The system can be discretized into a time-domain difference equation according to the system sampling period using the bilinear transform method or the zero-order hold method: ,in, , To identify the steady-state control output of the upstream pressure control loop. Indicates the first The incremental control output of the pressure control loop at the sampling time relative to the steady-state operating point. Indicates the first The actual control output value of the pressure control loop at the sampling time. and They represent the first Sampling time and the first Online estimate of the coupling interference between pressure and temperature at the sampling time. This represents the historical pressure increment control output after taking into account the coupling pure time delay.

[0053] The observer extracts the historical incremental control command sequence of the pressure control loop relative to the steady-state operating point in each operating cycle. By performing the aforementioned differential convolution operation, the coupling interference caused by pressure command fluctuations to the temperature channel is estimated online. For example, when the molding pressure control output command undergoes a 10% step change, the estimated coupling interference will gradually increase exponentially after a pure delay of 12 seconds, approaching the steady-state deviation of 4.2℃. After obtaining the coupling interference, the original feedback sequence of the current temperature control loop is read and a point-by-point algebraic subtraction compensation operation is performed with the corresponding coupling interference to generate the compensation amount. This process compensates for the main cross-disturbance components in the temperature sensor feedback caused by pressure fluctuations, thereby reconstructing the purified signal that mainly characterizes the main temperature channel characteristics.

[0054] Waveform analysis was performed on the purification signal, and a zero-crossing hysteresis dead zone with an amplitude of 0.5℃ was set to prevent misjudgment caused by minor noise. Quadratic polynomial fitting was used for zero-crossing time detection and local peak coordinate extraction. Assuming the program continuously detects three adjacent alternating positive and negative zero-crossing times of 10.0s, 35.0s, and 60.0s respectively, the average half-cycle duration was calculated to be 25s, and the critical period of the temperature control loop under isolated interference was calibrated. It takes 50.0 seconds.

[0055] The amplitude of the limiting cycle is obtained by accumulating the difference between the positive and negative peak values ​​of multiple cycles and averaging them. For example, calculated The temperature is 2.8℃. This is combined with the preset temperature-excitation bias relay limiting height. If set to 5%, and the limit cycle amplitude is... Substituting 2.8℃ into the equivalent gain formula of the describing function, we obtain the critical gain of the main temperature channel. ≈2.27, simultaneously obtaining the critical period of the main temperature channel. It takes 50.0 seconds.

[0056] S3: Construct the decoupling matrix and calculate the parameters to achieve perturbation-free switching updates.

[0057] Specifically, a transfer function is constructed based on the critical gain and critical period of the pressure main channel and the critical gain and critical period of the temperature main channel. This transfer function, together with the first and second coupled transfer functions, forms a feedforward compensation decoupling matrix. The feedforward compensation decoupling matrix is ​​then combined with the controlled object. The critical gain and critical period of the pressure main channel and the critical gain and critical period of the temperature main channel are converted into the transformed critical gain and transformed critical period of the decoupled dual-loop model. The proportional-integral-differential parameters of pressure and temperature are calculated based on the transformed critical gain and transformed critical period. A parameter fusion strategy is then used to seamlessly switch and update the newly calculated parameters with the current parameters.

[0058] Based on the critical gain and critical period of the pressure main channel and the critical gain and critical period of the temperature main channel, and combined with the output average value deviation caused by relay bias excitation or the process gain of each main channel obtained by small step test, the first-order inertial plus pure time delay transfer function model parameters of the pressure main channel and the temperature main channel are estimated using the two-point identification method based on frequency domain response characteristics. The transfer function of the main channel, as well as the first coupled transfer function and the second coupled transfer function, are combined into the transfer function matrix of the dual-input dual-output multivariable system.

[0059] Based on the principle of ideal decoupling control, the negative of the quotient of the first coupling transfer function from pressure to temperature and the transfer function of the main temperature channel is used as the feedforward compensator network parameter in the direction from pressure to temperature; the negative of the quotient of the second coupling transfer function from temperature to pressure and the transfer function of the main pressure channel is used as the feedforward compensator network parameter in the direction from temperature to pressure, thus forming a feedforward compensation decoupling matrix.

[0060] At the control algorithm layer, the feedforward compensation decoupling matrix is ​​cascaded between the multivariable controller and the actual controlled object to achieve diagonal decoupling of the system. The transfer function matrix of the controlled object is multiplied by the feedforward compensation decoupling matrix to obtain the overall equivalent decoupling object matrix, from which the fully decoupled pressure equivalent single-loop subsystem and temperature equivalent single-loop subsystem are extracted. The angular frequencies corresponding to the phase angle of the frequency response of these two equivalent single-loop subsystems reaching -180 degrees are calculated and used as the switching critical angular frequencies of their respective loops. Furthermore, the reciprocal of the amplitude response of each equivalent single-loop subsystem at the switching critical angular frequency is calculated, along with the time quotient obtained by dividing twice pi by the respective switching critical angular frequency. These two calculation results are used as the switching critical gain and switching critical period of the decoupled dual-loop model, i.e., the switching critical gain and switching critical period of the pressure control loop, and the switching critical gain and switching critical period of the temperature control loop.

[0061] The improved critical proportional gain tuning rule from the self-tuning algorithm library is invoked to calculate new parameter sets for the pressure and temperature control loops based on the switching critical gain and switching critical period. These new parameter sets consist of the proportional gain, integral time constant, and derivative time constant. A transition time window is set, and within this window, a fusion weighting coefficient is constructed that smoothly increases from 0 to 1. The old and new parameters are then weighted and summed according to the remaining weights and fusion weights, allowing the proportional, integral, and derivative parameters to smoothly transition from their current values ​​to the newly calculated values.

[0062] In some implementations, the critical gain is based on the pressure main channel. and critical period Critical gain of the main temperature channel and critical period Construct a transfer function, which, together with the first coupled transfer function and the second coupled transfer function, forms a feedforward compensation decoupling matrix, including: Critical gain based on the extracted pressure master channel and critical period Critical gain of the main temperature channel and critical period The transfer functions of the pressure main channel and the temperature main channel are constructed separately, and the first coupling transfer function of pressure to temperature and the second coupling transfer function of temperature to pressure are extracted. These four functions are then used to construct the transfer function matrix of a dual-input dual-output system. Based on the transfer function matrix, the diagonal matrix decoupling principle is adopted, with the transfer function of the main channel as the denominator and the negative value of the cross-coupling transfer function as the numerator. The realizability conditions are satisfied by using filters and delay compensation factors, and the feedforward compensator network parameters of pressure to temperature and temperature to pressure are calculated respectively. The feedforward compensator network parameters are combined to obtain the feedforward compensation decoupling matrix, so that the series transformation model of the transfer function matrix of the controlled object and the feedforward compensation decoupling matrix presents two equivalent single-loop subsystems.

[0063] The control system is based on the two-point identification method and the static process gain of the main channel, utilizing the obtained critical gain of the main pressure channel. and critical period Critical gain of the main temperature channel and critical period The transfer functions of the pressure main channel and the temperature main channel are calculated.

[0064] For example, through conversion and calculation, the process gain of the pressure main channel is 1.6, the inertial time constant is 2.0s, and the pure time delay is 1.0s; the process gain of the temperature main channel is 2.5, the inertial time constant is 25.0s, and the pure time delay is 8.0s. The first coupling transfer function from pressure to temperature, identified in the previous steps (e.g., a first-order inertial plus pure delay model with a gain of 0.42, an inertial time constant of 45.5s, and a pure time delay of 12s), and the second coupling transfer function from temperature to pressure (e.g., a model with a gain of 0.15, an inertial time constant of 5.0s, and a pure time delay of 3s), together with the transfer functions of the pressure main channel, the temperature main channel, the first coupling transfer function from pressure to temperature, and the second coupling transfer function from temperature to pressure, construct the dual-input dual-output transfer function matrix of the controlled object.

[0065] The design of a feedforward diagonal decoupler is based on the transfer function matrix of the controlled object. This feedforward compensation decoupling matrix is ​​cascaded between the pressure proportional-integral-derivative (PID) controller, the temperature PID controller, and the controlled object. Its inputs are the outputs of the pressure and temperature controllers, and its outputs are the pressure and temperature control quantities after cross-coupling compensation. The network model consists of four parallel feedforward transfer function computation nodes arranged in a two-row, two-column diagonal decoupling topology. The computation nodes on the main diagonal are designed as through-through computation nodes, while the internal logic of the computation nodes on the secondary diagonal is based on the negative value of the quotient of the corresponding coupled transfer function and the transfer function of the main channel. A low-pass inertial filter node and a delay computation node are connected in series to satisfy physical realizability conditions.

[0066] The pressure and temperature principal elements on the main diagonal of the decoupling matrix are set to be equal to 1, and then the decoupling compensation elements on the off-diagonal are calculated. The decoupling node transfer function in the pressure-to-temperature direction is set as the negative of the quotient of the first coupling transfer function from pressure to temperature and the transfer function of the main temperature channel, used to counteract the coupling effect of the pressure control input on the temperature output; the decoupling node transfer function in the temperature-to-pressure direction is set as the negative of the quotient of the second coupling transfer function from temperature to pressure and the transfer function of the main pressure channel, used to counteract the coupling effect of the temperature control input on the pressure output. Since algebraic division may result in the numerator order being greater than the denominator or the time lag term being unrealizable, a low-pass filter and a delay compensation factor are added to meet the feasibility conditions. Taking the compensator in the pressure-to-temperature direction as an example, substituting the model parameters from the previous example into the division logic, the resulting compensation characteristics not only include the lead-lag dynamic response characteristics obtained by recombining the gains and time constants of the two channels, but also have a net delay time. When calculating the compensator in the pressure-to-temperature direction, this net delay time is equal to the pure lag time of the first coupled transfer function minus the pure lag time of the main temperature channel; when calculating the compensator in the temperature-to-pressure direction, this net delay time is equal to the pure lag time of the second coupled transfer function minus the pure lag time of the main pressure channel. If the calculated net delay time is greater than zero, satisfying the causal realization condition, it can be directly discretized into compensator parameters.

[0067] Similarly, when calculating the compensator in the temperature-to-pressure direction, if the model parameters from the previous example are substituted, the net delay time is equal to 3s minus 1s, resulting in a delay of 2s. Since this is a positive value, it complies with the causal fulfillment condition. However, if, when calculating the compensator in the temperature-to-pressure direction, the pure lag time of the main pressure channel is greater than the pure lag time of the temperature-to-pressure coupling channel, causing the compensator to theoretically include an unrealizable physical lead term, then this lead term is forcibly discarded in the control algorithm, and an additional first-order low-pass inertial filter is added in series in the denominator of its transfer function. The filtering time constant of this first-order low-pass inertial filter is usually taken as 0.5 to 1 times the calculated non-compliant lead time, i.e., the required lag time compensation amount, such as 1.5s, to satisfy causality and suppress sudden changes in system output and high-frequency gain divergence.

[0068] The modified feedforward compensator network parameters are converted into discrete difference matrices and programmed into the control program to obtain a diagonal decoupling matrix. The diagonal decoupling matrix is ​​the aforementioned feedforward compensation decoupling matrix. After multiplying the transfer function matrix with the diagonal decoupling matrix, the original control object with strong crosstalk is converted into an equivalent decoupling object matrix dominated by two diagonal single-loop transmission characteristics.

[0069] In some implementations, the proportional-integral-differential parameters of pressure and temperature are calculated based on the critical gain and critical period of the transition, including: For the decoupled pressure equivalent single-loop subsystem and temperature equivalent single-loop subsystem, the improved gain margin and phase margin constraints are applied respectively. Based on the frequency domain characteristics of the equivalent decoupled object matrix at the critical oscillation point, and combined with the obtained critical gain and critical period of the pressure control loop, the improved formula of the frequency domain tuning rule is used to calculate the proportional coefficient, integral time, and derivative time that meet the gain margin requirements of the pressure control loop. Combined with the obtained critical gain and critical period of the temperature control loop, the proportional coefficient, integral time, and derivative time that meet the gain margin and response index of the temperature control loop are calculated, thus completing the solution of the dual-loop proportional-integral-derivative parameters of the decoupled model.

[0070] Because the feedforward compensation decoupling matrix introduces filters and delay compensation factors, their series connection with the controlled object causes changes in the amplitude and phase angle of the system's equivalent frequency domain characteristics. This renders the original critical parameters identified in the preceding steps inapplicable to the decoupled model. Therefore, it is necessary to recalculate the phase angles of the decoupled pressure equivalent single-loop subsystem and temperature equivalent single-loop subsystem, respectively. By analyzing the frequency response characteristics at different times, the critical gain and critical period of the pressure control loop, as well as the critical gain and critical period of the temperature control loop, can be obtained.

[0071] For the equivalent single-input single-output subsystem after compensation for the two coupling effects, the target boundary specifications need to be configured according to the process response requirements. Preferably, the amplitude margin Am should be between 2.5 and 3.5, and the phase margin Pm should be set between 45° and 60°. For example, since the pressure control loop requires a fast following speed, the constraints can be set to a response index Am equal to 2.5 and Pm equal to 45°. Temperature control requires a smooth heating response without overshoot, so the constraints are set to a conservative stability index Am equal to 3.0 and Pm equal to 60°. After clarifying the target requirements, the frequency response point state of the closed loop at the critical oscillation point of the limit loop, i.e., the angular frequency corresponding to that point, is used. The lower value is And with a phase angle of -180°, substituting these coordinates into the frequency domain gain margin compensation analysis equation, where, This is referred to here as the critical period of conversion. This is the critical gain for conversion.

[0072] Based on conversion critical gain and the critical period of conversion The proportional-integral-derivative parameters are calculated according to the preset improved critical proportional tuning rules, i.e. = , = , = ;in , and These are pre-calibrated setpoints based on the target amplitude margin, phase margin, and process response speed. The pressure control loop can be selected. The value ranges from 0.35 to 0.6. The value ranges from 0.4 to 0.6. The value is between 0.08 and 0.15; the temperature control loop can be selected. The value ranges from 0.2 to 0.45. The value ranges from 0.6 to 1.0. The value ranges from 0.05 to 0.12. Taking the pressure control loop as an example, the critical gain for switching the pressure control loop will be recalculated. With the critical period of conversion Substitute the preset tuning rules and select the corresponding tuning coefficient based on the target response speed of the pressure control loop. , and The proportional coefficient of the pressure control loop is obtained. Integral Time and differential time The temperature control loop selects a conservative tuning coefficient based on its stability requirements, and then calculates the proportional coefficient. Integral Time and differential time .

[0073] For example, when the pressure control loop switches to the critical cycle The time is 5.2 seconds. =0.5、 When = 0.125, the calculation yields , The same logic applies to the temperature channel, changing the switching critical gain of the temperature control loop. With the critical period of conversion Substituting the preset tuning rules and selecting a more conservative tuning coefficient based on the stability requirements of the temperature control loop, the proportional coefficient of the temperature control loop is obtained. Integral Time and differential time This step utilizes an analytical model closed-loop mechanism to calculate the dual-loop proportional-integral-derivative control parameters.

[0074] In some implementations, a parameter fusion strategy is used to seamlessly switch and update the newly calculated parameters with the current parameters, including: A transition time window and a fusion weight coefficient that increases over time are set. The fusion weight coefficient smoothly increases from 0 to 1 within the transition time window. In each control cycle within the transition time window, the old proportional-integral-derivative (PID) parameters and the newly calculated PID parameters are multiplied by the remaining weight and the fusion weight coefficient, respectively, and then summed to obtain the actual PID parameters executed in the current control cycle. When the transition time window ends, the old PID parameters are discarded, completing the online, disturbance-free update of the controller parameters.

[0075] The microprocessor calculates and obtains the new parameter set as follows: , as well as Then, set a transition time window, the total length of which is... Preferably, the control cycle should be between 10 and 20 times longer. Assuming the current base control cycle is set to 0.5 seconds, set... If there are 20 steps, the total transition time window is 10 seconds. Before parameter fusion begins, the integral state is initialized by back-calculating based on the current actuator output, current error, proportional term, and derivative term, and the differential filter state is inherited or reset. During fusion, output limiting and anti-integral saturation logic are enabled to ensure that the control output transitions continuously with parameter changes, avoiding a step jump in actuator output due to abrupt changes in the integral term.

[0076] During the duration of the time window, a smooth transition function is constructed to generate the fusion weight coefficient that increases over time. Typically, a cosine smoothing function with flexible start and stop characteristics is preferred. Specifically, it is 1 minus the cosine value based on the ratio of the current step number to the total length of the transition window, and then multiplied by 0.5, so that the fusion weight coefficient can achieve an S-shaped smooth transition trajectory from 0 to 1 as the number of iterations k increases.

[0077] After entering the 10-second transition time window, for each control cycle, the scheduler will extract the old parameter set currently in use. , as well as With the new parameter set, the actual proportional-integral-differential parameters for each cycle are calculated using interpolation. P generally refers to the proportionality coefficient Kp, the integral time constant Ti, and the differential time constant Td. If the proportionality coefficient needs to be derived from old parameters... The new calculated value is equal to 0.18. The transition value is equal to 0.12. When the execution reaches the window point, i.e., step 10, the fusion weight coefficient calculated by the cosine formula is equal to 0.5. At this point, the actual parameters used are... By calculating and executing control commands through a proportional mixing mechanism in each cycle, a smooth transition to the new adjustment trajectory is achieved.

[0078] When the 20th calculation step, i.e., the transition time window, ends, the fusion weight coefficient reaches 1, and the actual executed parameters are now equivalent to the new parameters. The time pointer is reset, the old parameter variables are released, and the current parameters are updated to... , as well as This enables seamless online updates of controller parameters.

[0079] Ablation experiments were conducted using the pressure and temperature dual closed-loop control system of a coke forming machine as the test platform. The initial steady-state operating conditions of the system were set as a pressure of 20 MPa and a temperature of 200 °C, with a basic sampling period of 0.5 s. The experimental group adopted the complete implementation architecture, including disturbance observer purification and extraction, feedforward compensation decoupling matrix, and a parameter non-disruptive update scheme with a transition time window. The ablation group removed the disturbance observer and feedforward compensation decoupling matrix structures, relying solely on single-loop linear relay identification and traditional proportional-integral-derivative control algorithms to handle operating condition disturbances. The experimental operation involved using a 10% step change command to instantaneously increase the pressure control loop setpoint from 20 MPa to 22 MPa under stable system operation, and simultaneously extracting various monitoring data from both schemes during the response process within a continuous test cycle.

[0080] In the ablation group test, the pressure command jump triggered cross disturbances, with the pressure main channel settling time reaching 68 seconds and the maximum overshoot reaching 18%. Simultaneously, the temperature feedback channel was affected by mechanical and thermal inertia-related disturbances, generating a forced fluctuation deviation peak of 7.5℃. The total adjustment recovery time for the dual-loop system to return to steady state was approximately 160 seconds, with an absolute pressure integral error of 320 and an absolute temperature integral error of 485. Figure 3 As shown, the ablation group exhibited temperature oscillations, with a peak exceeding 207℃ and a recovery time to steady state of approximately 160s. The experimental group employing the complete control scheme, after encountering the same disturbance, achieved only minor temperature fluctuations through decoupling compensation with the disturbance observer and feedforward, with a peak not exceeding 201.5℃ and a return to steady state within 40s. The time for the main pressure channel to smoothly enter the ±2% steady-state band for the first time was shortened to 25s, the overshoot was controlled at 3.5%, the pressure integral absolute error decreased to 125, and the temperature integral absolute error decreased to 78.

[0081] A disturbance observer combined with a feedforward compensation decoupling network constructs a reverse compensation term, which can reduce the hysteresis crosstalk effect of pressure step changes on the mold temperature channel and improve the temperature control loop's resistance to coupled disturbances. A cosine fusion smooth transition strategy using new and old parameters can reduce the abrupt change in control output during gain switching and suppress actuator chattering near the dead zone. Experimental results show that this combined scheme can shorten the pressure control loop settling time and improve control stability under time-varying coupled objects.

Claims

1. A method for coordinated adaptive control of coke forming pressure and temperature, characterized in that, include: S1: Obtain the setpoint and measured values ​​of pressure and temperature for coke forming, construct a performance index consisting of integral absolute error, overshoot, and settling time, and initiate online tuning when the performance index exceeds the trigger boundary; S2: System identification is performed using sequential relay excitation. In the pressure control loop, a limit loop signal is generated using a bias relay. After filtering, the critical gain and critical period of the main pressure channel are extracted. The first coupling transfer function of pressure to temperature is identified simultaneously. Excitation identification is performed in the temperature control loop. A disturbance observer is constructed based on the first coupling transfer function and the pressure control output. The disturbance is compensated to obtain the clean signal. The critical gain and critical period of the main temperature channel are obtained by analyzing the clean signal. The second coupling transfer function of temperature to pressure is identified simultaneously. S3: Based on the critical gain and critical period of the pressure main channel and the critical gain and critical period of the temperature main channel, a transfer function is constructed. Together with the first coupled transfer function and the second coupled transfer function, a feedforward compensation decoupling matrix is ​​constructed. The feedforward compensation decoupling matrix is ​​combined with the controlled object, and the critical gain and critical period of the pressure main channel and the temperature main channel are converted into the transformed critical gain and transformed critical period of the decoupled dual-loop model. The proportional integral differential parameters of pressure and temperature are calculated based on the transformed critical gain and transformed critical period. A parameter fusion strategy is used to seamlessly switch and update the newly calculated parameters with the current parameters.

2. The coking pressure and temperature adaptive control method according to claim 1, characterized in that, Online tuning is initiated when performance metrics exceed trigger boundaries, including: At the end of the response process after a step change in the setpoint or a disturbance recovery, the error sequence of the entire response process is collected; the integral absolute error, maximum overshoot, and settling time of the entire response process are calculated, and the baseline expected value and standard deviation of each index are established using weighted moving average filtering; the baseline expected value is added to the standard deviation by a preset multiple as the trigger boundary of the corresponding performance index; when the calculated value of any performance index in multiple consecutive response processes exceeds its corresponding trigger boundary, it is determined as performance degradation, a tuning trigger command is generated, the integral state of the original controller is frozen and the steady-state control output before identification is maintained, and the system identification is prepared.

3. The coking pressure and temperature adaptive control method according to claim 2, characterized in that, The critical gain and critical period of the main pressure channel are extracted through filtering, including: After entering the online identification mode, the integral state of the pressure proportional-integral-derivative controller is frozen, and the pressure control output is kept near the steady-state control output before identification. A bias relay excitation signal with asymmetric dead zone, output limiting and bias amount is superimposed on the steady-state control output. The limit loop signal generated by the pressure system after the bias relay is collected and smoothed by the state variable filter to separate the fundamental component and high-frequency noise. The in-phase component and quadrature component of the fundamental component are extracted by the quadrature demodulation algorithm, and the actual amplitude and oscillation angular frequency of the limit loop signal are calculated. Based on the describing function method, the critical gain and critical period of the main pressure channel are obtained by using the switching gain of the bias relay and the actual amplitude and oscillation angular frequency of the extracted limit loop signal.

4. The co-adaptive control method for coke forming pressure and temperature according to claim 1, characterized in that, Simultaneous identification of the first coupling transfer function of pressure on temperature includes: During the relay excitation of the pressure control loop to generate limit cycle oscillation, the control output sequence of the pressure control loop and the measurement feedback sequence of the temperature control loop are simultaneously acquired; the control output sequence is used as the excitation input, and the measurement feedback sequence is used as the forced response output. The recursive least squares method with a forgetting factor is used to perform online parameter estimation of the input-output sequence. A first-order inertial plus pure delay model consisting of gain coefficient, time constant and pure time delay is identified as the first coupling transfer function of pressure to temperature.

5. The coking pressure and temperature adaptive control method according to claim 1 or 4, characterized in that, Analyzing the purification signal yields the critical gain and critical period of the main temperature channel, including: The first coupling transfer function obtained by identification is discretized and convolved with the incremental control output of the current pressure control loop relative to the steady-state control output before identification to estimate the coupling interference of pressure control on the temperature object online. The original measurement feedback signal of the temperature control loop is synchronously subtracted from the coupling interference to compensate for the main superposition effect of pressure fluctuation on the temperature response and obtain the purification signal that mainly characterizes the characteristics of the main temperature channel. Zero-crossing detection and peak extraction are performed on the purification signal to calculate the amplitude and period of the limit loop of the temperature control loop and obtain the critical gain and critical period of the main temperature channel.

6. The coking pressure and temperature adaptive control method according to claim 1, characterized in that, Constructing the feedforward compensation decoupling matrix includes: Based on the extracted critical gain and critical period of the pressure main channel and the critical gain and critical period of the temperature main channel, the transfer functions of the pressure main channel and the temperature main channel are constructed respectively. These are then combined with the first coupling transfer function of pressure to temperature and the second coupling transfer function of temperature to pressure to construct the transfer function matrix of the dual-input dual-output system. Based on the transfer function matrix, the diagonal matrix decoupling principle is adopted. The transfer function of each main channel is used as the denominator, and the negative value of its corresponding coupling transfer function is used as the numerator. The realizability conditions are met by using the filter and the delay compensation factor. The feedforward compensator network parameters of pressure to temperature and temperature to pressure are calculated respectively. The feedforward compensator network parameters are combined to obtain the feedforward compensation decoupling matrix.

7. The coking pressure and temperature adaptive control method according to claim 1, characterized in that, Combining the feedforward compensation decoupling matrix with the controlled object includes: The transfer function matrix of the controlled object is multiplied with the feedforward compensation decoupling matrix to obtain the overall equivalent decoupling object matrix. The two single-loop subsystems after complete decoupling are then extracted from the matrix. These two single-loop subsystems are the pressure equivalent single-loop subsystem and the temperature equivalent single-loop subsystem, respectively.

8. The method for coordinated adaptive control of coke forming pressure and temperature according to claim 1, characterized in that, The proportional-integral-differential parameters of pressure and temperature are calculated based on the critical gain and critical period of the conversion, including: For the decoupled pressure equivalent single-loop subsystem and temperature equivalent single-loop subsystem, respectively, using the set improved gain margin and phase margin constraints, based on the frequency domain characteristics of the equivalent decoupled object matrix at the critical oscillation point, and combined with the obtained switching critical gain and switching critical period of the pressure control loop, the improved formula of the frequency domain tuning rule is used to calculate the proportional coefficient, integral time, and derivative time that meet the gain margin requirements of the pressure control loop; combined with the obtained switching critical gain and switching critical period of the temperature control loop, the proportional coefficient, integral time, and derivative time that meet the gain margin and response index of the temperature control loop are calculated, thus completing the solution of the dual-loop proportional-integral-derivative parameters of the decoupled model.

9. The method for coordinated adaptive control of coke forming pressure and temperature according to claim 1, characterized in that, A parameter fusion strategy is employed to seamlessly switch and update the newly calculated parameters with the current parameters, including: Set a transition time window and a fusion weight coefficient that increases over time. The fusion weight coefficient smoothly increases from 0 to 1 within the transition time window. In each control cycle within the transition time window, the old proportional-integral-derivative (PID) parameters and the newly calculated PID parameters are multiplied by the remaining weights and fused weights respectively, and then summed to obtain the actual PID parameters executed in the current control cycle. When the transition time window ends, the old PID parameters are discarded, and the online disturbance-free update of the controller parameters is completed.

10. A coordinated adaptive control system for coke forming pressure and temperature, comprising a processor and a memory, characterized in that, The memory stores a computer program, and the processor executes the computer program to implement the coking pressure and temperature adaptive control method as described in any one of claims 1-9.

Images