Self-adaptive adjusting method of intelligent pressure balance valve

Abstract

The invention relates to the technical field of automatic control of valves, and discloses a self-adaptive adjusting method of an intelligent pressure balance valve, which comprises the following steps of: performing robust preprocessing on valve front negative pressure and valve position signals in real time locally in a valve control cabinet, and identifying a valve-pipe network model online by using a normalized recursive least square method; residual low-frequency components are extracted through short-time spectrum analysis, disturbance estimation is generated in a self-adaptive mode, and meanwhile statistics is conducted on historical commands and position response quantization dead zones and friction indexes; the reverse opening degree is calculated based on an online model, a final control command is synthesized by combining disturbance feed-forward and dead zone reverse compensation, amplitude limiting and abnormity reporting are implemented, and therefore self-adaptive adjustment is conducted at each intelligent pressure balance valve of a bag-type dust collector pipe network, the fluctuation of the valve front negative pressure within the set target range is minimum, and the control accuracy is improved. And meanwhile, local self-adaptive compensation of sensor drift, slow disturbance and valve friction is realized, and the system is matched with a DCS (Distributed Control System) when an abnormal alarm is given.

Description

Technical Field

[0001] This invention relates to the field of valve automation control technology, specifically to an adaptive adjustment method for an intelligent pressure balancing valve. Background Technology

[0002] The steel industry is an important pillar industry for the development of the national economy and national defense. In recent years, the requirements for pollutant emissions in the steel industry have evolved from the treatment of sintering machines to the requirements for pollutant emissions throughout the entire steel industry process. Among these requirements, environmental dust removal for fugitive emissions has become more stringent, with particulate matter emissions required to be below 5 mg / m3. Correspondingly, dust removal technologies have become increasingly mature.

[0003] Baghouse dust collectors are widely used in the dust removal industry, accounting for approximately 70% of applications according to incomplete statistics. They work by installing dust collection hoods at emission points, with the dust collected from various branch points and then fed into the main pipe before entering the dust collector. Environmental particulate matter is collected and centrally processed through filter bags, and the treated air / flue gas is then discharged into the atmosphere via a fan. While dust collector technology is widely used, it also has certain drawbacks. These include large valve adjustment ranges in the dust collector's piping network, requiring significant manpower and resources; easy dust accumulation in the piping network; and poor dust collection efficiency at the suction points. To address the issues of large valve adjustment ranges, lack of intelligence, and dust escape from suction points, it is necessary to develop an intelligent valve control method for the piping network to improve the reliability of network control and contribute to the healthy and sustainable environmental development of enterprises. Summary of the Invention

[0004] This invention provides an adaptive adjustment method for an intelligent pressure balancing valve, which helps to solve the problems mentioned in the background art.

[0005] This invention provides the following technical solution: an adaptive adjustment method for an intelligent pressure balancing valve, comprising:

[0006] Set the sampling window and sampling interval, measure the valve inlet negative pressure and valve opening position at each sampling moment in real time, and obtain the last executed valve control command;

[0007] The median absolute deviation of the negative pressure sampling is calculated within a sliding window as a noise scale, and the original negative pressure is subjected to adaptive robust exponential smoothing based on this scale to obtain a purified negative pressure signal.

[0008] A regression vector with historical purification negative pressure and historical control commands as components is constructed, and the model parameters and parameter covariance are estimated online using the normalized recursive least squares method.

[0009] The identification residuals are calculated based on the online estimation model;

[0010] Spectral analysis of the residuals within a short time window is performed to extract the low-frequency energy ratio;

[0011] The smoothing coefficient is adaptively determined based on the low-frequency energy ratio, and the residual is weighted and smoothed based on the coefficient to generate a perturbation estimate.

[0012] Statistically analyze historical commands and location responses, extract the median of command amplitude in the case of no response as the dead zone threshold, and calculate the proportion of small command no-response events as a friction index.

[0013] The model parameters are estimated online and the inverse model is calculated based on the current purification negative pressure. The numerical lower bound is derived based on the parameter covariance to ensure the stability of the inverse solution.

[0014] The inverse model increment, disturbance feedforward term, and dead zone inverse compensation during friction anomalies are synthesized into the final control command, which implements actuator limiting and reports to the control system when the anomaly criteria are met.

[0015] Optionally, the step of calculating the median absolute deviation of the negative pressure samples within a sliding window as a noise scale, and performing adaptive robust exponential smoothing on the original negative pressure based on this scale to obtain a purified negative pressure signal, includes:

[0016] The median absolute deviation of the negative pressure samples is calculated as a noise scale within a preset sliding history window.

[0017] The current original negative pressure sample is subjected to exponential smoothing. The smoothing step size is determined by scaling the ratio of the absolute difference between the current sample and the previous filtered value to the noise scale using a robust constant. The robust constant is set to 3, thereby obtaining the purified negative pressure signal.

[0018] Optionally, the construction of a regression vector with historical purification negative pressure and historical control commands as components, and the online estimation of model parameters and parameter covariance using normalized recursive least squares method, includes:

[0019] Construct a regression vector whose components include the purification negative pressure of the previous time step and the corresponding control commands of the previous time step and the two time steps before that.

[0020] The parameter estimation vector and parameter covariance matrix are calculated using a normalized recursive least squares algorithm, and the update gain is determined in the normalized denominator form.

[0021] The bias term is obtained by slow smoothing the instantaneous residuals. The slow smoothing coefficient ranges from 0.95 to 0.99, and the smoothing result is limited by ±5 times the noise scale and used as the bias value for the model and control.

[0022] Optionally, the calculation of the identification residual based on the online estimation model includes:

[0023] Using the purified negative pressure as the observed value, subtract the model prediction value obtained from the regression vector and the online estimated parameters, and then subtract the bias value;

[0024] The resulting difference is used as the bias-corrected identification residual for subsequent spectrum analysis and perturbation estimation.

[0025] Optionally, the step of performing spectral analysis on the residuals within a short time window to extract the low-frequency energy ratio includes:

[0026] Obtain the identification residual sequence after bias correction, and perform discrete spectrum transformation on the residual sequence within a short time window. Calculate the low-frequency energy ratio according to the low-frequency sub-interval and the full-frequency interval of the spectrum.

[0027] The low-frequency sub-interval is defined as the set of frequency points below a preset boundary frequency.

[0028] Optionally, the step of adaptively determining a smoothing coefficient based on the low-frequency energy ratio, and then weighting and smoothing the residuals based on this coefficient to generate a perturbation estimate, includes:

[0029] The low-frequency energy ratio is linearly mapped to a smoothing coefficient in the range of 0.5 to 0.95 as the current filter weight, and the historical disturbance estimate and the current identification residual are weighted and smoothed with this weight to output an accurate disturbance estimate for feedforward compensation.

[0030] Optionally, the statistical analysis of historical commands and location responses includes extracting the median command amplitude in the absence of response as a dead zone threshold, and calculating the proportion of small command no-response events as a friction index, including:

[0031] Scan a preset short-term historical sampling window to identify events in which command changes exist but valve position changes are less than the device position resolution, and calculate the median of the command change amplitude corresponding to these events as the dead zone threshold.

[0032] The ratio of the number of events that meet the above non-response conditions to the total number of samples in the same window is used as the friction ratio indicator.

[0033] The historical median of the friction ratio index plus twice the interquartile range is used as the threshold for judging friction anomalies.

[0034] Optionally, the step-by-step opening increment required for calculating the inverse model based on the online estimated model parameters and the current purification negative pressure, and the lower bound of the value derived based on the parameter covariance to ensure the stability of the inverse solution, includes:

[0035] The model parameters estimated online, the current and previous purification negative pressure values, and the control command from the previous moment are substituted into the inverse mapping of the one-step prediction model to obtain the opening increment required to make the prediction output approach the expected reference value.

[0036] The lower bound of the value is determined by taking the square root of the largest diagonal element of the parameter covariance matrix and multiplying it by a coefficient of 3, and is used as a small guarantee quantity in the inverse solution operation;

[0037] The desired reference value is obtained by mixing the upper-level reference value with the short-term smoothed reference of the previous time step, according to a given smoothing coefficient, and adding a bias value. The smoothing coefficient is between 0 and 1.

[0038] Optionally, the step of synthesizing the inverse model increment, disturbance feedforward term, and dead-zone inverse compensation during friction anomalies into a final control command, implementing actuator limiting, and reporting to the control system when the anomaly criterion is met includes:

[0039] The step opening increment obtained by the inverse model is multiplied by the perturbation feedforward gain determined algebraically by the identification parameters and the perturbation estimate to obtain the perturbation feedforward term. When the friction index exceeds the threshold, the dead zone inverse compensation is constructed according to the dead zone threshold and the smoothed sign function. The three terms are combined to form the basic command.

[0040] The underlying command is truncated within the range allowed by the executor and used as the final control command;

[0041] The smoothing scale used to construct the smooth sign function takes the maximum value among the noise scale, the deterministic upper bound of the parameter uncertainty, and a local minimum value.

[0042] The perturbation feedforward gain is determined by a combination of the first input coefficient of the identification parameter, the sum of squares of each input coefficient, and the lower bound of the value.

[0043] Optionally, the step of synthesizing the inverse model increment, disturbance feedforward term, and dead-zone inverse compensation during friction anomalies into a final control command, implementing actuator limiting, and reporting to the control system when the anomaly criterion is met further includes:

[0044] During normal and stable system operation, a historical sample set is selected as the historical statistical window and a judgment window is set.

[0045] Obtain the friction ratio index sequence and its anomaly judgment threshold. If there are consecutive sampling points within the judgment window that satisfy the friction ratio index exceeding the threshold, trigger a friction anomaly alarm and report it to DCS.

[0046] Calculate the sequence of the largest eigenvalues ​​of the parameter covariance matrix within the historical statistics window, and set the threshold for abnormal identification uncertainty of the parameter based on its historical median and three times the absolute deviation of the median. If the current largest eigenvalue exceeds the limit continuously within the judgment window, an alarm for abnormal identification uncertainty will be triggered and reported.

[0047] The median and three times the absolute deviation of the median are calculated based on the absolute value sequence of the disturbance estimation using historical samples to set the abnormal threshold of the disturbance estimation. If the absolute value of the disturbance estimation exceeds the limit continuously within the judgment window, an abnormal disturbance estimation alarm is triggered and reported.

[0048] The present invention has the following beneficial effects:

[0049] 1. By synthesizing the inverse model increment, disturbance feedforward term, and dead-zone inverse compensation during friction anomalies into the final control command, implementing actuator limiting, and reporting to the control system when the anomaly criteria are met, this scheme integrates online identification, spectral residual decomposition, disturbance feedforward, dead-zone detection, and final limiting functions all locally in the valve control cabinet, achieving decentralized autonomous control. Its advantages include: local closed-loop reduces reliance on continuous communication with the DCS and wiring costs, facilitating distributed deployment and modular expansion; local real-time closed-loop has a shorter response path and lower latency, exhibiting stronger adaptability to sudden local disturbances, thereby reducing the risk of dust escape or instantaneous exceedances; integrated design facilitates single-point testing and fault isolation in the field, improving maintenance efficiency.

[0050] 2. By estimating the sensor noise scale using the median absolute deviation within a sliding window and using this scale as a normalization benchmark to drive adaptive exponential smoothing, noise spikes and real dynamic changes can be distinguished in real-time measurements. The robustness of MAD to outliers is much higher than that of variance estimation, and it can provide a stable noise scale estimate in field measurements with short-term spikes or impulse noise. This prevents subsequent filters from over-responding or distorting due to single-point anomalies, thereby improving the stability and reliability of the entire identification and control link. The adaptive step size based on the noise scale avoids the problem of the fixed filter constant being unsuitable for different operating conditions. The system has agile tracking capabilities under low-noise conditions, while automatically converging to a conservative step size to suppress transient errors when spikes or measurement changes occur. This directly reduces the risk of misoperation or incorrect compensation.

[0051] 3. By selecting a discrete second-order ARX regression vector and updating it with a normalized RLS, the aim is to describe the local dynamics of the valve-pipeline network with the lowest parameter dimension and achieve numerically stable online identification. Low-order parameterization keeps the model simple, reduces the variance of parameter estimation, and improves the convergence speed of online estimation, making it suitable for valve control cabinets with limited field computing resources. The normalized gain RLS structure introduces a dimensional adaptive term in the denominator, significantly reducing the numerical instability caused by near-singular or high condition number regression matrices, ensuring that the estimation remains stable even when the correlation of short-term data changes. By incorporating a bias term to absorb slow-varying drift, the drift can be avoided from causing systematic bias to the model parameters, thus making the estimation robust in the long term. This step ensures that the model can describe the dominant dynamics while taking into account both implementation complexity and numerical robustness, providing a reliable model parameter foundation for real-time control.

[0052] 4. By performing short-time spectrum analysis on the online identification residuals and calculating the proportion of low-frequency energy, slow-varying disturbances and high-frequency noise components can be distinguished: the spectrum domain discrimination can effectively distinguish slow-varying disturbances such as filter bag dust accumulation and gradual changes in pipeline resistance from measurement noise or short-term sudden changes in operating conditions, enabling the system to adopt smooth feedforward compensation for slow-varying disturbances and suppression strategies for high-frequency noise, thereby balancing steady-state performance and response speed; the short-time spectrum analysis itself can be calculated locally in real time, suitable for embedded execution, and can provide additional frequency domain indicators for alarm logic; overall, this step maps the time-domain residuals to frequency-domain features, giving disturbance discrimination a clearer statistical basis and improving the pertinence and accuracy of disturbance compensation;

[0053] 5. By mapping the low-frequency energy ratio obtained from the spectrum to a smoothing coefficient range, and using this coefficient to weight and smooth the residuals to output the disturbance estimate: mapping the low-frequency energy ratio linearly or nonlinearly to a smoothing coefficient, the system achieves the ability to adaptively adjust the filter bandwidth based on the observed data. This enables the system to smoothly extract steady-state disturbances when slow-varying disturbances dominate and avoid feeding noise forward as disturbances when high-frequency disturbances dominate. When the disturbance estimate is used as a feedforward compensation input, it can significantly reduce steady-state error and improve the stability of energy saving and pollution control. Overall, a data-driven disturbance identification and filtering strategy is realized, which enhances the controller's compensation capability for slow-varying operating conditions while avoiding noise amplification.

[0054] 6. By identifying events in historical samples where "the command changes but the valve position does not respond" and statistically analyzing the median of the command amplitude as the dead zone threshold, and using the frequency of such events as a friction index, this step utilizes historical statistics (median, frequency) instead of instantaneous thresholds. This helps suppress occasional false alarms and reflects the long-term behavior characteristics of the equipment. Furthermore, quantifying friction as a proportional measure makes maintenance criteria repeatable and verifiable, thus providing objective evidence for maintenance decisions. Finally, the statistical analysis of the dead zone threshold also supports progressive degradation detection and maintenance plan optimization, avoiding frequent false alarms caused by relying solely on threshold triggers. In summary, this step transforms mechanical nonlinearity (dead zone, jamming) from a qualitative judgment into a quantitative indicator, facilitating automatic compensation and engineering processing.

[0055] 7. By using the parameters obtained from online identification to perform a one-step inverse solution to obtain the desired opening increment, and using the parameter covariance to generate a numerical lower bound to prevent numerical instability of the inverse solution: the closed-loop inverse solution avoids complex online optimization, has low computational cost, and is suitable for real-time operation of embedded controllers; the lower bound derived from the covariance directly maps the identification uncertainty into numerical robustness assurance, preventing command amplification or denominator approaching zero caused by unreliable parameter estimation; the model uncertainty is transparently integrated into the control decision, enabling the controller to exhibit conservatism rather than reckless behavior under uncertain conditions, thus improving system safety; this step improves the robustness of the inverse solution while maintaining efficient computation, achieving real-time precise control;

[0056] 8. By synthesizing the inverse model increment, disturbance feedforward term, and dead-zone inverse compensation under abnormal friction conditions into a basic command, and then issuing it for execution after safety limiting, the combination of feedforward and inverse model takes into account both dynamic response and steady-state error elimination. Feedforward cancellation of slow-varying disturbances can significantly reduce closed-loop integral action and oscillation. Dead-zone inverse compensation is activated when friction indicators exceed limits, which can avoid excessive adjustment of the overall controller when dealing with valve body nonlinearity locally, thereby extending equipment life and avoiding control jitter. Finally, the limiting and anomaly reporting mechanism ensures that the system degrades to a safe mode under extreme conditions and notifies operation and maintenance, increasing engineering controllability. In summary, this synthesis strategy establishes a quantifiable trade-off mechanism between performance and safety, taking into account the reliability of field execution.

[0057] 9. By automatically setting anomaly thresholds and defining a decision window based on statistical measures such as historical median, interquartile range, or median absolute deviation for alarm triggering, the advantages are: statistical thresholds can adapt to different field baselines and long-term drift, thereby significantly reducing false alarm rates and the frequency of maintenance intervention; using historical statistics as a reference makes the alarm logic traceable, facilitating auditing and subsequent optimization; the decision window mechanism avoids false alarms caused by instantaneous noise and supports continuous over-limit strategies, thus more accurately reflecting the true fault trend; therefore, this statistical reporting logic improves the reliability of alarms and provides data support for asset management. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the basic process of the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] Example 1, refer to Figure 1 An adaptive adjustment method for an intelligent pressure balancing valve, comprising:

[0061] Set the sampling window and sampling interval, measure the valve inlet negative pressure and valve opening position at each sampling moment in real time, and obtain the last executed valve control command;

[0062] The median absolute deviation of the negative pressure sampling is calculated within a sliding window as a noise scale, and the original negative pressure is subjected to adaptive robust exponential smoothing based on this scale to obtain a purified negative pressure signal.

[0063] A regression vector with historical purification negative pressure and historical control commands as components is constructed, and the model parameters and parameter covariance are estimated online using the normalized recursive least squares method.

[0064] The identification residuals are calculated based on the online estimation model;

[0065] Spectral analysis of the residuals within a short time window is performed to extract the low-frequency energy ratio;

[0066] The smoothing coefficient is adaptively determined based on the low-frequency energy ratio, and the residual is weighted and smoothed based on the coefficient to generate a perturbation estimate.

[0067] Statistically analyze historical commands and location responses, extract the median of command amplitude in the case of no response as the dead zone threshold, and calculate the proportion of small command no-response events as a friction index.

[0068] The model parameters are estimated online and the inverse model is calculated based on the current purification negative pressure. The numerical lower bound is derived based on the parameter covariance to ensure the stability of the inverse solution.

[0069] By synthesizing the inverse model increment, disturbance feedforward term, and dead-zone inverse compensation during friction anomalies into the final control command, implementing actuator limiting, and reporting to the control system when anomaly criteria are met, this scheme integrates online identification, spectral residual decomposition, disturbance feedforward, dead-zone detection, and final limiting functions all locally in the valve control cabinet, achieving decentralized autonomous control. Its advantages include: local closed-loop reduces reliance on continuous communication with the DCS and wiring costs, facilitating distributed deployment and modular expansion; local real-time closed-loop has a shorter response path and lower latency, exhibiting stronger adaptability to sudden local disturbances, thereby reducing the risk of dust escape or instantaneous exceedances; and the integrated design facilitates single-point testing and fault isolation in the field, improving maintenance efficiency.

[0070] The step of calculating the median absolute deviation of the negative pressure sampling within a sliding window as a noise scale, and performing adaptive robust exponential smoothing on the original negative pressure based on this scale to obtain a purified negative pressure signal, includes:

[0071] The median absolute deviation of the negative pressure samples is calculated as a noise scale within a preset sliding history window.

[0072] The current original negative pressure sample is subjected to exponential smoothing. The smoothing step size is determined by scaling the ratio of the absolute difference between the current sample and the previous filtered value to the noise scale using a robust constant. The robust constant is set to 3, thereby obtaining the purified negative pressure signal.

[0073] The construction of a regression vector with historical purification negative pressure and historical control commands as components, and the online estimation of model parameters and parameter covariance using the normalized recursive least squares method, includes:

[0074] Construct a regression vector whose components include the purification negative pressure of the previous time step and the corresponding control commands of the previous time step and the two time steps before that.

[0075] The parameter estimation vector and parameter covariance matrix are calculated using a normalized recursive least squares algorithm, and the update gain is determined in the normalized denominator form.

[0076] The bias term is obtained by slow smoothing the instantaneous residuals. The slow smoothing coefficient ranges from 0.95 to 0.99, and the smoothing result is limited by ±5 times the noise scale and used as the bias value for the model and control.

[0077] The calculation of the identification residuals based on the online estimation model includes:

[0078] Using the purified negative pressure as the observed value, subtract the model prediction value obtained from the regression vector and the online estimated parameters, and then subtract the bias value;

[0079] The resulting difference is used as the bias-corrected identification residual for subsequent spectrum analysis and perturbation estimation.

[0080] The step of performing spectral analysis on the residuals within a short time window to extract the low-frequency energy ratio includes:

[0081] Obtain the identification residual sequence after bias correction, and perform discrete spectrum transformation on the residual sequence within a short time window. Calculate the low-frequency energy ratio according to the low-frequency sub-interval and the full-frequency interval of the spectrum.

[0082] The low-frequency sub-interval is defined as the set of frequency points below a preset boundary frequency.

[0083] The step of adaptively determining a smoothing coefficient based on the low-frequency energy ratio, and then using this coefficient to perform weighted smoothing of the residuals to generate a perturbation estimate, includes:

[0084] The low-frequency energy ratio is linearly mapped to a smoothing coefficient in the range of 0.5 to 0.95 as the current filter weight, and the historical disturbance estimate and the current identification residual are weighted and smoothed with this weight to output an accurate disturbance estimate for feedforward compensation.

[0085] The statistical historical commands and location responses are used to extract the median command amplitude in the absence of response as a dead zone threshold, and the proportion of small command no-response events is calculated as a friction index, including:

[0086] Scan a preset short-term historical sampling window to identify events in which command changes exist but valve position changes are less than the device position resolution, and calculate the median of the command change amplitude corresponding to these events as the dead zone threshold.

[0087] The ratio of the number of events that meet the above non-response conditions to the total number of samples in the same window is used as the friction ratio indicator.

[0088] The historical median of the friction ratio index plus twice the interquartile range is used as the threshold for judging friction anomalies.

[0089] The model parameters based on online estimation and the one-step opening increment required to calculate the inverse model of the current purification negative pressure, and the numerical lower bound derived based on the parameter covariance to ensure the stability of the inverse solution, include:

[0090] The model parameters estimated online, the current and previous purification negative pressure values, and the control command from the previous moment are substituted into the inverse mapping of the one-step prediction model to obtain the opening increment required to make the prediction output approach the expected reference value.

[0091] The lower bound of the value is determined by taking the square root of the largest diagonal element of the parameter covariance matrix and multiplying it by a coefficient of 3, and is used as a small guarantee quantity in the inverse solution operation;

[0092] The desired reference value is obtained by mixing the upper-level reference value with the short-term smoothed reference of the previous time step, according to a given smoothing coefficient, and adding a bias value. The smoothing coefficient is between 0 and 1.

[0093] The process of synthesizing the inverse model increment, disturbance feedforward term, and dead-zone inverse compensation during friction anomalies into a final control command, implementing actuator limiting, and reporting to the control system when an anomaly criterion is met includes:

[0094] The step opening increment obtained by the inverse model is multiplied by the perturbation feedforward gain determined algebraically by the identification parameters and the perturbation estimate to obtain the perturbation feedforward term. When the friction index exceeds the threshold, the dead zone inverse compensation is constructed according to the dead zone threshold and the smoothed sign function. The three terms are combined to form the basic command.

[0095] The underlying command is truncated within the range allowed by the executor and used as the final control command;

[0096] The smoothing scale used to construct the smooth sign function takes the maximum value among the noise scale, the deterministic upper bound of the parameter uncertainty, and a local minimum value.

[0097] The perturbation feedforward gain is determined by a combination of the first input coefficient of the identification parameter, the sum of squares of each input coefficient, and the lower bound of the value.

[0098] The process of synthesizing the inverse model increment, disturbance feedforward term, and dead-zone inverse compensation during friction anomalies into the final control command, implementing actuator limiting, and reporting to the control system when the anomaly criterion is met also includes:

[0099] During normal and stable system operation, a historical sample set is selected as the historical statistical window and a judgment window is set.

[0100] Obtain the friction ratio index sequence and its anomaly judgment threshold. If there are consecutive sampling points within the judgment window that satisfy the friction ratio index exceeding the threshold, trigger a friction anomaly alarm and report it to DCS.

[0101] Calculate the sequence of the largest eigenvalues ​​of the parameter covariance matrix within the historical statistics window, and set the threshold for abnormal identification uncertainty of the parameter based on its historical median and three times the absolute deviation of the median. If the current largest eigenvalue exceeds the limit continuously within the judgment window, an alarm for abnormal identification uncertainty will be triggered and reported.

[0102] The median and three times the absolute deviation of the median are calculated based on the absolute value sequence of the disturbance estimation using historical samples to set the abnormal threshold of the disturbance estimation. If the absolute value of the disturbance estimation exceeds the limit continuously within the judgment window, an abnormal disturbance estimation alarm is triggered and reported.

[0103] Example 2: An adaptive adjustment method for an intelligent pressure balancing valve, further comprising:

[0104] The step of calculating the median absolute deviation of the negative pressure sampling within a sliding window as a noise scale, and performing adaptive robust exponential smoothing on the original negative pressure based on this scale to obtain a purified negative pressure signal, includes:

[0105] In the preset sliding history window The noise scale for calculating the median absolute deviation estimate. :

[0106] ;

[0107] in: :window Inner One original negative pressure sample; Median operator; The median absolute deviation is mapped to a constant equivalent standard deviation under a Gaussian distribution; the sliding history window Defined as a data set consisting of several recent time points, used for calculating local statistical features;

[0108] For the current original sample Adaptive exponential smoothing is applied, which automatically reduces the step size when spikes occur, while quickly tracking during smooth or small fluctuations:

[0109] ;

[0110] in: The purified negative pressure signal obtained after adaptive exponential smoothing; The current sampling time point; Sampling interval; Sampling interval; Adaptive step size, with values ​​of:

[0111] ;

[0112] in: : Robustness constant, used to normalize the amplitude to the noise scale, taking The value represents the three-fold noise scale determination, a threshold commonly used in statistics. By estimating the sensor noise scale using the median absolute deviation within a sliding window and using this scale as a normalization benchmark to drive adaptive exponential smoothing, noise spikes and true dynamic changes can be distinguished in real-time measurements. MAD's robustness to outliers is far superior to variance estimation, providing stable noise scale estimates in field measurements with short-term spikes or impulse noise. This prevents subsequent filters from over-responding or distorting due to single-point anomalies, thus improving the stability and reliability of the entire identification and control chain. The adaptive step size based on the noise scale avoids the mismatch problem of fixed filter constants for different operating conditions. The system has agile tracking capabilities under low-noise conditions, while automatically converging to a conservative step size to suppress transient errors during spikes or measurement abrupt changes. This directly reduces the risk of misoperation or incorrect compensation.

[0113] The construction of a regression vector with historical purification negative pressure and historical control commands as components, and the online estimation of model parameters and parameter covariance using the normalized recursive least squares method, includes:

[0114] Constructing a regression vector and includes bias terms The regression model form:

[0115] , ;

[0116] in: , At the current moment The control signals for the balancing valve executed at the previous moment and the two moments before; : Parameter vector ;

[0117] Update parameters using NRLS With covariance :

[0118] ;

[0119] ;

[0120] ;

[0121] in: : NRLS gain vector; : Parameter estimation vector, which is the model updated via NRLS. The value; : Parametric covariance matrix;

[0122] Among them, the bias term Determined through slow smoothing of residuals:

[0123] Calculate the instantaneous residual: ;

[0124] in: The purified negative pressure signal obtained after adaptive exponential smoothing;

[0125] Bias slow smoothing estimation: ;

[0126] in: The slow smoothing coefficient ranges from 0.95 to 0.99. A larger value emphasizes historical data and reflects the capture of only slowly varying components.

[0127] Amplitude limiting protection based on noise scale:

[0128] ;

[0129] in: Noise scale; Limiting factor, take This is a commonly used conservative value in statistics for covering very few abnormalities; :Will Cut off to interval ; The final bias values ​​used in the model and control are determined by selecting a discrete second-order ARX regression vector and updating it with a normalized RLS. This aims to describe the local dynamics of the valve-pipeline network with the lowest parameter dimension and achieve numerically stable online identification. Low-order parameterization keeps the model simple, reduces the variance of parameter estimation, and improves the convergence speed of online estimation, making it suitable for valve control cabinets with limited field computing resources. The normalized gain RLS structure introduces a dimensional adaptive term in the denominator, significantly reducing the numerical instability caused by near-singular or high condition number regression matrices, ensuring that the estimation remains stable even when the correlation of short-term data changes. By incorporating a bias term to absorb slow-varying drift, the drift can be avoided from causing systematic biases to the model parameters, thus making the estimation robust in the long term. This step ensures that the model can describe the dominant dynamics while balancing implementation complexity and numerical robustness, providing a reliable model parameter foundation for real-time control.

[0130] The calculation of the identification residuals based on the online estimation model includes:

[0131] The residuals were calculated and identified after bias correction:

[0132] ;

[0133] in: The purified negative pressure signal obtained after adaptive exponential smoothing; The constructed regression vector; : Parameter estimation vector; : Bias term.

[0134] The step of performing spectral analysis on the residuals within a short time window to extract the low-frequency energy ratio includes:

[0135] Obtain the identification residual after bias correction ;

[0136] For identifying residuals Perform a short-time FFT to obtain the short-time FFT spectrum of the residuals. The short-time FFT mentioned above is a standard algorithm that is currently widely used.

[0137] Calculate the low-frequency energy ratio:

[0138] ;

[0139] in: The index variable for the frequency component is taken from the set of discrete frequency points obtained by short-time FFT and is a calculation variable; Residual spectrum The full frequency domain set, that is, the entire frequency range of the FFT output; : The set of low-frequency sub-intervals in the middle is defined as those below the preset boundary frequency. A set of frequency points is used to extract the low-frequency components of the system. By performing short-time spectrum analysis on the online identification residuals and calculating the proportion of low-frequency energy, slow-varying disturbances and high-frequency noise components can be distinguished. The spectrum domain discrimination can effectively distinguish slow-varying disturbances such as filter bag dust accumulation and gradual changes in pipeline resistance from measurement noise or short-term abrupt changes in operating conditions. This allows the system to adopt smooth feedforward compensation for slow-varying disturbances and suppression strategies for high-frequency noise, thus balancing steady-state performance and response speed. The short-time spectrum analysis itself can be calculated locally in real time, making it suitable for embedded execution and providing additional frequency domain indicators for alarm logic. Overall, this step maps the time-domain residuals to frequency-domain features, giving the disturbance discrimination a clearer statistical basis and improving the pertinence and accuracy of disturbance compensation.

[0140] The step of adaptively determining a smoothing coefficient based on the low-frequency energy ratio, and then using this coefficient to perform weighted smoothing of the residuals to generate a perturbation estimate, includes:

[0141] The smoothing coefficient is determined by the low-frequency energy ratio:

[0142] ;

[0143] Weighted smoothing of the residuals yields the disturbance estimate:

[0144] ;

[0145] in: The identification residual after bias correction. The low-frequency energy ratio derived from the spectrum is mapped to a smoothing coefficient range, and this coefficient is used to weight and smooth the residual to output a disturbance estimate. Mapping the low-frequency energy ratio linearly or nonlinearly to a smoothing coefficient enables the adaptive adjustment of the filter bandwidth based on observation data. This allows the system to smoothly extract steady-state disturbances when slow-varying disturbances dominate and avoid feeding noise forward as disturbance when high-frequency disturbances dominate. When the disturbance estimate is used as a feedforward compensation input, it significantly reduces steady-state error and improves energy-saving and pollution control stability. Overall, a data-driven disturbance identification and filtering strategy is implemented, enhancing the controller's compensation capability for slow-varying operating conditions while avoiding noise amplification.

[0146] The statistical historical commands and location responses are used to extract the median command amplitude in the absence of response as a dead zone threshold, and the proportion of small command no-response events is calculated as a friction index, including:

[0147] The system identifies historical command records of instances where control commands were issued but the balance valve failed to respond, and sets a dead zone threshold accordingly. :

[0148] ;

[0149] in: Historical command changes ; The position of the balance valve changes. ; : Preset short-term historical sampling window; Balance valve position resolution;

[0150] Calculate the friction ratio index:

[0151] ;

[0152] Set the threshold for judging friction anomalies:

[0153] ;

[0154] in: : The count of events that satisfy the condition; The span of the middle 50% frequency samples. , This represents the frequency value at the 75th percentile. This represents the frequency value at the 25th percentile. By identifying events in historical samples where "the command changes but the valve position does not respond" and statistically analyzing the median of the command amplitude as the dead zone threshold, and using the frequency of such events as a friction index, this step utilizes historical statistics (median, frequency) instead of instantaneous thresholds. This helps suppress occasional false alarms and reflects the long-term behavior characteristics of the equipment. Furthermore, quantifying friction as a proportional measure makes maintenance criteria repeatable and verifiable, thus providing objective evidence for maintenance decisions. Finally, the statistical analysis of the dead zone threshold also supports progressive degradation detection and maintenance plan optimization, avoiding frequent false alarms caused by relying solely on threshold triggers. In summary, this step transforms mechanical nonlinearity (dead zone, jamming) from a qualitative judgment into a quantitative indicator, facilitating automatic compensation and engineering processing.

[0155] The model parameters based on online estimation and the one-step opening increment required to calculate the inverse model of the current purification negative pressure, and the numerical lower bound derived based on the parameter covariance to ensure the stability of the inverse solution, include:

[0156] Calculate the required aperture increment for one-step prediction based on the identification parameters:

[0157] ;

[0158] in: , , , : Identify parameters, derived from parameter estimation vectors ; The purified negative pressure signal obtained after adaptive exponential smoothing; Expected reference value;

[0159] Derive the lower bound using covariance. :

[0160] ;

[0161] in: : Parametric covariance matrix;

[0162] The expected reference value is:

[0163] ;

[0164] in: Current moment Real-time pressure measurement value; : Preset smoothing coefficient, value range ; : Bias term; The smoothed reference value from the previous moment, calculated using a short-term moving average:

[0165] ;

[0166] in: The preset historical sample length. A one-step inverse solution is performed using parameters obtained from online identification to obtain the desired opening increment, and a lower bound is generated using parameter covariance to prevent numerical instability in the inverse solution. The closed-loop inverse solution avoids complex online optimization, has low computational cost, and is suitable for real-time operation of embedded controllers. The lower bound derived from the covariance directly maps identification uncertainty to numerical robustness assurance, preventing command amplification or near-zero denominator problems caused by unreliable parameter estimation. Transparently integrating model uncertainty into control decisions allows the controller to exhibit conservatism rather than reckless behavior under uncertain conditions, improving system safety. This step enhances the robustness of the inverse solution while maintaining efficient computation, achieving real-time precise control.

[0167] The process of synthesizing the inverse model increment, disturbance feedforward term, and dead-zone inverse compensation during friction anomalies into a final control command, implementing actuator limiting, and reporting to the control system when an anomaly criterion is met includes:

[0168] The basic command for generating the inverse model increment, disturbance feedforward, and dead zone inverse compensation is as follows:

[0169] ;

[0170] in: At the current moment The control signal executed on the balancing valve at the previous moment; : Predict the required opening increment in one step; : Perturbation feedforward gain; Perturbation estimation for residual weighted smoothing; ;

[0171] in: Indicator functions are used for logical decisions; Friction ratio index; Threshold for determining friction anomalies; : Smoothing sign function; Dead zone threshold;

[0172] The perturbation feedforward gain is determined by the identification parameters: ;

[0173] in: , , , : Identify parameters, derived from parameter estimation vectors ; Derive the lower bound using covariance;

[0174] The smoothing sign function ,in The value is taken as the deterministic upper bound of the measurement noise scale and parameter uncertainty:

[0175] ;in Noise scale; : Parametric covariance matrix; The lower bound of the value is set to a very small value. ;

[0176] Basic commands Limiting the amplitude:

[0177] ;

[0178] in: : Truncation operator, This defines the allowable control range for the actuator. A basic command is synthesized by combining the inverse model increment, disturbance feedforward term, and dead-zone inverse compensation under abnormal friction conditions, and then executed after safety limiting. The combination of feedforward and inverse model simultaneously considers dynamic response and steady-state error elimination. Feedforward cancellation of slowly varying disturbances significantly reduces closed-loop integral action and oscillation. Dead-zone inverse compensation is activated when friction parameters exceed limits, preventing excessive adjustment of the overall controller when dealing with valve body nonlinearity locally, thereby extending equipment life and avoiding control jitter. Finally, the limiting and anomaly reporting mechanism ensures that the system degrades to a safe mode under extreme conditions and notifies maintenance, increasing engineering controllability. In summary, this synthesis strategy establishes a quantifiable trade-off between performance and safety, while also considering on-site execution reliability.

[0179] The process of synthesizing the inverse model increment, disturbance feedforward term, and dead-zone inverse compensation during friction anomalies into the final control command, implementing actuator limiting, and reporting to the control system when the anomaly criterion is met also includes:

[0180] A historical sample set was selected to form a historical statistical window during the normal and stable operation of the system. The judgment window is set as Length is ;

[0181] Obtain the calculated friction ratio index And obtain the threshold for determining friction anomalies. ;

[0182] If in the decision window Internal satisfaction ,

[0183] This is considered an abnormal friction event, triggering an alarm and reporting it to the DCS.

[0184] In historical statistics window Internal calculation of the maximum eigenvalue sequence of historical samples ,in The parameter is the covariance matrix;

[0185] Set the threshold for identifying anomalies in parameter uncertainty:

[0186] ;

[0187] like Continuous existence If a sample is taken, the parameter identification uncertainty is determined to be abnormal, an alarm is triggered and reported to the DCS;

[0188] Set the threshold for determining anomalies in disturbance estimation:

[0189] ;

[0190] If in the decision window absolute value of perturbation estimate If the disturbance estimation is deemed abnormal, an alarm is triggered and reported to the DCS. The system automatically sets anomaly thresholds and defines a decision window for alarm triggering using statistical measures such as historical median, interquartile range, or median absolute deviation. Its advantages include: statistical thresholds can adapt to different field baselines and long-term drift, significantly reducing false alarm rates and maintenance intervention frequency; using historical statistics as a reference ensures traceability of the alarm logic, facilitating auditing and subsequent optimization; the decision window mechanism avoids false alarms caused by instantaneous noise and supports continuous over-limit strategies, thus more accurately reflecting the true fault trend; therefore, this statistical reporting logic improves alarm reliability and provides data support for asset management.

[0191] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0192] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An adaptive adjustment method for an intelligent pressure balancing valve, characterized in that, include: Set the sampling window and sampling interval, measure the valve inlet negative pressure and valve opening position at each sampling moment in real time, and obtain the last executed valve control command; The median absolute deviation of the negative pressure sampling is calculated within a sliding window as a noise scale, and the original negative pressure is subjected to adaptive robust exponential smoothing based on this scale to obtain a purified negative pressure signal. A regression vector with historical purification negative pressure and historical control commands as components is constructed, and the model parameters and parameter covariance are estimated online using the normalized recursive least squares method. The identification residuals are calculated based on the online estimation model; Spectral analysis of the residuals within a short time window is performed to extract the low-frequency energy ratio; The smoothing coefficient is adaptively determined based on the low-frequency energy ratio, and the residual is weighted and smoothed based on the coefficient to generate a perturbation estimate. Statistically analyze historical commands and location responses, extract the median of command amplitude in the case of no response as the dead zone threshold, and calculate the proportion of small command no-response events as a friction index. The model parameters are estimated online and the inverse model is calculated based on the current purification negative pressure. The numerical lower bound is derived based on the parameter covariance to ensure the stability of the inverse solution. The inverse model increment, disturbance feedforward term, and dead zone inverse compensation during friction anomalies are synthesized into the final control command, which implements actuator limiting and reports to the control system when the anomaly criteria are met.

2. The adaptive adjustment method for an intelligent pressure balancing valve according to claim 1, characterized in that, The step of calculating the median absolute deviation of the negative pressure sampling within a sliding window as a noise scale, and performing adaptive robust exponential smoothing on the original negative pressure based on this scale to obtain a purified negative pressure signal, includes: The median absolute deviation of the negative pressure samples is calculated as a noise scale within a preset sliding history window. The current original negative pressure sample is subjected to exponential smoothing. The smoothing step size is determined by scaling the ratio of the absolute difference between the current sample and the previous filtered value to the noise scale using a robust constant. The robust constant is set to 3, thereby obtaining the purified negative pressure signal.

3. The adaptive adjustment method for an intelligent pressure balancing valve according to claim 1, characterized in that, The construction of a regression vector with historical purification negative pressure and historical control commands as components, and the online estimation of model parameters and parameter covariance using the normalized recursive least squares method, includes: Construct a regression vector whose components include the purification negative pressure of the previous time step and the corresponding control commands of the previous time step and the two time steps before that. The parameter estimation vector and parameter covariance matrix are calculated using a normalized recursive least squares algorithm, and the update gain is determined in the normalized denominator form. The bias term is obtained by slow smoothing the instantaneous residuals. The slow smoothing coefficient ranges from 0.95 to 0.99, and the smoothing result is limited by ±5 times the noise scale and used as the bias value for the model and control.

4. The adaptive adjustment method for an intelligent pressure balancing valve according to claim 1, characterized in that, The calculation of the identification residuals based on the online estimation model includes: Using the purified negative pressure as the observed value, subtract the model prediction value obtained from the regression vector and the online estimated parameters, and then subtract the bias value; The resulting difference is used as the bias-corrected identification residual for subsequent spectrum analysis and perturbation estimation.

5. The adaptive adjustment method for an intelligent pressure balancing valve according to claim 1, characterized in that, The step of performing spectral analysis on the residuals within a short time window to extract the low-frequency energy ratio includes: Obtain the identification residual sequence after bias correction, and perform discrete spectrum transformation on the residual sequence within a short time window. Calculate the low-frequency energy ratio according to the low-frequency sub-interval and the full-frequency interval of the spectrum. The low-frequency sub-interval is defined as the set of frequency points below a preset boundary frequency.

6. The adaptive adjustment method for an intelligent pressure balancing valve according to claim 1, characterized in that, The step of adaptively determining a smoothing coefficient based on the low-frequency energy ratio, and then using this coefficient to perform weighted smoothing of the residuals to generate a perturbation estimate, includes: The low-frequency energy ratio is linearly mapped to a smoothing coefficient in the range of 0.5 to 0.95 as the current filter weight, and the historical disturbance estimate and the current identification residual are weighted and smoothed with this weight to output an accurate disturbance estimate for feedforward compensation.

7. The adaptive adjustment method for an intelligent pressure balancing valve according to claim 1, characterized in that, The statistical historical commands and location responses are used to extract the median command amplitude in the absence of response as a dead zone threshold, and the proportion of small command no-response events is calculated as a friction index, including: Scan a preset short-term historical sampling window to identify events in which command changes exist but valve position changes are less than the device position resolution, and calculate the median of the command change amplitude corresponding to these events as the dead zone threshold. The ratio of the number of events that meet the above non-response conditions to the total number of samples in the same window is used as the friction ratio indicator. The historical median of the friction ratio index plus twice the interquartile range is used as the threshold for judging friction anomalies.

8. The adaptive adjustment method for an intelligent pressure balancing valve according to claim 1, characterized in that, The model parameters estimated online and the one-step opening increment required to calculate the inverse model based on the current purification negative pressure, and the numerical lower bound derived based on the parameter covariance to ensure the stability of the inverse solution, include: The model parameters estimated online, the current and previous purification negative pressure values, and the control command from the previous moment are substituted into the inverse mapping of the one-step prediction model to obtain the opening increment required to make the prediction output approach the expected reference value. The lower bound of the value is determined by taking the square root of the largest diagonal element of the parameter covariance matrix and multiplying it by a coefficient of 3, and is used as a small guarantee quantity in the inverse solution operation; The desired reference value is obtained by mixing the upper-level reference value with the short-term smoothed reference of the previous time step, according to a given smoothing coefficient, and adding a bias value. The smoothing coefficient is between 0 and 1.

9. The adaptive adjustment method for an intelligent pressure balancing valve according to claim 1, characterized in that, The process of synthesizing the inverse model increment, disturbance feedforward term, and dead-zone inverse compensation during friction anomalies into a final control command, implementing actuator limiting, and reporting to the control system when an anomaly criterion is met includes: The step opening increment obtained by the inverse model is multiplied by the perturbation feedforward gain determined algebraically by the identification parameters and the perturbation estimate to obtain the perturbation feedforward term. When the friction index exceeds the threshold, the dead zone inverse compensation is constructed according to the dead zone threshold and the smoothed sign function. The three terms are combined to form the basic command. The underlying command is truncated within the range allowed by the executor and used as the final control command; The smoothing scale used to construct the smooth sign function takes the maximum value among the noise scale, the deterministic upper bound of the parameter uncertainty, and a local minimum value. The perturbation feedforward gain is determined by a combination of the first input coefficient of the identification parameter, the sum of squares of each input coefficient, and the lower bound of the value.

10. The adaptive adjustment method for an intelligent pressure balancing valve according to claim 9, characterized in that, The process of synthesizing the inverse model increment, disturbance feedforward term, and dead-zone inverse compensation during friction anomalies into the final control command, implementing actuator limiting, and reporting to the control system when the anomaly criterion is met also includes: During normal and stable system operation, a historical sample set is selected as the historical statistical window and a judgment window is set. Obtain the friction ratio index sequence and its anomaly judgment threshold. If there are consecutive sampling points within the judgment window that satisfy the friction ratio index exceeding the threshold, trigger a friction anomaly alarm and report it to DCS. Calculate the sequence of the largest eigenvalues ​​of the parameter covariance matrix within the historical statistics window, and set the threshold for abnormal determination of parameter identification uncertainty based on its historical median and three times the median absolute deviation. If the current largest eigenvalue exceeds the limit continuously within the determination window, an alarm for abnormal identification uncertainty will be triggered and reported. The median and three times the absolute deviation of the median are calculated based on the absolute value sequence of the disturbance estimation using historical samples to set the abnormal threshold of the disturbance estimation. If the absolute value of the disturbance estimation exceeds the limit continuously within the judgment window, an abnormal disturbance estimation alarm is triggered and reported.

Images