Butterfly valve opening degree self-adaptive control method with deep learning model

Abstract

This invention relates to the field of intelligent adaptive control technology and proposes a butterfly valve opening adaptive control method with a deep learning model. The method includes: identifying the stable operating range of the butterfly valve based on its historical dynamic response characteristics; applying a limiting constraint to the opening command value when it is outside the stable operating range; inputting the constrained target opening of the butterfly valve and the current pipeline fluid parameters into a pre-trained deep learning prediction model, and outputting a feedforward control quantity for the butterfly valve opening; performing an algebraic superposition of the deviation comparison result of the actual flow response value in the butterfly valve and the feedforward control quantity, and driving the butterfly valve to perform opening adjustment based on the algebraically superimposed butterfly valve opening command value; this invention can improve the opening control stability and long-cycle adaptive adjustment accuracy of the butterfly valve under time-varying operating conditions.

Description

Technical Field

[0001] This invention relates to the field of intelligent adaptive control technology, and in particular to an adaptive control method for butterfly valve opening with a deep learning model. Background Technology

[0002] Butterfly valves are widely used regulating elements in industrial fluid control systems. They adjust the flow and pressure of the medium in the pipeline by changing the rotation angle of the butterfly plate. The accuracy of their opening control directly affects the system's operating efficiency and equipment safety. In time-varying nonlinear fluid transportation scenarios such as secondary pipe networks for district heating, butterfly valves often need to operate continuously within a small opening range. The valve plate is subjected to severe fluid impact, which can easily cause valve body vibration and cavitation on the sealing surface, leading to a decrease in equipment life and deterioration of system stability.

[0003] In existing technologies, butterfly valve opening control methods based on PID or fuzzy PID lack the ability to identify the critical state of vibration and cavitation under low opening conditions online. They struggle to determine the stable operating boundary of the butterfly valve in real time based on dynamic response characteristics such as actuator current fluctuations and downstream pressure pulsations. When the opening command falls into this unstable range, existing methods cannot actively apply amplitude limiting constraints to avoid dangerous conditions, leading to mechanical oscillations and frequent actuator movements during regulation, further exacerbating valve wear and deteriorating control accuracy. Simultaneously, existing deep learning control schemes based on fixed parameters or offline training lack incremental adaptation mechanisms to time-varying pipeline characteristics. They struggle to utilize online operating data for local fine-tuning and iterative updates of prediction model parameters, causing model prediction errors to gradually increase over the operating cycle, failing to maintain long-term stable flow regulation quality. Therefore, there is an urgent need to develop an adaptive butterfly valve opening control method that integrates online identification of stable ranges and incremental model self-learning to address the insufficient operational reliability under low opening conditions and poor adaptability to drift in long-term operating conditions, thereby improving the stability and intelligence level of butterfly valve control. Summary of the Invention

[0004] To achieve the above objectives, this invention provides an adaptive control method for butterfly valve opening with a deep learning model, comprising: D1: Based on the historical dynamic response characteristics of the butterfly valve, identify the stable operating range of the butterfly valve. When the opening command value of the butterfly valve is outside the stable operating range, apply a limiting constraint to the opening command value. D2: Input the constrained target opening of the butterfly valve and the current pipeline fluid parameters into the pre-trained deep learning prediction model, and output the feedforward control quantity of the butterfly valve opening; D3: The deviation comparison result of the actual flow response value in the butterfly valve and the feedforward control quantity are algebraically superimposed, and the butterfly valve is driven to perform opening adjustment based on the butterfly valve opening command value after algebraic superposition.

[0005] In a preferred embodiment, the step of identifying the stable operating range of the butterfly valve based on its historical dynamic response characteristics, and applying a limiting constraint to the opening command value when the opening command value of the butterfly valve is outside the stable operating range, includes: The dynamic response characteristics of the butterfly valve during continuous adjustment at multiple different opening ranges are obtained, including actuator current signal and valve downstream pressure signal; The root mean square value of the actuator current signal after removing the DC component within the corresponding opening interval is taken as the time-domain fluctuation amplitude of the actuator current. Perform a fast Fourier transform on the downstream pressure signal to obtain the amplitude of the characteristic frequency component of the downstream pressure signal within a preset frequency band. The preset frequency band is the frequency band corresponding to the characteristic frequency of pressure pulsation induced by cavitation or vibration of the butterfly valve. When the time-domain fluctuation amplitude exceeds the first preset threshold, or the characteristic frequency component amplitude exceeds the second preset threshold, the boundary opening value of the opening interval is recorded as the boundary value of the stable operating interval of the butterfly valve, and the stable operating interval of the butterfly valve is obtained. The boundary value of the stable operating range is compared with the opening command value of the butterfly valve to determine whether to apply a limiting constraint to the opening command value.

[0006] In a preferred embodiment, acquiring the dynamic response characteristics of the butterfly valve during continuous adjustment across multiple different opening ranges includes: Extract the original monitoring data records of the butterfly valve during continuous adjustment in multiple different opening ranges. The original monitoring data records include the original sampled values ​​of actuator current and valve downstream pressure. The dynamic response characteristics are obtained by time-synchronized correlation of the original sampled values ​​of actuator current and valve downstream pressure after removing high-frequency noise.

[0007] In a preferred embodiment, comparing the boundary value of the stable operating range with the opening command value of the butterfly valve to determine whether to impose a limiting constraint on the opening command value includes: Obtain the original butterfly valve opening command value in the current control cycle; The original butterfly valve opening command value is compared with the stable operating range boundary value, which includes the lower limit boundary value and the upper limit boundary value of the stable operating range. When the original butterfly valve opening command value is within the stable operating range, the original butterfly valve opening command value is used as the target opening value of the butterfly valve after constraint correction. When the original butterfly valve opening command value is less than the lower limit boundary value, the lower limit boundary value is used as the target opening of the butterfly valve after constraint correction. When the original butterfly valve opening command value is greater than the upper limit boundary value, the upper limit boundary value is used as the target opening of the butterfly valve after constraint correction.

[0008] In a preferred embodiment, the step of inputting the constrained target opening of the butterfly valve and the current pipeline fluid parameters into a pre-trained deep learning prediction model, and outputting a feedforward control quantity for the butterfly valve opening, includes: Obtain the current pipeline fluid parameters of the butterfly valve, including the current upstream pressure value, the current downstream pressure value, and the current fluid temperature value; The normalized current pipeline fluid parameters and the constrained target opening of the butterfly valve are input into the pre-trained deep learning prediction model. The output layer of the deep learning prediction model outputs a normalized flow response prediction value. The normalized flow response prediction value is denormalized to obtain the actual flow response prediction value; The actual flow response prediction value is compared with the target flow value of the butterfly valve to obtain the flow prediction deviation, and the feedforward control value of the butterfly valve opening is obtained based on the flow prediction deviation.

[0009] In a preferred embodiment, obtaining the feedforward control quantity of the butterfly valve based on the flow prediction deviation includes: The feedforward gain is applied to the flow prediction deviation to obtain the initial feedforward control quantity; The initial feedforward control quantity is subjected to amplitude limiting and rate limiting to obtain the feedforward control quantity of the butterfly valve.

[0010] In a preferred embodiment, after comparing the actual flow response prediction value with the target flow value of the butterfly valve to obtain the flow prediction deviation, the method further includes: After this opening adjustment, when the actual flow response value of the butterfly valve exceeds the flow prediction deviation, the parameter incremental update of the deep learning prediction model is triggered. Extract the steady-state online operating condition data of the butterfly valve, which includes the constrained target opening degree of the butterfly valve, the upstream pressure value, the downstream pressure value, the fluid temperature value, and the corresponding actual flow response value. For each set of steady-state online operating condition data and actual flow response value, data normalization is performed to construct an incremental update training sample set; The incrementally updated training sample set is input into the current version of the deep learning prediction model, and the network weight parameters and bias terms of the last two fully connected layers in the deep learning prediction model are iteratively updated to obtain the updated deep learning prediction model. Replace the current version of the deep learning prediction model with the updated deep learning prediction model.

[0011] In a preferred embodiment, the step of inputting the incrementally updated training sample set into the current version of the deep learning prediction model, and iteratively updating the network weight parameters and bias terms of the last two fully connected layers in the deep learning prediction model to obtain the updated deep learning prediction model includes: The normalized input feature vector corresponding to the small batch sample subset in the incremental update training sample set is input into the current version of the deep learning prediction model to obtain the normalized traffic response prediction value in the small batch sample subset. The difference between the normalized flow response prediction value and the normalized pipeline actual flow value in the small batch sample subset is used as the prediction bias. The mean of the squared prediction bias of each sample in the small batch sample subset is used as the loss metric. The adjustment direction and amount of the network weight parameters and bias terms of the last two fully connected layers in the deep learning prediction model are determined with the goal of reducing the loss metric. Based on the adjustment direction and adjustment amount, the network weight parameters and bias terms of the last two fully connected layers are numerically updated until all samples in the incremental update training sample set are traversed. After each round of parameter update iteration, the overall loss metric after this round of iteration is compared with the overall loss metric corresponding to the previous round of iteration. When the reduction of the overall loss metric is lower than the preset threshold, the iteration ends, and the deep learning prediction model under the current parameter state is used as the deep learning prediction model for the next round of butterfly valve opening adaptive control.

[0012] In a preferred embodiment, the step of algebraically superimposing the deviation comparison result of the actual flow response value in the butterfly valve and the feedforward control quantity, and driving the butterfly valve to perform opening adjustment based on the algebraically superimposed butterfly valve opening command value, includes: By comparing the deviation between the target flow rate value of the butterfly valve and the actual flow rate response value of the butterfly valve, a deviation comparison result is obtained. Based on the current actual opening value of the butterfly valve, the feedforward control quantity and the deviation comparison result are algebraically superimposed to calculate the total adjustment of the butterfly valve opening. The current actual opening value is algebraically added to the total butterfly valve opening adjustment value to obtain the final butterfly valve opening command value, and the butterfly valve is driven to perform opening adjustment based on the final butterfly valve opening command value.

[0013] In a preferred embodiment, driving the butterfly valve to perform opening adjustment based on the final butterfly valve opening command value includes: The final butterfly valve opening command value is converted into a control signal that matches the butterfly valve actuator. Based on the control signal, the butterfly valve is driven to rotate to the opening position corresponding to the final butterfly valve opening command value, thus completing this opening adjustment.

[0014] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention extracts the time-domain fluctuation amplitude of the actuator current signal and the frequency-domain characteristic components of the valve downstream pressure signal from a historical operating condition database. It then identifies the critical states of cavitation or vibration of the butterfly valve in different opening ranges online, obtains the boundary values ​​of the stable operating range, and actively applies amplitude limiting constraints when the opening command exceeds this stable operating range. This method transforms the traditional fixed safe opening range, which relies on experience or offline calibration, into an adaptive stable boundary identified in real time based on dynamic response characteristics. This effectively avoids problems such as valve body vibration, sealing surface cavitation, and frequent actuator movements caused by continuous operation of the butterfly valve in the unstable low-opening range, significantly improving the operational stability and service life of the butterfly valve in variable operating conditions such as regional centralized heating.

[0015] 2. After adjusting the valve opening, this invention uses the accumulated deviation between the actual flow response value and the model prediction value after the pipeline reaches steady state as a trigger condition. Only the network weight parameters and bias terms of the last two fully connected layers in the deep learning prediction model are updated in small batches, while the parameters of the remaining layers remain unchanged. This incremental update strategy enables the prediction model to quickly adapt to the slow drift of pipeline resistance characteristics, medium temperature, and pressure distribution over time with minimal computational overhead. This maintains the long-term tracking accuracy of the flow response prediction value to the actual operating conditions, thereby ensuring the continuous effectiveness of the feedforward control quantity. It effectively overcomes the problem of decreased control accuracy caused by time-varying system characteristics in existing offline training models, achieving adaptive optimization and intelligent maintenance of butterfly valve opening control under long-term operating conditions. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below refer to only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0017] Figure 1 This is a flowchart illustrating an adaptive control method for butterfly valve opening with a deep learning model, provided in an embodiment of the present invention. Detailed Implementation To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the present invention and are not intended to limit the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] Reference Figure 1 The diagram shown is a flowchart illustrating an adaptive control method for butterfly valve opening with a deep learning model according to an embodiment of the present invention. In this embodiment, the adaptive control method for butterfly valve opening with a deep learning model includes: D1: Based on the historical dynamic response characteristics of the butterfly valve, identify the stable operating range of the butterfly valve. When the opening command value of the butterfly valve is outside the stable operating range, apply a limiting constraint to the opening command value. In this embodiment of the invention, the step of identifying the stable operating range of the butterfly valve based on its historical dynamic response characteristics, and applying a limiting constraint to the opening command value when the opening command value of the butterfly valve is outside the stable operating range, includes: The dynamic response characteristics of the butterfly valve during continuous adjustment at multiple different opening ranges are obtained, including actuator current signal and valve downstream pressure signal; The root mean square value of the actuator current signal after removing the DC component within the corresponding opening interval is taken as the time-domain fluctuation amplitude of the actuator current. Perform a fast Fourier transform on the downstream pressure signal to obtain the amplitude of the characteristic frequency component of the downstream pressure signal within a preset frequency band. The preset frequency band is the frequency band corresponding to the characteristic frequency of pressure pulsation induced by cavitation or vibration of the butterfly valve. When the time-domain fluctuation amplitude exceeds the first preset threshold, or the characteristic frequency component amplitude exceeds the second preset threshold, the boundary opening value of the opening interval is recorded as the boundary value of the stable operating interval of the butterfly valve, and the stable operating interval of the butterfly valve is obtained. The boundary value of the stable operating range is compared with the opening command value of the butterfly valve to determine whether to apply a limiting constraint to the opening command value.

[0019] The acquisition of the dynamic response characteristics of the butterfly valve during continuous adjustment in multiple different opening ranges includes: Extract the original monitoring data records of the butterfly valve during continuous adjustment in multiple different opening ranges. The original monitoring data records include the original sampled values ​​of actuator current and valve downstream pressure. The dynamic response characteristics are obtained by time-synchronized correlation of the original sampled values ​​of actuator current and valve downstream pressure after removing high-frequency noise.

[0020] The step of comparing the boundary value of the stable operating range with the opening command value of the butterfly valve to determine whether to apply a limiting constraint to the opening command value includes: Obtain the original butterfly valve opening command value in the current control cycle; The original butterfly valve opening command value is compared with the stable operating range boundary value, which includes the lower limit boundary value and the upper limit boundary value of the stable operating range. When the original butterfly valve opening command value is within the stable operating range, the original butterfly valve opening command value is used as the target opening value of the butterfly valve after constraint correction. When the original butterfly valve opening command value is less than the lower limit boundary value, the lower limit boundary value is used as the target opening of the butterfly valve after constraint correction. When the original butterfly valve opening command value is greater than the upper limit boundary value, the upper limit boundary value is used as the target opening of the butterfly valve after constraint correction.

[0021] The raw sampled values ​​of actuator current and downstream pressure were retrieved synchronously from historical operation records during continuous adjustment of the butterfly valve within different opening ranges. The raw sampled values ​​of actuator current were subjected to sliding window filtering to remove high-frequency noise components, resulting in a filtered actuator current signal. The raw sampled values ​​of downstream pressure were detrended to remove low-frequency drift components caused by macroscopic pressure changes in the pipeline system, resulting in a detrended downstream pressure signal.

[0022] The filtered actuator current signal and the detrended downstream pressure signal are synchronously correlated at the same acquisition time, establishing a one-to-one correspondence between the actuator current value and the downstream pressure value at the same acquisition time, forming a synchronous dynamic response feature sequence. This synchronous dynamic response feature sequence is then matched and correlated with the corresponding opening command value according to the acquisition time to obtain dynamic response characteristics under different opening ranges. These dynamic response characteristics include both the actuator current signal and the downstream pressure signal.

[0023] Time-domain analysis is performed on the actuator current signal within each opening interval. The average value of the current signal within that interval is subtracted from the actuator current signal within that opening interval to obtain the AC fluctuation component after removing the DC component. The root mean square value of the AC fluctuation component is calculated. This root mean square value is the time-domain fluctuation amplitude of the actuator current. This time-domain fluctuation amplitude reflects the severity of load fluctuations experienced by the actuator within that opening interval.

[0024] Frequency domain analysis is performed on the downstream pressure signal within each opening interval. Through time-domain to frequency-domain conversion, the downstream pressure signal is transformed from a waveform that varies with time to an amplitude distribution that varies with frequency. The amplitude of the characteristic frequency component within a preset frequency band is extracted. This preset frequency band corresponds to the frequency range in which the downstream pressure pulsation is concentrated when the butterfly valve experiences cavitation or vibration. The amplitude of this characteristic frequency component reflects the intensity of the pressure pulsation induced by cavitation or vibration within the opening interval.

[0025] The time-domain fluctuation amplitude corresponding to each opening interval is compared with a first preset threshold, which is statistically determined based on the upper limit of the actuator current fluctuation range under known stable operating conditions in historical operating conditions. The amplitude of the characteristic frequency component corresponding to each opening interval is compared with a second preset threshold, which is statistically determined based on the upper limit of the valve downstream pressure pulsation amplitude range under known stable operating conditions in historical operating conditions.

[0026] When the time-domain fluctuation amplitude corresponding to a certain opening interval exceeds a first preset threshold or the characteristic frequency component amplitude exceeds a second preset threshold, the operating condition corresponding to that opening interval is determined to be in a critical unstable state, and the boundary opening value of that opening interval is recorded as the boundary value of the stable operating range of the butterfly valve. The recorded lower and upper boundary values ​​are jointly determined as the stable operating range of the butterfly valve. This stable operating range indicates that when the butterfly valve operates within this opening range, the actuator load is stable and the downstream pressure pulsation does not reach the critical state of cavitation or vibration.

[0027] Obtain the original butterfly valve opening command value generated by the upper-level flow controller in the current control cycle. This original butterfly valve opening command value represents the target opening position to be executed by the butterfly valve actuator in the current control cycle. Compare the original butterfly valve opening command value with the lower and upper boundary values ​​of the stable operating range, respectively.

[0028] When the original butterfly valve opening command value is greater than or equal to the lower limit boundary value and less than or equal to the upper limit boundary value, it is determined that the original butterfly valve opening command value is within the stable operating range. In this case, no limiting constraint is applied to the original butterfly valve opening command value, and the original butterfly valve opening command value is directly used as the target opening value output after constraint correction. When the original butterfly valve opening command value is less than the lower limit boundary value, it is determined that the original butterfly valve opening command value is below the stable operating range. In this case, a lower limit limiting constraint is applied to the original butterfly valve opening command value, and the lower limit boundary value is used as the target opening value output after constraint correction.

[0029] When the original butterfly valve opening command value is greater than the upper limit boundary value, it is determined that the original butterfly valve opening command value is above the stable operating range. At this time, an upper limit constraint is applied to the original butterfly valve opening command value, and the upper limit boundary value is used as the target opening value of the butterfly valve after constraint correction. The target opening value of the butterfly valve after constraint correction is sent to the subsequent control loop to replace the original butterfly valve opening command value to generate feedforward control quantity to ensure that the butterfly valve performs opening adjustment action within the stable operating range.

[0030] The beneficial effects are as follows: By extracting the time-domain fluctuation amplitude of the actuator current signal and the frequency-domain characteristic components of the downstream pressure signal from historical operating condition records, the critical states of cavitation or vibration of the butterfly valve under different opening ranges can be identified online. This allows for the acquisition of stable operating range boundary values, and the active application of amplitude limiting constraints when the opening command exceeds these stable operating ranges. This method transforms the traditional fixed safe opening range, which relies on experience or offline calibration, into an adaptive stable boundary identified in real time based on dynamic response characteristics. It effectively avoids problems such as valve body vibration, sealing surface cavitation, and frequent actuator movements caused by continuous operation of the butterfly valve in unstable low-opening ranges, significantly improving the operational stability and service life of the butterfly valve in variable operating conditions such as regional centralized heating.

[0031] D2: Input the constrained target opening of the butterfly valve and the current pipeline fluid parameters into the pre-trained deep learning prediction model, and output the feedforward control quantity of the butterfly valve opening; In this embodiment of the invention, the step of inputting the constrained target opening of the butterfly valve and the current pipeline fluid parameters into a pre-trained deep learning prediction model, and outputting a feedforward control quantity for the butterfly valve opening, includes: Obtain the current pipeline fluid parameters of the butterfly valve, including the current upstream pressure value, the current downstream pressure value, and the current fluid temperature value; The normalized current pipeline fluid parameters and the constrained target opening of the butterfly valve are input into the pre-trained deep learning prediction model. The output layer of the deep learning prediction model outputs a normalized flow response prediction value. The normalized flow response prediction value is denormalized to obtain the actual flow response prediction value; The actual flow response prediction value is compared with the target flow value of the butterfly valve to obtain the flow prediction deviation, and the feedforward control value of the butterfly valve opening is obtained based on the flow prediction deviation.

[0032] The step of obtaining the feedforward control quantity of the butterfly valve based on the flow prediction deviation includes: The feedforward gain is applied to the flow prediction deviation to obtain the initial feedforward control quantity; The initial feedforward control quantity is subjected to amplitude limiting and rate limiting to obtain the feedforward control quantity of the butterfly valve.

[0033] After comparing the actual flow response prediction value with the target flow value of the butterfly valve to obtain the flow prediction deviation, the method further includes: After this opening adjustment, when the actual flow response value of the butterfly valve exceeds the flow prediction deviation, the parameter incremental update of the deep learning prediction model is triggered. Extract the steady-state online operating condition data of the butterfly valve, which includes the constrained target opening degree of the butterfly valve, the upstream pressure value, the downstream pressure value, the fluid temperature value, and the corresponding actual flow response value. For each set of steady-state online operating condition data and actual flow response value, data normalization is performed to construct an incremental update training sample set; The incrementally updated training sample set is input into the current version of the deep learning prediction model, and the network weight parameters and bias terms of the last two fully connected layers in the deep learning prediction model are iteratively updated to obtain the updated deep learning prediction model. Replace the current version of the deep learning prediction model with the updated deep learning prediction model.

[0034] The step of inputting the incrementally updated training sample set into the current version of the deep learning prediction model, and iteratively updating the network weight parameters and bias terms of the last two fully connected layers in the deep learning prediction model to obtain the updated deep learning prediction model includes: The normalized input feature vector corresponding to the small batch sample subset in the incremental update training sample set is input into the current version of the deep learning prediction model to obtain the normalized traffic response prediction value in the small batch sample subset. The difference between the normalized flow response prediction value and the normalized pipeline actual flow value in the small batch sample subset is used as the prediction bias. The mean of the squared prediction bias of each sample in the small batch sample subset is used as the loss metric. The adjustment direction and amount of the network weight parameters and bias terms of the last two fully connected layers in the deep learning prediction model are determined with the goal of reducing the loss metric. Based on the adjustment direction and adjustment amount, the network weight parameters and bias terms of the last two fully connected layers are numerically updated until all samples in the incremental update training sample set are traversed. After each round of parameter update iteration, the overall loss metric after this round of iteration is compared with the overall loss metric corresponding to the previous round of iteration. When the reduction of the overall loss metric is lower than the preset threshold, the iteration ends, and the deep learning prediction model under the current parameter state is used as the deep learning prediction model for the next round of butterfly valve opening adaptive control.

[0035] At the outset of constructing the deep learning prediction structure, multiple sets of steady-state operating data under different conditions need to be selected from the historical operating condition database as training materials. Each set of training materials includes the butterfly valve opening command value, the inlet pressure value, the outlet pressure value, the fluid temperature value, and the actual flow rate measured by the flow sensor after the pipeline reaches steady state under that operating condition. The butterfly valve opening command value, inlet pressure value, outlet pressure value, and fluid temperature value from each set of training materials are used as the input, and the corresponding actual flow rate value is used as the expected output.

[0036] The input and expected output components of all training materials are normalized separately. The normalization process involves dividing each physical quantity by its maximum value in the historical operating condition database, ensuring that the normalized values ​​all fall within the dimensionless range of zero to one. After normalization, the normalized input components and normalized expected output components are matched one-to-one to form a complete training sample set.

[0037] The training sample set is fed one by one into the deep learning prediction structure to be trained. Initially, the connection strength between each layer of neurons and the bias value of each neuron are preset values ​​based on experience. As each set of training samples is fed into the normalized input portion, the structure propagates the input value layer by layer forward by weighting the connection strength between neurons and adjusting the bias values, ultimately outputting a normalized flow response prediction value at the final layer. This normalized flow response prediction value is compared with the normalized expected actual flow value corresponding to that sample; the difference is used as the output error for the current sample.

[0038] During training, all samples from the training sample set are repeatedly fed into the structure. After each round of feeding all samples, the decrease in the total output error of the current round is compared with that of the previous round. When the decrease in the total output error is less than a pre-set threshold for several consecutive rounds, it indicates that the structure has sufficiently learned the pipeline flow characteristics reflected by the training samples. At this point, the training process is terminated, and all connection strength values ​​and bias values ​​between neural nodes in each layer of the structure at that moment are fixed and saved, forming a pre-trained deep learning prediction model for use in subsequent online control stages.

[0039] The target opening degree of the butterfly valve after constraint correction, together with the current inlet pressure value collected by the pressure sensor installed in the inlet pipeline, the current outlet pressure value collected by the pressure sensor installed in the outlet pipeline, and the current fluid temperature value collected by the temperature sensor installed in the pipeline, are used as the input information for the current control cycle.

[0040] These inputs are normalized to eliminate differences in dimensions and numerical ranges between different physical quantities, enabling unified processing in subsequent stages. The normalized constraint correction target opening value, the normalized inlet valve pressure value, the normalized outlet valve pressure value, and the normalized fluid temperature value are then fed into a deep learning prediction structure pre-trained on historical operating data.

[0041] This deep learning prediction structure consists of multiple layers of sequentially connected processing nodes. Each node receives values ​​from the node above it, transforms the received values ​​based on internally stored connection strength and bias values, and then passes the result to the next layer node. After multiple layers of transmission and transformation, a normalized value is finally output at the last layer. This normalized value is the normalized flow response prediction value, representing the normalized representation of the expected flow rate after the pipeline reaches a steady state, given a target opening and pipeline fluid parameters.

[0042] The normalized flow response prediction value output from the last layer of the deep learning prediction structure is denormalized to restore its original physical dimensions, thus obtaining the actual flow response prediction value with flow units. This actual flow response prediction value represents the steady-state flow value theoretically calculated based on the current pipeline conditions and the opening command to be executed.

[0043] The actual flow response prediction is compared with the target flow value preset in the control system. The difference between the target flow value and the actual flow response prediction is taken as the flow prediction deviation. This flow prediction deviation reflects the flow difference that cannot be completely eliminated by relying solely on the currently set target opening degree.

[0044] The flow prediction deviation is fed into the feedforward gain scheduling stage. Based on the opening range of the butterfly valve target opening after constraint correction, the feedforward gain coefficient that matches the opening range is selected from the pre-calibrated correspondence.

[0045] The calibration process for the feedforward gain coefficient is also based on the steady-state operating data stored in the historical operating condition database. Multiple sets of butterfly valve operating point records at different opening degrees in steady state are selected from the historical operating condition database, including the opening command value for each operating point and the actual flow rate value of the pipeline after reaching steady state at that opening degree.

[0046] Arrange the opening command values ​​in ascending order. Calculate the opening change between any two adjacent opening command values, which is the difference between the previous and subsequent opening command values. Simultaneously, calculate the corresponding flow change under these two opening degrees, which is the difference between the actual flow rate under the subsequent opening degree and the actual flow rate under the previous opening degree. The ratio of the flow rate change to the opening degree change is used as the feedforward gain coefficient within that opening degree interval.

[0047] The entire opening adjustment range is divided into multiple consecutive opening intervals. For each opening interval, a feedforward gain coefficient is calculated using the method described above. The upper and lower boundaries of the opening interval are associated with the corresponding feedforward gain coefficients and stored to form a feedforward gain scheduling lookup table. During actual control, the feedforward gain scheduling loop reads the corresponding feedforward gain coefficient from this lookup table based on the opening interval where the target opening of the butterfly valve after current constraint correction is located, and uses it for subsequent calculation of feedforward control quantities.

[0048] The flow prediction deviation is multiplied by the feedforward gain coefficient to obtain the initial feedforward control quantity. This initial feedforward control quantity represents the amount of butterfly valve opening change that theoretically requires further adjustment to compensate for the aforementioned flow difference. An amplitude limiting process is applied to the initial feedforward control quantity. When the absolute value of the initial feedforward control quantity exceeds the maximum allowable opening adjustment range of the butterfly valve actuator within a single control cycle, the amplitude of the initial feedforward control quantity is truncated to within this maximum allowable adjustment range, thus obtaining the limited initial feedforward control quantity.

[0049] The initial feedforward control quantity after amplitude limiting is further subjected to rate limiting processing. The change range of the initial feedforward control quantity after amplitude limiting calculated in the current cycle is compared with the actual butterfly valve opening feedforward control quantity used in the previous control cycle. When the absolute value of the difference between the two exceeds the maximum allowable opening change rate of the actuator in a single control cycle, the feedforward control quantity of the current cycle is corrected with the constraint of not exceeding the maximum allowable change rate. That is, based on the feedforward control quantity of the previous cycle, only the opening change corresponding to the maximum allowable change rate is increased or decreased, so as to finally obtain the butterfly valve opening feedforward control quantity that will be actually used in the current control cycle.

[0050] After each opening adjustment, once the fluid flow in the pipeline has reached a steady state again after a sufficiently long period, the actual flow response value measured by the flow sensor installed on the pipeline is collected. This actual flow response value is subtracted from the target flow value set by the control system, and the difference is recorded as the actual flow deviation. The absolute value of this actual flow deviation indicates the degree to which the current pipeline flow deviates from the target value. Simultaneously, the predicted actual flow response value output by the aforementioned deep learning prediction structure is subtracted from the target flow value, and the difference is recorded as the flow prediction deviation. The absolute value of this flow prediction deviation indicates the magnitude of the flow deviation that cannot be theoretically completely eliminated by feedforward control alone.

[0051] The absolute value of the actual flow deviation is numerically compared with the absolute value of the flow prediction deviation. When monitoring for multiple consecutive control cycles shows that the absolute value of the actual flow deviation is greater than the absolute value of the flow prediction deviation, it indicates that the deep learning prediction structure is significantly inadequate in describing the current pipeline flow characteristics. At this point, the trigger condition for incremental parameter updates is determined to be met, and the subsequent incremental update process is initiated.

[0052] After the triggering conditions for incremental parameter updates are met, operating condition data records showing steady-state fluid flow in the pipeline are selected from the current control cycle and several previous control cycles. Each set of these operating condition data records includes the target opening degree of the butterfly valve after amplitude limiting constraint correction, the corresponding inlet and outlet pressure values, fluid temperature value, and the actual flow response value measured by the flow sensor. The constrained target opening degree of the butterfly valve, inlet and outlet pressure values, and fluid temperature value in each set of steady-state online operating condition data are normalized in the same way as before, and the corresponding actual flow response value is normalized in the same way. The normalized input data and the normalized actual flow response value are paired one by one to form an incremental update training sample set containing multiple sets of paired data.

[0053] From the constructed incremental update training sample set, a portion of samples are randomly selected to form a mini-batch sample subset. The normalized target opening, normalized inlet valve pressure, normalized outlet valve pressure, and normalized fluid temperature contained in each group of samples in this mini-batch sample subset are combined into a normalized input feature vector.

[0054] These normalized input feature vectors are sequentially fed into the deep learning prediction structure currently in use during the control cycle. After weighted combination and nonlinear transformation of the connection strength values ​​and bias values ​​between multiple neural nodes within the structure, and passed layer by layer for computation, the normalized flow response prediction value corresponding to each sample in the mini-batch sample subset is obtained at the output.

[0055] For each sample in the mini-batch subset, the normalized flow response prediction value corresponding to that sample is subtracted from the normalized actual pipeline flow value. The result of this subtraction is taken as the prediction bias for that sample. Subsequently, the prediction biases of all samples in the mini-batch subset are squared, and the squared values ​​of the prediction biases of each sample are summed. The sum is then divided by the number of samples in the subset to obtain the loss metric for that mini-batch subset. The magnitude of this loss metric characterizes the overall deviation between the output of the current deep learning prediction structure and the actual measurement results on this mini-batch.

[0056] With the aim of reducing the aforementioned loss metric, the connection strength and bias values ​​between the neural nodes of the last two fully connected layers in the deep learning prediction structure are specifically adjusted. The direction and magnitude of these adjustments are determined. The direction of adjustment is to gradually reduce the loss metric, while the magnitude of the adjustment is controlled by a pre-set update step size based on experience.

[0057] Based on the determined direction and magnitude of modification, the connection strength and bias values ​​of the last two fully connected layers are updated. The connection strength and bias values ​​between neurons in all other layers of the deep learning prediction structure, excluding the last two fully connected layers, remain unchanged during this adjustment. The entire process—from extracting a small subset of samples to calculating the loss metric, and then adjusting the connection strength and bias values ​​of the last two fully connected layers—is repeated until all samples in the incrementally updated training sample set have been extracted and participated in at least one adjustment, completing one round of parameter update iteration.

[0058] After completing one round of parameter update iteration, the overall loss metric of the current deep learning prediction structure after this round of adjustment is calculated on all samples of the incremental update training sample set. That is, for each sample in the incremental update training sample set, the square of the prediction deviation between its normalized flow response prediction value and the normalized pipeline actual flow value is calculated, and the mean of the square of the prediction deviation of all samples is taken.

[0059] The overall loss metric obtained after the current iteration is compared with the overall loss metric obtained after the previous iteration. When the decrease in the overall loss metric of the current iteration compared to the previous iteration is less than a pre-set threshold, it indicates that continuing the iteration has no significant effect on improving the accuracy of the prediction structure. At this point, the iteration process is terminated, and all connection strength values ​​and bias values ​​of the deep learning prediction structure at the time of iteration termination are saved as the updated deep learning prediction structure.

[0060] The existing deep learning prediction structure currently in use in the control system is completely replaced with the updated deep learning prediction structure. This ensures that when generating the feedforward control quantity for the butterfly valve opening in the next control cycle and subsequent control cycles, the deep learning prediction structure that has been adjusted and updated based on online steady-state operating data is used. This ensures that the feedforward control quantity can continuously adapt to the slow time-varying characteristics of the pipeline.

[0061] The beneficial effects are as follows: After each opening adjustment action, the cumulative deviation between the actual flow response value after the pipeline enters steady state and the predicted value output by the deep learning prediction structure is used as the basis for determining the incremental update of the starting parameters. Only the connection strength and bias values ​​of the last two fully connected layers in the deep learning prediction structure are adjusted in small batches and at specific times, while the connection strength and bias values ​​of the remaining layers remain stable. This adjustment strategy allows the prediction structure to quickly adapt to changes in pipeline resistance characteristics, medium temperature distribution, and other operating condition drift phenomena that occur gradually over time with relatively low computational resource consumption. It maintains the long-term accuracy of the flow response prediction value in tracking the actual pipeline flow, thereby ensuring the effectiveness of the feedforward control quantity during long-term continuous operation and maintaining the long-term stability and intelligent adaptive level of the butterfly valve opening adjustment quality.

[0062] D3: The deviation comparison result of the actual flow response value in the butterfly valve and the feedforward control quantity are algebraically superimposed, and the butterfly valve is driven to perform opening adjustment based on the butterfly valve opening command value after algebraic superposition.

[0063] In this embodiment of the invention, the step of algebraically superimposing the deviation comparison result of the actual flow response value in the butterfly valve and the feedforward control quantity, and driving the butterfly valve to perform opening adjustment based on the algebraically superimposed butterfly valve opening command value, includes: By comparing the deviation between the target flow rate value of the butterfly valve and the actual flow rate response value of the butterfly valve, a deviation comparison result is obtained. Based on the current actual opening value of the butterfly valve, the feedforward control quantity and the deviation comparison result are algebraically superimposed to calculate the total adjustment of the butterfly valve opening. The current actual opening value is algebraically added to the total butterfly valve opening adjustment value to obtain the final butterfly valve opening command value, and the butterfly valve is driven to perform opening adjustment based on the final butterfly valve opening command value.

[0064] The step of driving the butterfly valve to perform opening adjustment based on the final butterfly valve opening command value includes: The final butterfly valve opening command value is converted into a control signal that matches the butterfly valve actuator. Based on the control signal, the butterfly valve is driven to rotate to the opening position corresponding to the final butterfly valve opening command value, thus completing this opening adjustment.

[0065] The target flow rate value preset by the control system is compared with the actual flow response value of the butterfly valve measured by the flow sensor installed on the pipeline. The difference between the target flow rate value and the actual flow response value is taken as the deviation comparison result. This deviation comparison result reflects the magnitude and direction of the gap between the actual flow rate delivered in the pipeline and the expected target flow rate, whether it is higher or lower than the target value.

[0066] The deviation comparison result is sent to the flow control controller, which generates a feedback correction amount needed to eliminate the aforementioned flow difference based on the magnitude and direction of the deviation comparison result and the trend of the deviation changing over time. This feedback correction amount indicates the extent by which the butterfly valve opening needs to be further increased or decreased in order to adjust the actual flow to the target flow.

[0067] The current actual opening value of the butterfly valve is obtained in real time through a position feedback element installed on the butterfly valve actuator. This current actual opening value represents the actual rotation angle position of the butterfly valve disc before the start of this adjustment action.

[0068] Using the current actual opening value of the butterfly valve as the superposition benchmark, the butterfly valve opening feedforward control quantity, generated by the deep learning prediction structure and processed by amplitude and rate limiting in the aforementioned steps, is algebraically added to the feedback correction quantity generated by the flow control controller based on the deviation comparison result, to obtain the total butterfly valve opening adjustment. This total butterfly valve opening adjustment represents the net opening change that the butterfly valve actuator needs to perform from the current actual opening position within the current control cycle. Its sign indicates the direction of rotation, and its absolute value indicates the rotation amplitude.

[0069] The current actual opening value of the butterfly valve is algebraically added to the total opening adjustment amount. The result of this addition is the final butterfly valve opening command value. This final butterfly valve opening command value indicates the target opening position that the butterfly plate should reach after the butterfly valve actuator completes its adjustment action in the current control cycle.

[0070] The final butterfly valve opening command value is converted from a numerical form representing the opening percentage into a control signal format that the butterfly valve actuator can directly recognize and respond to. For electric actuators, this control signal is an analog current or voltage signal proportional to the target opening position; for actuators using digital communication, this control signal is a digital communication protocol command frame containing the target opening information.

[0071] The converted control signal is sent to the drive control unit equipped with the butterfly valve actuator via a signal transmission cable or field communication bus. After receiving the control signal, the drive control unit drives the internal servo motor to rotate in the direction and stroke specified by the control signal. The rotational motion of the servo motor is transmitted to the valve stem of the butterfly valve through a reduction gear mechanism, causing the butterfly plate, which is fixedly connected to the valve stem, to rotate around the valve stem axis.

[0072] Driven by a servo motor, the butterfly plate rotates continuously until it reaches the target opening position corresponding to the final butterfly valve opening command value. At this point, the drive control unit stops outputting driving torque to the servo motor, and the butterfly plate stops at the target opening position, completing this opening adjustment action. At this time, the flow area of ​​the medium in the pipeline changes accordingly, causing the actual flow rate in the pipeline to approach the target flow rate value.

[0073] The beneficial effects are as follows: By algebraically superimposing the feedforward control quantity generated by the deep learning prediction structure and the feedback correction quantity generated by the flow regulation controller based on the actual flow deviation, using the current actual opening value as a benchmark, the final opening command value of the butterfly valve is determined together. The feedforward control quantity compensates in advance for the impact of pipeline characteristic changes on the flow response based on the prediction model, thereby accelerating the regulation response speed and reducing the overshoot during the flow regulation process; the feedback correction quantity performs closed-loop correction based on the real-time deviation between the actual flow and the target flow, thereby eliminating the steady-state residual error caused by feedforward prediction error and external disturbances. The synergistic cooperation of the two control actions enables the butterfly valve opening regulation to have both the ability to quickly predict and respond to changes in operating conditions and the ability to guarantee the regulation accuracy in a closed loop. In pipeline systems with large inertial time delay characteristics, such as district centralized heating, it can significantly improve the dynamic response quality and steady-state control accuracy of flow regulation.

[0074] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0075] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0076] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A butterfly valve opening degree adaptive control method with a deep learning model, characterized by, The method includes: D1: Based on the historical dynamic response characteristics of the butterfly valve, identify the stable operating range of the butterfly valve. When the opening command value of the butterfly valve is outside the stable operating range, apply a limiting constraint to the opening command value. D2: Input the constrained target opening of the butterfly valve and the current pipeline fluid parameters into the pre-trained deep learning prediction model, and output the feedforward control quantity of the butterfly valve opening; D3: The deviation comparison result of the actual flow response value in the butterfly valve and the feedforward control quantity are algebraically superimposed, and the butterfly valve is driven to perform opening adjustment based on the butterfly valve opening command value after algebraic superposition.

2. The adaptive control method for butterfly valve opening with a deep learning model as described in claim 1, characterized in that, The method, based on the historical dynamic response characteristics of the butterfly valve, identifies the stable operating range of the butterfly valve. When the opening command value of the butterfly valve is outside the stable operating range, a limiting constraint is applied to the opening command value, including: The dynamic response characteristics of the butterfly valve during continuous adjustment at multiple different opening ranges are obtained, including actuator current signal and valve downstream pressure signal; The root mean square value of the actuator current signal after removing the DC component within the corresponding opening interval is taken as the time-domain fluctuation amplitude of the actuator current. Perform a fast Fourier transform on the downstream pressure signal to obtain the amplitude of the characteristic frequency component of the downstream pressure signal within a preset frequency band. The preset frequency band is the frequency band corresponding to the characteristic frequency of pressure pulsation induced by cavitation or vibration of the butterfly valve. When the time-domain fluctuation amplitude exceeds the first preset threshold, or the characteristic frequency component amplitude exceeds the second preset threshold, the boundary opening value of the opening interval is recorded as the boundary value of the stable operating interval of the butterfly valve, and the stable operating interval of the butterfly valve is obtained. The boundary value of the stable operating range is compared with the opening command value of the butterfly valve to determine whether to apply a limiting constraint to the opening command value.

3. The adaptive control method for butterfly valve opening with a deep learning model as described in claim 2, characterized in that, The acquisition of the dynamic response characteristics of the butterfly valve during continuous adjustment in multiple different opening ranges includes: Extract the original monitoring data records of the butterfly valve during continuous adjustment in multiple different opening ranges. The original monitoring data records include the original sampled values ​​of actuator current and valve downstream pressure. The dynamic response characteristics are obtained by time-synchronized correlation of the original sampled values ​​of actuator current and valve downstream pressure after removing high-frequency noise.

4. The adaptive control method for butterfly valve opening with a deep learning model as described in claim 2, characterized in that, The step of comparing the boundary value of the stable operating range with the opening command value of the butterfly valve to determine whether to apply a limiting constraint to the opening command value includes: Obtain the original butterfly valve opening command value in the current control cycle; The original butterfly valve opening command value is compared with the stable operating range boundary value, which includes the lower limit boundary value and the upper limit boundary value of the stable operating range. When the original butterfly valve opening command value is within the stable operating range, the original butterfly valve opening command value is used as the target opening value of the butterfly valve after constraint correction. When the original butterfly valve opening command value is less than the lower limit boundary value, the lower limit boundary value is used as the target opening of the butterfly valve after constraint correction. When the original butterfly valve opening command value is greater than the upper limit boundary value, the upper limit boundary value is used as the target opening of the butterfly valve after constraint correction.

5. The adaptive control method for butterfly valve opening with a deep learning model as described in claim 1, characterized in that, The step of inputting the constrained target opening of the butterfly valve and the current pipeline fluid parameters into a pre-trained deep learning prediction model, and outputting a feedforward control quantity for the butterfly valve opening, includes: Obtain the current pipeline fluid parameters of the butterfly valve, including the current upstream pressure value, the current downstream pressure value, and the current fluid temperature value; The normalized current pipeline fluid parameters and the constrained target opening of the butterfly valve are input into the pre-trained deep learning prediction model. The output layer of the deep learning prediction model outputs a normalized flow response prediction value. The normalized flow response prediction value is denormalized to obtain the actual flow response prediction value; The actual flow response prediction value is compared with the target flow value of the butterfly valve to obtain the flow prediction deviation, and the feedforward control value of the butterfly valve opening is obtained based on the flow prediction deviation.

6. The adaptive control method for butterfly valve opening with a deep learning model as described in claim 5, characterized in that, The step of obtaining the feedforward control quantity of the butterfly valve based on the flow prediction deviation includes: The feedforward gain is applied to the flow prediction deviation to obtain the initial feedforward control quantity; The initial feedforward control quantity is subjected to amplitude limiting and rate limiting to obtain the feedforward control quantity of the butterfly valve.

7. The adaptive control method for butterfly valve opening with a deep learning model as described in claim 5, characterized in that, After comparing the actual flow response prediction value with the target flow value of the butterfly valve to obtain the flow prediction deviation, the method further includes: After this opening adjustment, when the actual flow response value of the butterfly valve exceeds the flow prediction deviation, the parameter incremental update of the deep learning prediction model is triggered. Extract the steady-state online operating condition data of the butterfly valve, which includes the constrained target opening degree of the butterfly valve, the upstream pressure value, the downstream pressure value, the fluid temperature value, and the corresponding actual flow response value. For each set of steady-state online operating condition data and actual flow response value, data normalization is performed to construct an incremental update training sample set; The incrementally updated training sample set is input into the current version of the deep learning prediction model, and the network weight parameters and bias terms of the last two fully connected layers in the deep learning prediction model are iteratively updated to obtain the updated deep learning prediction model. Replace the current version of the deep learning prediction model with the updated deep learning prediction model.

8. The adaptive control method for butterfly valve opening with a deep learning model as described in claim 7, characterized in that, The step of inputting the incrementally updated training sample set into the current version of the deep learning prediction model, and iteratively updating the network weight parameters and bias terms of the last two fully connected layers in the deep learning prediction model to obtain the updated deep learning prediction model includes: The normalized input feature vector corresponding to the small batch sample subset in the incremental update training sample set is input into the current version of the deep learning prediction model to obtain the normalized traffic response prediction value in the small batch sample subset. The difference between the normalized flow response prediction value and the normalized pipeline actual flow value in the small batch sample subset is used as the prediction bias. The mean of the squared prediction bias of each sample in the small batch sample subset is used as the loss metric. The adjustment direction and amount of the network weight parameters and bias terms of the last two fully connected layers in the deep learning prediction model are determined with the goal of reducing the loss metric. Based on the adjustment direction and adjustment amount, the network weight parameters and bias terms of the last two fully connected layers are numerically updated until all samples in the incremental update training sample set are traversed. After each round of parameter update iteration, the overall loss metric after this round of iteration is compared with the overall loss metric corresponding to the previous round of iteration. When the reduction of the overall loss metric is lower than the preset threshold, the iteration ends, and the deep learning prediction model under the current parameter state is used as the deep learning prediction model for the next round of butterfly valve opening adaptive control.

9. The adaptive control method for butterfly valve opening with a deep learning model as described in claim 1, characterized in that, The process of algebraically superimposing the deviation comparison result of the actual flow response value in the butterfly valve with the feedforward control quantity, and driving the butterfly valve to perform opening adjustment based on the algebraically superimposed butterfly valve opening command value, includes: By comparing the deviation between the target flow rate value of the butterfly valve and the actual flow rate response value of the butterfly valve, a deviation comparison result is obtained. Based on the current actual opening value of the butterfly valve, the feedforward control quantity and the deviation comparison result are algebraically superimposed to calculate the total adjustment of the butterfly valve opening. The current actual opening value is algebraically added to the total butterfly valve opening adjustment value to obtain the final butterfly valve opening command value, and the butterfly valve is driven to perform opening adjustment based on the final butterfly valve opening command value.

10. The adaptive control method for butterfly valve opening with a deep learning model as described in claim 9, characterized in that, The step of driving the butterfly valve to perform opening adjustment based on the final butterfly valve opening command value includes: The final butterfly valve opening command value is converted into a control signal that matches the butterfly valve actuator. Based on the control signal, the butterfly valve is driven to rotate to the opening position corresponding to the final butterfly valve opening command value, thus completing this opening adjustment.

Images