Energy-saving optimization control method and system for variable-frequency centrifugal fan

Abstract

The invention relates to the technical field of variable-frequency centrifugal fans, and discloses an energy-saving optimization control method and system for a variable-frequency centrifugal fan, and the method comprises the steps: collecting a real-time parameter data set of the variable-frequency centrifugal fan through a sensor network, and carrying out the digital filtering and feature extraction, and obtaining a target feature vector; inputting the target feature vector into a trained VentNet dual-network deep neural network for prediction calculation to obtain an optimal rotating speed prediction value and an energy consumption prediction value under the current load condition; calculating control error data and inputting the control error data into a hyperbolic law of arrival controller for processing to obtain a frequency adjustment instruction of the frequency converter; the frequency adjusting instruction is sent to the frequency converter, the actual rotating speed of the variable-frequency centrifugal fan is adjusted, and the variable-frequency centrifugal fan operates in the optimal energy-saving state, real-time energy-saving optimization control over the variable-frequency centrifugal fan under different load conditions is achieved, the energy consumption amplitude is reduced, and meanwhile the stability and reliability of operation of the fan are guaranteed.

Description

Technical Field

[0001] This application relates to the technical field of variable-frequency centrifugal fans, and particularly to an energy-saving optimization control method and system for variable-frequency centrifugal fans. Background Art

[0002] As a key equipment in the industrial and construction fields, the energy consumption optimization of variable-frequency centrifugal fans has been increasingly emphasized. With the rising energy costs and stricter environmental protection requirements, the fan system, as a major industrial electricity consumer, its energy-saving optimization has become a key issue that the industry urgently needs to solve. Although the currently widely used variable-frequency control technology has certain energy-saving effects compared with the traditional valve regulation and inlet guide vane regulation methods, the existing variable-frequency control methods usually rely on simple PID control strategies or preset variable-frequency curves, and it is difficult to adapt to complex and changeable load conditions, resulting in the fan frequently operating in a non-optimal state during actual operation, and there is significant energy waste.

[0003] The main challenge faced by the traditional variable-frequency centrifugal fan control system is the inability to accurately predict the fan performance under multiple load conditions simultaneously. During the industrial production process, the fan load conditions change frequently with the process requirements. It is difficult for the existing technologies to establish an accurate fan performance model. Especially when multiple parameters such as flow rate, pressure, and system resistance change simultaneously, the control system can often only optimize for a single working condition, and it is difficult to achieve global optimal control. In addition, there are uncertain factors such as non-linear interference and load mutations in the system, and the anti-interference ability of traditional control methods is insufficient, resulting in unstable control performance and it is difficult to ensure that the fan can maintain efficient and energy-saving operation under various working conditions. Summary of the Invention

[0004] This application provides an energy-saving optimization control method and system for variable-frequency centrifugal fans. The present invention realizes real-time energy-saving optimization control of variable-frequency centrifugal fans under different load conditions, reduces the energy consumption amplitude, and at the same time ensures the stability and reliability of the fan operation.

[0005] The first aspect of this application provides an energy-saving optimization control method for variable-frequency centrifugal fans. The energy-saving optimization control method for variable-frequency centrifugal fans includes: Collect the inlet static pressure, outlet static pressure, flow rate, motor power, system resistance coefficient, and ambient temperature of the variable-frequency centrifugal fan through a sensor network to obtain a real-time parameter data set; Perform digital filtering and feature extraction on the real-time parameter data set to obtain a target feature vector; Input the target feature vector into the trained VentNet dual-network deep neural network for prediction calculation to obtain the optimal speed prediction value and energy consumption prediction value under the current load condition; Calculate the control error data according to the predicted optimal rotational speed value and the predicted energy consumption value, and input the control error data into a hyperbolic reaching law controller for processing to obtain a frequency adjustment command for the frequency converter; Send the frequency adjustment command to the frequency converter to adjust the actual rotational speed of the variable-frequency centrifugal fan, so that the variable-frequency centrifugal fan operates in the optimal energy-saving state.

[0006] The second aspect of the present application provides an energy-saving optimization control system for a variable-frequency centrifugal fan. The energy-saving optimization control system for the variable-frequency centrifugal fan includes: An acquisition module, configured to acquire the inlet static pressure, outlet static pressure, flow rate, motor power, system resistance coefficient, and ambient temperature of the variable-frequency centrifugal fan through a sensor network to obtain a real-time parameter data set; A feature extraction module, configured to perform digital filtering and feature extraction on the real-time parameter data set to obtain a target feature vector; A prediction calculation module, configured to input the target feature vector into a trained VentNet dual-network deep neural network for prediction calculation to obtain a predicted optimal rotational speed value and a predicted energy consumption value under the current load condition; A processing module, configured to calculate control error data according to the predicted optimal rotational speed value and the predicted energy consumption value, and input the control error data into a hyperbolic reaching law controller for processing to obtain a frequency adjustment command for the frequency converter; An adjustment module, configured to send the frequency adjustment command to the frequency converter to adjust the actual rotational speed of the variable-frequency centrifugal fan, so that the variable-frequency centrifugal fan operates in the optimal energy-saving state.

[0007] Compared with the prior art, the present application has the following beneficial effects: By constructing the VentNet dual-network deep neural network, the present invention realizes the simultaneous prediction of the optimal speed and energy consumption of a variable-frequency centrifugal fan under different load conditions, overcomes the defect that the prior art cannot accurately predict the performance of a fan under multiple load conditions simultaneously, and enables the control system to more precisely find the optimal operating point of the fan under the current working conditions. High-frequency noise filtering, pulsating noise and interference signal filtering, power grid fluctuation interference elimination, and slow variable trend extraction are performed on the real-time parameter data set. Combining with the adaptive weighted fusion of multi-scale feature data significantly improves the accuracy and anti-interference ability of fan performance prediction. Applying the hyperbolic reaching law to the control of variable-frequency centrifugal fans effectively reduces the width of the quasi-sliding mode region of the system and suppresses the chattering phenomenon compared with traditional discrete sliding mode control, making the fan speed regulation smoother and avoiding mechanical losses caused by frequent starting and stopping and large-scale speed changes. By estimating and compensating the system disturbance through a disturbance observer and combining with the inherent robustness of the hyperbolic reaching law controller, the system has a stronger ability to suppress external disturbances (such as sudden loads and airflow fluctuations), ensuring control stability under complex working conditions. The control system can make the fan speed reach the optimal energy-saving state within a theoretically calculable finite number of steps, greatly improving the system response speed and accelerating the working condition adaptation ability. The problem of scarce historical operation data is solved by using a two-stage training method of coarse and fine. First, an initial model is established using general data of multiple fans, and then fine-tuning is performed using specific data of the target fan, enabling the control system to have better generalization ability and personalized adaptation ability. The present invention realizes the real-time energy-saving optimization control of variable-frequency centrifugal fans under different load conditions, reduces the energy consumption amplitude, and at the same time ensures the stability and reliability of fan operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0009] The structures, ratios, sizes, etc. shown in the drawings of this specification are only used to cooperate with the content disclosed in the specification for those familiar with this technology to understand and read, and are not used to limit the limited conditions under which the present invention can be implemented. Therefore, they do not have a substantial technical meaning. Any modification of the structure, change of the proportional relationship, or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope that can be covered by the technical content disclosed by the present invention.

[0010] Figure 1It is a schematic flowchart of the energy-saving optimization control method for a variable-frequency centrifugal fan provided by an embodiment of the present invention; Figure 2 It is a schematic block diagram of the structure of the energy-saving optimization control system for a variable-frequency centrifugal fan provided by an embodiment of the present invention. Specific embodiments

[0011] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0012] The flowchart shown in the accompanying drawings is only an example for illustration, and does not necessarily include all the contents and operations / steps, nor does it necessarily need to be executed in the described order. For example, some operations / steps can also be decomposed, combined, or partially merged, so the actual execution order may change according to the actual situation.

[0013] It should also be understood that the terms used in the specification of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. As used in the specification of this application and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an", and "the" are intended to include the plural forms.

[0014] It should be further understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination and all possible combinations of one or more of the related listed items, and includes these combinations. Please refer to Figure 1 , an embodiment of the energy-saving optimization control method for a variable-frequency centrifugal fan in the embodiments of this application includes: Step 100, collect the inlet static pressure, outlet static pressure, flow rate, motor power, system resistance coefficient, and ambient temperature of the variable-frequency centrifugal fan through a sensor network to obtain a real-time parameter dataset; It can be understood that the execution subject of this application can be the energy-saving optimization control system of the variable-frequency centrifugal fan, or it can also be a terminal or a server, and specifically it is not limited here. In the embodiments of this application, the server is used as the execution subject for illustration.

[0015] Specifically, a static pressure sensor is installed at the inlet of the fan to sense the static pressure state of the gas at the fan suction port, convert this physical signal into a continuous electrical signal, and then the data acquisition system completes signal acquisition and numerical conversion to form the inlet static pressure parameter, representing the actual air pressure level at the suction end of the fan. At the same time, a static pressure sensor is installed at the outlet of the fan to detect the static pressure state of the gas after being compressed by the fan, and through the same data processing process, the outlet static pressure parameter is formed, so as to jointly reflect the pressure difference at both ends of the fan with the inlet static pressure parameter. A high-precision flow sensor is installed in the air duct connected to the fan. This sensor continuously senses the air volume passing through the pipeline under changing working conditions, outputs continuous flow data, and after filtering and conversion, obtains a standardized flow parameter, reflecting the conveying capacity and load size of the fan. A power analyzer is configured at the input end of the fan motor to monitor the electrical energy actually consumed by the motor. It simultaneously records electrical indicators such as voltage, current, frequency, and power factor, calculates the input power of the motor, and outputs it as the motor power parameter for evaluating the current energy consumption level of the fan operation. By using the embedded data processing logic with the above-mentioned inlet static pressure, outlet static pressure, and flow data, the resistance coefficient of the fan system is calculated. This resistance coefficient represents the flow resistance intensity of the system where the fan is located, reflecting the degree of air flow blockage caused by pipeline elbows, valve openings, system loads, or other non-linear factors. By integrating the variation relationship between the static pressure difference and the flow rate, the system resistance coefficient parameter is obtained, reflecting the health status and load characteristics during the system operation. Temperature sensors are installed around the area where the fan equipment is located to detect the temperature condition of the environment where the fan is located in real time. Temperature changes will affect the density and viscosity of air, thereby affecting the system pressure and air flow pattern. Therefore, the data collected by the temperature sensors is standardized and recorded as the ambient temperature parameter to support the comprehensive judgment of the model on external disturbance conditions. Unified time synchronization and alignment processing are performed on the inlet static pressure parameter, outlet static pressure parameter, flow parameter, motor power parameter, system resistance coefficient parameter, and ambient temperature parameter to ensure that there is a consistent time reference between all data points. These parameters are combined into a structured real-time parameter data set, reflecting the current operating state, load characteristics, and environmental impact of the fan.

[0016] Step 200: Perform digital filtering and feature extraction on the real-time parameter data set to obtain the target feature vector; Specifically, high-frequency noise filtering is performed on the inlet static pressure parameter and the outlet static pressure parameter in the real-time parameter dataset. Since the fan is affected by external factors such as electromagnetic interference, air flow disturbance, and structural vibration during operation, the collected static pressure data contains irregular high-frequency noise. To exclude the interference of non-operating condition characteristic signals on subsequent calculations, a low-pass digital filter is embedded to perform real-time processing on such pressure signals, thereby retaining the true and stable pressure change trend and forming filtered pressure data. At the same time, the flow rate parameter is also affected by pipeline air flow pulsation and local reflection interference during actual acquisition, especially in scenarios with sudden load changes or frequent starts and stops. To remove the periodic noise and non-stable interference in the flow rate parameter, a pulsating noise suppression and interference signal rejection algorithm is applied to the flow rate data to extract filtered flow rate data that is continuous, smooth, and reflects the actual air flow behavior of the operating condition, so as to truly reflect the conveying load of the fan. Considering that the measured value of the fan motor power is greatly affected by external power factors such as grid voltage fluctuations and frequency offsets, especially when there are high-frequency start-stop operations or severe fluctuations in the power supply system, it is more likely to have abnormal energy consumption data. An identification and elimination processing mechanism for grid fluctuation interference is adopted for the motor power parameter to remove energy consumption anomalies caused by non-typical signals such as voltage jumps and current spikes, so that the filtered power data can more truly express the energy consumption level during the operation of the fan. For the system resistance coefficient parameter and the environmental temperature parameter, these two types of variables change relatively slowly, with significant low-frequency trend and long-period disturbance characteristics. Therefore, it is not suitable to directly use the traditional high-pass filtering method for processing. Instead, through the slow variable trend extraction technology, the steady components representing the system structure and environmental change trends are stripped from the long time series to avoid affecting the overall characteristic judgment due to short-term fluctuations. After the above process, the filtered resistance temperature data is obtained. Based on the filtered pressure data and flow rate data, dynamic characteristics describing the working state response of the fan are constructed, including pressure fluctuation amplitude, flow rate change speed, pressure difference stability, etc., to depict the dynamic behavior of the fan under complex load conditions; at the same time, energy consumption characteristics are extracted based on the filtered power data, such as energy consumption per unit volume, instantaneous power fluctuation range, and load factor, etc., to reflect the energy efficiency performance of the fan; combined with the filtered resistance temperature data, environmental characteristics are extracted, such as temperature change rate, system impedance trend, and heat-resistance interaction factor, etc., to assist the model in understanding the influence of the external environment on the system behavior. The above dynamic characteristics, energy consumption characteristics, and environmental characteristics differ in terms of dimension, time granularity, change speed, etc. The three types of characteristics are unified and combined into a composite characteristic data set with multi-scale characteristics, enabling it to simultaneously have the description ability of short-term response and long-term trend.To improve the expression ability of features in the subsequent neural network model, according to the performance curve generated from the historical operation data of the fan, adaptive weighted fusion is performed on the multi-scale feature data. During this process, different types of features are assigned dynamic weights according to their correlation with the fan operation state, so as to emphasize the key features strongly related to the fan speed and energy consumption, and at the same time weaken the role of redundant or low-impact features. All feature values are transformed into a unified feature space through a non-linear mapping function, so that features with different dimensions and different numerical ranges are uniformly represented in a comparable coordinate system, eliminating the interference of model input caused by factors such as unit differences and distribution offsets. Finally, the target feature vector is formed.

[0017] Step 300: Input the target feature vector into the trained VentNet dual-network deep neural network for prediction calculation to obtain the optimal speed prediction value and energy consumption prediction value under the current load condition; It should be noted that the target feature vector is input into the trained VentNet neural network model. In this structure, the target feature vector is passed to the feature extraction network part. The feature extraction network consists of three fully connected hidden layers. Each layer maps the low-dimensional and original-structured data into a higher-dimensional space that is more abstract and can better reflect the internal laws of the data. After receiving the feature vector at the input layer, through the weight transformation and activation operations of the first hidden layer, the linear and non-linear coupling relationships between features are initially captured, and then it is passed to the second hidden layer for further refinement and compression of its representation. Through re-mapping and non-linear activation, the ability to model the cross-response patterns between features is enhanced. When reaching the third hidden layer, the representation of the features has been transformed from the original physical space into a highly condensed representation form, forming a hidden feature vector that can be used for high-level semantic judgment. To improve the training stability and accelerate the model convergence speed, batch normalization is performed on this hidden feature vector. In each small batch, the feature distributions of each dimension are standardized to have a unified mean and variance, eliminating the differences in the order of magnitude and distribution form between features, and forming a normalized feature vector. The feature-level attention mechanism is used to calculate the weight vector of the original features based on the normalized feature vector. The attention mechanism inputs the normalized feature vector into a special linear mapping structure, and generates a weight vector representing the importance degree of each-dimensional feature through a weight calculation module. This weight vector reflects the influence intensity of different features on the final prediction target under the current working conditions. Subsequently, this weight vector is multiplied element by element with the original input target feature vector to achieve weighted processing of the features, emphasizing the expression of key features and weakening the interference of redundant features, and obtaining a feature-weighted vector with context adaptation ability. The feature-weighted vector and the normalized hidden feature vector are concatenated in dimension to form an enhanced feature vector. The enhanced feature vector is input into two parallel sub-networks in the trained VentNet model: the rotational speed prediction branch and the energy consumption prediction branch. In the rotational speed prediction branch, the enhanced feature vector is input into the first fully connected layer, where feature fusion and dimensionality reduction are performed, and output transformation is performed through a non-linear activation function to enhance its ability to model non-linear relationships; the output result is passed to the second fully connected layer, where the neurons are further mapped to the final single output node, and the optimal rotational speed prediction value of the fan under the current load condition is output. This prediction value represents the optimal operating speed of the fan under the existing environmental parameters and system state after passing through the normalized inverse transformation process, and is used as a reference for subsequent control strategies. In parallel with the rotational speed prediction branch, the energy consumption prediction branch also receives the same enhanced feature vector as input and is processed through a two-layer fully connected network. The first layer compresses the high-dimensional features into an intermediate representation space and extracts non-linear energy consumption correlation information through an activation function. The second layer further integrates the feature responses, and finally outputs the energy consumption prediction value, representing the expected energy consumption level corresponding to the fan at the previously predicted rotational speed.

[0018] Collect the first sample operation data from multiple variable-frequency centrifugal fan systems. These data cover the fan operation state parameters under different times, different loads, and environmental conditions. After normalization processing with a unified standard, a standardized training dataset is constructed. To ensure the generalization ability of model training and the scientific nature of evaluation, this dataset is divided into training set data, validation set data, and test set data according to a set ratio. Among them, the training set is used for the update and iteration of model parameters, the validation set is used for the adjustment and selection of model structure and hyperparameters, and the test set is used for the final performance evaluation to ensure that the model has a stable and reliable promotion ability. After completing the dataset division, enter the rough training stage. It mainly faces the training set data. By constructing a basic version of the VentNet network model, carry out model structure initialization and parameter training. The VentNet model includes a feature extraction network and two output branches, namely the rotational speed prediction branch and the energy consumption prediction branch. Among them, the feature extraction network consists of multiple fully connected layers and is used to extract potential high-dimensional representations from the original features; the two prediction branches respectively perform output calculations for the target rotational speed and target energy consumption of the fan. During the model training process, the commonly used mini-batch gradient descent method is combined with the Adam optimizer for parameter update. The Adam optimizer improves the training efficiency by dynamically adjusting the learning rate and avoids falling into local optima. In terms of training objectives, the weighted sum of the rotational speed prediction error and the energy consumption prediction error is used as the loss function. The prediction deviation of each branch is calculated through the mean square error function, and the total loss value is synthesized according to a fixed weight to guide the backpropagation and update of the entire network parameters. After multiple rounds of training, a basic model with good performance on the general data of multiple fans is obtained, that is, the general VentNet model. On the basis of the general model, a structure optimization mechanism is introduced to improve the expression ability and training stability of the model under complex working conditions. This optimization process includes embedding the feature-level attention mechanism into the VentNet model, enabling the network to adaptively identify and enhance the importance of key features when processing feature inputs, suppressing the interference of redundant signals, and thus improving the overall modeling accuracy. At the same time, residual connections are added to the network structure. By adding direct skip connections in the feature propagation path, the problems of gradient disappearance and feature attenuation in deep networks are effectively alleviated, thereby improving the training depth and expression efficiency. To ensure the stability of data distribution during the training process, a batch normalization layer is added after the output of each main hidden layer, so that the feature output of each small batch has a unified mean and variance before transmission, enhancing the training stability. After the above structure enhancement processing, a stronger enhanced VentNet model structure is obtained. Then enter the fine training stage, specifically for the second sample operation data of a certain target variable-frequency centrifugal fan for training and fine-tuning. Since the operation data of a single fan is usually limited in quantity, direct training will lead to overfitting or insufficient learning of the model. Therefore, the original samples are expanded through data augmentation techniques.The data augmentation methods include introducing random Gaussian noise to simulate measurement disturbances, using the time sliding window technique to generate more training segments, and slightly perturbing the eigenvalue to improve the model's robustness to small changes. These augmented samples are used to continue training the enhanced VentNet model, and an L2 regularization term is further introduced into the loss function as a constraint term for the parameter complexity to prevent overfitting during the fine-tuning process. At this time, the loss function is modified to a linear combination of the weighted sum of the original rotational speed prediction error and energy consumption prediction error and the L2 norm of the model parameters. The regularization term ensures that the model maintains a reasonable range of parameter values while pursuing prediction accuracy, thereby obtaining the target model parameters. Based on the validation set data and the target model parameters, the performance of the enhanced VentNet structure and training parameters is optimized and fine-tuned. By comparing the performance of the model on the validation set under different parameter combinations, the model structure and parameter configuration with the smallest error and the strongest stability are selected as the final output, and the trained VentNet dual-network deep neural network is obtained.

[0019] Step 400: Calculate the control error data according to the optimal rotational speed prediction value and energy consumption prediction value, and input the control error data into the hyperbolic reaching law controller for processing to obtain the frequency adjustment command of the frequency converter; Specifically, in the control system, the current actual speed of the variable-frequency centrifugal fan is obtained and compared with the optimal speed predicted by the model. By calculating the difference between the two, the speed control error at the current moment is obtained, which reflects the deviation between the current operation of the fan and the ideal state. On this basis, the change rate of the error is calculated, that is, the difference between the speed control error at the current moment and the speed control error at the previous moment, to quantify the change trend of the fan speed. The speed control error and the error change rate are combined to form comprehensive control error data. Based on this control error data and a preset positive coefficient parameter, a sliding surface equation is constructed. The sliding surface equation is obtained by multiplying the error at the current moment by the coefficient and adding the difference between the error at the current moment and the error at the previous moment, resulting in the sliding surface state quantity. The sliding surface state quantity can reflect the dynamic response of the control system and ensure the stability and robustness of the system. By applying the hyperbolic reaching law to the sliding surface state quantity, the basic control quantity at the current moment is calculated. The hyperbolic reaching law adjusts the control law so that the system can quickly converge to the optimal state within a finite number of steps, avoiding the problems of excessive oscillation or slow adjustment in traditional methods. At the same time, to enhance the stability of the control system, a Lyapunov function is constructed based on the sliding surface state quantity and the energy consumption prediction value, which is used to analyze the stability of the system. By calculating the difference in the Lyapunov function values at the previous and current moments, the increment of the function is determined, and by analyzing the sign condition of this increment, the parameter constraint conditions for ensuring the stable convergence of the system are obtained. These parameter constraint conditions play a role in restricting the range of change of the control input, ensuring that the system will not become unstable or exceed physical limits due to excessive control input. Based on the parameter constraint conditions and the relationship between the actual speed of the variable-frequency centrifugal fan at the previous moment, the actual speed at the current moment, and the control input at the previous moment, the difference between the actual state change and the theoretical state change is calculated. Through this calculation, the disturbance at the current moment is estimated, and the control signal is compensated to offset the influence of external disturbances. The basic control quantity at the current moment is subtracted from the disturbance compensation quantity to obtain the comprehensive control quantity. The comprehensive control quantity combines the control error, the system dynamic response, and the effect of disturbance compensation, and can more precisely adjust the operating state of the fan. To ensure the physical feasibility of the comprehensive control quantity, upper and lower limit constraints are imposed on it, restricting it within the frequency adjustment range of the frequency converter, and finally the frequency adjustment instruction for the frequency converter is obtained.

[0020] The sliding surface state variable is an important parameter for describing system error, reflecting the deviation between the current state of the system and the ideal target. The purpose of inputting the sliding surface state variable into the hyperbolic reaching law is to dynamically adjust the control input so that the system can quickly converge to the optimal solution within a finite number of steps, thereby ensuring the energy-saving operation of the variable-frequency centrifugal fan under different loads and operating conditions. To achieve this goal, a scale transformation is performed on the sliding surface state variable. When applying the hyperbolic reaching law, the sliding surface state variable is multiplied by a preset shape parameter, which is used to adjust the transition region of the system response. The role of the shape parameter is to control the speed and amplitude of system convergence, thereby affecting the smoothness and stability of the control signal. Through this scale transformation, the scaled state variable is obtained. The hyperbolic tangent function is applied to the scaled state variable for non-linear transformation. The hyperbolic tangent function is a non-linear activation function used for system control, which can smooth the changes in the state variable and avoid excessive oscillation of the system. Through this transformation, the hyperbolic tangent output value is between -1 and 1, which means that the control signal can be maintained within a reasonable range, avoiding excessive control amounts that may cause system instability. The hyperbolic tangent output value is multiplied by the first control gain parameter to obtain the non-linear control component. The first control gain parameter determines the weight of the non-linear part in the overall control, reflecting the sensitivity of the system to non-linear errors. By adjusting this parameter, it can better adapt to the dynamic response requirements of the fan under different operating conditions. At the same time, the sliding surface state variable is multiplied by the second control gain parameter to obtain the linear control component. The second control gain parameter is used to adjust the linear response of the system to errors, ensuring that the control system can provide appropriate adjustment force when the error is small or the system is close to the target state. The non-linear control component and the linear control component are summed and inverted to form the hyperbolic reaching law control law. This control law comprehensively considers the influences of the non-linear and linear parts, can achieve fast convergence and effectively avoid the chattering phenomenon of the system. Through the hyperbolic reaching law, the controller ensures that the system maintains a smooth transition while making rapid adjustments, thereby avoiding unnecessary mechanical losses or excessive fluctuations. On this basis, according to the absolute value of the sliding surface state variable and the absolute value of the change in the state variable at the current moment, combined with the preset learning rate, the adaptive update value of the control gain parameter is calculated. The learning rate is a key factor for adjusting the change rate of the control gain, which can automatically adjust the size of the control gain parameter according to the change of the current error. Through this adaptive update mechanism, the control system dynamically adjusts the gain parameter to adapt to the influences of different load changes and external disturbances. Specifically, according to the change of the sliding surface state variable, the adjustment amount of the control gain is calculated, and combined with the learning rate to determine the update step of the gain parameter, ensuring that the system can flexibly respond to complex operating conditions while remaining stable. The updated control gain parameter is substituted into the previously formed hyperbolic reaching law control law to recalculate the control input value. The control input value determines the frequency adjustment amplitude of the frequency converter, and thus adjusts the rotational speed of the fan.By precisely calculating and controlling the input value, the control system ensures that the operation of the variable-frequency centrifugal fan always remains in the optimal energy-saving state according to the real-time working conditions and load demands.

[0021] Perform a subtraction operation on the basic control quantity at the current moment and the disturbance compensation quantity at the current moment to obtain a comprehensive control quantity, which represents the adjustment quantity required by the system considering the disturbance. Compare the comprehensive control quantity with the maximum adjustment amplitude allowed by the frequency converter. The maximum amplitude is determined by the design limitations of the frequency converter. Through this comparison, ensure that the control quantity does not exceed the range that the frequency converter hardware can withstand, and avoid equipment damage or system instability caused by excessive adjustment signals. Based on the comparison result, the control system uses a limiting function to constrain and ensure that the amplitude of the control signal is within the allowed physical range. According to the physical characteristics of the variable-frequency centrifugal fan and the fan performance curve, perform time-domain smoothing on the control quantity after amplitude constraint, evaluate the long-term impact of control signals at different moments on the fan state, and smooth out instantaneous fluctuations. Add the smoothed control quantity to the current operating frequency to calculate the target adjustment frequency, which represents the expected frequency after the fan is adjusted. Conduct a physical limit check on the target adjustment frequency to ensure that the frequency value is within the physical adjustment range of the frequency converter. The adjustment range of the frequency converter is physically limited by electrical equipment and the mechanical structure of the fan. If the frequency is too high or too low, it will cause equipment failures. Therefore, conduct a physical limit check to ensure that the calculated frequency value conforms to the actual adjustment ability of the frequency converter and obtain a safe frequency value. Based on the number of pole pairs of the fan motor and the relationship between frequency and speed, convert the safe frequency value into the corresponding target speed value. There is a certain mathematical relationship between the number of pole pairs of the motor and the frequency, enabling the speed of the fan to be directly calculated through the frequency. Through this conversion, the target frequency is accurately converted into the target speed. To ensure the stability of the system and avoid mechanical losses of the fan during frequent adjustments, perform an acceleration limit process on the target speed value. Acceleration limit means that the rate of change of the fan speed is restricted, that is, the system only allows a certain amplitude of speed change per unit time. This limit is based on the mechanical design of the fan and is set as the maximum speed change rate to avoid excessive wear or damage of mechanical components caused by too fast speed changes. By calculating the maximum allowable speed change amount per unit time, limit the speed change rate between adjacent control cycles to obtain a frequency adjustment value with a limited speed change rate. Encode the frequency adjustment value with a limited speed change rate into a control instruction format recognizable by the frequency converter. To ensure the accuracy and security of data transmission, the control instruction is encoded into a format that the frequency converter can interpret and additional necessary communication verification information and security identification are added. The verification information is used to ensure that no data loss or damage occurs during the transmission of the control instruction, and the security identification ensures that the instruction complies with the system's security protocol throughout the execution process. Obtain the frequency adjustment instruction of the frequency converter.

[0022] Step 500: Send a frequency adjustment command to the frequency converter to adjust the actual speed of the variable-frequency centrifugal fan so that the variable-frequency centrifugal fan operates in the optimal energy-saving state.

[0023] Specifically, the frequency adjustment command is encapsulated according to a specific communication protocol to form a standardized communication data packet. The standardized communication data packet is sent to the control unit of the frequency converter. After receiving the data packet, the control unit parses out the frequency adjustment command and adjusts the output frequency of the frequency converter based on this command. The frequency converter controls the speed of the fan according to the adjusted frequency, so that the variable-frequency centrifugal fan can be adjusted to the optimal energy-saving state. The control of the frequency converter is not only based on the current frequency adjustment command, but also further optimizes the operating state according to real-time feedback. During this process, the frequency converter returns its operating state information to the control system through the feedback signal interface. Through these feedback messages, the control system monitors the working state of the frequency converter in real time, especially the change of the fan speed, to ensure that the fan operates under the optimal working conditions. During the real-time monitoring of the fan operating state, the control system calculates the change rate of the actual speed and compares it with the expected speed change rate to obtain the deviation value of the adjustment process, which reflects the difference between the actual speed and the expected speed and is an important indicator for evaluating the stability of the control process. If a large deviation is found between the change of the actual speed and the expected target during the adjustment process, it indicates that there are unstable factors in the control process, and then fine-tuning is performed to ensure that the fan can operate stably in the optimal energy-saving state. Based on the deviation value of the adjustment process, the control system determines whether the adjustment process meets the preset stability index. If the deviation is too large and the system cannot meet the preset stability requirements, the control system will generate a fine-tuning command to further optimize the operating state of the fan. The fine-tuning commands are dynamically adjusted according to the current system error and stability index. They can refine the parameters during the fan operation process, further smooth the speed change, and reduce unnecessary energy consumption and mechanical losses. Through this fine-tuning process, the system ensures that the fan can work in a more stable state, thereby effectively improving the energy-saving effect and extending the equipment life. The control system collects and calculates the energy consumption data in the optimized fan state. By real-time monitoring and analyzing the optimized operating state, accurate actual energy-saving effect data is obtained. The energy-saving effect data reflects the energy efficiency performance of the fan after optimized adjustment, including the difference between the actual energy consumption and the theoretical energy consumption. The actual energy-saving effect data is stored in the historical database as the basis for parameter optimization in the next control cycle. Through the continuous accumulation and analysis of historical data, the system gradually adjusts the control strategy, so that the energy efficiency optimization level of the variable-frequency centrifugal fan is continuously improved, and the energy-saving performance and stability of the system are improved.

[0024] In the embodiments of the present application, the present invention realizes the simultaneous prediction of the optimal speed and energy consumption of a variable-frequency centrifugal fan under different load conditions by constructing a VentNet dual-network deep neural network, overcomes the defect of the prior art that it is impossible to accurately predict the performance of the fan under multiple load conditions simultaneously, and enables the control system to more accurately find the optimal operating point of the fan under the current working conditions. High-frequency noise filtering, pulsating noise and interference signal filtering, power grid fluctuation interference elimination and slow variable trend extraction are performed on the real-time parameter data set, and combined with the adaptive weighted fusion of multi-scale feature data, the accuracy and anti-interference ability of the fan performance prediction are significantly improved. Applying the hyperbolic reaching law to the control of the variable-frequency centrifugal fan, compared with the traditional discrete sliding mode control, effectively reduces the width of the quasi-sliding mode region of the system, suppresses the chattering phenomenon, makes the fan speed regulation more stable, and avoids the mechanical losses caused by frequent start-stop and large-amplitude speed changes. The system disturbance is estimated and compensated by a disturbance observer, and combined with the inherent robustness of the hyperbolic reaching law controller, the system has a stronger ability to suppress external disturbances (such as sudden loads and airflow fluctuations), ensuring the control stability under complex working conditions. The control system can make the fan speed reach the optimal energy-saving state within a limited number of steps that can be theoretically calculated, greatly improving the system response speed and accelerating the working condition adaptation ability. The problem of scarce historical operation data is solved by using a two-stage training method of coarse and fine. First, an initial model is established using the general data of multiple fans, and then the specific data of the target fan is used for fine adjustment, enabling the control system to have better generalization ability and personalized adaptation ability. The present invention realizes the real-time energy-saving optimization control of the variable-frequency centrifugal fan under different load conditions, reduces the energy consumption amplitude, and at the same time ensures the stability and reliability of the fan operation.

[0025] In a specific embodiment, the process of executing step 100 may specifically include the following steps: Install a static pressure sensor at the inlet of the variable-frequency centrifugal fan to collect the inlet static pressure data and obtain the inlet static pressure parameter; install a static pressure sensor at the outlet of the variable-frequency centrifugal fan to collect the outlet static pressure data and obtain the outlet static pressure parameter; install a flow sensor on the pipeline of the variable-frequency centrifugal fan to collect the gas flow data and obtain the flow parameter; install a power analyzer at the motor input end of the variable-frequency centrifugal fan to collect the motor power data and obtain the motor power parameter; calculate the resistance coefficient of the variable-frequency centrifugal fan through the inlet static pressure parameter, the outlet static pressure parameter and the flow parameter to obtain the system resistance coefficient parameter; install a temperature sensor in the surrounding environment of the variable-frequency centrifugal fan to collect the ambient temperature data and obtain the ambient temperature parameter; Combine the inlet static pressure parameter, the outlet static pressure parameter, the flow parameter, the motor power parameter, the system resistance coefficient parameter and the ambient temperature parameter to obtain a real-time parameter data set.

[0026] Specifically, install a static pressure sensor at the inlet of the fan to measure the static pressure of the air flow entering the fan in real time and record the pressure state of the fluid at the fan inlet. Static pressure is an important parameter of the air flow and can reflect the load condition of the fan suction port. If the inlet static pressure changes significantly, it means a change in the system load or an increase in the air flow resistance. By continuously monitoring this parameter, the control system can evaluate the working efficiency of the fan in real time under different load conditions, thereby helping to adjust the operating state of the fan so that it always remains in the optimal working range. Install a static pressure sensor at the outlet of the fan to measure the static pressure of the air flow after being compressed by the fan. The outlet static pressure is a key parameter for measuring the pressure rise capacity of the fan and the air flow compression effect inside the fan. If the difference between the outlet static pressure and the inlet static pressure is large, it indicates that the pressure rise effect of the fan is good and the air flow has been effectively accelerated. If the outlet static pressure is low, it means that the working efficiency of the fan is not high and the air flow has not been fully accelerated, and the fan speed or load state needs to be adjusted. By monitoring the inlet static pressure and the outlet static pressure simultaneously, evaluate the overall performance of the fan and detect whether it is operating in the optimal state. Install a flow sensor on the pipeline of the fan to measure the gas flow rate passing through the fan. This parameter directly reflects the conveying capacity of the fan. The flow sensor can record the volume or mass flow rate of the air flow in real time. The change in the flow rate is caused by changes in the load, fan speed adjustment, or pipeline resistance change. By continuously monitoring the flow rate, dynamically adjust the fan speed and load to ensure that the fan can operate efficiently according to the actual demand, and at the same time avoid energy consumption waste caused by air flow fluctuations. Install a power analyzer at the input end of the fan motor to measure the input power of the motor in real time, and then evaluate the energy consumption level of the fan. By collecting the motor power in real time, determine whether the fan is currently in an energy-saving working state. After installing the static pressure sensor, flow sensor, and power analyzer, calculate the resistance coefficient of the fan system based on the collected data. The resistance coefficient is an important parameter reflecting the air flow resistance inside the system and is jointly determined by the inlet static pressure, outlet static pressure, and flow rate. By the changes in these parameters, judge whether the resistance faced by the fan is reasonable. For example, if the resistance coefficient of the fan is high, it means that the fan is operating under a high load condition, the air flow is subject to a large resistance, resulting in a decrease in energy efficiency. By calculating and monitoring the resistance coefficient in real time, adjust the fan operating parameters in a timely manner to optimize the working state of the fan and avoid energy waste caused by excessive resistance. To more comprehensively evaluate the operating environment of the fan, install a temperature sensor. The ambient temperature directly affects the density of the air flow and the operating efficiency of the fan. A higher ambient temperature will increase the power demand of the fan because the density of the gas decreases and the compression of the air flow becomes more difficult. Install a temperature sensor around the fan to collect ambient temperature data in real time and provide important auxiliary information for the control system. By combining the temperature data with other operating data, judge the energy efficiency performance of the fan in a specific environment and make appropriate adjustments according to the temperature change to ensure that the fan always operates in the optimal energy-saving state.The data collected by the above sensors form a real-time parameter dataset, including information in multiple dimensions such as inlet static pressure, outlet static pressure, flow rate, motor power, system resistance coefficient, and ambient temperature. Each parameter reflects the working state of the fan at a specific moment.

[0027] In a specific embodiment, the process of executing step 200 may specifically include the following steps: Filter the high-frequency noise of the inlet static pressure parameter and the outlet static pressure parameter in the real-time parameter dataset to obtain the filtered pressure data; Filter the pulsating noise and interference signals of the flow rate parameter in the real-time parameter dataset to obtain the filtered flow rate data; Eliminate the power grid fluctuation interference of the motor power parameter in the real-time parameter dataset to obtain the filtered power data; Extract the slow variable trend of the system resistance coefficient parameter and the ambient temperature parameter in the real-time parameter dataset to obtain the filtered resistance temperature data; Calculate the dynamic characteristics based on the filtered pressure data and the filtered flow rate data, calculate the energy consumption characteristics based on the filtered power data, calculate the environmental characteristics based on the filtered resistance temperature data, and combine the dynamic characteristics, energy consumption characteristics, and environmental characteristics into multi-scale feature data; Perform adaptive weighted fusion on the multi-scale feature data according to the fan performance curve, and transform each feature into a unified feature space through a non-linear mapping function to obtain the target feature vector.

[0028] Specifically, high-frequency noise filtering is performed on the inlet static pressure and outlet static pressure parameters in the real-time parameter dataset. The static pressure data of the fan is affected by factors such as system vibration, external interference, and sensor errors. These high-frequency noises can cause data instability and inaccuracy. Through digital filtering methods, the static pressure signal is smoothed. Digital filtering methods include Butterworth low-pass filters and Kalman filters, which can effectively remove high-frequency noises, maintain the main characteristics of the signal, and ensure the accuracy of subsequent data analysis. After this process, the filtered pressure data obtained more truly reflects the working state of the fan, avoiding misjudgment caused by noise interference. For the flow rate parameters in the real-time parameter dataset, pulsating noises and interference signals are filtered. The flow rate is a key parameter for measuring the performance of the fan. However, due to factors such as air flow pulsation, pipeline reflection, and external disturbances during the operation of the fan, the flow rate signal often shows irregular fluctuations. In order to eliminate these pulsating noises and interference signals, time-domain filtering techniques are used, such as moving window filtering or wavelet transform. By smoothing the signal, high-frequency fluctuations and abnormal data are removed, and the flow rate change trend during the normal operation of the fan is retained. For the motor power parameters, due to the influence of power grid fluctuations and electrical interference, the power data fluctuates greatly. Therefore, the power data of the motor is processed to eliminate power grid fluctuation interference. Problems such as voltage fluctuations, current mutations, and frequency fluctuations in the power grid will affect the readings of the power analyzer, resulting in abnormal fluctuations in the power signal. Filtering algorithms, such as band-pass filters or Kalman filters, are used to effectively remove the frequency interference caused by power grid fluctuations. After processing, the filtered power data is obtained. The changes in the system resistance coefficient and environmental temperature parameters are relatively gentle, showing a trend of slow variables. When processing these parameters, their long-term change trends are extracted. Through time series analysis methods, such as moving average method or trend decomposition method, the long-term trends of the system resistance coefficient and environmental temperature are extracted, and the filtered resistance and temperature data are obtained. Based on the filtered pressure data, flow rate data, power data, and resistance temperature data, the dynamic characteristics, energy consumption characteristics, and environmental characteristics of the fan are calculated. The dynamic characteristics reflect the real-time changes during the operation of the fan. By analyzing the changes in pressure and flow rate, the load fluctuation and response ability of the fan are evaluated. The energy consumption characteristics are revealed through the analysis of power data, showing the energy efficiency performance of the fan under different load states, and helping to determine the optimal energy consumption state of the fan operation. The environmental characteristics, through the extraction of resistance temperature data, can reflect the influence of the external environment on the performance of the fan, especially the influence of temperature changes on the energy efficiency of the fan. These characteristics are combined to evaluate the performance of the fan under different working conditions. In order to improve the expression ability and prediction effect of the characteristic data, adaptive weighted fusion is performed on the multi-scale characteristic data according to the fan performance curve. The fan performance curve describes the operating efficiency and performance of the fan under different working conditions. By combining these characteristic data with the performance curve, the weight of each characteristic is dynamically adjusted.Adaptive weighted fusion can automatically identify the most important features according to the actual working conditions and environmental conditions, thereby improving the modeling accuracy of the system for the performance of the fan. This process is achieved through a variety of machine learning methods, such as weighted regression or the weighted mechanism in neural networks, so that the mutual relationships between different features are fully explored, thus achieving the optimal data fusion effect. The various features are transformed into a unified feature space through a non-linear mapping function. This mapping process can eliminate the dimensional differences between different features, enabling feature data from different sources to be compared and analyzed in the same space. Through non-linear mapping, the complex fan operation data is transformed into a standardized feature vector.

[0029] In a specific embodiment, the process of executing step 300 may specifically include the following steps: Input the target feature vector into the feature extraction network in the trained VentNet dual-network deep neural network, and perform feature transformation through three hidden layers in the feature extraction network to obtain a hidden feature vector; Perform batch normalization on the hidden feature vector to obtain a normalized feature vector; Adopt a feature-level attention mechanism to calculate the weight vector of the original features according to the normalized feature vector, and weight the weight vector with the original features to obtain a feature-weighted vector; Concatenate the feature-weighted vector and the normalized feature vector to obtain an enhanced feature vector; Input the enhanced feature vector into the rotational speed prediction branch in the trained VentNet dual-network deep neural network, and perform rotational speed prediction calculations through two fully connected layers in the rotational speed prediction branch to obtain the optimal rotational speed prediction value; Input the enhanced feature vector into the energy consumption prediction branch in the trained VentNet dual-network deep neural network, and perform energy consumption prediction calculations through two fully connected layers in the energy consumption prediction branch to obtain the energy consumption prediction value.

[0030] Specifically, the target feature vector is input into the feature extraction network in the trained VentNet dual-network deep neural network. The feature extraction network consists of multiple hidden layers, including three fully connected hidden layers. After receiving the input feature vector, each layer performs a linear transformation on it and processes it through a non-linear activation function (such as ReLU), enabling the network to capture complex relationships between features. As the data passes through the hidden layers layer by layer, the features gradually become more abstract in each transformation, forming a more expressive hidden feature vector that captures the high-order features of the data and represents the comprehensive performance of the fan under different operating conditions. Batch normalization is performed on the hidden feature vector to ensure the stability of the features and the efficiency of the training process. Batch normalization can ensure that each batch of data has a uniform distribution when passed to the next layer of the network, avoiding the problems of vanishing gradients or exploding gradients caused by data distribution shifts. By adjusting the mean and variance of the features, batch normalization makes their distribution relatively stable, thereby accelerating the training process of the neural network and making the model more robust. Through this step, a normalized feature vector is obtained. The feature-level attention mechanism is adopted to enhance the expressive power of the feature vector. The feature-level attention mechanism calculates the importance of each original feature based on the normalized feature vector and weights the original features according to these weights. In this way, the network adaptively identifies the features that are most important for fan performance prediction under different operating environments and conditions, thereby enhancing the prediction accuracy of the model. Through the attention mechanism, the network dynamically adjusts the weight of each feature according to the current data situation. The weight vector of the original features is weighted with the original features to obtain a feature-weighted vector. The feature-weighted vector is concatenated with the normalized feature vector to obtain an enhanced feature vector. The enhanced feature vector is input into the rotational speed prediction branch in the VentNet dual-network deep neural network. This branch performs rotational speed prediction calculations through two fully connected layers. Each layer maps the input feature vector to the rotational speed prediction space through a linear transformation of weighted sum and bias. Behind each layer, a non-linear activation function (such as ReLU) is used to enhance the expressive power of the model and enable it to capture the complex non-linear relationships between the fan rotational speed and various environmental and load conditions. After being processed by two fully connected layers, the model outputs a predicted value, which is the optimal rotational speed prediction value of the fan under the current operating conditions. At the same time, the enhanced feature vector is input into the energy consumption prediction branch in the VentNet dual-network deep neural network to predict the energy consumption level of the fan at a given rotational speed. Similar to the rotational speed prediction branch, the energy consumption prediction branch also includes two fully connected layers and performs calculations through weighted sum and non-linear activation functions in each layer. Through the processing of these layers, the system accurately predicts the energy consumption of the fan under different load and rotational speed conditions.

[0031] In a specific embodiment, the energy-saving optimization control method for the variable-frequency centrifugal fan further includes the following steps: Collect and organize the first sample operation data of multiple variable-frequency centrifugal fans to obtain a standardized training data set, and divide the standardized training data set into training set data, validation set data, and test set data; Perform the rough training stage processing based on the training set data. Use the gradient descent method and the Adam optimizer to construct a rotation speed prediction branch and an energy consumption prediction branch that include a feature extraction network, and set the weighted sum of the rotation speed prediction error and the energy consumption prediction error as the loss function to train the network parameters and obtain a general VentNet model; Add a feature-level attention mechanism and a residual connection structure to the general VentNet model, and use a batch normalization layer to process the features to obtain an enhanced VentNet model structure; Perform the fine training stage processing on the second sample operation data of the variable-frequency centrifugal fan. Use data augmentation technology to expand the samples, and add an L2 regularization term to the loss function. The loss function is modified to a combination of the weighted sum of the rotation speed prediction error and the energy consumption prediction error and the L2 norm of the model parameters for fine tuning to obtain the target model parameters; Optimize the parameters of the enhanced VentNet model structure based on the validation set data and the target model parameters to obtain a trained VentNet dual-network deep neural network.

[0032] Specifically, the operation data of multiple variable-frequency centrifugal fans are collected and sorted. By obtaining the operation data of the first sample from multiple fans, it is ensured that the model has good generalization ability and can adapt to the operation characteristics of different fans. These operation data are standardized to eliminate the dimensional differences between various features, making the ranges of all data unified, so as to ensure the smooth progress of the training of the neural network model. The standardized data set is divided into a training set, a validation set, and a test set. The training set is used for model training, the validation set is used to adjust model parameters and prevent overfitting, and the test set is used for the final performance evaluation. After the data set is prepared, it enters the rough training stage for processing. The training set data is used to initially train the model. The purpose of rough training is to quickly obtain a general model that can handle basic tasks and avoid falling into local optima. During the training process, the gradient descent method and the Adam optimizer are used to optimize the network parameters. The gradient descent method adjusts the network parameters by calculating the gradient of the loss function, so as to minimize the loss function. The Adam optimizer further improves the training efficiency by adaptively adjusting the learning rate of each parameter and avoids the difficulty of learning rate selection in the standard gradient descent method. In order to construct the general VentNet model, a dual-branch structure including a feature extraction network is designed, which is used for rotational speed prediction and energy consumption prediction respectively. The rotational speed prediction branch and the energy consumption prediction branch process data through the network and output the predicted values of the rotational speed and energy consumption of the fan. During the training process, the weighted sum of the rotational speed prediction error and the energy consumption prediction error is used as the loss function. This setting can balance the influence of the two, enabling the model to accurately predict the rotational speed and effectively evaluate the energy consumption of the fan. After multiple rounds of training, the obtained model is the general VentNet model. The general VentNet model is optimized and enters the construction stage of the enhanced model. The structure of the enhanced VentNet model is improved by adding a feature-level attention mechanism and a residual connection structure to enhance the performance of the model. The feature-level attention mechanism enables the network to adaptively adjust the weights of each input feature, so that the model can pay more attention to key features under specific working conditions and reduce the interference of redundant information. At the same time, the residual connection structure can alleviate the problem of gradient disappearance in deep networks, allowing signals to be directly transmitted to deeper layers, thereby enhancing the learning ability of the model. To improve the stability of the network, a batch normalization layer is used to process features after each hidden layer. Batch normalization standardizes each batch of features to ensure that the data does not fluctuate too much during transmission, thereby accelerating the training process and enhancing the robustness of the model. The model structure optimization enters the fine training stage. The goal of the fine training stage is to fine-tune according to the operation data of the second sample to adapt to the characteristics of specific fans. Since the second sample data is relatively small, data augmentation techniques are used to expand the sample size.Data augmentation methods include means such as randomly perturbing the input data, adding noise, and sliding the time window, thereby improving the generalization ability of the model and avoiding overfitting caused by insufficient data volume. In the fine-tuning training stage, an L2 regularization term is added to the loss function. This regularization term can limit the magnitude of the model's parameters and prevent the network parameters from being too large, resulting in an overly complex model that cannot be generalized. The loss function is modified to a combination of the weighted sum of the rotational speed prediction error and the energy consumption prediction error and the L2 norm of the model parameters, enabling the model to not only optimize the prediction results but also maintain the simplicity of the parameters. After training in this stage, the target model parameters are obtained. Based on the validation set data and the target model parameters, the parameters of the enhanced VentNet model are optimized. The validation set data serves as an important basis for model tuning. By calculating the performance on the validation set, the model is fine-tuned, and parameters such as the learning rate and regularization coefficient are adjusted to ensure the stability and accuracy of the model. After parameter optimization and training, the trained VentNet dual-network deep neural network is obtained.

[0033] In a specific embodiment, the process of executing step 400 may specifically include the following steps: Calculate the rotational speed control error at the current moment according to the difference between the current actual rotational speed of the variable-frequency centrifugal fan and the optimal rotational speed prediction value, and calculate the error change rate according to the difference between the rotational speed control error at the current moment and the rotational speed control error at the previous moment. Combine the rotational speed control error and the error change rate to obtain control error data; Based on the control error data and a preset positive coefficient parameter, construct a sliding surface equation. Multiply the error at the current moment by the coefficient and add it to the difference between the error at the current moment and the error at the previous moment to obtain the sliding surface state quantity at the current moment; Apply the hyperbolic reaching law to the sliding surface state quantity to calculate the basic control quantity at the current moment; Construct a Lyapunov function based on the sliding surface state quantity and the energy consumption prediction value. Calculate the difference between the Lyapunov function values at the previous and current moments to determine the function increment, and obtain the parameter constraint conditions to ensure the stable convergence of the system by analyzing the sign conditions of the function increment; Based on the parameter constraint conditions and the relationship between the actual rotational speed of the variable-frequency centrifugal fan at the previous moment, the actual rotational speed at the current moment, and the control input at the previous moment, calculate the difference between the actual state change and the theoretical state change to obtain the disturbance compensation quantity at the current moment; Perform a subtraction operation on the basic control quantity at the current moment and the disturbance compensation quantity at the current moment to obtain the comprehensive control quantity, and perform upper and lower limit constraint processing on the comprehensive control quantity to ensure that the comprehensive control quantity meets the physical limitations of the frequency adjustment range of the frequency converter, and obtain the frequency adjustment instruction for the frequency converter in the variable-frequency centrifugal fan.

[0034] Specifically, calculate the rotational speed control error at the current moment based on the difference between the current actual rotational speed of the fan and the predicted value of the optimal rotational speed, which represents the deviation between the actual rotational speed and the theoretical optimal rotational speed, and evaluate whether the current operating state of the fan is close to the ideal state. When there is a gap between the actual rotational speed of the fan and the predicted optimal rotational speed, this error needs to be monitored and adjusted in real time. By calculating this control error, it is determined whether the fan is in the optimal operating state, and then appropriate adjustments are made to optimize its energy efficiency and stability. To better capture the dynamic characteristics of the fan's rotational speed change, calculate the error change rate. The error change rate is the difference between the rotational speed control error at the current moment and the rotational speed control error at the previous moment, which can reflect the change trend of the rotational speed error over a period of time. The introduction of the error change rate helps the system determine whether the change in rotational speed is tending to be stable and avoid unstable operation of the fan caused by excessive error fluctuations. When the error change rate is large, it indicates that the rotational speed change is unstable, and more stringent control measures need to be taken to avoid drastic fluctuations in the fan's rotational speed. Combine the rotational speed control error at the current moment and the error change rate into control error data. Based on the control error data and the preset positive coefficient parameter, construct a sliding surface equation. The sliding surface equation defines a reference standard for adjusting the state of the control system. By multiplying the error at the current moment by the coefficient and adding the difference between the error at the current moment and the error at the previous moment, the sliding surface state quantity at the current moment is obtained, which reflects the dynamic performance of the current error of the system. Apply the hyperbolic reaching law to the sliding surface state quantity to calculate the basic control quantity at the current moment. The hyperbolic reaching law adjusts the control input according to the change of the sliding surface state quantity to make the system converge to the ideal state as soon as possible. Construct a Lyapunov function based on the sliding surface state quantity and the energy consumption prediction value. The Lyapunov function is a mathematical tool for analyzing the stability of dynamic systems. By calculating the difference between the Lyapunov function values at the previous and current moments, the function increment is obtained, and the sign condition of the increment is analyzed to determine whether the control strategy can ensure the stable convergence of the system. If the increment of the Lyapunov function is negative, it indicates that the system is converging stably, while if the increment is positive, it indicates that the system will become unstable. Based on the obtained parameter constraint conditions and the relationship between the actual rotational speed of the fan, the control input at the previous moment, and the actual rotational speed at the current moment, calculate the difference between the actual state change and the theoretical state change, and obtain the disturbance compensation quantity at the current moment. The role of the disturbance compensation quantity is to correct the control error caused by external disturbances or internal uncertainties, so that the system can more accurately track the target state. During actual operation, the fan will be affected by external factors such as air flow fluctuations and power grid fluctuations, resulting in changes in rotational speed or load. By introducing the disturbance compensation quantity, adjustments are made in a timely manner when disturbances occur, thereby maintaining the stability and high efficiency of the system. Perform a subtraction operation on the basic control quantity at the current moment and the disturbance compensation quantity at the current moment to obtain the comprehensive control quantity.Perform upper and lower limit constraint processing on the comprehensive control quantity to ensure that the comprehensive control quantity does not exceed the physical limit of the frequency adjustment of the frequency converter, and finally obtain the frequency adjustment instruction of the frequency converter in the variable frequency centrifugal fan.

[0035] In a specific embodiment, the process of applying the hyperbolic reaching law to the sliding surface state quantity and calculating the basic control quantity at the current moment may specifically include the following steps: Apply the hyperbolic reaching law, perform scale transformation on the sliding surface state quantity by multiplying it with a preset shape parameter to obtain the scaled state quantity; Input the scaled state quantity into the hyperbolic tangent function for non-linear transformation to obtain the hyperbolic tangent output value; Multiply the first control gain parameter by the hyperbolic tangent output value to obtain the non-linear control component, and multiply the sliding surface state quantity by the second control gain parameter to obtain the linear control component; Sum and invert the non-linear control component and the linear control component to form the hyperbolic reaching law control law; Based on the absolute value of the sliding surface state quantity and the absolute value of the change in the state quantity at the current moment, combined with the preset learning rate, calculate the adaptive update value of the control gain parameter to obtain the updated control gain parameter; Substitute the updated control gain parameter into the hyperbolic reaching law control law to recalculate the control input value to obtain the basic control quantity at the current moment.

[0036] Specifically, multiply the sliding surface state quantity by a preset shape parameter for scale transformation to adjust the response rate of the control system to errors. The preset shape parameter can determine the speed and accuracy of system convergence. Through this scale transformation, the scaled state quantity is obtained. By adjusting the shape parameter, the system can flexibly change the sensitivity of the control strategy under different control objectives, ensuring that the fan can be quickly and smoothly adjusted to the optimal state under various operating conditions. Input the scaled state quantity into the hyperbolic tangent function for non-linear transformation to obtain the hyperbolic tangent output value. The hyperbolic tangent function is a non-linear activation function that can effectively map the input signal to the range between -1 and 1. The purpose of non-linear transformation is to enable the system's response to not only accurately track the target value but also suppress large errors. The characteristics of the hyperbolic tangent function are such that when the error is small, the control system provides a small adjustment amount, and when the error is large, the response of the control system becomes more aggressive, thereby accelerating system convergence. Through this non-linear transformation, the control input of the system can more flexibly and intelligently adapt to different error situations, ensuring that the system does not produce excessive fluctuations during the rapid adjustment process. Multiply the first control gain parameter by the hyperbolic tangent output value to obtain the non-linear control component, which represents the dynamic response part caused by errors in the control system and helps the system quickly adjust to the target state. At the same time, multiply the sliding surface state quantity by the second control gain parameter to obtain the linear control component. The second control gain parameter is used to adjust the linear response of the system to errors, ensuring that when the error is small or close to the target, the system can maintain an appropriate adjustment force. The role of the linear control component is to balance the error in the stable state of the system, so that the control signal is not too drastic, thereby ensuring that the fan speed or energy consumption is within a stable range. Sum the non-linear control component and the linear control component and take the negative of the result to form the hyperbolic reaching law control law. This control law balances the dynamic response and stability of the system through the combination of non-linear and linear parts. By summing the two, the balance between the rapid adjustment of the fan speed and the long-term stability is comprehensively considered, avoiding oscillations or energy waste caused by over-regulation. Based on the absolute value of the sliding surface state quantity at the current moment and the absolute value of the change in the state quantity, combined with a preset learning rate, calculate the adaptive update value of the control gain parameter. The learning rate is a key factor in controlling the gain adjustment and determines the amplitude of each update. An overly large learning rate will cause the system to over-adjust, while an overly small learning rate will cause the system to respond sluggishly. By combining the absolute value of the sliding surface state quantity at the current moment and the absolute value of the change in the state quantity, the most suitable control gain update value for the current working condition is dynamically calculated. The preset learning rate enables the system to maintain an appropriate adjustment rate under different control objectives and load conditions, thereby improving the control accuracy and efficiency. After adaptively adjusting the control gain parameter, substitute the updated gain parameter into the hyperbolic reaching law control law to recalculate the control input value to obtain the basic control quantity at the current moment.The basic control quantity is the frequency adjustment command for the frequency converter, which determines the adjustment range of the fan speed. With the adjustment of the basic control quantity, the system makes rapid adjustments according to the real-time error and control target to ensure that the fan can always operate in the best energy-saving state. For example, if the current speed error is large, the control system increases the basic control quantity to accelerate the adjustment process; if the error is small, the control system reduces the control quantity to avoid unnecessary mechanical damage and energy waste caused by frequent speed changes.

[0037] In a specific embodiment, the process of performing a subtraction operation on the basic control quantity at the current moment and the disturbance compensation quantity at the current moment to obtain a comprehensive control quantity, and performing upper and lower limit constraint processing on the comprehensive control quantity to ensure that the physical limits of the frequency adjustment range of the frequency converter are met, and obtaining the frequency adjustment command of the frequency converter may specifically include the following steps: Perform a subtraction operation on the basic control quantity at the current moment and the disturbance compensation quantity at the current moment to obtain a comprehensive control quantity; Compare the comprehensive control quantity with the maximum allowable adjustment amplitude of the frequency converter to obtain a comparison result, and perform constraint through a clipping function based on the comparison result to obtain a control quantity after amplitude constraint; Perform time-domain smoothing processing on the control quantity after amplitude constraint according to the physical characteristics of the variable-frequency centrifugal fan and the fan performance curve to obtain a smoothed control quantity; Add the smoothed control quantity to the current operating frequency, calculate the target adjustment frequency, and perform a physical limit check on the target adjustment frequency to ensure that the physical limits of the frequency adjustment range of the frequency converter are met, and obtain a safe frequency value; Based on the number of pole pairs of the fan motor and the relationship between frequency and speed, convert the safe frequency value into the corresponding target speed value; Perform acceleration limit processing on the target speed value, calculate the maximum allowable speed change per unit time, and limit the speed change rate between adjacent control cycles to obtain a frequency adjustment value with limited speed change rate; Encode the frequency adjustment value with limited speed change rate into the control instruction format recognizable by the frequency converter, and add communication verification information and safety identification to obtain the frequency adjustment command of the frequency converter.

[0038] Specifically, subtract the disturbance compensation amount from the basic control amount at the current moment to obtain the comprehensive control amount. The basic control amount is an adjustment signal calculated based on the current state of the fan and the preset target, while the disturbance compensation amount is used to correct the errors caused by external disturbances or uncertainty factors. By subtracting these two amounts, the influence of external disturbances is removed, and a more accurate control signal is obtained. Compare the comprehensive control amount with the maximum adjustment range allowed by the frequency converter to ensure that the control signal does not exceed the adjustment ability of the device. If the comprehensive control amount exceeds the maximum adjustment range, the control amount is constrained based on the comparison result through a limiting function to ensure that the final control signal is always within the physical limit range of the frequency converter. The limiting function limits the control signal within a reasonable range to prevent system instability or mechanical damage caused by excessive adjustment. Due to the influence of factors such as load changes and environmental fluctuations during the operation of the fan, the output of the control system will show certain fluctuations. If the control amount changes too violently, it will cause unnecessary wear to the fan and even affect the energy efficiency. Therefore, perform time-domain smoothing processing on the control amount after amplitude constraint. Through filtering technology, the fluctuations of the control signal are effectively suppressed, avoiding excessive changes in the fan speed, and thus ensuring that the fan operates in a stable state. Through time-domain smoothing, the control signal is smoother, avoiding excessive frequent adjustments, which helps the fan to operate efficiently and stably. Add the smoothed control amount to the current operating frequency to calculate the target adjustment frequency. The target adjustment frequency is the optimal frequency value calculated based on the current control requirements and the operating state of the fan. To ensure that the frequency converter can correctly adjust the frequency of the fan, perform a physical limit check on the target adjustment frequency to ensure that the target frequency does not exceed the physical limit range of the frequency converter, and obtain the safe frequency value. Based on the number of pole pairs of the fan motor and the relationship between frequency and speed, convert the safe frequency value into the corresponding target speed value. There is a clear proportional relationship between the speed of the fan and the frequency. Through this relationship, the corresponding speed value is deduced from the target adjustment frequency. The target speed value is the core index of the fan operation, which determines the conveying capacity and energy efficiency level of the fan. Perform acceleration limit processing on the target speed value to control the rate of speed change, ensuring that within each control cycle, the speed change of the fan does not exceed the predetermined range. By calculating the maximum allowable speed change amount per unit time, limit the speed change rate between adjacent control cycles, thereby avoiding the influence of violent speed fluctuations on the fan. Encode the frequency adjustment value with limited speed change rate into the control instruction format recognizable by the frequency converter. And add communication verification information and safety identification. The communication verification information is used to ensure that the control signal is not lost or in error during transmission, while the safety identification is used to identify the safety and legality of the instruction to avoid illegal operations or equipment failures, and finally obtain the frequency converter frequency adjustment instruction.

[0039] In a specific embodiment, the process of executing step 500 may specifically include the following steps: Encapsulate the frequency adjustment instruction with a communication protocol to obtain a standardized communication data packet, and send the standardized communication data packet to the control unit of the frequency converter; Adjust the actual speed of the variable-frequency centrifugal fan through the control unit, so that the variable-frequency centrifugal fan operates in the best energy-saving state, and at the same time obtain the operation status information of the frequency converter through the feedback signal interface of the frequency converter; Monitor the stability of the speed change process of the variable-frequency centrifugal fan based on the operation status information, calculate the deviation between the actual speed change rate and the expected speed change rate, and obtain the adjustment process deviation value; Judge whether the adjustment process meets the preset stability index according to the adjustment process deviation value. If not, generate a fine-tuning instruction to obtain an optimized fan operation status; Collect and calculate the energy consumption data under the optimized fan operation status to obtain the actual energy-saving effect data, and store the actual energy-saving effect data in the historical database for parameter optimization in the next control cycle.

[0040] Specifically, the frequency adjustment instruction is encapsulated by a communication protocol to ensure the integrity and accuracy of the instruction during transmission. Through the communication protocol encapsulation, the instruction is converted into a normalized communication data packet that conforms to a specific data format, including the frequency information, timestamp, and other relevant parameters that the fan needs to adjust. After the control unit receives the normalized communication data packet, it adjusts the actual speed of the variable-frequency centrifugal fan according to the information in the data packet, so that the fan operates in the best energy-saving state. The role of the frequency converter is to convert the electrical signal into the frequency for adjusting the fan speed, and control the speed and energy efficiency of the fan by changing the frequency. The frequency converter adjusts the speed of the motor according to the received frequency adjustment instruction, so that the fan can operate efficiently under different load conditions. At the same time, the frequency converter continuously monitors the operating state of the fan through the feedback signal interface and returns the operating data to the control system. These feedback signals include the actual speed, power consumption, and other key operating parameters of the fan. Based on the operating state information returned by the frequency converter, the control system monitors the stability of the fan speed change process. The speed change rate is a key dynamic index in the fan operation, which reflects the change rate of the fan speed from one control cycle to the next. By calculating the deviation between the actual speed change rate and the expected speed change rate, the stability of the fan during operation is evaluated. The expected speed change rate is the ideal speed change curve obtained through system prediction, while the actual speed change rate is the actual speed change situation obtained through real-time monitoring. If the deviation between the two is large, it indicates that there are unstable factors in the system, and the speed change is too drastic, which will affect the operating efficiency and lifespan of the fan. After obtaining the deviation value of the adjustment process, it is judged whether the current adjustment process meets the preset stability index. The preset stability index is a standard set according to the design specifications, operating requirements, and energy-saving goals of the fan. These standards include the maximum allowable speed change amplitude, the smoothness of the speed change, etc. If the system finds that the adjustment process deviation is too large, it means that the fan fails to meet the set stability requirements during the current adjustment process, then the control system generates a fine-tuning instruction to make a subtle correction to the current adjustment process to ensure that the fan can stably reach the best energy-saving state. The generation of the fine-tuning instruction takes into account factors such as control error and environmental changes, optimizes the adjustment amplitude based on real-time data, makes the system more accurately adjust the operating state of the fan, and thus realizes the optimization of the stability and energy efficiency of the fan operation. After the fine-tuning instruction is executed, the operating state of the fan is optimized, and the control system continues to monitor and collect the optimized fan energy consumption data. By real-time collecting these energy consumption data, the actual energy-saving effect data is obtained, which reflects the amount of energy saved by the fan after optimized control. By comparing the energy consumption levels before and after optimization, the effectiveness of the control strategy is evaluated to ensure that the energy-saving effect of the fan meets the expected goal. The energy-saving effect data is stored in the historical database for reference in future control cycles. At the beginning of the next control cycle, the control parameters are optimized through the data in the historical database.

[0041] The energy-saving optimization control method of the variable-frequency centrifugal fan in the embodiment of the present application has been described above. Next, the energy-saving optimization control system 10 of the variable-frequency centrifugal fan in the embodiment of the present application will be described. Please refer to Figure 2 , an embodiment of the energy-saving optimization control system 10 of the variable-frequency centrifugal fan in the embodiment of the present application includes: An acquisition module 11, configured to collect the inlet static pressure, outlet static pressure, flow rate, motor power, system resistance coefficient, and ambient temperature of the variable-frequency centrifugal fan through a sensor network, and obtain a real-time parameter data set; A feature extraction module 12, configured to perform digital filtering and feature extraction on the real-time parameter data set to obtain a target feature vector; A prediction calculation module 13, configured to input the target feature vector into the trained VentNet dual-network deep neural network for prediction calculation, and obtain the best speed prediction value and energy consumption prediction value under the current load condition; A processing module 14, configured to calculate control error data according to the best speed prediction value and energy consumption prediction value, and input the control error data into a hyperbolic reaching law controller for processing to obtain a frequency adjustment instruction for the frequency converter; An adjustment module 15, configured to send the frequency adjustment instruction to the frequency converter to adjust the actual speed of the variable-frequency centrifugal fan, so that the variable-frequency centrifugal fan operates in the best energy-saving state.

[0042] Through the collaborative cooperation of the above-mentioned various components, the present invention realizes the simultaneous prediction of the optimal speed and energy consumption of a variable-frequency centrifugal fan under different load conditions by constructing a VentNet dual-network deep neural network, overcomes the defect that the prior art cannot accurately predict the performance of a fan under multiple load conditions simultaneously, and enables the control system to more precisely find the optimal operating point of the fan under the current working conditions. High-frequency noise filtering, pulsating noise and interference signal filtering, power grid fluctuation interference elimination, and slow-variable trend extraction are performed on the real-time parameter data set. Combined with the adaptive weighted fusion of multi-scale feature data, the accuracy and anti-interference ability of fan performance prediction are significantly improved. Applying the hyperbolic reaching law to the control of a variable-frequency centrifugal fan effectively reduces the width of the quasi-sliding mode region of the system and suppresses the chattering phenomenon compared with the traditional discrete sliding mode control, making the fan speed regulation more stable and avoiding mechanical losses caused by frequent starting and stopping and large-scale speed changes. The system disturbance is estimated and compensated by a disturbance observer. Combined with the inherent robustness of the hyperbolic reaching law controller, the system has a stronger ability to suppress external disturbances (such as sudden loads and airflow fluctuations), ensuring control stability under complex working conditions. The control system can make the fan speed reach the optimal energy-saving state within a theoretically calculable finite number of steps, greatly improving the system response speed and accelerating the working condition adaptation ability. The problem of scarce historical operation data is solved by using a two-stage training method of coarse and fine. First, an initial model is established using general data of multiple fans, and then fine-tuning is performed using specific data of the target fan, enabling the control system to have better generalization ability and personalized adaptation ability. The present invention realizes real-time energy-saving optimization control of a variable-frequency centrifugal fan under different load conditions, reduces the energy consumption amplitude, and at the same time ensures the stability and reliability of the fan operation.

[0043] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described system, system, and unit can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0044] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0045] As described above, the above embodiments are only used to illustrate the technical solution of this application, rather than to limit it; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of various embodiments of this application.

Claims

1. An energy-saving optimization control method for a variable-frequency centrifugal fan, characterized in that, Including: Collect the inlet static pressure, outlet static pressure, flow rate, motor power, system resistance coefficient, and ambient temperature of the variable-frequency centrifugal fan through a sensor network to obtain a real-time parameter dataset; Perform digital filtering and feature extraction on the real-time parameter dataset to obtain a target feature vector; Input the target feature vector into the trained VentNet dual-network deep neural network for prediction calculation to obtain the optimal rotational speed prediction value and energy consumption prediction value under the current load condition; Calculate the control error data based on the optimal rotational speed prediction value and the energy consumption prediction value, and input the control error data into a hyperbolic reaching law controller for processing to obtain the frequency adjustment command for the frequency converter; Send the frequency adjustment command to the frequency converter to adjust the actual rotational speed of the variable-frequency centrifugal fan so that the variable-frequency centrifugal fan operates in the optimal energy-saving state.

2. The energy-saving optimization control method of the variable-frequency centrifugal fan according to claim 1, characterized in that The collecting the inlet static pressure, outlet static pressure, flow rate, motor power, system resistance coefficient, and ambient temperature of the variable-frequency centrifugal fan through a sensor network to obtain a real-time parameter dataset includes: Install a static pressure sensor at the inlet of the variable-frequency centrifugal fan to collect inlet static pressure data and obtain the inlet static pressure parameter; install a static pressure sensor at the outlet of the variable-frequency centrifugal fan to collect outlet static pressure data and obtain the outlet static pressure parameter; install a flow sensor on the pipeline of the variable-frequency centrifugal fan to collect gas flow data and obtain the flow rate parameter; install a power analyzer at the motor input end of the variable-frequency centrifugal fan to collect motor power data and obtain the motor power parameter; calculate the resistance coefficient of the variable-frequency centrifugal fan through the inlet static pressure parameter, outlet static pressure parameter, and flow rate parameter to obtain the system resistance coefficient parameter; install a temperature sensor in the surrounding environment of the variable-frequency centrifugal fan to collect ambient temperature data and obtain the ambient temperature parameter; Combine the inlet static pressure parameter, the outlet static pressure parameter, the flow rate parameter, the motor power parameter, the system resistance coefficient parameter, and the ambient temperature parameter to obtain a real-time parameter dataset.

3. The energy-saving optimization control method for the variable-frequency centrifugal fan according to claim 1, wherein, The performing digital filtering and feature extraction on the real-time parameter dataset to obtain a target feature vector includes: Filter the high-frequency noise of the inlet static pressure parameter and the outlet static pressure parameter in the real-time parameter dataset to obtain the filtered pressure data; Filter the pulsating noise and interference signals of the flow rate parameter in the real-time parameter dataset to obtain the filtered flow rate data; Eliminate the power grid fluctuation interference of the motor power parameter in the real-time parameter dataset to obtain the filtered power data; Extract the slow variable trend of the system resistance coefficient parameter and the ambient temperature parameter in the real-time parameter dataset to obtain the filtered resistance temperature data; Calculate the dynamic feature based on the filtered pressure data and the filtered flow rate data, calculate the energy consumption feature based on the filtered power data, calculate the environmental feature based on the filtered resistance temperature data, and combine the dynamic feature, the energy consumption feature, and the environmental feature into multi-scale feature data; Adaptive weighted fusion is performed on the multi-scale feature data according to the fan performance curve, and each feature is transformed into a unified feature space through a non-linear mapping function to obtain a target feature vector.

4. The energy-saving optimization control method of the variable-frequency centrifugal fan according to claim 1, characterized in that Inputting the target feature vector into the trained VentNet dual-network deep neural network for prediction calculation to obtain the optimal rotational speed prediction value and energy consumption prediction value under the current load condition includes: Inputting the target feature vector into the feature extraction network in the trained VentNet dual-network deep neural network, and performing feature transformation through three hidden layers in the feature extraction network to obtain a hidden feature vector; Performing batch normalization processing on the hidden feature vector to obtain a normalized feature vector; Using a feature-level attention mechanism to calculate the weight vector of the original feature according to the normalized feature vector, and weighting the weight vector with the original feature to obtain a feature-weighted vector; Concatenating the feature-weighted vector and the normalized feature vector to obtain an enhanced feature vector; Inputting the enhanced feature vector into the rotational speed prediction branch in the trained VentNet dual-network deep neural network, and performing rotational speed prediction calculation through two fully connected layers in the rotational speed prediction branch to obtain the optimal rotational speed prediction value; Inputting the enhanced feature vector into the energy consumption prediction branch in the trained VentNet dual-network deep neural network, and performing energy consumption prediction calculation through two fully connected layers in the energy consumption prediction branch to obtain the energy consumption prediction value.

5. The energy-saving optimization control method of the variable-frequency centrifugal fan according to claim 4, wherein, The energy-saving optimization control method for the variable-frequency centrifugal fan further includes: Collecting and organizing the first sample operation data of multiple variable-frequency centrifugal fans to obtain a standardized training data set, and dividing the standardized training data set into training set data, validation set data, and test set data; Performing rough training stage processing based on the training set data, using the gradient descent method and the Adam optimizer, constructing a rotational speed prediction branch and an energy consumption prediction branch including a feature extraction network, and setting the weighted sum of the rotational speed prediction error and the energy consumption prediction error as the loss function to perform network parameter training to obtain a general VentNet model; Adding a feature-level attention mechanism and a residual connection structure to the general VentNet model, and using a batch normalization layer to process the features to obtain an enhanced VentNet model structure; Performing fine training stage processing on the second sample operation data of the variable-frequency centrifugal fan, using data augmentation technology to expand the samples, and adding an L2 regularization term to the loss function. The loss function is modified to a combination of the weighted sum of the rotational speed prediction error and the energy consumption prediction error and the L2 norm of the model parameters to perform fine tuning to obtain the target model parameters; Based on the validation set data and the target model parameters, optimizing the parameters of the enhanced VentNet model structure to obtain the trained VentNet dual-network deep neural network.

6. The energy-saving optimization control method of the variable-frequency centrifugal fan according to claim 1, characterized in that, Calculating control error data according to the optimal rotational speed prediction value and the energy consumption prediction value, and inputting the control error data into a hyperbolic reaching law controller for processing to obtain a control signal, including: Calculate the rotational speed control error at the current moment according to the difference between the current actual rotational speed of the variable-frequency centrifugal fan and the predicted value of the optimal rotational speed, and calculate the error change rate according to the difference between the rotational speed control error at the current moment and the rotational speed control error at the previous moment. Combine the rotational speed control error and the error change rate to obtain control error data; Based on the control error data and a preset positive coefficient parameter, construct a sliding surface equation. Multiply the error at the current moment by a coefficient and add it to the difference between the error at the current moment and the error at the previous moment to obtain the sliding surface state quantity at the current moment; Apply the hyperbolic reaching law to the sliding surface state quantity to calculate the basic control quantity at the current moment; Construct a Lyapunov function according to the sliding surface state quantity and the predicted value of the energy consumption. Calculate the difference between the Lyapunov function values at the previous and current moments to determine the function increment, and obtain the parameter constraint conditions for ensuring the stable convergence of the system by analyzing the sign conditions of the function increment; Based on the parameter constraint conditions and the relationship between the actual rotational speed of the variable-frequency centrifugal fan at the previous moment, the actual rotational speed at the current moment, and the control input at the previous moment, calculate the difference between the actual state change and the theoretical state change to obtain the disturbance compensation quantity at the current moment; Perform a subtraction operation on the basic control quantity at the current moment and the disturbance compensation quantity at the current moment to obtain a comprehensive control quantity, and perform upper and lower limit constraint processing on the comprehensive control quantity to ensure that the comprehensive control quantity meets the physical limitations of the frequency adjustment range of the frequency converter, thereby obtaining the frequency adjustment instruction for the frequency converter in the variable-frequency centrifugal fan.

7. The energy-saving optimization control method of the variable-frequency centrifugal fan according to claim 6, characterized in that The step of applying the hyperbolic reaching law to the sliding surface state quantity to calculate the basic control quantity at the current moment includes: Apply the hyperbolic reaching law, and perform a scale transformation on the sliding surface state quantity by multiplying it by a preset shape parameter to obtain a scaled state quantity; Input the scaled state quantity into the hyperbolic tangent function for non-linear transformation to obtain a hyperbolic tangent output value; Multiply the first control gain parameter by the hyperbolic tangent output value to obtain a non-linear control component, and multiply the sliding surface state quantity by the second control gain parameter to obtain a linear control component; Sum and invert the non-linear control component and the linear control component to form a hyperbolic reaching law control law; Based on the absolute value of the sliding surface state quantity at the current moment and the absolute value of the change in the state quantity, and in combination with a preset learning rate, calculate the adaptive update value of the control gain parameter to obtain the updated control gain parameter; Substitute the updated control gain parameter into the hyperbolic reaching law control law to recalculate the control input value to obtain the basic control quantity at the current moment.

8. The energy-saving optimization control method of the variable-frequency centrifugal fan according to claim 6, wherein The step of performing a subtraction operation on the basic control quantity at the current moment and the disturbance compensation quantity at the current moment to obtain a comprehensive control quantity, and performing upper and lower limit constraint processing on the comprehensive control quantity to ensure that it meets the physical limitations of the frequency adjustment range of the frequency converter to obtain the frequency adjustment instruction for the frequency converter includes: Perform a subtraction operation on the basic control quantity at the current moment and the disturbance compensation quantity at the current moment to obtain a comprehensive control quantity; Compare the comprehensive control quantity with the maximum adjustment range allowed by the frequency converter to obtain a comparison result, and perform constraints through a limiting function based on the comparison result to obtain the controlled quantity after amplitude constraint; Perform time-domain smoothing processing on the controlled quantity after amplitude constraint according to the physical characteristics of the variable-frequency centrifugal fan and the fan performance curve to obtain a smoothed control quantity; Add the smoothed control quantity to the current operating frequency, calculate the target adjustment frequency, and perform a physical limit check on the target adjustment frequency to ensure that the physical limit of the frequency adjustment range of the frequency converter is satisfied to obtain a safe frequency value; Based on the relationship between the number of pole pairs of the fan motor and the relationship between frequency and speed, convert the safe frequency value into the corresponding target speed value; Perform acceleration limit processing on the target speed value, calculate the maximum allowable speed change per unit time, and limit the speed change rate between adjacent control cycles to obtain a frequency adjustment value with limited speed change rate; Encode the frequency adjustment value with limited speed change rate into a control instruction format recognizable by the frequency converter, and add communication verification information and a safety identifier to obtain a frequency adjustment instruction for the frequency converter.

9. The energy-saving optimization control method of the variable-frequency centrifugal fan according to claim 1, wherein, Sending the frequency adjustment instruction to the frequency converter to adjust the actual speed of the variable-frequency centrifugal fan so that the variable-frequency centrifugal fan operates in the best energy-saving state includes: Perform communication protocol encapsulation on the frequency adjustment instruction to obtain a standardized communication data packet, and send the standardized communication data packet to the control unit of the frequency converter; Adjust the actual speed of the variable-frequency centrifugal fan through the control unit so that the variable-frequency centrifugal fan operates in the best energy-saving state, and obtain the operating state information of the frequency converter through the feedback signal interface of the frequency converter; Monitor the stability of the speed change process of the variable-frequency centrifugal fan based on the operating state information, calculate the deviation between the actual speed change rate and the expected speed change rate to obtain an adjustment process deviation value; Judge whether the adjustment process meets the preset stability index according to the adjustment process deviation value, and if not, generate a fine-tuning instruction to obtain an optimized operating state of the fan; Collect and calculate the energy consumption data under the optimized operating state of the fan to obtain actual energy-saving effect data, and store the actual energy-saving effect data in the historical database for parameter optimization in the next control cycle.

10. An energy-saving optimization control system for a variable-frequency centrifugal fan, characterized in that, For implementing the energy-saving optimization control method of the variable-frequency centrifugal fan according to any one of claims 1-9, the energy-saving optimization control system of the variable-frequency centrifugal fan includes: An acquisition module for acquiring the inlet static pressure, outlet static pressure, flow rate, motor power, system resistance coefficient, and ambient temperature of the variable-frequency centrifugal fan through a sensor network to obtain a real-time parameter data set; A feature extraction module for performing digital filtering and feature extraction on the real-time parameter data set to obtain a target feature vector; A prediction calculation module for inputting the target feature vector into the trained VentNet dual-network deep neural network for prediction calculation to obtain the best speed prediction value and energy consumption prediction value under the current load condition; A processing module, configured to calculate control error data according to the predicted optimal rotational speed value and the predicted energy consumption value, and input the control error data into a hyperbolic reaching law controller for processing to obtain a frequency adjustment instruction for the frequency converter; An adjustment module, configured to send the frequency adjustment instruction to the frequency converter to adjust the actual rotational speed of the variable-frequency centrifugal fan, so that the variable-frequency centrifugal fan operates in an optimal energy-saving state.

Images