Large fan air volume real-time dynamic compensation measurement method and system based on deep learning

Abstract

The invention discloses a large fan air volume real-time dynamic compensation measurement method and system based on deep learning. The large fan air volume real-time dynamic compensation measurement method comprises the steps that S1, fan initial data of a large fan are collected in real time through a preset multi-type sensor; s2, performing optimization synchronous alignment on the initial data of the fan through edge nodes of the industrial Internet of Things to obtain a multi-dimensional time sequence data matrix with working condition labels; s3, on the basis of the multi-dimensional time sequence data matrix, through an improved CNN-LSTM hybrid network, generating a dynamic compensation feature containing an eddy current suppression coefficient; and S4, based on the dynamic compensation characteristics, the optimal dynamic compensation air volume is decoded through an adaptive enhanced integrated learning framework, a PID controller is triggered based on the optimal dynamic compensation air volume to achieve dynamic compensation measurement of the air volume, and more accurate prediction can be achieved for the fan air volume which is the data which is influenced by various factors and has certain complexity.

Description

Technical Field

[0001] The present invention relates to the field of air volume measurement technology, and in particular to a real-time dynamic compensation measurement method and system for large-scale fan air volume based on deep learning. Background Art

[0002] In modern industrial production and energy supply systems, fans, as key equipment, are widely used in numerous fields, including power generation, chemical engineering, mining, and construction. For example, in thermal power plants, fans are responsible for providing the large quantities of air required for boiler combustion. Precise control of air volume is directly related to combustion efficiency and power generation costs. In mine ventilation systems, fans deliver fresh air to ensure the safety of underground workers, making a stable air supply crucial. At present, traditional fan air volume measurement and control methods have many limitations. In terms of air volume measurement, most of them adopt the method of measuring differential pressure or thermoelectric potential, and then convert it into air volume signal through a transmitter. However, this method is difficult to obtain accurate air volume data in complex duct environments (such as the presence of adjustable dampers, supporting structures, elbows, baffles, reducers, etc.). For example, at the control level, traditional control strategies represented by PID control have fixed control parameters. In the face of frequent fluctuations in external environmental conditions (such as temperature, humidity, duct pressure, etc.), it is difficult to flexibly and quickly adjust the fan air volume. Manual intervention or excessive preset parameters are often required, resulting in low adjustment efficiency. Therefore, a real-time dynamic compensation measurement method and system for large-scale fan air volume based on deep learning are proposed here. Summary of the Invention

[0003] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned objectives, the present invention proposes the following technical solutions: A real-time dynamic compensation measurement method for large fan air volume based on deep learning, including: S1: Collect the initial data of large wind turbines in real time through pre-set multi-type sensors; S2: Optimize and synchronize the initial data of the wind turbine through the edge nodes of the Industrial Internet of Things to obtain a multi-dimensional time series data matrix with operating condition labels; S3: Based on the multi-dimensional time series data matrix, a dynamic compensation feature including eddy current suppression coefficient is generated through an improved CNN-LSTM hybrid network; The improved CNN-LSTM hybrid network is realized by embedding a physical constraint layer based on the NS fluid mechanics equation into the original CNN-LSTM hybrid network, so that the CNN-LSTM hybrid network outputs dynamic compensation features that conform to the laws of fluid dynamics. S4: Based on the dynamic compensation features, the optimal dynamic compensation air volume is decoded through an adaptive enhanced ensemble learning framework, and the PID controller is triggered based on the optimal dynamic compensation air volume to realize dynamic compensation measurement of air volume.

[0004] The multiple types of sensors include a wind speed sensor, a vortex sensor, a motor speed sensor and a torque sensor.

[0005] The process of optimizing synchronization alignment is as follows: Set a unified data collection frequency f for each sensor, and Collect data internally and set up sensors At the moment The data collected is ,in Indicates the sensor number, Indicates the number of collections; Set as slave sensor The received data sequence is After receiving data from n sensors, the industrial IoT edge node organizes the data into a two-dimensional matrix , where m is the length of the time series, and the optimized synchronization alignment is completed.

[0006] The process of obtaining the multi-dimensional time series data matrix with working condition labels is as follows: Assume that the operating conditions of the fan are divided into k types, namely ; Assume that there are l external environmental factors, namely ; According to the operation mode of the fan and the external environmental conditions, the working condition label is added to the collected data. Assume that at a certain time t, the fan is in the working condition , external environmental factors are , then the data collected at this moment The corresponding working condition label is ; Finally, a multi-dimensional time series data matrix with working condition labels is obtained: , m is the length of the time series, and n is the quantity index.

[0007] The improved CNN-LSTM hybrid network architecture includes a convolutional neural network, a long short-term memory network, and a physical constraint layer generated based on NS fluid mechanics.

[0008] The process of constructing the physical constraint layer based on NS fluid mechanics is as follows: Based on the NS equation: ; Where u is the velocity vector, t is the time, ρ is the fluid density, and p is the pressure. is the kinematic viscosity, For external force; The NS equations are deformed to solve the related equations of vorticity to obtain the eddy suppression coefficient. The physical constraints generated based on NS fluid mechanics are embedded into the original CNN-LSTM hybrid network. The discretized form of the NS equations is used as part of the loss function to complete the construction of the physical constraint layer. Let the predicted vortex-related characteristics be , the actual physical feature is y, the physical constraint loss function Expressed as: ,in, is the number of features associated with the physical constraint; After completing the construction of the physical constraint layer, the physical constraint loss function of the physical constraint layer is The trained improved CNN-LSTM hybrid network outputs the dynamic compensation feature Z containing the vortex suppression coefficient.

[0009] The process of acquiring the adaptive enhanced ensemble learning framework is as follows: Set up a set of training samples and initialize the weight of each sample. After each round of iteration, train a weak learner , get the error of the weak learner on the training set , according to the error Get weak learners Weight , according to the weak learner and corresponding weights Obtain a strong learner , where T is the number of iterations, For the nth preset dynamic compensation feature, the adaptive reinforcement integrated learning framework is constructed.

[0010] The process of obtaining the optimal dynamic compensation air volume is as follows: The dynamic compensation feature is input. During the training process of the adaptive reinforcement ensemble learning framework, after T rounds of iteration, T weak learners h and their corresponding weights α are obtained. For each dynamic compensation feature Z, the optimal dynamic compensation air volume is obtained by decoding it through the strong learner of the adaptive reinforcement ensemble learning framework.

[0011] A real-time dynamic compensation measurement system for large-scale fan air volume based on deep learning, including: Data acquisition module: collects the initial data of large wind turbines in real time through preset multiple types of sensors; Time series acquisition module: This module optimizes and synchronizes the initial data of wind turbines through the edge nodes of the Industrial Internet of Things to obtain a multi-dimensional time series data matrix with operating condition labels; Dynamic compensation module: Based on the multi-dimensional time series data matrix, an improved CNN-LSTM hybrid network is used to generate dynamic compensation features including eddy current suppression coefficients; Optimization measurement module: Based on the dynamic compensation features, the optimal dynamic compensation air volume is decoded through an adaptive enhanced integrated learning framework, and the PID controller is triggered based on the optimal dynamic compensation air volume to realize dynamic compensation measurement of air volume.

[0012] The present invention has the following beneficial effects: In this invention, first, through an improved CNN-LSTM hybrid network, deep dynamic compensation features can be mined from complex wind turbine operating data, effectively capturing the correlation of this data at different temporal and spatial scales. The network can adaptively learn and adjust to data changes generated by the wind turbine under different operating conditions (such as startup, stable operation, and variable speed operation). This avoids the limitations of traditional feature extraction methods when processing complex operating condition data. Secondly, by embedding the NS equations (Navier-Stokes equations) into the network as physical constraints, the dynamic compensation features extracted by the network are ensured to conform to the laws of fluid dynamics, which makes the generated features more reasonable and reliable in a physical sense. For example, when predicting the air volume characteristics related to the internal flow field of the fan, the intrinsic relationship between the viscosity, pressure and velocity of the fluid can be taken into account, thereby improving the accuracy of the air volume prediction.

[0013] Finally, multiple weak learners (simple decision trees) are iteratively trained through an adaptive reinforcement ensemble learning framework, and the sample weights are dynamically adjusted according to their performance during the training process. Finally, these weak learners are combined into a strong learner. This mechanism can fully utilize the advantages of different weak learners and achieve more accurate predictions for fan air volume data, which is affected by multiple factors and has a certain degree of complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a method step diagram of the real-time dynamic compensation measurement method and system for large-scale fan air volume based on deep learning proposed by the present invention.

[0015] Figure 2 This is a system block diagram of the deep learning-based real-time dynamic compensation measurement method and system for large-scale fan air volume proposed in the present invention. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] Example 1: Figure 1As shown, the present invention proposes a real-time dynamic compensation measurement method for large-scale fan air volume based on deep learning, including: S1: Collect the initial data of large wind turbines in real time through pre-set multi-type sensors; Wind speed sensor data: Wind speed sensors are installed at the inlet and outlet of the fan. The inlet wind speed data can reflect the airflow entering the fan, and the outlet wind speed data can reflect the fan's acceleration effect on the airflow. The wind direction data helps analyze the flow path of the airflow and the airflow interaction between the fan and the surrounding environment.

[0018] Assume the inlet wind speed is , the outlet wind speed is According to the law of conservation of mass, the air volume Q of the fan is related to the wind speed: ,in and are the effective cross-sectional areas of the fan inlet and outlet respectively; Vortex sensor data: Vortex sensors are installed inside and around the fan to detect vortices in the airflow. The generation and development of vortices will affect the fan's air volume and efficiency. The vortex intensity and frequency are monitored to obtain the vortex vortex intensity ω. Motor speed and torque data: Install the motor speed sensor and torque sensor. Assume the motor speed is n. The relationship between the fan air volume Q and the motor speed is expressed as follows: ,That is the proportionality coefficient related to the fan characteristics and is obtained through prior experience; The relationship between motor torque T and motor power P is: ,in, is the angular velocity of the motor; Specifically, the fan's air volume is directly related to the motor speed. The fan's air volume is proportional to the motor speed. The torque data can reflect the motor's load condition and indirectly reflect the fan's working intensity. The motor power is related to the fan's air volume and pressure head. Through these relationships, the fan's operating status can be comprehensively analyzed.

[0019] S2: Optimize and synchronize the initial data of the wind turbine through the edge nodes of the Industrial Internet of Things to obtain a multi-dimensional time series data matrix with operating condition labels; The process of optimizing synchronization alignment is: Set a unified data acquisition frequency f (f = 100 Hz) for each sensor, and Internal data collection; Set up sensor At the moment The data collected is ,in Indicates the sensor number, Indicates the number of collections; The IIoT edge node is responsible for receiving data from various sensors and performing preliminary processing on the data: Set as slave sensor The received data sequence is After receiving data from n sensors, the industrial IoT edge node organizes the data into a two-dimensional matrix , where m is the length of the time series, completing the optimized synchronization alignment; The process of obtaining a multi-dimensional time series data matrix with working condition labels is as follows: First, the edge node will sort the data with timestamps to ensure that the data is arranged in chronological order. Then, the edge node will perform integrity checks on the data to ensure that there are complete data records in each collection cycle. If data is missing or abnormal, the edge node will mark it and try to repair the data, such as filling the missing data points through interpolation or other data recovery algorithms; Assume that the fan operating conditions are divided into k types, respectively At the same time, suppose there are l external environmental factors, use express; According to the operation mode of the fan and the external environmental conditions, the working condition label is added to the collected data. Assume that at a certain time t, the fan is in the working condition , external environmental factors are , then the data collected at this moment The corresponding working condition label is ; Finally, a multi-dimensional time series data matrix with working condition labels is obtained: , which includes multi-dimensional data such as pressure, temperature, vibration, etc. of the fan under different working conditions.

[0020] S3: Based on the multi-dimensional time series data matrix, a dynamic compensation feature including eddy current suppression coefficient is generated through an improved CNN-LSTM hybrid network; Normalize the obtained multidimensional time series data matrix to the interval [0,1]; Through an original CNN-LSTM hybrid network, the normalized multidimensional time series data matrix is ​​segmented according to the time series. Using the sliding window method, the window size is set to 100 data points and the step size is 10 data points. The continuous time series data is segmented into multiple data subsets, and each subset is used as the input sample of the original CNN-LSTM hybrid network. Furthermore, through sliding window processing, we can fully utilize the previous and next correlation information in time series data and improve the network's ability to extract data features; The improved CNN-LSTM hybrid network is realized by embedding a physical constraint layer based on the NS fluid mechanics equation into the original CNN-LSTM hybrid network, so that the CNN-LSTM hybrid network outputs dynamic compensation features that conform to the laws of fluid dynamics. Improved CNN-LSTM hybrid network architecture includes: The improved CNN-LSTM hybrid network consists of a convolutional neural network (CNN), a long short-term memory network (LSTM), and a physical constraint layer generated based on NS fluid dynamics. The CNN part is mainly responsible for extracting the spatial features of the input data, while the LSTM part is used to process the long-term dependencies in time series data. In the CNN part, multiple convolutional layers and pooling layers are set, and the convolutional layer uses a small-size 3×3 convolution kernel; The local features in the data can be extracted through the convolution operation; The pooling layer uses the maximum pooling or average pooling method to reduce the dimensionality of the convolutional features, reducing the amount of data while retaining the main features; The LSTM part is composed of multiple LSTM units. Each LSTM unit contains an input gate, a forget gate, and an output gate. LSTM can remember information in long time series, avoiding the gradient vanishing problem in traditional recurrent neural networks (RNNs). By performing time series processing on features processed by CNN through LSTM units, it can capture the dynamic change characteristics in the data and obtain vortex-related features. ; The construction process of the physical constraint layer generated based on NS fluid mechanics is as follows: Embed a physical constraint layer generated based on NS fluid mechanics in the improved CNN-LSTM hybrid network; The NS equations (Navier-Stokes equations) are expressed as: ; Where u is the velocity vector, t is the time, ρ is the fluid density, and p is the pressure. is the kinematic viscosity, is the external force, which describes the motion law of the fluid; Specifically, by discretizing the NS equation and embedding it into the network as a physical constraint, the dynamic compensation characteristics of the network output can be made to conform to the laws of fluid dynamics; The NS equation is deformed and the vortex suppression coefficient is obtained by solving the related equations of vorticity. Specifically, the vortex equation is obtained by taking the curl of the NS equation: ; In this equation, we can define the relationship between the eddy current suppression coefficient and each term in the equation by analyzing the generation, diffusion and dissipation mechanism of vorticity. The eddy current suppression coefficient is obtained by adjusting the coefficients of these terms; The discretized form of the NS equation is used as part of the loss function to complete the construction of the physical constraint layer; In specific implementation, during the network training process, the network output is made to satisfy the physical relationship described by the NS equation as much as possible. For example, when calculating the loss function, in addition to traditional loss terms such as the mean square error (MSE), physical constraint terms based on the NS equation are also added. By adjusting the network weights, the vortex suppression coefficient output by the network is made more consistent with the actual fluid dynamic characteristics. Network training and optimization: Use a subset of multidimensional time series data with working condition labels to train the improved CNN-LSTM hybrid network. Divide the data subset into a training set, a validation set, and a test set, for example, at a ratio of 70%, 20%, and 10%. During training, the Adam optimization algorithm is used. The Adam algorithm combines the advantages of the adaptive gradient algorithm (Adagrad) and the root mean square propagation (RMSProp). It can adaptively adjust the learning rate, accelerate network convergence, and continuously adjust the network's hyperparameters, including the number of convolutional layers, the number of LSTM units, and the learning rate. By evaluating the network's performance on the validation set, the optimal hyperparameter combination is selected. Specifically, the performance evaluation index of the network uses the mean absolute error (MAE) to ensure that the network can accurately generate dynamic compensation features containing vortex suppression coefficients; The physical constraints generated based on NS fluid mechanics are embedded into the original CNN-LSTM hybrid network, and the discretized form of the NS equation is used as part of the loss function; Assume that the predicted vortex correlation characteristics are (obtained from LSTM output), the actual physical feature (calculated according to the NS equation) is y, and the physical constraint loss function is It can be expressed as: ; in, is the number of features associated with the physical constraint; After physical constraint loss function The trained improved CNN-LSTM hybrid network outputs the dynamic compensation feature Z containing the vortex suppression coefficient; Specifically, assuming the network's final output layer has d neurons, the dimension of the dynamic compensation feature Z is [d], and its elements are the vortex suppression-related eigenvalues ​​predicted by the network based on the input multidimensional time series data. These eigenvalues ​​can be used in subsequent dynamic airflow compensation calculations.

[0021] S4: Based on the dynamic compensation features, the optimal dynamic compensation air volume is decoded through an adaptive enhanced ensemble learning framework, and the PID controller is triggered based on the optimal dynamic compensation air volume to achieve dynamic compensation measurement of the air volume; The dynamic compensation features output from the improved CNN-LSTM hybrid network have different numerical ranges and dimensions. These feature vectors are normalized so that the mean of each feature is 0 and the variance is 1. Perform feature selection and dimensionality reduction on the standardized dynamic compensation features, and design an adaptive reinforcement ensemble learning framework to decode the dynamic compensation features; The process of building the adaptive reinforcement ensemble learning framework is as follows: Suppose a set of training samples ,in, is the nth preset dynamic compensation feature, is the true value of the wind volume corresponding to the nth one; First, initialize the weights of each sample , after each round of iteration, train a weak learner (such as a simple decision tree stump), obtain the error of the weak learner on the training set ,in is the indicator function; According to the error Get weak learners Weight ; Specifically, the process of updating sample weights is expressed as ,in, is the normalization factor, Represents the sample weight of the next iteration; According to the weak learner and corresponding weights Obtain a strong learner. The final strong learner is a linear combination of multiple weak learners, expressed as: , where T is the number of iterations, completing the construction of the adaptive reinforcement ensemble learning framework; Specifically, when classifying new samples, the strong learner H will comprehensively consider the prediction results of each weak learner and perform a weighted sum based on their weights to obtain the final classification result, completing the construction of the adaptive reinforcement ensemble learning framework; In the constructed adaptive reinforcement ensemble learning framework, the optimal dynamic compensation air volume is decoded by the strong learner of the adaptive reinforcement ensemble learning framework. The specific process is as follows: The input data is the dynamic compensation feature Z obtained through the previous processing. During the training process of the adaptive enhanced ensemble learning framework, after T rounds of iteration, T weak learners h and their corresponding weights α are obtained. For each dynamic compensation feature Z, according to the strong learner formula of the adaptive enhanced ensemble learning framework Decoding the optimal dynamic compensation air volume ; Specifically, each weak learner h processes the feature vector based on a specific algorithm (decision tree stump) during training to obtain an intermediate prediction value, and then combines it with its weight α for weighted summation to obtain the final optimal dynamic compensation air volume. ; The process of realizing dynamic compensation measurement of air volume is as follows: In the dynamic compensation of fan air volume, the set value is the desired air volume value, and the actual measured value is the optimal dynamic compensation air volume obtained through the above steps. The PID controller adjusts the control signal according to the size and change trend of the error, so that the actual air volume of the fan approaches the desired air volume; Adjustment of PID control parameters based on optimal dynamic compensation air volume Based on the selected optimal dynamic compensation air volume, the air volume error e(t) is calculated, and the air volume error is input into the PID controller. Dynamic compensation of air volume is achieved by adjusting the parameters of the PID controller. When the air volume error is large, the proportional coefficient can be appropriately increased, such as: Assume that when the air volume error e(t) is greater than or equal to (cubic meters per hour), it is considered that the air volume error is large. In this case, the proportional coefficient of the PID controller can be appropriately increased. , set the initial , when the air volume error is greater than or equal to When Increase to 1.0; This is because a larger proportional coefficient enables the controller to respond more strongly to larger errors, thereby quickly reducing the air volume error; Assume that when the air volume error e(t) is less than When the air volume error is small, the proportional coefficient of the PID controller can be appropriately reduced. ,Will Reduced from 0.5 to 0.3; A smaller proportional coefficient helps avoid excessive adjustments in the controller when the error is small, thus preventing the system from oscillating. By reasonably adjusting the parameters of the PID controller, accurate dynamic compensation of the fan air volume can be achieved, ensuring that the fan can stably output the desired air volume under different working conditions.

[0022] Example 2: Figure 2 As shown in the figure, the large-scale fan air volume real-time dynamic compensation measurement system based on deep learning includes: Data acquisition module: collects the initial data of large wind turbines in real time through preset multiple types of sensors; Time series acquisition module: This module optimizes and synchronizes the initial data of wind turbines through the edge nodes of the Industrial Internet of Things to obtain a multi-dimensional time series data matrix with operating condition labels; Dynamic compensation module: Based on the multi-dimensional time series data matrix, an improved CNN-LSTM hybrid network is used to generate dynamic compensation features including eddy current suppression coefficients; Optimization measurement module: Based on the dynamic compensation features, the optimal dynamic compensation air volume is decoded through an adaptive enhanced integrated learning framework, and the PID controller is triggered based on the optimal dynamic compensation air volume to realize dynamic compensation measurement of air volume.

[0023] In the application, several formulas involved are calculated by taking their numerical values ​​after removing the dimensions, and the formulas are established by collecting a large amount of data and performing software simulation to obtain a formula for the most recent real situation. Some coefficients or weights in the formulas are set by technical personnel in this field according to actual conditions, so they will not be elaborated here.

[0024] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0025] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A real-time dynamic compensation measurement method for large-scale fan air volume based on deep learning, characterized in that: include: S1: Collect the initial data of large wind turbines in real time through pre-set multi-type sensors; S2: Optimize and synchronize the initial data of the wind turbine through the edge nodes of the Industrial Internet of Things to obtain a multi-dimensional time series data matrix with operating condition labels; S3: Based on the multi-dimensional time series data matrix, a dynamic compensation feature including eddy current suppression coefficient is generated through an improved CNN-LSTM hybrid network; The improved CNN-LSTM hybrid network is realized by embedding a physical constraint layer based on the NS fluid mechanics equation into the original CNN-LSTM hybrid network, so that the CNN-LSTM hybrid network outputs dynamic compensation features that conform to the laws of fluid dynamics. S4: Based on the dynamic compensation features, the optimal dynamic compensation air volume is decoded through an adaptive enhanced ensemble learning framework, and the PID controller is triggered based on the optimal dynamic compensation air volume to realize dynamic compensation measurement of air volume.

2. The method for real-time dynamic compensation measurement of large-scale fan air volume based on deep learning according to claim 1 is characterized in that: The multiple types of sensors include a wind speed sensor, a vortex sensor, a motor speed sensor and a torque sensor.

3. The method for real-time dynamic compensation measurement of large-scale wind turbine air volume based on deep learning according to claim 1 is characterized in that: The process of optimizing synchronization alignment is as follows: Set a unified data collection frequency f for each sensor, and Collect data internally and set up sensors At the moment The data collected is ,in Indicates the sensor number, Indicates the number of collections; Set as slave sensor The received data sequence is After receiving data from n sensors, the industrial IoT edge node organizes the data into a two-dimensional matrix , where m is the length of the time series, and the optimized synchronization alignment is completed.

4. The method for real-time dynamic compensation measurement of large-scale fan air volume based on deep learning according to claim 1 is characterized in that: The process of obtaining the multi-dimensional time series data matrix with working condition labels is as follows: Assume that the operating conditions of the fan are divided into k types, namely ; Assume that there are l external environmental factors, namely ; According to the operation mode of the fan and the external environmental conditions, the working condition label is added to the collected data. Assume that at a certain time t, the fan is in the working condition , external environmental factors are , then the data collected at this moment The corresponding working condition label is ; Finally, a multi-dimensional time series data matrix with working condition labels is obtained: , m is the length of the time series, and n is the quantity index.

5. The method for real-time dynamic compensation measurement of large-scale wind turbine air volume based on deep learning according to claim 4 is characterized in that: The improved CNN-LSTM hybrid network architecture includes a convolutional neural network, a long short-term memory network, and a physical constraint layer generated based on NS fluid mechanics.

6. The method for real-time dynamic compensation measurement of large-scale wind turbine air volume based on deep learning according to claim 5 is characterized in that: The process of constructing the physical constraint layer based on NS fluid mechanics is as follows: Based on the NS equation: ; Where u is the velocity vector, t is the time, ρ is the fluid density, and p is the pressure. is the kinematic viscosity, For external forces; The NS equations are deformed to solve the related equations of vorticity to obtain the eddy suppression coefficient. The physical constraints generated based on NS fluid mechanics are embedded into the original CNN-LSTM hybrid network. The discretized form of the NS equations is used as part of the loss function to complete the construction of the physical constraint layer. Let the predicted vortex-related characteristics be , the actual physical feature is y, the physical constraint loss function Expressed as: ,in, is the number of features associated with the physical constraint; After completing the construction of the physical constraint layer, the physical constraint loss function of the physical constraint layer is The trained improved CNN-LSTM hybrid network outputs the dynamic compensation feature Z containing the vortex suppression coefficient.

7. The method for real-time dynamic compensation measurement of large-scale wind turbine air volume based on deep learning according to claim 6 is characterized in that: The process of acquiring the adaptive enhanced ensemble learning framework is as follows: Set up a set of training samples and initialize the weight of each sample. After each round of iteration, train a weak learner , get the error of the weak learner on the training set , according to the error Get weak learners Weight , according to the weak learner and corresponding weights Obtain a strong learner , where T is the number of iterations, For the nth preset dynamic compensation feature, the adaptive reinforcement integrated learning framework is constructed.

8. The method for measuring the air volume of a large-scale fan in real-time dynamic compensation based on deep learning according to claim 7 is characterized in that: The process of obtaining the optimal dynamic compensation air volume is as follows: The dynamic compensation feature is input. During the training process of the adaptive reinforcement ensemble learning framework, after T rounds of iteration, T weak learners h and their corresponding weights α are obtained. For each dynamic compensation feature Z, the optimal dynamic compensation air volume is obtained by decoding it through the strong learner of the adaptive reinforcement ensemble learning framework.

9. A real-time dynamic compensation measurement system for large-scale fan air volume based on deep learning, implemented according to the method according to any one of claims 1 to 8, characterized in that: include: Data acquisition module: collects the initial data of large wind turbines in real time through preset multiple types of sensors; Time series acquisition module: This module optimizes and synchronizes the initial data of wind turbines through the edge nodes of the Industrial Internet of Things to obtain a multi-dimensional time series data matrix with operating condition labels; Dynamic compensation module: Based on the multi-dimensional time series data matrix, an improved CNN-LSTM hybrid network is used to generate dynamic compensation features including eddy current suppression coefficients; Optimization measurement module: Based on the dynamic compensation features, the optimal dynamic compensation air volume is decoded through an adaptive enhanced integrated learning framework, and the PID controller is triggered based on the optimal dynamic compensation air volume to realize dynamic compensation measurement of air volume.

Images