Wind speed control method, device and equipment of current collector and medium

Abstract

The application discloses a wind speed control method, device, equipment and medium of a flow concentrator, relates to the technical field of wind tunnel experiments, and the flow concentrator has a main-aid dual-channel structure, main-aid dual-channel regulating valves and sensors for collecting environmental parameters; the method comprises the following steps: determining the initial main channel opening degree of the flow concentrator according to the expected wind speed of the flow concentrator and the mapping relationship between the main channel opening degree and the wind speed; predicting the target main channel opening degree interval by using a target prediction model, the rough main channel opening degree obtained by searching based on the initial main channel opening degree and the current environmental parameters; searching in the target main channel opening degree interval to obtain a target main-aid dual-channel opening degree combination; and adjusting the current main-aid dual-channel regulating valves to control the wind speed; the target prediction model is a model that fuses a convolutional neural network, a bidirectional long short-term memory network and an attention mechanism, and the training data of the target prediction model is the main-aid dual-channel opening degree combination and flow field dynamic data of the flow concentrator. The wind speed of the flow concentrator is accurately and low-cost controlled under various environments.

Description

Technical Field

[0001] This invention relates to the field of wind tunnel testing technology, and in particular to a method, device, equipment and medium for controlling wind speed in a combiner. Background Technology

[0002] In modern wind tunnel testing systems, precise control of airflow parameters is a crucial prerequisite for achieving high-quality aerodynamic testing. However, the complex nonlinear characteristics of airflow and its coupling relationship with the confluencer make it difficult for traditional theoretical calculation methods to accurately characterize this physical process.

[0003] Currently, the common method used in engineering applications is to obtain the characteristic curves of the merger channels through a large amount of wind tunnel experimental data, and then approximate the nonlinear mapping relationship between the opening of the main and auxiliary channels and the airflow parameters. However, this method has many limitations. It requires a large number of repeated wind tunnel calibration tests, which is labor-intensive, consumes a lot of air, and has a long calibration cycle. More importantly, the traditional channel characteristic curves are only applicable to specific calibration conditions and lack the ability to adapt to changing environmental conditions, making it difficult to flexibly meet the control requirements of different flow field parameters and aerodynamic conditions.

[0004] In summary, how to accurately and cost-effectively control the wind speed of the combiner under different environments is a problem that needs to be solved in this field. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for controlling the wind speed of a combiner, which can accurately and cost-effectively control the wind speed of the combiner under different environments. The specific solution is as follows:

[0006] In a first aspect, this application discloses a wind speed control method for a combiner, wherein the combiner has a main and auxiliary dual-channel structure and includes a main and auxiliary dual-channel regulating valve and sensors for collecting various environmental parameters; the method includes:

[0007] The initial main channel opening of the combiner is determined based on the expected wind speed and target mapping relationship of the combiner; wherein, the target mapping relationship is the mapping relationship between each historical main channel opening of the combiner and the wind speed of the combiner;

[0008] Wind speed is predicted using the target prediction model, the coarse main channel openings obtained from the first search based on the initial main channel opening, and the current environmental parameters, so as to obtain the target main channel opening range.

[0009] The target prediction model is used to perform a second search from each combination of main and auxiliary dual-channel openings within the target main channel opening range, so as to obtain the target main and auxiliary dual-channel opening combination corresponding to the expected wind speed.

[0010] Based on the target main and auxiliary dual-channel opening combination control ground jet test device, adjust the current main and auxiliary dual-channel regulating valve to control the wind speed of the combiner to correspond to the expected wind speed;

[0011] The target prediction model is a deep learning model that integrates convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms. The training data for the target prediction model consists of the historical main and auxiliary dual-channel opening combinations and historical flow field dynamic data of the merger.

[0012] Optionally, obtain the target prediction model, including:

[0013] The historical main and auxiliary dual-channel opening combinations and historical flow field dynamic data of the merger are used as training data; wherein, the historical flow field dynamic data includes historical environmental parameters and historical wind speed, and the environmental parameters include the temperature, air pressure and air density of the merger.

[0014] An initial prediction model is established, which includes a convolutional neural network, a bidirectional long short-term memory network, and an attention mechanism. The initial prediction model is then iteratively trained using the training data to obtain the target prediction model.

[0015] Optionally, the step of iteratively training the initial prediction model using the training data to obtain the target prediction model includes:

[0016] The initial prediction model is determined as the current prediction model;

[0017] The current prediction model is trained using the training data to obtain the prediction results for the current round; wherein, the prediction results include wind speed prediction results and pressure difference coefficient prediction results;

[0018] The first mean square error of the wind speed prediction result and the second mean square error of the pressure difference coefficient prediction result are determined by using a multi-task loss function, and the weighted mean square error of the first mean square error and the second mean square error is obtained.

[0019] The parameters of the current prediction model are fine-tuned using the weighted mean square error to obtain the next prediction model;

[0020] The next prediction model is determined as the new current prediction model, and the process jumps back to the step of training the current prediction model using the training data to obtain the prediction result for the current round, until a target prediction model that meets the preset training stopping condition is obtained.

[0021] Optionally, the step of fine-tuning the parameters of the current prediction model using the weighted mean square error to obtain the next prediction model includes:

[0022] Under the gradient pruning mechanism, the parameters of the current prediction model are tuned using the weighted mean square error and the Adam optimizer, and the learning rate of the current prediction model is dynamically adjusted using the scheduler to obtain the next prediction model.

[0023] Optionally, training the current prediction model using the training data to obtain the prediction result for the current round includes:

[0024] Input the training data into the current prediction model;

[0025] The current prediction model uses the convolutional neural network to extract feature vectors from the training data, and uses the bidirectional long short-term memory network to obtain feature sequences containing temporal dependencies based on the feature vectors. The attention mechanism is used to obtain context vectors from the feature sequences, and the prediction result for the current round is generated based on the shared feature representations extracted from the context vectors.

[0026] Optionally, the step of using the target prediction model, the coarse main channel openings obtained from the first search based on the initial main channel opening, and the current environmental parameters to predict wind speed, in order to obtain the target main channel opening range, includes:

[0027] Using the initial main channel opening as the directional cutting point, and based on the first preset step size and preset opening constraint conditions, perform the first search in the high opening direction and the low opening direction respectively to obtain each coarse main channel opening;

[0028] Wind speed is predicted using the target prediction model, the rough main channel opening, and current environmental parameters to obtain the target main channel opening range corresponding to the predicted wind speed.

[0029] Optionally, the step of using the target prediction model to perform a second search from each combination of main and auxiliary dual-channel openings within the target main channel opening range to obtain the target main and auxiliary dual-channel opening combination corresponding to the expected wind speed includes:

[0030] The opening combinations of each main and auxiliary dual channel under the target main channel opening range are obtained based on a second preset step size; wherein, the second preset step size is smaller than the first preset step size;

[0031] The target prediction model is used to obtain the wind speed prediction results under each of the main and auxiliary dual-channel opening combinations, and the wind speed error between each of the wind speed prediction results and the expected wind speed is determined.

[0032] A second search is performed in each of the main and auxiliary dual-channel opening combinations to obtain the target main and auxiliary dual-channel opening combination with the smallest wind speed error.

[0033] Secondly, this application discloses a wind speed control device for a combiner, wherein the combiner has a main and auxiliary dual-channel structure and includes a main and auxiliary dual-channel regulating valve and a sensor for collecting various environmental parameters; the device includes:

[0034] An initial opening determination module is used to determine the initial main channel opening of the combiner based on the expected wind speed and target mapping relationship of the combiner; wherein, the target mapping relationship is the mapping relationship between each historical main channel opening of the combiner and the wind speed of the combiner;

[0035] The opening interval acquisition module is used to predict wind speed using the target prediction model, the rough main channel openings obtained by the first search based on the initial main channel opening, and the current environmental parameters, so as to obtain the target main channel opening interval.

[0036] The opening combination screening module is used to perform a second search from each main and auxiliary dual-channel opening combination under the target main channel opening range using the target prediction model, so as to obtain the target main and auxiliary dual-channel opening combination corresponding to the expected wind speed.

[0037] The wind speed control module is used to adjust the current main and auxiliary dual-channel regulating valve of the ground jet test device based on the target main and auxiliary dual-channel opening combination control, so as to control the wind speed of the combiner to correspond to the expected wind speed;

[0038] The target prediction model is a deep learning model that integrates convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms. The training data for the target prediction model consists of the historical main and auxiliary dual-channel opening combinations and historical flow field dynamic data of the merger.

[0039] Thirdly, this application discloses an electronic device, including:

[0040] Memory, used to store computer programs;

[0041] A processor is used to execute the computer program to implement the steps of the aforementioned wind speed control method for the combiner.

[0042] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned wind speed control method for a combiner.

[0043] The beneficial effects of this application are as follows: The combiner of this application has a main and auxiliary dual-channel structure and includes a main and auxiliary dual-channel regulating valve and a sensor for collecting various environmental parameters; the method includes: determining the initial main channel opening of the combiner according to the expected wind speed and target mapping relationship of the combiner; wherein, the target mapping relationship is the mapping relationship between each historical main channel opening of the combiner and the wind speed of the combiner; using the target prediction model, each coarse main channel opening obtained by a first search based on the initial main channel opening, and the current environmental parameters to predict the wind speed, so as to obtain the target main channel opening range; using the target prediction model to predict the wind speed. The prediction model performs a second search within each main and auxiliary dual-channel opening combination under the target main channel opening range to obtain the target main and auxiliary dual-channel opening combination corresponding to the expected wind speed. Based on the target main and auxiliary dual-channel opening combination, the ground jet flow testing device is controlled to adjust the current main and auxiliary dual-channel regulating valve to control the wind speed of the merger to correspond to the expected wind speed. The target prediction model is a deep learning model that integrates convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms. The training data of the target prediction model are the historical main and auxiliary dual-channel opening combinations of the merger and historical flow field dynamic data. Therefore, the combiner of this application adopts a dual-channel structure and a dual-channel regulating valve, which can perform macro-level wind speed range control through the main channel and fine adjustment through the auxiliary channel. Combined with real-time sensing data from sensors that collect environmental parameters, it can control wind speed more accurately than single-channel or simple mechanical control. This application determines the initial main channel opening based on the target mapping relationship between the expected wind speed and the historical main channel opening and wind speed, which is more efficient than the existing method that relies on empirical calibration points. It utilizes a target prediction model, a coarse main channel opening obtained based on the initial main channel opening, and environmental parameters to obtain a rough main channel opening. The model predicts wind speed to determine the main channel opening range and selects the main and auxiliary channel opening combinations corresponding to the expected wind speed from the range to control the regulating valve. Because the training data includes historical main and auxiliary channel opening combinations and flow field dynamic data, and integrates convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms, this model has stronger nonlinear feature extraction and key parameter focusing capabilities. It can also adapt to changes in environmental parameters. Compared with traditional models, it can predict wind speed more accurately, realize the corresponding control of the combiner wind speed and the expected wind speed, and improve control accuracy, efficiency, and flexibility. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of the wind speed control process of a combiner disclosed in this application;

[0046] Figure 2 This is a schematic diagram of a specific prediction model structure disclosed in this application;

[0047] Figure 3 This application discloses a specific flowchart for obtaining the target primary and secondary dual-channel opening combination;

[0048] Figure 4 This is a schematic diagram of a specific combiner wind speed control disclosed in this application;

[0049] Figure 5 This is a specific overall loss diagram disclosed in this application;

[0050] Figure 6 This is a schematic diagram of a specific wind speed loss disclosed in this application;

[0051] Figure 7 This is a schematic diagram of a specific differential pressure coefficient loss disclosed in this application;

[0052] Figure 8 This application discloses a specific wind speed curve under 7°C conditions.

[0053] Figure 9 This application discloses a specific wind speed curve under 9°C conditions.

[0054] Figure 10 This application discloses a specific wind speed curve under 11°C conditions.

[0055] Figure 11 This application discloses a specific pressure differential coefficient curve under 7°C conditions.

[0056] Figure 12 This application discloses a specific pressure differential coefficient curve under 9°C conditions.

[0057] Figure 13 This application discloses a specific pressure differential coefficient curve under 11°C conditions.

[0058] Figure 14 This is a schematic diagram of a specific channel parameter search path disclosed in this application;

[0059] Figure 15 This is a schematic diagram illustrating wind speed changes during a specific search process disclosed in this application.

[0060] Figure 16 This is a schematic diagram illustrating a specific search error variation disclosed in this application;

[0061] Figure 17 This application discloses a specific wind speed heat map;

[0062] Figure 18 This is a schematic diagram of the wind speed control device for a combiner disclosed in this application;

[0063] Figure 19 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0064] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0065] In modern wind tunnel testing systems, precise control of airflow parameters is a crucial prerequisite for achieving high-quality aerodynamic testing. However, the complex nonlinear characteristics of airflow and its coupling relationship with the confluencer make it difficult for traditional theoretical calculation methods to accurately characterize this physical process.

[0066] Currently, the common method used in engineering applications is to obtain the characteristic curves of the merger channels through a large amount of wind tunnel experimental data, and then approximate the nonlinear mapping relationship between the opening of the main and auxiliary channels and the airflow parameters. However, this method has many limitations. It requires a large number of repeated wind tunnel calibration tests, which is labor-intensive, consumes a lot of air, and has a long calibration cycle. More importantly, the traditional channel characteristic curves are only applicable to specific calibration conditions and lack the ability to adapt to changing environmental conditions, making it difficult to flexibly meet the control requirements of different flow field parameters and aerodynamic conditions.

[0067] Therefore, this application provides a wind speed control scheme for a combiner, which can accurately and cost-effectively control the wind speed of the combiner under different environments.

[0068] See Figure 1 As shown in the figure, this application discloses a wind speed control method for a combiner, wherein the combiner has a main and auxiliary dual-channel structure and includes a main and auxiliary dual-channel regulating valve and a sensor for collecting various environmental parameters; the method includes:

[0069] Step S11: Determine the initial main channel opening of the combiner based on the expected wind speed and target mapping relationship of the combiner; wherein, the target mapping relationship is the mapping relationship between each historical main channel opening of the combiner and the wind speed of the combiner.

[0070] The combiner has a main and auxiliary dual-channel structure and includes a main and auxiliary dual-channel regulating valve and sensors for collecting various environmental parameters; that is, the combiner in this embodiment is a coaxial intelligent combiner. The coaxial intelligent combiner achieves full-speed range wind speed regulation through its main and auxiliary dual-channel structure. By coordinating the output pressure of the front-end pressure reducing valve with the opening of the main channel regulating valve of the coaxial intelligent combiner, the wind speed range of the ground jet test device can be controlled from 0 to 200 m / s. Precise control relies on the coordinated operation of the main channel regulating valve and the auxiliary channel regulating valve.

[0071] In modern wind tunnel testing systems, precise control of airflow parameters is a crucial prerequisite for achieving high-quality aerodynamic testing. However, the complex nonlinear characteristics of airflow and its coupling relationship with coaxial intelligent confluencers make it difficult for traditional theoretical calculation methods to accurately characterize this physical process. Currently, the commonly used method in engineering applications is to obtain channel characteristic curves through iterative correction using a large amount of wind tunnel experimental data, thereby approximating the nonlinear mapping relationship between the main and auxiliary channel openings and airflow parameters. This approach often relies on empirical solutions, which have many limitations. It requires numerous repetitive wind tunnel calibration tests, resulting in a heavy workload, huge air consumption, and long calibration cycles. More importantly, traditional channel characteristic curves are only applicable to specific calibration conditions and lack adaptability under varying environmental conditions, making it difficult to flexibly address the control requirements of different flow field parameters and aerodynamic conditions.

[0072] In this embodiment, the channel calibration module establishes a target mapping relationship between wind speed and the main channel opening of the coaxial intelligent combiner. Specifically, it can establish a target mapping relationship between the historical main channel opening at 9 key calibration points and the wind speed of the corresponding combiner. For example, the target mapping relationship between historical main channel opening O1 and wind speed S1, the target mapping relationship between historical main channel opening O2 and wind speed S2, ..., the target mapping relationship between historical main channel opening O9 and wind speed S9.

[0073] Once the expected wind speed of the combiner is determined, the channel calibration module determines the initial main channel opening of the combiner based on the expected wind speed, the target mapping relationship, and an extrapolation algorithm. Specifically, after obtaining the expected wind speed of the combiner, if the expected wind speed is within the existing calibration point wind speed range, the corresponding opening can be estimated using an interpolation algorithm based on the mapping relationship between adjacent calibration points. If the expected wind speed exceeds the calibration point range, an extrapolation algorithm is used, such as calculating the rate of change of the boundary calibration points based on a linear relationship. Combining the rate of change with the difference between the expected wind speed and the boundary calibration point wind speed, the opening of the portion exceeding the range is calculated, thus obtaining the initial main channel opening of the combiner, providing basic parameters for subsequent control and adjustment.

[0074] Step S12: Utilize the target prediction model, the coarse main channel openings obtained from the first search based on the initial main channel opening, and current environmental parameters to predict wind speed, thereby obtaining the target main channel opening range. The target prediction model is a deep learning model integrating a convolutional neural network, a bidirectional long short-term memory network, and an attention mechanism. The training data for the target prediction model consists of the historical main and auxiliary dual-channel opening combinations of the merger and historical flow field dynamic data.

[0075] In this embodiment, obtaining the target prediction model includes: determining the historical main and auxiliary dual-channel opening combinations and historical flow field dynamic data of the merger as training data; wherein, the historical flow field dynamic data includes historical environmental parameters and historical wind speed, and the environmental parameters include the temperature, air pressure and air density of the merger; establishing an initial prediction model including a convolutional neural network, a bidirectional long short-term memory network and an attention mechanism, and using the training data to iteratively train the initial prediction model to obtain the target prediction model.

[0076] For example Figure 2 The diagram shows a specific predictive model structure. It collects historical main and auxiliary dual-channel opening combinations and historical flow field dynamic data of the merger. The historical flow field dynamic data includes historical environmental parameters and historical wind speed. The environmental parameters include the temperature, air pressure, and air density of the merger. The air pressure is the difference between dynamic pressure and static pressure. In other words, the historical main valve opening, historical auxiliary valve opening, temperature, air density, dynamic pressure, and static pressure of the merger are determined as training data.

[0077] An initial prediction model is established, which includes Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM) network, and Attention mechanism. The prediction model is a multi-output deep learning model (Multi Output-CNN-BiLSTM-Attention).

[0078] The initial prediction model is iteratively trained using training data, continuously optimizing the parameters and hyperparameters of the prediction model until a preset stopping condition is met, thus obtaining the target prediction model. The preset stopping condition can be that the convergence of the target prediction model is greater than a preset convergence threshold, or that the number of iterations is not less than a preset number of iterations threshold.

[0079] In this embodiment, the step of iteratively training the initial prediction model using the training data to obtain the target prediction model includes: determining the initial prediction model as the current prediction model; training the current prediction model using the training data to obtain the prediction result for the current round; wherein the prediction result includes wind speed prediction result and pressure difference coefficient prediction result; determining the first mean square error of the wind speed prediction result and the second mean square error of the pressure difference coefficient prediction result using a multi-task loss function, and obtaining the weighted mean square error of the first mean square error and the second mean square error; optimizing the parameters of the current prediction model using the weighted mean square error to obtain the next prediction model; determining the next prediction model as the new current prediction model, and re-jumping to the step of training the current prediction model using the training data to obtain the prediction result for the current round, until a target prediction model that meets the preset training stopping condition is obtained.

[0080] It is important to note that the prediction model in this embodiment differs from traditional deep learning models in its output. Specifically, the prediction model in this embodiment is a multi-output deep learning model, meaning it outputs not only the predicted wind speed but also the predicted pressure coefficient. The initial prediction model is designated as the current prediction model, initiating the iterative training process. Training data is input into the current prediction model for the current training iteration, yielding the wind speed and pressure coefficient predictions for that iteration. The model parameters are then fine-tuned using a multi-task loss function and the prediction results. The tuning process involves: using the multi-task loss function to determine the first mean square error (MSE) of the wind speed prediction and the second mean square error (MSE) of the pressure coefficient prediction; obtaining the weighted mean square error (MSE) of the first and second MSEs; and using this weighted MSE to fine-tune the parameters of the current prediction model to obtain the next prediction model. This next prediction model is then designated as the new current prediction model, and the process returns to training the current prediction model using the training data to obtain the prediction results for the current iteration, thus initiating the next training iteration. This continues until a target prediction model that meets the preset training termination conditions is obtained.

[0081] In this embodiment, the step of using the weighted mean square error to fine-tune the parameters of the current prediction model to obtain the next prediction model includes: fine-tuning the parameters of the current prediction model using the weighted mean square error and the Adam optimizer under the gradient pruning mechanism, and dynamically adjusting the learning rate of the current prediction model using the scheduler to obtain the next prediction model.

[0082] During training, a gradient pruning mechanism is applied to limit the gradient norm to within 1.0 to prevent gradient explosion. The training process employs a multi-task loss function, which weights and combines the mean squared errors of wind speed prediction and pressure coefficient prediction. The Adam optimizer is used with an initial learning rate of 0.001, and the learning rate is automatically adjusted using the ReduceLROnPlateau learning rate scheduler. In other words, under the gradient pruning mechanism, the parameters of the current prediction model are tuned using the weighted mean squared error and the Adam optimizer, and the learning rate of the current prediction model is dynamically adjusted by the scheduler to obtain the next prediction model.

[0083] In this embodiment, training the current prediction model using the training data to obtain the prediction result for the current round includes: inputting the training data into the current prediction model; using the current prediction model to extract feature vectors from the training data using the convolutional neural network, and using the bidirectional long short-term memory network to obtain feature sequences containing temporal dependencies based on the feature vectors; obtaining context vectors from the feature sequences based on the attention mechanism; and generating the prediction result for the current round based on the shared feature representations extracted from the context vectors.

[0084] The prediction model includes convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms. The prediction model also includes an input projection layer and a shared feature layer.

[0085] The input projection layer expands the training data at a single time point into a sequence form, creating artificial time series data with a length of 5, which facilitates subsequent time series processing.

[0086] The feature extraction part consists of three layers of convolutional neural networks, each layer using 32, 64 and 64 convolutional kernels respectively, with a kernel size of 3, and is combined with batch normalization and ReLU activation function. This structural design can capture the local correlation between parameters of coaxial smart merger and extract key features, that is, the convolutional neural network extracts the feature vector of training data.

[0087] Furthermore, the feature vectors need to be dimensionally transformed to obtain the dimensionally transformed feature vectors. The bidirectional long short-term memory network obtains the feature sequence containing temporal dependencies based on the dimensionally transformed feature vectors. The bidirectional long short-term memory network adopts a bidirectional structure with a hidden layer dimension of 64. A dropout (temporary removal) mechanism is set to prevent overfitting. The bidirectional design enables the model to capture temporal dependencies from both forward and backward directions, and to model the dynamic characteristics of the coaxial intelligent combiner.

[0088] The attention mechanism consists of two fully connected network layers, with the Tanh activation function in between. After the output layer calculates the attention weights, it is normalized by softmax to generate a context vector. In other words, the attention mechanism obtains the context vector from the feature sequence. This mechanism identifies and focuses on the most critical time steps in the sequence, enhancing the ability to perceive changes in the channel flow of the coaxial smart combiner.

[0089] The shared feature layer following the attention layer is processed by a fully connected network, ReLU activation, and dropout to extract shared feature representations from the context vector. This multi-task learning design enables wind speed and pressure differential coefficient predictions to share underlying features while maintaining their respective prediction characteristics. In other words, the prediction result for the current round is generated based on the shared feature representations extracted from the context vector.

[0090] The prediction model is ultimately divided into two parallel output branches: wind speed prediction and pressure difference coefficient prediction. Each branch consists of two fully connected networks. This dual-output structure enables the model to learn two tasks simultaneously, improving the control accuracy of the coaxial smart combiner.

[0091] In this embodiment, the step of using the target prediction model, the coarse main channel openings obtained from the first search based on the initial main channel opening, and the current environmental parameters to predict wind speed and obtain the target main channel opening range includes: using the initial main channel opening as the direction cutting point, and performing a first search in the high opening direction and the low opening direction based on a first preset step size and a preset opening constraint condition to obtain the coarse main channel openings; using the target prediction model, the coarse main channel openings, and the current environmental parameters to predict wind speed and obtain the target main channel opening range corresponding to the predicted result and the expected wind speed.

[0092] In obtaining the target main channel opening range, the main channel is responsible for macroscopic wind speed range control, so a first search, i.e., a coarse search, is performed: using the initial main channel opening as the directional cutting point, i.e., the initial main channel opening is the starting point for both the high and low opening directions. Specifically, a first preset step size and preset opening constraints are set. Since this is a coarse search, the first preset step size is relatively large and will not be very small. The preset opening constraints can include the maximum and minimum values ​​of the main channel opening. The first search is performed starting from the initial main channel opening and moving towards directions greater than the initial main channel opening; the first search is also performed starting from the initial main channel opening and moving towards directions less than the initial main channel opening, thus obtaining the coarse main channel openings. Next, wind speed is predicted using the target prediction model, the approximate opening of each main channel, and the current environmental parameters. This yields the target main channel opening range corresponding to the predicted wind speed. The current environmental parameters, including temperature, pressure difference, and air density, are collected using sensors. The target prediction model learns the relationship between environmental parameters, the opening of the main and auxiliary channels, and wind speed. Once the resulting parameters are determined, the target prediction model can also predict other unknown parameters. Therefore, the approximate opening of each main channel and the current environmental parameters are input into the target prediction model. Based on the input data and the expected wind speed, the target prediction model can determine the target main channel opening range from the approximate main channel openings.

[0093] Step S13: Use the target prediction model to perform a second search from each main and auxiliary dual-channel opening combination under the target main channel opening range to obtain the target main and auxiliary dual-channel opening combination corresponding to the expected wind speed.

[0094] In this embodiment, the step of using the target prediction model to perform a second search from each main and auxiliary dual-channel opening combination under the target main channel opening range to obtain the target main and auxiliary dual-channel opening combination corresponding to the expected wind speed includes: obtaining each main and auxiliary dual-channel opening combination under the target main channel opening range based on a second preset step size; wherein, the second preset step size is smaller than the first preset step size; using the target prediction model to obtain the wind speed prediction result under each main and auxiliary dual-channel opening combination, and determining the wind speed error between each wind speed prediction result and the expected wind speed; performing a second search in each main and auxiliary dual-channel opening combination to obtain the target main and auxiliary dual-channel opening combination with the smallest wind speed error.

[0095] The generation of wind speed presents a problem of multiple solutions, meaning that the same wind speed can be generated by different main and auxiliary channel configurations of the coaxial intelligent combiner. The main channel is responsible for macro-level wind speed range control, while the auxiliary channel performs fine-tuning. This division of labor makes full use of the structural characteristics of the coaxial intelligent combiner and effectively prevents inconsistencies or conflicts in channel configurations during the wind speed increase process.

[0096] For example Figure 3 The flowchart illustrates a specific process for obtaining the target primary and secondary dual-channel opening combination. It utilizes a target prediction model to perform a second search (i.e., a precise search) on each primary and secondary dual-channel opening combination within the target primary channel opening range, thereby obtaining the target primary and secondary dual-channel opening combination corresponding to the expected wind speed. Specifically: because this is a precise search, a second preset step size is set, smaller than the first preset step size. Based on the second preset step size, each primary and secondary dual-channel opening combination within the target primary channel opening range is obtained. It can be understood that the primary and secondary dual-channel opening combination is equivalent to the primary and secondary dual-channel opening. The target prediction model is used to obtain wind speed prediction results under the current environmental parameters and each primary and secondary dual-channel opening combination, and the wind speed error between each wind speed prediction result and the expected wind speed is determined. A second search is performed on each primary and secondary dual-channel opening combination to obtain the target primary and secondary dual-channel opening combination with the smallest wind speed error. For example, the current main and auxiliary dual-channel opening combination is determined within the target main channel opening range. Then, the wind speed prediction result R1 is obtained using the target prediction model under the current environmental parameters and the main and auxiliary dual-channel opening combination C1. Based on the second preset step size and the main and auxiliary dual-channel opening combination C1, the main and auxiliary dual-channel opening combination C2 is obtained. The wind speed prediction result R2 is obtained using the target prediction model under the current environmental parameters and the main and auxiliary dual-channel opening combination C2. The wind speed prediction results R1 and R2 are compared with the expected wind speed to obtain the wind speed errors E1 and E2, respectively. The main and auxiliary dual-channel opening combination with the smaller wind speed error is retained. Based on the second preset step size and the main and auxiliary dual-channel opening combination C2, the main and auxiliary dual-channel opening combination C3 is obtained. According to the above steps, the wind speed error E3 between the wind speed prediction result R3 of the main and auxiliary dual-channel opening combination C3 and the expected wind speed is determined. This process is repeated until the target main and auxiliary dual-channel opening combination with the smallest wind speed error is found within the target main channel opening range.

[0097] The core architecture of the intelligent control system based on the coaxial smart combiner consists of two parts: a channel calibration module and a reverse channel optimizer. These two parts implement the reverse derivation process from the target wind speed to the optimal channel configuration through an algorithmic architecture. The channel calibration module establishes the mapping relationship between wind speed and the main channel opening of the coaxial smart combiner. By storing nine key calibration points and applying an extrapolation algorithm, it provides a reliable initial estimate of the main channel for any target wind speed. The reverse channel optimizer adopts a multi-stage search strategy, including bidirectional search and fine-tuning, to systematically explore the main and auxiliary channel space of the coaxial smart combiner and locate the optimal configuration. The control system also provides practical functions such as generating wind speed-channel configuration lookup tables, interpolation calculation, and simulation of the adjustment process. It also supports visualization of the search process, providing a comprehensive solution for wind tunnel operation and making the complex wind speed control process efficient, accurate, and visualized.

[0098] Wind speed generation suffers from ambiguity, meaning the same wind speed can be generated by different main and auxiliary channel configurations of a coaxial intelligent combiner. Lack of calibration leads to an excessively large search space for the control algorithm, significantly reducing efficiency. Calibration, by clearly defining the functional division of the main channel's responsibility for macroscopic wind speed range control and the auxiliary channels' responsibility for fine-tuning, fully utilizes the structural characteristics of the coaxial intelligent combiner and effectively prevents inconsistencies or conflicts in channel configurations during wind speed increases. More importantly, this calibration data provides crucial prior knowledge for the inverse control algorithm, enabling the system to quickly locate a reasonable initial search range. This significantly improves the control accuracy, stability, and response speed of the coaxial intelligent combiner, providing a fundamental guarantee for high-quality wind tunnel experiments.

[0099] The reverse channel optimizer employs a hierarchical structure and a progressive search strategy, forming a complete control solution based on a coaxial intelligent combiner. In the initialization phase, this module integrates two core functional components: a predictive model and channel calibration, and sets key parameters such as channel physical constraints and wind speed tolerance, laying the foundation for the subsequent search process. Its core objective is to find the optimal channel configuration. Through three organically linked search stages, it achieves a precise mapping from the target wind speed to the optimal channel parameters: first, an initial main channel estimate is obtained based on calibration data; then, a bidirectional coarse search (downward and upward) is performed to determine the approximate range; and finally, a fine search is conducted near the optimal point to further optimize the results. This multi-stage search strategy of coarse positioning and fine tuning not only balances search efficiency and accuracy but also highly matches the physical structure of the coaxial intelligent combiner's main and auxiliary dual channels, effectively avoiding the trap of local optima and ensuring that the system can find the optimal combination of channel control parameters under various operating conditions.

[0100] Step S14: Based on the target main and auxiliary dual-channel opening combination control ground jet test device, adjust the current main and auxiliary dual-channel regulating valve to control the wind speed of the combiner to correspond to the expected wind speed.

[0101] For example Figure 4 The diagram shows a specific example of manifold wind speed control. After determining the target main and auxiliary dual-channel opening combination, the ground jet test device can be controlled to adjust the current main and auxiliary dual-channel regulating valves according to the target main and auxiliary dual-channel opening combination, thereby controlling the wind speed of the manifold to correspond to the expected wind speed.

[0102] Furthermore, the wind speed of the combiner can be visualized under different environmental parameters and the opening degree of the main and auxiliary dual channels. That is, the relationship curve between environmental parameters, the opening degree of the main and auxiliary dual channels and the wind speed of the combiner can be constructed, so that users can understand the wind speed of the combiner more intuitively.

[0103] The beneficial effects of this application are as follows: The combiner of this application has a main and auxiliary dual-channel structure and includes a main and auxiliary dual-channel regulating valve and a sensor for collecting various environmental parameters; the method includes: determining the initial main channel opening of the combiner according to the expected wind speed and target mapping relationship of the combiner; wherein, the target mapping relationship is the mapping relationship between each historical main channel opening of the combiner and the wind speed of the combiner; using the target prediction model, each coarse main channel opening obtained by a first search based on the initial main channel opening, and the current environmental parameters to predict the wind speed, so as to obtain the target main channel opening range; using the target prediction model to predict the wind speed. The prediction model performs a second search within each main and auxiliary dual-channel opening combination under the target main channel opening range to obtain the target main and auxiliary dual-channel opening combination corresponding to the expected wind speed. Based on the target main and auxiliary dual-channel opening combination, the ground jet flow testing device is controlled to adjust the current main and auxiliary dual-channel regulating valve to control the wind speed of the merger to correspond to the expected wind speed. The target prediction model is a deep learning model that integrates convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms. The training data of the target prediction model are the historical main and auxiliary dual-channel opening combinations of the merger and historical flow field dynamic data. Therefore, the combiner of this application adopts a dual-channel structure and a dual-channel regulating valve, which can perform macro-level wind speed range control through the main channel and fine adjustment through the auxiliary channel. Combined with real-time sensing data from sensors that collect environmental parameters, it can control wind speed more accurately than single-channel or simple mechanical control. This application determines the initial main channel opening based on the target mapping relationship between the expected wind speed and the historical main channel opening and wind speed, which is more efficient than the existing method that relies on empirical calibration points. It utilizes a target prediction model, a coarse main channel opening obtained based on the initial main channel opening, and environmental parameters to obtain a rough main channel opening. The model predicts wind speed to determine the main channel opening range and selects the main and auxiliary channel opening combinations corresponding to the expected wind speed from the range to control the regulating valve. Because the training data includes historical main and auxiliary channel opening combinations and flow field dynamic data, and integrates convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms, this model has stronger nonlinear feature extraction and key parameter focusing capabilities. It can also adapt to changes in environmental parameters. Compared with traditional models, it can predict wind speed more accurately, realize the corresponding control of the combiner wind speed and the expected wind speed, and improve control accuracy, efficiency, and flexibility.

[0104] The analysis focuses on the learning and convergence characteristics of the multi-output CNN-BiLSTM-Attention model on the coaxial smart merging circuit. These loss curves include... Figure 5 , Figure 6 and Figure 7Each group simultaneously displays the performance changes on both the training and validation sets. The curves exhibit typical deep learning model training characteristics: a steep decline occurs within the first 5 epochs, with the loss value rapidly decreasing from approximately 0.35 initially to around 0.1 as training progresses, indicating that the model quickly captures the main patterns in the coaxial smart merger main / auxiliary channel configuration data; this is followed by a stable decline period from epochs 5 to 15, during which the learning rate slows down but continues to optimize, showing that the model is refining its understanding of the complex nonlinear response relationships of the coaxial smart merger; finally, a plateau is reached between epochs 15 and 32, with the loss curve fluctuating less and tending to converge, and the training loss remaining at around 0.07. Notably, the validation loss (orange line) is consistently significantly lower than the training loss (blue line) throughout the entire training process, and is close to zero (around 0.003), indicating that the model has extremely strong generalization ability, accurately predicting the flow field parameters under unseen coaxial smart merger configurations, without overfitting. The loss contributions of the two sub-tasks (wind speed prediction and pressure coefficient prediction) are roughly equivalent, indicating that the multi-task learning framework reasonably balances the dual control objectives of the coaxial intelligent combiner. An early stopping mechanism was triggered at the 32nd epoch during training, and the final model achieved nearly identical performance metrics on the independent test set (wind speed R² = 0.9997, pressure coefficient R² = 0.9992). These results fully demonstrate that the deep learning architecture can accurately model the complex relationships between parameters such as the opening of the main and auxiliary channels, temperature, and pressure in the coaxial intelligent combiner and wind speed and pressure differential coefficient, revealing the intrinsic working mechanism of the coaxial intelligent combiner as a structure-control cooperative unit. The model provides a reliable algorithmic foundation for constructing a high-precision, low-turbulence, real-time adaptive wind tunnel control system, significantly reducing the cost of traditional wind tunnel calibration, improving the adaptability of the coaxial intelligent combiner to changes in environmental variables, and providing strong intelligent support for the stable operation of this device in complex aerodynamic environments.

[0105] For example Figure 8 The diagram shows a specific wind speed curve under 7°C conditions. Figure 9 The diagram shows a specific wind speed curve under 9°C conditions. Figure 10 The diagram shows a specific wind speed curve under 11°C conditions. Figure 11 The figure shown is a specific pressure differential coefficient curve under 7°C conditions. Figure 12 The figure shown is a specific pressure differential coefficient curve under 9°C conditions. Figure 13The figure shows a specific pressure differential coefficient curve under 11°C conditions, illustrating the influence of the main and auxiliary channel openings of a coaxial smart combiner on wind speed and pressure differential coefficient under three different temperature conditions (7°C, 9°C, and 11°C). This set of three-dimensional response surfaces demonstrates the systematic impact of temperature changes on the control characteristics of the coaxial smart combiner, revealing its internal fluid dynamics mechanism and structure-control coupling relationship. The figure shows nonlinear characteristics of wind speed and pressure differential coefficient changes with the main and auxiliary channel openings: wind speed exhibits an "S"-shaped growth curve with increasing channel opening, increasing slowly in the low opening range (0-30%), rapidly in the medium opening range (30-70%), and tending to level off in the high opening range (70-100%); the pressure differential coefficient shows a characteristic of first rising slowly and then increasing rapidly, with more significant changes after the main channel opening exceeds 40%. The auxiliary channel has a relatively small impact on the two parameters, but its adjustment effect becomes more obvious in the high-opening region of the main channel, reflecting the interactive coupling between the main and auxiliary channels of the coaxial intelligent combiner, which is consistent with its "coarse adjustment + fine adjustment" structure-control design. As the temperature increases (from 7°C to 11°C), the three-dimensional surfaces of wind speed and pressure differential coefficient exhibit systematic changes: the surface shifts upwards overall, indicating that under the same channel configuration, higher temperatures produce greater wind speed and pressure differential coefficient output; the shape and curvature characteristics of the surface also change, at 11°C ( Figure 10 , Figure 13 The curved surface exhibits a smoother transition characteristic in the high channel opening region, while at 7°C ( Figure 8 , Figure 11 The corresponding region exhibits a steeper rate of change. This temperature dependence is consistent with gas dynamics theory, namely, that increased temperature leads to decreased gas density and increased kinematic viscosity, affecting Reynolds number and pressure distribution, thereby altering the flow field characteristics inside the coaxial smart combiner. Temperature changes not only cause surface displacement but also alter surface morphology, indicating that the effect of temperature on the coaxial smart combiner is nonlinear. This multi-parameter coupling characteristic makes it difficult for traditional control methods based on physical models or lookup tables to accurately describe and predict the system behavior of the coaxial smart combiner. The multi-output CNN-BiLSTM-Attention deep learning architecture, by fusing feature extraction from convolutional neural networks, temporal modeling from bidirectional LSTM, and key information focusing through an attention mechanism, achieves modeling of the complex nonlinear system of the coaxial smart combiner (wind speed R² = 0.9997, pressure differential R² = 0.9992). Visualization results verify the model's ability to capture temperature effects, providing operators with a reference for the dynamic behavior of the coaxial smart combiner under different environmental conditions. This method has application value in improving wind tunnel testing accuracy, reducing calibration costs, and realizing adaptive control of coaxial intelligent combiners, and provides a reference for the collaborative design of fluid dynamics testing devices.

[0106] For example Figure 14 The diagram illustrates a specific channel parameter search path, showing the search trajectories of the main and auxiliary channel openings. The three stages—downward search, upward search, and fine-grained search—are marked with blue dots, green triangles, and purple squares, respectively. The search begins at an initial estimate of 39% for the main channel and 5% for the auxiliary channel. Through a coarse-fine search strategy, it eventually converges to the optimal configuration point (40% for the main channel and 2% for the auxiliary channel, marked with a red asterisk). The three gray guide lines demonstrate the algorithm's process of gradually approaching the optimal solution by exploring the parameter space. Figure 15 The diagram illustrates wind speed changes during a specific search process, tracking the dynamic changes in wind speed throughout the entire search. Starting from the initial point (blue dot in the downlink search) and the starting point in the uplink search (green triangle), after 18 steps of refined searching (purple squares), the wind speed gradually approaches the target value of 110 m / s (red dashed line), finally reaching the optimal solution of 110.21 m / s (red asterisk). This process demonstrates the algorithm's adaptive adjustment in the wind speed space, avoiding local optima. Figure 16 The diagram illustrates a specific search error variation, recording the trend of search error change over the steps. The error gradually decreases from an initial 0.26 m / s, fluctuating during the search process (rising and then falling between steps 6-9), and finally stabilizing and converging to a minimum error of 0.21 m / s (marked with a red asterisk). This non-monotonic decrease in error curve reflects the algorithm's balance strategy between exploration and exploitation. Figure 17 The diagram illustrates a specific wind speed heatmap, visualizing the influence of the main and auxiliary channel opening combinations of a coaxial smart combiner on wind speed. The color change from dark blue to yellow corresponds to a wind speed variation from 110.2 m / s to 110.344 m / s. Figure 14 and Figure 17 The image shows the optimal region centered on the main channel (40%) and auxiliary channel (2%), marked with a red asterisk, surrounded by a wind speed gradient distribution. The heatmap reveals the sensitivity of wind speed to channel parameters—wind speed is more sensitive to the main channel opening than to the auxiliary channel, consistent with the physical characteristics of the coaxial smart combiner. This method, combining channel calibration data and the predictive capabilities of the CNN-BiLSTM-Attention model, achieves control of the nonlinear system of the coaxial smart combiner. In a test case with a target wind speed of 110 m / s, the error was controlled within 0.21 m / s (relative error of approximately 0.19%) through 18 iterations, outperforming traditional control methods. This visualized search path verifies the effectiveness of the algorithm in optimizing the parameters of the coaxial smart combiner and provides a reference for testing procedures and control strategies.

[0107] See Figure 18As shown in the figure, this application discloses a wind speed control device for a combiner. The combiner has a main and auxiliary dual-channel structure and includes a main and auxiliary dual-channel regulating valve and a sensor for collecting various environmental parameters. The device includes:

[0108] The initial opening determination module 11 is used to determine the initial main channel opening of the combiner based on the expected wind speed and target mapping relationship of the combiner; wherein, the target mapping relationship is the mapping relationship between each historical main channel opening of the combiner and the wind speed of the combiner;

[0109] The opening interval acquisition module 12 is used to predict wind speed using the target prediction model, the rough main channel openings obtained by the first search based on the initial main channel opening, and the current environmental parameters, so as to obtain the target main channel opening interval.

[0110] The opening combination screening module 13 is used to perform a second search from each main and auxiliary dual-channel opening combination under the target main channel opening range using the target prediction model, so as to obtain the target main and auxiliary dual-channel opening combination corresponding to the expected wind speed.

[0111] The wind speed control module 14 is used to adjust the current main and auxiliary dual-channel regulating valve of the ground jet test device based on the target main and auxiliary dual-channel opening combination control, so as to control the wind speed of the combiner to correspond to the expected wind speed.

[0112] The target prediction model is a deep learning model that integrates convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms. The training data for the target prediction model consists of the historical main and auxiliary dual-channel opening combinations and historical flow field dynamic data of the merger.

[0113] The beneficial effects of this application are as follows: The combiner of this application has a main and auxiliary dual-channel structure and includes a main and auxiliary dual-channel regulating valve and a sensor for collecting various environmental parameters; the method includes: determining the initial main channel opening of the combiner according to the expected wind speed and target mapping relationship of the combiner; wherein, the target mapping relationship is the mapping relationship between each historical main channel opening of the combiner and the wind speed of the combiner; using the target prediction model, each coarse main channel opening obtained by a first search based on the initial main channel opening, and the current environmental parameters to predict the wind speed, so as to obtain the target main channel opening range; using the target prediction model to predict the wind speed. The prediction model performs a second search within each main and auxiliary dual-channel opening combination under the target main channel opening range to obtain the target main and auxiliary dual-channel opening combination corresponding to the expected wind speed. Based on the target main and auxiliary dual-channel opening combination, the ground jet flow testing device is controlled to adjust the current main and auxiliary dual-channel regulating valve to control the wind speed of the merger to correspond to the expected wind speed. The target prediction model is a deep learning model that integrates convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms. The training data of the target prediction model are the historical main and auxiliary dual-channel opening combinations of the merger and historical flow field dynamic data. Therefore, the combiner of this application adopts a dual-channel structure and a dual-channel regulating valve, which can perform macro-level wind speed range control through the main channel and fine adjustment through the auxiliary channel. Combined with real-time sensing data from sensors that collect environmental parameters, it can control wind speed more accurately than single-channel or simple mechanical control. This application determines the initial main channel opening based on the target mapping relationship between the expected wind speed and the historical main channel opening and wind speed, which is more efficient than the existing method that relies on empirical calibration points. It utilizes a target prediction model, a coarse main channel opening obtained based on the initial main channel opening, and environmental parameters to obtain a rough main channel opening. The model predicts wind speed to determine the main channel opening range and selects the main and auxiliary channel opening combinations corresponding to the expected wind speed from the range to control the regulating valve. Because the training data includes historical main and auxiliary channel opening combinations and flow field dynamic data, and integrates convolutional neural networks, bidirectional long short-term memory networks, and attention mechanisms, this model has stronger nonlinear feature extraction and key parameter focusing capabilities. It can also adapt to changes in environmental parameters. Compared with traditional models, it can predict wind speed more accurately, realize the corresponding control of the combiner wind speed and the expected wind speed, and improve control accuracy, efficiency, and flexibility.

[0114] Furthermore, embodiments of this application also provide an electronic device. Figure 19 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0115] Figure 19This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Specifically, it may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the wind speed control method for a combiner performed by the electronic device disclosed in any of the foregoing embodiments.

[0116] In this embodiment, the power supply 23 is used to provide operating voltage for various hardware devices on the electronic device; the communication interface 24 can create a data transmission channel between the electronic device and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0117] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0118] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored on it include operating system 221, computer program 222 and data 223, etc., and the storage method can be temporary storage or permanent storage.

[0119] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. The operating system can be Windows, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the wind speed control method of the combiner executed by the electronic device as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0120] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned wind speed control method for a combiner. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0121] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0122] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly in hardware, software modules executed by a processor, or a combination of both. The software module may be located in random access memory (RAM), memory, read-only memory (ROM), electrically programmable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), register, hard disk, removable disk, CD-ROM (Compact Disc Read-Only Memory), or any other form of storage medium known in the art.

[0123] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0124] The wind speed control method, device, equipment, and medium of a combiner provided by the present invention have been described in detail above. Specific examples have been used to illustrate the principle and implementation of the present invention. The description of the above embodiments is only for the purpose of helping to understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation and application scope based on the idea of ​​the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A wind speed control method of a concentrator, characterized by, The collector has a main-aid dual-channel structure and comprises a main-aid dual-channel regulating valve and sensors for collecting various environmental parameters; the method comprises: determining an initial main-channel opening degree of the collector according to an expected wind speed of the collector and a target mapping relationship; wherein the target mapping relationship is a mapping relationship between various historical main-channel opening degrees of the collector and wind speeds of the collector; performing wind speed prediction based on the initial main-channel opening degree, various rough main-channel opening degrees obtained through first searching, and current environmental parameters, to obtain a target main-channel opening degree interval; performing second searching from various main-aid dual-channel opening degree combinations under the target main-channel opening degree interval by using the target prediction model, to obtain a target main-aid dual-channel opening degree combination corresponding to the expected wind speed; controlling a ground jet test device to adjust the main-aid dual-channel regulating valve based on the target main-aid dual-channel opening degree combination, so as to control the wind speed of the collector to correspond to the expected wind speed. The target prediction model is a deep learning model that fuses a convolutional neural network, a bidirectional long short-term memory network, and an attention mechanism, and training data of the target prediction model is historical main-aid dual-channel opening degree combinations and historical flow field dynamic data of the collector.

2. The wind speed control method of the concentrator according to claim 1, wherein, The target prediction model is obtained by: determining historical main-aid dual-channel opening degree combinations and historical flow field dynamic data of the collector as training data; wherein the historical flow field dynamic data comprises historical environmental parameters and historical wind speeds, and the environmental parameters comprise temperature, air pressure, and air density of the collector; establishing an initial prediction model comprising a convolutional neural network, a bidirectional long short-term memory network, and an attention mechanism, and iteratively training the initial prediction model by using the training data, to obtain the target prediction model.

3. The wind speed control method of the concentrator according to claim 2, wherein, The target prediction model is obtained by iteratively training the initial prediction model by using the training data, comprising: determining the initial prediction model as a current prediction model; training the current prediction model by using the training data, to obtain a prediction result in a current round; wherein the prediction result comprises a wind speed prediction result and a differential pressure coefficient prediction result; determining a first mean square error of the wind speed prediction result and a second mean square error of the differential pressure coefficient prediction result by using a multi-task loss function, and obtaining a weighted mean square error of the first mean square error and the second mean square error; optimizing parameters of the current prediction model by using the weighted mean square error, to obtain a next prediction model; determining the next prediction model as a new current prediction model, and returning to the step of training the current prediction model by using the training data, to obtain a prediction result in a current round, until a target prediction model that meets a preset training stop condition is obtained.

4. The wind speed control method of the concentrator according to claim 3, wherein, The target prediction model is obtained by iteratively training the initial prediction model by using the training data, comprising: The parameters of the current prediction model are optimized by using the weighted mean square error and the Adam optimizer under a gradient clipping mechanism, and a learning rate of the current prediction model is dynamically adjusted by using a scheduler to obtain a next prediction model.

5. The wind speed control method of the concentrator according to claim 3, wherein, The training of the current prediction model by using the training data to obtain a prediction result in a current round comprises: inputting the training data into the current prediction model; extracting a feature vector of the training data by using the convolutional neural network and obtaining a feature sequence containing a time sequence dependency relationship based on the feature vector by using the bidirectional long short-term memory network, obtaining a context vector from the feature sequence based on the attention mechanism, and generating a prediction result in a current round according to a shared feature representation extracted from the context vector by using the current prediction model.

6. The wind speed control method of the concentrator according to claim 1, wherein, The wind speed prediction by using a target prediction model, each rough main channel opening obtained by performing a first search based on the initial main channel opening, and a current environmental parameter to obtain a target main channel opening interval comprises: cutting the initial main channel opening as a direction cutting point, and performing a first search in a high opening direction and a low opening direction based on a first preset step length and a preset opening constraint condition to obtain each rough main channel opening; The wind speed prediction by using a target prediction model, each rough main channel opening, and a current environmental parameter to obtain a target main channel opening interval corresponding to a prediction result and the expected wind speed comprises:

7. The wind speed control method of a concentrator according to claim 6, wherein The second search from each main auxiliary double-channel opening combination under the target main channel opening interval by using the target prediction model to obtain a target main auxiliary double-channel opening combination corresponding to the expected wind speed comprises: obtaining each main auxiliary double-channel opening combination under the target main channel opening interval based on a second preset step length; wherein the second preset step length is smaller than the first preset step length; obtaining wind speed prediction results under each main auxiliary double-channel opening combination by using the target prediction model, and determining wind speed errors between each wind speed prediction result and the expected wind speed; performing a second search in each main auxiliary double-channel opening combination to obtain a target main auxiliary double-channel opening combination with the minimum wind speed error.

8. A wind speed control device for a concentrator, characterized by, The current collector has a main auxiliary double-channel structure and comprises a main auxiliary double-channel regulating valve and a sensor for collecting each environmental parameter; the device comprises: An initial opening determination module is configured to determine an initial main channel opening of the current collector according to an expected wind speed of the current collector and a target mapping relationship; wherein the target mapping relationship is a mapping relationship between each historical main channel opening of the current collector and a wind speed of the current collector. An opening interval obtaining module is configured to perform wind speed prediction by using a target prediction model, each rough main channel opening obtained by performing a first search based on the initial main channel opening, and a current environmental parameter to obtain a target main channel opening interval. An opening combination screening module is configured to perform a second search from each main auxiliary double-channel opening combination under the target main channel opening interval by using the target prediction model to obtain a target main auxiliary double-channel opening combination corresponding to the expected wind speed. A wind speed control module is configured to control the ground jet test device to adjust the main and auxiliary dual-channel regulating valve based on the target main and auxiliary dual-channel opening combination, so that the wind speed of the current flow collector corresponds to the expected wind speed. The target prediction model is a deep learning model integrating a convolutional neural network, a bidirectional long short-term memory network, and an attention mechanism, and the training data of the target prediction model is historical main and auxiliary dual-channel opening combinations and historical flow field dynamic data of the flow collector.

9. An electronic device, comprising: The method comprises the following steps: a memory for storing a computer program; a processor for executing the computer program to implement the steps of the wind speed control method of the flow collector according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, a memory for storing a computer program; wherein the computer program, when executed by a processor, implements the steps of the wind speed control method of the flow collector according to any one of claims 1 to 7.

Images