Satellite antenna exchange method and system

Abstract

The invention relates to the technical field related to satellite communication, in particular to a satellite antenna exchange method and system. The satellite antenna exchange method comprises the following steps: acquiring positioning information of a satellite; inputting the positioning information into a set neural network model; calculating azimuth information of the satellite through the set neural network model; and according to the azimuth information of the satellite, adjusting the direction and the inclination angle of a pan-tilt of a satellite antenna, so that the satellite antenna is aligned with the satellite. According to the invention, the automatic and accurate alignment of the satellite antenna is realized, the satellite orientation features can be directly learned and extracted from the complex positioning information, the dynamic influence caused by the satellite orbit change and the receiving end motion can be processed in real time, and the stability and reliability of a satellite communication system are improved. The automatic alignment process reduces the requirement of manual operation, reduces the operation difficulty and personal error, and improves the efficiency and precision of satellite antenna alignment.

Description

Technical Field

[0001] The present invention relates to the technical field related to satellite communications, and in particular to a satellite antenna swapping method and system. Background Art

[0002] Satellite communications, as an important communication method, plays an irreplaceable role in modern society, providing people with more convenient and reliable communication services. Satellite signals need to be transmitted through satellite antennas, and this type of antenna communication often requires adjustment to ensure accurate signal reception.

[0003] However, satellite communication signals typically have a narrow angle. Ground-based receivers must adjust the direction and tilt of the satellite antenna to ensure it remains accurately aligned with the satellite signal. Manual outdoor operations require simultaneous adjustment of both the antenna's direction and tilt, and each adjustment requires waiting for satellite signal transmission. This results in lengthy debugging times, low efficiency, and difficulty achieving optimal alignment.

[0004] Furthermore, existing adjustment methods often rely on manual experience and lack precise calculations and automated control, resulting in a time-consuming adjustment process and difficulty in achieving optimal results. In response to the above issues, existing technologies urgently need to be improved. Summary of the Invention

[0005] The object of the present invention is to provide a satellite antenna alignment method and system to improve adjustment efficiency and accuracy and reduce manual intervention.

[0006] In order to solve the above technical problems, the present invention provides a satellite antenna swap method and system.

[0007] The satellite antenna alignment method of the present application includes:

[0008] Obtain satellite positioning information;

[0009] Inputting the positioning information into a set neural network model;

[0010] Calculating the satellite's position information using the set neural network model;

[0011] According to the azimuth information of the satellite, the direction and inclination of the pan / tilt of the satellite antenna are adjusted so that the satellite antenna is aligned with the satellite.

[0012] Furthermore, the satellite positioning information includes pseudo-moment, signal strength and reception time.

[0013] Furthermore, the neural network model is constructed in the following manner:

[0014] Get sample data;

[0015] Build an initial neural network model;

[0016] The initial neural network model is trained using the sample data to obtain the neural network model.

[0017] Furthermore, the sample data includes satellite positioning information and a satellite position information tag corresponding to the sample data.

[0018] Furthermore, before constructing the initial neural network model, the sample data is preprocessed to remove outliers and noise.

[0019] Furthermore, the initial neural network model includes an input layer, a hidden layer and an output layer. The input layer is used to receive the input of the sample data and the position information label, and convert the data into a format that can be processed by the initial neural network model; the hidden layer is used to extract features from the input data and obtain the inherent laws of the sample data and the position information label; the output layer is used to map the features extracted by the hidden layer to specific mission objectives to output the position information of the satellite.

[0020] Furthermore, the initial neural network model also includes a loss function and an optimizer, wherein the loss function is used to calculate the average value of the square of the difference between the predicted value and the true value, and the optimizer is used to select stochastic gradient descent and its variants to update the weights of the initial neural network model to minimize the loss function.

[0021] Furthermore, the process of training the initial neural network model includes:

[0022] Dividing the sample data into a training set, a validation set, and a test set, wherein the training set is used to train the initial neural network model, the validation set is used to adjust the hyperparameters of the initial neural network model and prevent overfitting, and the test set is used to evaluate the generalization ability of the initial neural network model;

[0023] Iteratively training the initial neural network model using the training set data, wherein in each training step, input data is passed into the initial neural network model, forward propagation is calculated to obtain a prediction result, and then a loss value is calculated according to the loss function, and gradients are calculated by a backpropagation algorithm to update the weights of the initial neural network model;

[0024] The trained initial neural network model is evaluated, and the one that meets the evaluation criteria is the set neural network model.

[0025] Furthermore, during the training process, the performance of the model is regularly evaluated on the validation set, and the hyperparameters of the initial neural network model are adjusted according to the loss value of the validation set.

[0026] The present application also provides a satellite antenna alignment system, including a positioning information acquisition module, a positioning information input module, a neural network model setting module and an adjustment module;

[0027] The positioning information acquisition module is used to obtain satellite positioning information;

[0028] The positioning information input module is used to input the positioning information into a set neural network model;

[0029] The set neural network model is used to calculate the position information of the satellite;

[0030] The adjustment module adjusts the direction and tilt of the pan / tilt of the satellite antenna according to the azimuth information of the satellite, so that the satellite antenna is aligned with the satellite.

[0031] Compared with the prior art, the present invention has at least the following beneficial effects:

[0032] This application achieves automated and precise alignment of satellite antennas, enabling direct learning and extraction of satellite orientation features from complex positioning information. The proposed satellite antenna alignment method also features rapid response, enabling real-time processing of dynamic effects from satellite orbit changes and receiver motion, thereby improving the stability and reliability of satellite communication systems. The automated alignment process reduces the need for manual operation, reduces operational difficulty, and reduces human error, thereby improving the efficiency and accuracy of satellite antenna alignment. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 The figure is a flow chart of an embodiment of a satellite antenna swapping method according to the present invention. DETAILED DESCRIPTION

[0034] The following describes the satellite antenna alignment method and system of the present invention with reference to schematic diagrams, which illustrate preferred embodiments of the present invention. It should be understood that those skilled in the art may modify the invention described herein while still achieving the beneficial effects of the present invention. Therefore, the following description should be understood as a general guideline for those skilled in the art and not as a limitation of the present invention. Based on the teachings of this specification, those skilled in the art may, without creating any technical contradictions, create new technical solutions by combining different embodiments, and such variations should be considered to fall within the scope of protection of this patent.

[0035] The serial numbers of the components herein, such as "first", "second", etc., are only used to distinguish the objects described and do not have any order or technical meaning. The "connection" and "coupling" mentioned in this application, unless otherwise specified, include direct and indirect connections (couplings). In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "upper", "lower", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", etc. are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention.

[0036] In the present invention, unless otherwise expressly specified or limited, when a first feature is "above" or "below" a second feature, it may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediary. Furthermore, when a first feature is "above," "above," or "above" a second feature, it may mean that the first feature is directly above or diagonally above the second feature, or simply means that the first feature is at a higher level than the second feature. When a first feature is "below," "below," or "below" a second feature, it may mean that the first feature is directly below or diagonally below the second feature, or simply means that the first feature is at a lower level than the second feature.

[0037] The present invention is described in more detail in the following paragraphs by way of example with reference to the accompanying drawings. The advantages and features of the present invention will become more apparent from the following description and claims. It should be noted that the drawings are greatly simplified and not to exact scale, and are provided solely for the purpose of assisting in the description of the embodiments of the present invention.

[0038] In traditional existing satellite communication systems, ground receivers rely on manual operation to adjust the azimuth and elevation angles of the satellite antenna to achieve alignment. Due to the narrow and dynamic changes in satellite signal angles, operators need to repeatedly adjust the gimbal attitude based on experience and reversely verify the alignment effect through signal reporting results. This process has multi-parameter coupling problems. For example, the nonlinear relationship between signal strength, reception time and satellite position is not quantitatively modeled, resulting in redundant adjustments and significantly increased time consumption. At the same time, manual operation cannot respond in real time to dynamic deviations caused by satellite orbit perturbations or the motion of the receiving carrier, resulting in a decrease in the stability of the communication link, which directly affects the signal-to-noise ratio and bit error rate indicators of signal reception.

[0039] When faced with the above problems, this application first analyzed the limitations of traditional manual adjustment methods and found that the nonlinear relationship between multi-source positioning information and satellite azimuth is difficult to accurately model through empirical formulas. Taking into account the combined influence of the dynamic changes in satellite orbit parameters and the motion of the receiving carrier, an attempt was made to analyze the relationship between pseudorange, signal strength and azimuth by establishing a mathematical model, but it was found that the computational complexity increased exponentially with the increase in parameter dimension. Further research was conducted on sensor fusion solutions, attempting to compensate for attitude disturbances through inertial navigation data, but it was difficult to balance hardware costs and real-time performance. Finally, turning to the field of machine learning, it was found that neural networks have the advantage of processing high-dimensional nonlinear mappings and can directly learn the inherent laws of positioning information and azimuth from historical data without explicitly modeling complex physical relationships. By comparing supervised learning and unsupervised learning paradigms, a supervised training method based on sample labels was selected to ensure that the model output can directly drive the actuator.

[0040] In this regard, this application proposes a satellite antenna swap method, such as Figure 1 As shown, the following steps are included:

[0041] S100: Obtain satellite positioning information;

[0042] S200: Inputting the positioning information into a set neural network model; calculating the satellite's position information through the set neural network model;

[0043] S300: According to the satellite's position information, adjust the direction and tilt of the satellite antenna's pan / tilt to align the satellite antenna with the satellite.

[0044] Among them, the satellite positioning information refers to the data obtained by receiving satellite signals for determining the satellite position. Specifically, it can be achieved by using a receiver to obtain pseudorange, signal strength and reception time to provide basic data for the subsequent calculation of the satellite orientation. Among them, setting a neural network model refers to a machine learning model that has been pre-trained with sample data. Specifically, it can be implemented using a deep neural network structure including an input layer, a hidden layer and an output layer. It is used to extract features from the positioning information and output orientation information. Among them, the satellite orientation information refers to the direction angle and position parameters of the satellite in three-dimensional space. Specifically, it can be achieved through the azimuth and inclination parameters output by the neural network model, which are used to guide the gimbal to adjust the antenna to align with the satellite. Among them, adjusting the direction and inclination of the satellite antenna gimbal refers to controlling the mechanical structure of the gimbal to change the pointing angle of the antenna. Specifically, it can be achieved by using a motor to drive the gimbal to rotate and pitch to establish a stable communication link between the antenna and the satellite.

[0045] This application sets a neural network model to directly calculate the azimuth information based on the satellite's positioning information, and automatically adjusts the direction and inclination of the antenna gimbal based on the calculation results, thereby replacing the traditional manual repeated debugging method and solving the problems of low adjustment efficiency and poor accuracy.

[0046] The working process and principle of this application are as follows: first, the satellite's positioning information is obtained, including parameters such as pseudorange, signal strength, and reception time. These parameters reflect the relative position relationship between the satellite and the receiver. The obtained positioning information is then input into a pre-set neural network model. This neural network model is trained with a large amount of historical data and can extract features from the input positioning information and establish a nonlinear mapping relationship. The neural network model receives the positioning information as input, and after calculations and activation function processing by multiple layers of neurons, it ultimately outputs the satellite's azimuth and elevation information. Based on the satellite azimuth information calculated by the neural network model, the control system drives the motor of the satellite antenna pan-tilt platform to adjust the antenna's direction and tilt. The pan-tilt platform achieves precise control of the antenna's pointing through horizontal rotation and pitch movement. During the adjustment process, the system continuously obtains the latest positioning information and repeats the above steps until the satellite antenna is accurately aligned with the satellite position. This neural network-based method can quickly process complex nonlinear relationships and achieve automated and precise alignment of the satellite antenna.

[0047] As a preferred embodiment, the solution of this application is specifically implemented as follows:

[0048] First, a satellite receiver obtains satellite positioning information, including pseudorange, signal strength, and reception time. Pseudorange reflects the distance between the satellite and the receiver, signal strength indicates the strength of the received signal, and reception time records the exact moment the signal arrives.

[0049] Furthermore, the acquired positioning information is converted into a standard format, such as converting pseudoranges into meters, signal strengths into decibels, and reception times into a unified timestamp format. This pre-processes the positioning information into a data format that can be directly processed by the neural network.

[0050] Next, the preprocessed positioning information is input into a pre-trained neural network model. This neural network model utilizes a multi-layer perceptron architecture, consisting of an input layer, multiple hidden layers, and an output layer. The input layer receives positioning information, the hidden layers extract features using a nonlinear activation function, and the output layer generates the satellite's azimuth and elevation angles. Specifically, during the training phase, the neural network model uses a large amount of historical data to learn the mapping between positioning information and satellite positions. During the operational phase, the model rapidly processes the input positioning information and outputs the corresponding position information.

[0051] Finally, based on the azimuth information output by the neural network model, the control system drives the stepper motors of the satellite antenna pan / tilt to adjust the antenna's direction and tilt. For example, the azimuth is controlled by the pan motor, and the elevation is controlled by the pitch motor. During this adjustment process, the system continuously acquires the latest positioning information and repeats the above steps until the satellite antenna is accurately aligned with the satellite's position.

[0052] Through the above scheme, the present application realizes the automated and precise alignment of satellite antennas, and can directly learn and extract satellite orientation features from complex positioning information. The satellite antenna adjustment method of the present application also has a rapid response capability, and can handle the dynamic effects of satellite orbit changes and receiving end movement in real time, thereby improving the stability and reliability of the satellite communication system. The automated alignment process reduces the need for manual operation, reduces operational difficulty and human error, and improves the efficiency and accuracy of satellite antenna alignment. In addition, it has good adaptability and generalization capabilities, and can meet the needs of satellite communication under different environmental conditions.

[0053] Furthermore, in some of the embodiments, the satellite positioning information includes pseudo moment, signal strength and reception time.

[0054] Pseudorange is a distance measurement calculated by multiplying the satellite signal transmission time by the speed of light, reflecting the relative spatial position of the satellite and the ground receiver. Signal strength measures the power level of the satellite signal in decibel milliwatts and characterizes signal transmission quality and environmental interference. Reception time records the precise moment the satellite signal reaches the ground terminal and is used to synchronize multi-source data and establish time series correlations. These three parameters correspond to spatial position relationships, signal stability, and temporal continuity, respectively. Through multidimensional data fusion, they provide more comprehensive input features for setting up neural network models.

[0055] Specifically, the positioning information acquisition module simultaneously collects pseudorange, signal strength, and reception time, forming a composite data set containing distance, power, and timestamps that is input into the neural network model. Pseudorange serves as a basic spatial parameter, calculating the initial position through geometric relationships; signal strength serves as an auxiliary parameter to correct pseudorange errors caused by atmospheric attenuation or multipath effects; and reception time serves as a timing parameter, combining satellite orbit data to predict dynamic position changes. During the model training phase, the pseudorange, signal strength, and reception time in the sample data are jointly used in feature extraction, enabling the hidden layer to learn the nonlinear relationship between different parameters. The output layer then maps the fused features into more accurate position information.

[0056] By simultaneously considering three key pieces of information: pseudo-moment, signal strength, and reception time, this method can more comprehensively describe the satellite's position and signal characteristics, reducing the potential errors introduced by a single piece of information. This multi-dimensional positioning information helps the neural network model more accurately calculate the satellite's position, thereby improving the accuracy of satellite antenna alignment.

[0057] Furthermore, in some embodiments, the neural network model is constructed in the following manner:

[0058] Get sample data;

[0059] Build an initial neural network model;

[0060] The initial neural network model is trained using the sample data to obtain the neural network model.

[0061] The sample data includes the satellite's positioning information and the satellite's orientation information corresponding to the sample data as a label, which is used to provide the model with a mapping relationship between input and target output.

[0062] The initial neural network model includes an input layer, a hidden layer, and an output layer. The input layer is used to receive the sample data and the position information labels and convert the data into a format that the initial neural network model can process. The hidden layer is used to extract features from the input data and obtain the inherent patterns of the sample data and the position information labels. The output layer is used to map the features extracted by the hidden layer to specific mission objectives to output the satellite's position information. During training, the sample data is divided into a training set, a validation set, and a test set. The training set is used to update the model weights, the validation set is used to monitor overfitting and adjust hyperparameters, and the test set is used to evaluate the generalization ability of the final model.

[0063] Specifically, the initial neural network model can be a multi-layer perceptron structure. In the model construction stage, the input layer receives preprocessed sample data, including pseudorange, signal strength and reception time, the hidden layer performs feature transformation on the data through the activation function, and the output layer generates predicted values ​​of azimuth and inclination. During training, the gradient of the loss function is calculated by the backpropagation algorithm, and the optimizer updates the model weights to minimize the error between the predicted value and the true value. For example, the mean square error is used as the loss function, the stochastic gradient descent optimizer is used to update the weights, the training cycle is set to 100 rounds, and the batch size is 32. The model performance is regularly evaluated through the validation set. If the validation loss does not decrease for 5 consecutive rounds, the training is terminated early to avoid overfitting. Ultimately, the azimuth information prediction error on the test set is reduced to within ±0.5 degrees, significantly shortening the antenna adjustment time and improving the alignment accuracy.

[0064] Through the above scheme, this application utilizes machine learning methods to automatically construct a neural network model suitable for satellite positioning and orientation prediction. This model can learn the complex relationship between satellite positioning information and orientation from a large amount of historical data, avoiding the tedious process of manually designing features and rules. Furthermore, through continuous training and optimization, the model's prediction accuracy can be continuously improved, providing more accurate orientation information support for subsequent satellite antenna alignment.

[0065] In some embodiments, the sample data includes satellite positioning information and a satellite position information tag corresponding to the sample data.

[0066] The satellite position information tags corresponding to the sample data include the satellite's elevation and azimuth angles, recorded synchronously by high-precision measuring instruments. This positioning information and position information are mapped one-to-one by timestamp, forming structured training data. The data is stored in a table format, with each row containing a set of positioning information and its corresponding position information tags, ensuring a clear mapping between input and output.

[0067] Specifically, the correspondence between positioning information and azimuth information labels is achieved through a time synchronization module. When the receiving device collects pseudorange, signal strength, and time, the high-precision measuring instrument synchronously records the elevation and azimuth angles of the satellite. During the model training process, positioning information is used as input features and azimuth information labels are used as supervisory signals. By minimizing the error between the predicted value and the label, the model learns the nonlinear mapping between positioning information and azimuth information. For example, pseudorange and signal strength reflect the relative position change between the satellite and the antenna, and the reception time is used to match the dynamic trajectory. The model outputs an accurate prediction value of the azimuth information through multi-dimensional data association. As a result, the trained neural network model can quickly infer the satellite position based on real-time positioning information, reducing the trial and error time of manual adjustment and improving the efficiency of antenna alignment.

[0068] In some embodiments, before constructing the initial neural network model, the sample data is preprocessed to remove outliers and noise.

[0069] Preprocessing involves a data cleaning phase, where outliers are identified by setting thresholds or using statistical methods, and filtering algorithms are used to eliminate signal noise. This data cleaning phase, along with the sample collection phase, forms a data quality assurance chain. This cleaned data, when fed into model training, results in a more concentrated feature distribution. For example, the three-sigma (3σ) principle is used to filter out outliers that deviate beyond three standard deviations from the mean, and a Kalman filter is applied to eliminate Gaussian noise in satellite signals.

[0070] Specifically, before entering the model training process, sample data first undergoes data cleaning in the preprocessing module. This module's built-in anomaly detection algorithm scans the raw satellite positioning information, removing pseudorange data that falls outside a reasonable range. A sliding window mean filter is then used to process the signal strength data to eliminate transient interference noise. The cleaned sample data forms a structured dataset for input into the model construction phase, effectively preventing noise from interfering with hidden layer feature extraction and improving the stability of the model's output position information. This preprocessing step ensures that the training set data distribution more closely reflects real-world scenarios, reducing parameter oscillations caused by data quality issues during model training.

[0071] In some embodiments, the initial neural network model includes an input layer, a hidden layer and an output layer, wherein the input layer is used to receive the input of the sample data and the position information label and convert the data into a format that can be processed by the initial neural network model; the hidden layer is used to extract features from the input data and obtain the inherent laws of the sample data and the position information label; the output layer is used to map the features extracted by the hidden layer to specific mission objectives to output the position information of the satellite.

[0072] The input layer can convert the raw positioning information into a standardized numerical matrix through normalization. The hidden layer can include at least three fully connected layers, each of which is followed by a linear activation function to gradually extract the nonlinear relationship between positioning and orientation information. The output layer uses a linear activation function to map the feature vector output by the hidden layer into azimuth and elevation values.

[0073] Specifically, the input layer receives the pseudorange, signal strength, and reception time values ​​in the sample data, and maps them to the range of 0 to 1 through normalization, forming a 32-dimensional vector as input. The first fully connected layer of the hidden layer expands the input vector dimension to 128 dimensions, the second fully connected layer compresses it to 64 dimensions, and the third fully connected layer further compresses it to 16 dimensions. Each layer uses a linear activation function to eliminate negative interference. The output layer multiplies the 16-dimensional feature vector with two trainable weight matrices and outputs the azimuth deviation value and elevation deviation value respectively. This structure eliminates redundant information through layer-by-layer dimensionality reduction, avoiding overfitting while ensuring feature extraction capabilities. During the training process, the data processed by the input layer undergoes three nonlinear transformations in the hidden layer. The output layer then compares the final features with the azimuth information label. The weight parameters of each layer are continuously optimized through backpropagation, so that the model gradually establishes an accurate mapping relationship between positioning information and azimuth adjustment.

[0074] In some embodiments, the initial neural network model also includes a loss function and an optimizer, wherein the loss function is used to calculate the average of the square of the difference between the predicted value and the true value, and the optimizer is used to select stochastic gradient descent and its variants, such as Adagrad, Adadelta, RMSProp or Adam optimizers to update the weights of the initial neural network model and minimize the loss function.

[0075] The loss function uses the mean squared error algorithm, quantifying the model's prediction deviation by calculating the average squared error between the predicted value and the label value. The optimizer uses the stochastic gradient descent algorithm or its improved versions, such as stochastic gradient descent with momentum or the adaptive learning rate optimizer, to calculate the gradient and update the model parameters through the backpropagation algorithm. The loss function and the optimizer work together to dynamically adjust the network weights through an error feedback mechanism during training, allowing the model to gradually approach the inherent laws of the sample data.

[0076] Specifically, during training, the training set data is fed into the initial neural network model, and forward propagation generates predictions. The loss function calculates the mean squared error between the predicted and true position information, and the optimizer adjusts the model weights based on the error gradient. With each iteration, the optimizer calculates the partial derivative of the loss function with respect to the weights and updates the weight parameters along the gradient descent path, gradually reducing the prediction error. Through multiple iterations of optimization, the model weights converge to an optimal state, and the error in the final output satellite position information is significantly reduced, thereby ensuring the accuracy and efficiency of satellite antenna gimbal adjustments.

[0077] This application effectively calculates the error between predicted and true values ​​and adaptively adjusts model parameters through the optimizer, accelerating model convergence and improving training efficiency. Furthermore, the optimizer adaptively adjusts the learning rate of each parameter, adapting to gradients of varying scales. This helps handle sparse or noisy gradients, improving the model's robustness and generalization capabilities.

[0078] In some embodiments, the process of training the initial neural network model includes:

[0079] Dividing the sample data into a training set, a validation set, and a test set, wherein the training set is used to train the initial neural network model, the validation set is used to adjust the hyperparameters of the initial neural network model and prevent overfitting, and the test set is used to evaluate the generalization ability of the initial neural network model;

[0080] Iteratively training the initial neural network model using the training set data, wherein in each training step, input data is passed into the initial neural network model, forward propagation is calculated to obtain a prediction result, and then a loss value is calculated according to the loss function, and gradients are calculated by a backpropagation algorithm to update the weights of the initial neural network model;

[0081] The trained initial neural network model is evaluated, and the one that meets the evaluation criteria is the set neural network model.

[0082] Among them, the training set uses 60% to 70% of the total sample data to perform iterative training for weight update; the validation set uses 15% to 20% of the total sample data, which is input into the model after each complete training cycle, and the validation loss value is calculated and compared with the difference in the training loss value. When the validation loss value does not decrease for three consecutive cycles, the early stopping mechanism is triggered; the test set uses 15% to 20% of the total sample data, which is input into the model after the model training is completed to calculate the mean square error index. When the index is lower than the preset threshold, the model is confirmed to be qualified.

[0083] Specifically, the sample data is first divided into training, validation, and test sets in a ratio of 7:2:1. The training data is fed into the model in batches, with each batch containing 32 to 128 samples. The data is converted into normalized tensors by the input layer and then passed to the hidden layer for feature extraction. The feature vector output by the hidden layer is mapped to the output layer through a fully connected layer to generate a predicted value for the orientation information. The loss function calculates the mean squared error between the predicted orientation and the true orientation. Learning rate decay is triggered when the loss value exceeds 1.05 times the historical minimum. The validation set is called every 1000 iterations to calculate the validation loss and record a snapshot of the model weights. After training, the test set is fed into the final model, and the mean absolute error between the predicted orientation and the true orientation is output as a generalization indicator. When this indicator is less than 0.5 degrees, the model is considered to meet deployment requirements.

[0084] The training set is used to update the model weights through multiple iterations of training. The validation set is used to periodically evaluate model performance during training, adjusting hyperparameters such as the learning rate and batch size based on the validation set loss. The test set is used to ultimately evaluate the model's performance on unseen data. Forward propagation converts input data into predictions. The loss function calculates the difference between the prediction and the true value. Backward propagation adjusts the weights based on the loss gradient to gradually reduce the loss.

[0085] This application ensures the model's accuracy in azimuth predictions in practical applications through phased data partitioning and dynamic learning rate adjustment. Hyperparameter optimization, performed under validation set monitoring, significantly improves the model's generalization capabilities to unseen data, avoiding repeated debugging issues caused by model bias and enabling the antenna gimbal to achieve optimal pointing with a single adjustment.

[0086] The present application also provides a satellite antenna alignment system, including a positioning information acquisition module, a positioning information input module, a neural network model setting module and an adjustment module;

[0087] The positioning information acquisition module is used to obtain satellite positioning information;

[0088] The positioning information input module is used to input the positioning information into a set neural network model;

[0089] The set neural network model is used to calculate the position information of the satellite;

[0090] The adjustment module adjusts the direction and tilt of the pan / tilt of the satellite antenna according to the azimuth information of the satellite, so that the satellite antenna is aligned with the satellite.

[0091] Among them, the positioning information acquisition module extracts pseudo-moment, signal strength and reception time by receiving satellite signals. The positioning information input module converts the data into a format that can be processed by the set neural network model and transmits it to the input layer. The neural network model is set based on the data received by the input layer, extracts features through the hidden layer and calculates the orientation information. The output layer passes the calculation results to the adjustment module; after receiving the orientation information, the adjustment module drives the pan-tilt motor to adjust the direction and inclination so that the antenna pointing is consistent with the calculation result.

[0092] Specifically, the positioning information acquisition module collects the pseudo-moment, signal strength and reception time in the satellite signal in real time, and generates standardized data after filtering out interference through the signal processing unit; the positioning information input module transmits the standardized data to the input layer of the set neural network model in a set format, and the input layer maps the data into a multi-dimensional vector; the hidden layer of the set neural network model performs a nonlinear transformation on the multi-dimensional vector through an activation function, extracts the correlation between the signal characteristics and the azimuth information, and the output layer maps the transformed feature vector into an azimuth angle value; the adjustment module parses the azimuth angle value, converts it into a gimbal drive instruction, and controls the stepper motor to adjust the horizontal direction angle and pitch angle of the antenna.

[0093] This application achieves automated and precise alignment of satellite antennas, can directly learn and extract satellite orientation features from complex positioning information, and has rapid response capabilities, capable of processing the dynamic effects of satellite orbit changes and receiver motion in real time, thereby improving the stability and reliability of satellite communication systems. The automated alignment process reduces the need for manual operation, reduces operational difficulty and human error, and improves the efficiency and accuracy of satellite antenna alignment. In addition, it has good adaptability and generalization capabilities, capable of meeting satellite communication needs in different environmental conditions.

[0094] The satellite antenna alignment system of the present application can be used to implement the satellite antenna alignment method described in any one of the above technical solutions.

[0095] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A satellite antenna swap method, characterized in that: include: Obtain satellite positioning information; Inputting the positioning information into a set neural network model; Calculating the satellite's position information using the set neural network model; According to the azimuth information of the satellite, the direction and inclination of the pan / tilt of the satellite antenna are adjusted so that the satellite antenna is aligned with the satellite.

2. The satellite antenna alignment method according to claim 1, wherein: The satellite positioning information includes pseudo moment, signal strength and reception time.

3. The satellite antenna alignment method according to claim 1, wherein: The neural network model is constructed in the following way: Get sample data; Build an initial neural network model; The initial neural network model is trained using the sample data to obtain the neural network model.

4. The satellite antenna alignment method according to claim 3, wherein: The sample data includes satellite positioning information and a satellite position information tag corresponding to the sample data.

5. The satellite antenna alignment method according to claim 3, wherein: Before constructing the initial neural network model, the sample data is preprocessed to remove outliers and noise.

6. The satellite antenna adjustment method according to claim 3, wherein: The initial neural network model includes an input layer, a hidden layer and an output layer. The input layer is used to receive the input of the sample data and the position information label and convert the data into a format that can be processed by the initial neural network model; the hidden layer is used to extract features from the input data and obtain the inherent laws of the sample data and the position information label; the output layer is used to map the features extracted by the hidden layer to specific mission objectives to output the position information of the satellite.

7. The satellite antenna adjustment method according to claim 6, characterized in that: The initial neural network model also includes a loss function and an optimizer, wherein the loss function is used to calculate the average of the square of the difference between the predicted value and the true value, and the optimizer is used to select stochastic gradient descent and its variants to update the weights of the initial neural network model to minimize the loss function.

8. The satellite antenna adjustment method according to claim 7, characterized in that: The process of training the initial neural network model includes: Dividing the sample data into a training set, a validation set, and a test set, wherein the training set is used to train the initial neural network model, the validation set is used to adjust the hyperparameters of the initial neural network model and prevent overfitting, and the test set is used to evaluate the generalization ability of the initial neural network model; Iteratively training the initial neural network model using the training set data, wherein in each training step, input data is passed into the initial neural network model, forward propagation is calculated to obtain a prediction result, and then a loss value is calculated according to the loss function, and gradients are calculated by a backpropagation algorithm to update the weights of the initial neural network model; The trained initial neural network model is evaluated, and the one that meets the evaluation criteria is the set neural network model.

9. The satellite antenna swap method according to claim 8, characterized in that: During the training process, the performance of the model is regularly evaluated on the validation set, and the hyperparameters of the initial neural network model are adjusted according to the loss value of the validation set.

10. A satellite antenna alignment system, characterized in that: It includes a positioning information acquisition module, a positioning information input module, a neural network model setting module and an adjustment module; The positioning information acquisition module is used to obtain satellite positioning information; The positioning information input module is used to input the positioning information into a set neural network model; The set neural network model is used to calculate the position information of the satellite; The adjustment module adjusts the direction and tilt of the pan / tilt of the satellite antenna according to the azimuth information of the satellite, so that the satellite antenna is aligned with the satellite.

Images