A satellite signal quality prediction model construction method, device and prediction method

Abstract

The application provides a satellite signal quality prediction model construction method and device and a prediction method, and relates to the technical field of artificial intelligence. The construction method comprises the following steps: acquiring a sky image and corresponding satellite historical signal-to-noise ratio data; performing feature extraction on the sky image to obtain sky image features, and performing normalization processing on the satellite historical signal-to-noise ratio data to obtain numerical data features; splicing the image feature data and the numerical data features to obtain a comprehensive feature vector; and training and optimizing an original quality prediction model according to the comprehensive feature vector and the satellite historical signal-to-noise ratio data to obtain a quality prediction model, wherein the quality prediction model is used for predicting satellite signal quality, and the quality prediction model comprises a random convolution kernel conversion module and a ridge regression module. The application improves the accuracy and stability of satellite signal quality prediction.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and more specifically, to a method, apparatus, and prediction method for constructing a satellite signal quality prediction model. Background Technology

[0002] With the development and application of the new generation of satellite mobile communication systems, satellite mobile communication is moving towards broadband and high speed, which poses a huge challenge to radio monitoring and prediction. At the same time, due to the wide range and special nature of satellite mobile terminal applications, researchers have been conducting research on satellite signal quality prediction in recent years.

[0003] Currently, satellite signal quality prediction methods can be categorized into physical methods, statistical methods, and combined methods. Physical methods rely on numerical weather predictions provided by meteorological departments and require the establishment of physical models for specific satellite stations. These methods have limited prediction accuracy and poor robustness. Statistical methods are data-driven approaches that utilize prediction models, the relationship between input variables and predicted values, and statistical analysis of historical satellite signal quality data and related factors. However, the accuracy of these methods is limited by weather conditions and other environmental factors, resulting in low prediction accuracy and stability, insufficient to meet the needs of various application areas. Summary of the Invention

[0004] The problem addressed by this invention is how to improve the accuracy and stability of satellite signal quality prediction.

[0005] To address the above problems, this invention provides a method for constructing a satellite signal quality prediction model, comprising the following steps:

[0006] Acquire sky images and corresponding historical satellite signal-to-noise ratio data;

[0007] Feature extraction is performed on the sky image to obtain sky image features, and the historical signal-to-noise ratio data of the satellite is normalized to obtain numerical data features;

[0008] The image feature data and the numerical data features are concatenated to obtain a comprehensive feature vector;

[0009] The original quality prediction model is trained and optimized based on the comprehensive feature vector and the historical signal-to-noise ratio data of the satellite to obtain a quality prediction model. The quality prediction model is used to predict the satellite signal quality and includes a random convolution kernel transformation module and a ridge regression module.

[0010] Optionally, the step of extracting features from the sky image to obtain sky image features includes:

[0011] Obtain a feature extraction model for sky images;

[0012] The sky image is input into the sky image feature extraction model to obtain the sky image features. The sky image feature extraction model is based on an adversarial autocoding system.

[0013] Optionally, the method for constructing the sky image feature extraction model includes:

[0014] Obtain historical sky images and corresponding historical feature images.

[0015] The original sky image feature extraction model is trained and optimized based on the historical sky image and the historical feature image to obtain the sky image feature extraction model.

[0016] Optionally, the adversarial autocoding system includes an encoder, a decoder, and a discriminator. The step of training and optimizing the original sky image feature extraction model based on the historical sky image and the historical feature image to obtain the sky image feature extraction model includes:

[0017] The historical sky image is input into the encoder to obtain a temporary feature vector;

[0018] The temporary feature vector is input into the discriminator to obtain the target feature vector;

[0019] The temporary feature vector and the target feature vector are input into the decoder to obtain a temporary sky feature image.

[0020] The original sky image feature extraction model is optimized based on the temporary sky feature image and the historical feature image to obtain the sky image feature extraction model.

[0021] Optionally, the step of training and optimizing the original quality prediction model based on the integrated feature vector and the historical signal-to-noise ratio data of the satellite to obtain the quality prediction model includes:

[0022] The comprehensive feature vector is input into the random convolution kernel conversion module to obtain intermediate feature data;

[0023] The intermediate feature data is input into the ridge regression module to obtain temporary signal quality data;

[0024] The original quality prediction model is optimized based on the temporary signal quality data and the satellite's historical signal-to-noise ratio data to obtain the quality prediction model.

[0025] Optionally, the step of optimizing the original quality prediction model based on the temporary signal quality data and the satellite's historical signal-to-noise ratio data to obtain the quality prediction model includes:

[0026] Based on the temporary signal quality data and the historical signal-to-noise ratio data of the satellite, the loss of the original wind turbine power prediction model is calculated to obtain the loss function output.

[0027] The model parameters of the original quality prediction model are adjusted according to the output of the loss function until the input of the loss function meets the preset conditions. The original quality prediction model after parameter adjustment is then used as the quality prediction model.

[0028] Optionally, the historical signal-to-noise ratio (SNR) data of the satellite includes multiple sub-SNR data, and the normalization processing of the historical SNR data of the satellite to obtain numerical data features includes:

[0029] Based on all the aforementioned sub-signal-to-noise ratio data, obtain the maximum and minimum values ​​of the satellite's historical signal-to-noise ratio data;

[0030] Based on the maximum and minimum values, the historical signal-to-noise ratio data of the satellite is processed using Equation 1 to obtain the numerical data characteristics;

[0031] Wherein, Equation 1 is:

[0032]

[0033] Among them, X N X is the numerical data feature corresponding to the Nth sub-signal-to-noise ratio data. N For the Nth sub-signal-to-noise ratio data, X max X is the maximum value of the historical signal-to-noise ratio data of the satellite. min This is the minimum value of the historical signal-to-noise ratio data of the satellite.

[0034] The advantages of the satellite signal quality prediction model construction method described in this invention compared to existing technologies are as follows: First, it acquires sky images and corresponding historical satellite signal-to-noise ratio (SNR) data. The sky images are observed by satellite sensors, and the historical SNR data consists of a series of measurements related to satellite signal quality. Features are extracted from the sky images to obtain their feature representations. Simultaneously, the historical SNR data is normalized and transformed into numerical features. The image feature data and numerical features are then concatenated to generate a comprehensive feature vector. This comprehensive feature vector contains two aspects of information: the feature representation of the sky images and the numerical features of the historical SNR data. This comprehensive feature vector and the historical SNR data are used to train and optimize the original quality prediction model. The quality prediction model employs a random convolution kernel transformation module and a ridge regression module. The random convolution kernel transformation module uses random convolution kernel transformation technology, which can extract key information from image features and has strong expressive power and anti-interference ability. The ridge regression module is a regression algorithm capable of modeling high-dimensional data and has good generalization ability. By using integrated feature vectors and historical satellite signal-to-noise ratio (SNR) data during training, the satellite signal quality prediction model can more fully utilize the correlation information between image features and numerical features to more accurately predict satellite signal quality. Furthermore, the use of random convolution kernel transformation techniques and ridge regression algorithms can fully exploit the feature information of the integrated feature vectors, improving the model's predictive performance and generalization ability. Therefore, this application, by combining image features and numerical features, makes fuller use of multifaceted information. By concatenating image feature data and numerical feature data, the correlation between image features and historical satellite SNR data can be fully considered, thus better representing the diversity of signal quality. Simultaneously, the use of random convolution kernel transformation and ridge regression modules can improve the model's expressive and generalization abilities, adapting to the signal quality prediction needs of different scenarios. Furthermore, by fully utilizing multi-source heterogeneous data input, satellite signal quality prediction can be effectively achieved, contributing to improved satellite communication performance.

[0035] To address the aforementioned technical problems, the present invention also provides a satellite signal quality prediction model construction apparatus, comprising:

[0036] The acquisition unit is used to acquire sky images and corresponding satellite historical signal-to-noise ratio data, and to extract features from the sky images to obtain sky image features;

[0037] The processing unit is used to extract features from the sky image to obtain sky image features, and to normalize the historical signal-to-noise ratio data of the satellite to obtain numerical data features.

[0038] The processing unit is also used to concatenate the image feature data with the numerical data features to obtain a comprehensive feature vector;

[0039] The processing unit is also used to train and optimize the original quality prediction model based on the comprehensive feature vector and the satellite historical signal-to-noise ratio data to obtain a quality prediction model. The quality prediction model is used to predict satellite signal quality, and the quality prediction model includes a random convolution kernel transformation module and a ridge regression module.

[0040] The satellite signal quality prediction model construction device and the satellite signal quality prediction model construction method described in this invention have the same advantages over the prior art, and will not be repeated here.

[0041] To address the aforementioned technical problems, the present invention also provides a satellite signal quality prediction method, comprising:

[0042] Acquire target sky images and corresponding satellite signal-to-noise ratio data;

[0043] Feature extraction is performed on the target sky image to obtain target image features, and the satellite signal-to-noise ratio data is normalized to obtain target numerical data features;

[0044] The target image features are concatenated with the target numerical data features to obtain the target comprehensive feature vector;

[0045] The target integrated feature vector is input into the quality prediction model obtained by the satellite signal quality prediction model construction method to obtain the satellite signal quality prediction result.

[0046] The satellite signal quality prediction method and the satellite signal quality prediction model construction method described in this invention have the same advantages over the prior art, and will not be repeated here.

[0047] To address the aforementioned technical problems, the present invention also provides a satellite signal quality prediction method apparatus, comprising:

[0048] The acquisition module is used to acquire target sky images and corresponding satellite signal-to-noise ratio data;

[0049] The processing module is used to extract features from the target sky image to obtain target image features, and to normalize the satellite signal-to-noise ratio data to obtain target numerical data features.

[0050] The processing module is also used to concatenate the target image features with the target numerical data features to obtain a target comprehensive feature vector;

[0051] The processing module is also used to input the target integrated feature vector into the quality prediction model obtained by the satellite signal quality prediction model construction method to obtain the satellite signal quality prediction result.

[0052] The satellite signal quality prediction device and the satellite signal quality prediction model construction method described in this invention have the same advantages over the prior art, and will not be repeated here. Attached Figure Description

[0053] Figure 1 This is a flowchart of the satellite signal quality prediction model construction method in an embodiment of the present invention;

[0054] Figure 2 This is one of the flowcharts for the sky image feature extraction model construction method in this embodiment of the invention;

[0055] Figure 3 This is the second flowchart of the sky image feature extraction model construction method in this embodiment of the invention;

[0056] Figure 4 This is a structural diagram of the satellite signal quality prediction model construction device in an embodiment of the present invention;

[0057] Figure 5 This is one of the flowcharts for the satellite signal quality prediction method in the embodiments of the present invention;

[0058] Figure 6 This is the second flowchart of the satellite signal quality prediction method in this embodiment of the invention;

[0059] Figure 7 This is an internal structural diagram of a computer device in an embodiment of the present invention. Detailed Implementation

[0060] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0061] like Figure 1 As shown, in one embodiment, a method for constructing a satellite signal quality prediction model is provided, including the following steps:

[0062] Step S1: Obtain sky images and corresponding historical signal-to-noise ratio data for satellites.

[0063] Specifically, sky images are acquired using satellite sensors, image sensors, or other suitable data sources. All collected and used sky images are unlabeled. Simultaneously, historical signal-to-noise ratio (SNR) data from low-Earth orbit (LEO) satellites corresponding to the sky images is collected. This historical SNR data, obtainable from satellite terminals, covers signal quality information at different time points. The collected sky images are then precisely mapped to the corresponding historical SNR data from LEO satellites, preparing for subsequent training and validation of the original quality prediction model.

[0064] Step S2: Extract features from the sky image to obtain sky image features, and normalize the historical signal-to-noise ratio data of the satellite to obtain numerical data features.

[0065] Specifically, feature extraction is performed on sky images to obtain their feature representations. Simultaneously, historical satellite signal-to-noise ratio (SNR) data is normalized. Normalization maps SNR data of different ranges and scales to a unified scale. Normalization is a method of simplifying calculations; it transforms dimensional expressions into dimensionless expressions, making them scalars. Normalization is a dimensionless processing technique that transforms the absolute values ​​of physical system values ​​into relative values. This is an effective way to simplify calculations and reduce the magnitude of quantities.

[0066] Step S3: Concatenate the image feature data with the numerical data features to obtain a comprehensive feature vector.

[0067] Specifically, the image feature data is concatenated with the numerical data features to obtain a comprehensive feature vector. This comprehensive feature vector contains two aspects of information: the feature representation of the sky image and the numerical features of the satellite's historical signal-to-noise ratio data. It can fully consider the correlation between the image features and the satellite's historical signal-to-noise ratio data, thereby better representing the diversity of signal quality and providing training data for subsequent model training.

[0068] Step S4: Train and optimize the original quality prediction model based on the comprehensive feature vector and the historical signal-to-noise ratio data of the satellite to obtain a quality prediction model. The quality prediction model is used to predict the satellite signal quality and includes a random convolution kernel transformation module and a ridge regression module.

[0069] Specifically, the random convolution kernel transformation module employs random convolution kernel transformation technology, which can extract key information from image features and has strong expressive power and anti-interference ability. The ridge regression module is a regression algorithm capable of modeling high-dimensional data and has good generalization ability. By using comprehensive feature vectors and historical satellite signal-to-noise ratio data during training, the quality prediction model can more fully utilize the correlation information between image features and numerical features to more accurately predict satellite signal quality. The comprehensive feature vector serves as the input to the quality prediction model, and the historical satellite signal-to-noise ratio data serves as label data, used for comparison with the output of the quality prediction model and for loss calculation during training.

[0070] The satellite signal quality prediction model construction method described in this embodiment first acquires sky images and corresponding historical satellite signal-to-noise ratio (SNR) data. The sky images are observed by satellite sensors, and the historical SNR data consists of a series of measurements related to satellite signal quality. Feature extraction is performed on the sky images to obtain their feature representation. Simultaneously, the historical SNR data is normalized, transforming it into numerical features. The image feature data and numerical features are then concatenated to generate a comprehensive feature vector. This comprehensive feature vector contains two aspects of information: the feature representation of the sky images and the numerical features of the historical SNR data. This comprehensive feature vector and the historical SNR data are used to train and optimize the original quality prediction model. The quality prediction model employs a random convolution kernel transformation module and a ridge regression module. The random convolution kernel transformation module uses random convolution kernel transformation technology, which can extract key information from image features and has strong expressive power and anti-interference ability. The ridge regression module is a regression algorithm capable of modeling high-dimensional data and has good generalization ability. By using integrated feature vectors and historical satellite signal-to-noise ratio (SNR) data during training, the satellite signal quality prediction model can more fully utilize the correlation information between image features and numerical features to more accurately predict satellite signal quality. Furthermore, the use of random convolution kernel transformation techniques and ridge regression algorithms can fully exploit the feature information of the integrated feature vectors, improving the model's predictive performance and generalization ability. Therefore, this application, by combining image features and numerical features, makes fuller use of multifaceted information. By concatenating image feature data and numerical feature data, the correlation between image features and historical satellite SNR data can be fully considered, thus better representing the diversity of signal quality. Simultaneously, the use of random convolution kernel transformation and ridge regression modules can improve the model's expressive and generalization abilities, adapting to the signal quality prediction needs of different scenarios. Furthermore, by fully utilizing multi-source heterogeneous data input, satellite signal quality prediction can be effectively achieved, contributing to improved satellite communication performance.

[0071] In some embodiments, step S2, extracting features from the sky image to obtain sky image features, includes:

[0072] Step S21: Obtain the sky image feature extraction model.

[0073] Step S22: Input the sky image into the sky image feature extraction model to obtain the sky image features. The sky image feature extraction model is based on an adversarial autocoding system.

[0074] Specifically, a sky image feature extraction model based on an adversarial autocoding system is constructed to achieve automatic extraction of features from unlabeled sky image data. An adversarial autocoding system is a deep learning model whose training process includes training the encoder and decoder, as well as adversarial training of the discriminator. By inputting unlabeled sky images (sky images) into the model, corresponding latent feature vectors (sky image features) can be obtained, which are then used for subsequent signal quality prediction.

[0075] In some preferred embodiments, a set of unlabeled sky images, including pictures under different weather conditions such as sunny, cloudy, and overcast, are used. An adversarial autocoding system is used to extract latent features from these images. Input images: The unlabeled images are one of three images: a sunny sky image, a cloudy sky image, and an overcast sky image. The adversarial autocoding system converts each input image into a latent feature vector (sky image feature). This latent feature vector is the output of the encoder, capturing important features of the input image, such as the sky's color, texture, and brightness. Each input image, after being processed by the encoder, yields a corresponding latent feature vector. These latent feature vectors are typically numerical vectors with low dimensionality, such as 100 dimensions or less. For example, the latent feature vectors (simplified to 3D) are as follows: Latent feature vector for a sunny image: [0.2, 0.8, 0.4], latent feature vector for a cloudy image: [0.5, 0.3, 0.7], latent feature vector for an overcast image: [0.1, 0.2, 0.6]. These latent feature vectors capture different features of each input image; for example, a sunny image might have bright features, a cloudy image might have soft features, and an overcast image might have darker features. These feature vectors will be used in subsequent tasks.

[0076] In some embodiments, the method for constructing the sky image feature extraction model includes:

[0077] Step A1: Obtain historical sky images and corresponding historical feature images;

[0078] Step A2: Train and optimize the original sky image feature extraction model based on the historical sky image and the historical feature image to obtain the sky image feature extraction model.

[0079] In some embodiments, such as Figure 2 As shown, the adversarial autocoding system includes an encoder, a decoder, and a discriminator. In step A2, training and optimizing the original sky image feature extraction model based on the historical sky image and the historical feature image to obtain the sky image feature extraction model includes:

[0080] Step A21: Input the historical sky image into the encoder to obtain a temporary feature vector.

[0081] Step A22: Input the temporary feature vector into the discriminator to obtain the target feature vector.

[0082] Step A23: Input the temporary feature vector and the target feature vector into the decoder to obtain a temporary sky feature image.

[0083] Step A24: Optimize the original sky image feature extraction model based on the temporary sky feature image and the historical feature image to obtain the sky image feature extraction model.

[0084] Specifically, an adversarial autocoding system is a deep learning model that includes an encoder, a decoder, and a discriminator. The training process of the model includes training the encoder and decoder, as well as adversarial training of the discriminator.

[0085] The encoder, decoder, and discriminator are all composed of deep neural networks, with the input h of the m-th layer of the network. m-1 The x-th output feature map of this layer for:

[0086]

[0087] Where, σ m It is the activation function of the m-th layer, C m-1 It is the number of output channels of layer m-1. Let be the connection weights between the i-th channel and the x-th channel in the m-th layer. It is the bias of the x-th channel in layer m.

[0088] Specifically, the adversarial autoencoder system is an unsupervised algorithm model whose training process alternates between adversarial training and reconstruction training. First, raw sky image data without label information is sampled, and the model is trained in two phases using this sampled dataset. During adversarial training, the encoder and discriminator are trained, updating their parameters to improve the discriminator's discriminative ability and the encoder's feature extraction ability. During reconstruction training, the encoder and decoder are jointly trained. The training objective is to minimize the difference between the generated feature vectors and the true feature vectors, effectively extracting meaningful features from the sky images. This completes the self-supervised training of the adversarial autoencoder system, using the encoder in the model to transform the unlabeled sky images into latent feature vectors for subsequent signal quality prediction.

[0089] like Figure 3 As shown, the adversarial autoencoder system converts the input sky image into a latent feature vector. The training process is as follows: the input sky image is the starting point of the module; the encoder is the first part of the module; the feature vector is the output of the encoder, also known as the representation in the latent space, and is a numerical vector containing abstract feature information of the input image; the discriminator is the second part of the module, used to evaluate the quality of the generated feature vector; the decoder is the third part of the module, which takes the outputs of the encoder and discriminator and reconstructs the input sky image. The reconstructed image is generated by decoding the feature vector into pixel values; the reconstruction loss is the loss function used to train the model, measuring the difference between the reconstructed sky image and the original input sky image, with the goal of minimizing the reconstruction loss.

[0090] In some embodiments, step S4, which involves training and optimizing the original quality prediction model based on the integrated feature vector and the historical signal-to-noise ratio data of the satellite to obtain the quality prediction model, includes:

[0091] Step S41: Input the comprehensive feature vector into the random convolution kernel conversion module to obtain intermediate feature data.

[0092] Step S42: Input the intermediate feature data into the ridge regression module to obtain temporary signal quality data.

[0093] Step S43: The original quality prediction model is optimized based on the temporary signal quality data and the satellite historical signal-to-noise ratio data to obtain the quality prediction model.

[0094] In some embodiments, step S43, which involves optimizing the original quality prediction model based on the temporary signal quality data and the satellite historical signal-to-noise ratio data to obtain the quality prediction model, includes:

[0095] Step S431: Calculate the loss of the original wind turbine power prediction model based on the temporary signal quality data and the satellite historical signal-to-noise ratio data to obtain the loss function output.

[0096] Step S432: Adjust the model parameters of the original quality prediction model according to the output of the loss function until the input of the loss function meets the preset conditions, and use the original quality prediction model after parameter adjustment as the quality prediction model.

[0097] Specifically, a comprehensive feature vector is created by concatenating the feature vectors of the sky image and the numerical data. This comprehensive feature vector contains information from different sources and includes two aspects: the feature representation of the sky image and the numerical features of the historical signal-to-noise ratio (SNR) data from satellites. The numerical data features are normalized SNR data, an ordered set of measurements and timestamps, X = {(t1,x1),(t2,x2),…(t3,x3)}, where t…(t3,x3)…(t1,x1),…(t2,x2),…(t3,x3)} ... i It is x i The corresponding timestamp. And the feature vector of the sky image (Y = {x} l+1 ,x l+2 ,…,x n After concatenation, the comprehensive feature vector is defined as a univariate time series as a sample of the dataset:

[0098] X′={(x1,x2,…,x),(x2,x3,…,x l +1),…(x n-1 ,x n-l+1 ,x n-1 )},Y=

[0099] {x l+1 ,x l+2 ,…,x n}:

[0100] Here, l is the time window, and the next value is predicted using consecutive values ​​of the time window size; that is, x is predicted using (x1, x2, ..., x). l+1 wait.

[0101] Specifically, the comprehensive feature vector is input into the random convolution kernel conversion module to obtain intermediate feature data; the intermediate feature data is input into the ridge regression module to obtain temporary signal quality data; based on the loss function, loss is calculated according to the temporary signal quality data and the satellite historical signal-to-noise ratio data, and the model parameters of the original quality prediction model are tuned by backpropagation until the loss function input meets the preset conditions, and the original quality prediction model after parameter adjustment is used as the quality prediction model.

[0102] In some embodiments, in step S2, the historical signal-to-noise ratio (SNR) data of the satellite includes multiple sub-SNR data, and the normalization processing of the historical SNR data of the satellite to obtain numerical data features includes:

[0103] Step S23: Obtain the maximum and minimum values ​​of the satellite's historical signal-to-noise ratio data based on all the sub-signal-to-noise ratio data.

[0104] Step S24: Based on the maximum value and the minimum value, process the historical signal-to-noise ratio data of the satellite using Equation 1 to obtain the numerical data characteristics.

[0105] Wherein, Equation 1 is:

[0106]

[0107] Among them, X N X is the numerical data feature corresponding to the Nth sub-signal-to-noise ratio data. N For the Nth sub-signal-to-noise ratio data, X max X is the maximum value of the historical signal-to-noise ratio data of the satellite. min This is the minimum value of the historical signal-to-noise ratio data of the satellite.

[0108] Specifically, the normalization process uses Min-Max Normalization. This is a common normalization method used to map data to a specified range, typically between 0 and 1. This method performs a linear transformation of the data by calculating a minimum and a maximum value, distributing the data within the specified range. The uniform scale obtained through normalization is the range to which the data is mapped.

[0109] In some preferred embodiments, historical satellite signal-to-noise ratio (SNR) data, with values ​​ranging from 20 to 100, is scaled to a range between 0 and 1. Min-max normalization is used to scale the data to this range. The minimum and maximum values ​​of the sub-SNR data are X. min For 20, X max The value is 100. A scaling transformation is applied to each sub-signal-to-noise ratio (SNR) data point. For example, the characteristics of numerical data with a SNR of 30 are as follows:

[0110]

[0111] For numerical data with a sub-signal-to-noise ratio of 80, the characteristics are as follows:

[0112]

[0113] The satellite signal quality prediction model construction method described in this embodiment first acquires sky images and corresponding historical satellite signal-to-noise ratio (SNR) data. The sky images are observed by satellite sensors, and the historical SNR data consists of a series of measurements related to satellite signal quality. Feature extraction is performed on the sky images to obtain their feature representation. Simultaneously, the historical SNR data is normalized, transforming it into numerical features. The image feature data and numerical features are then concatenated to generate a comprehensive feature vector. This comprehensive feature vector contains two aspects of information: the feature representation of the sky images and the numerical features of the historical SNR data. This comprehensive feature vector and the historical SNR data are used to train and optimize the original quality prediction model. The quality prediction model employs a random convolution kernel transformation module and a ridge regression module. The random convolution kernel transformation module uses random convolution kernel transformation technology, which can extract key information from image features and has strong expressive power and anti-interference ability. The ridge regression module is a regression algorithm capable of modeling high-dimensional data and has good generalization ability. By using integrated feature vectors and historical satellite signal-to-noise ratio (SNR) data during training, the satellite signal quality prediction model can more fully utilize the correlation information between image features and numerical features to more accurately predict satellite signal quality. Furthermore, the use of random convolution kernel transformation techniques and ridge regression algorithms can fully exploit the feature information of the integrated feature vectors, improving the model's predictive performance and generalization ability. Therefore, this application, by combining image features and numerical features, makes fuller use of multifaceted information. By concatenating image feature data and numerical feature data, the correlation between image features and historical satellite SNR data can be fully considered, thus better representing the diversity of signal quality. Simultaneously, the use of random convolution kernel transformation and ridge regression modules can improve the model's expressive and generalization abilities, adapting to the signal quality prediction needs of different scenarios. Furthermore, by fully utilizing multi-source heterogeneous data input, satellite signal quality prediction can be effectively achieved, contributing to improved satellite communication performance.

[0114] like Figure 4 As shown, another embodiment of the present invention provides a satellite signal quality prediction model construction apparatus, comprising:

[0115] The acquisition unit is used to acquire sky images and corresponding satellite historical signal-to-noise ratio data, and to extract features from the sky images to obtain sky image features.

[0116] The processing unit is used to extract features from the sky image to obtain sky image features, and to normalize the historical signal-to-noise ratio data of the satellite to obtain numerical data features.

[0117] The processing unit is also used to concatenate the image feature data with the numerical data features to obtain a comprehensive feature vector.

[0118] The processing unit is also used to train and optimize the original quality prediction model based on the comprehensive feature vector and the satellite historical signal-to-noise ratio data to obtain a quality prediction model. The quality prediction model is used to predict satellite signal quality, and the quality prediction model includes a random convolution kernel transformation module and a ridge regression module.

[0119] like Figure 5 As shown, another embodiment of the present invention provides a satellite signal quality prediction method, comprising:

[0120] Step T1: Obtain the target sky image and the corresponding satellite signal-to-noise ratio data.

[0121] Step T2: Extract features from the target sky image to obtain target image features, and normalize the satellite signal-to-noise ratio data to obtain target numerical data features.

[0122] Step T3: The target image features and the target numerical data features are concatenated to obtain the target comprehensive feature vector.

[0123] Step T4: Input the target integrated feature vector into the quality prediction model obtained by the satellite signal quality prediction model construction method to obtain the satellite signal quality prediction result.

[0124] Specifically, such as Figure 6 As shown, the satellite signal quality prediction model (quality prediction model) is built based on a random convolution kernel transformation module and a ridge regression module, and the prediction process is divided into two stages. In the first stage, the random convolution kernel transformation module is used to process the concatenated feature vector (target comprehensive feature vector) to generate an intermediate representation (intermediate features). This first stage generates a large number of random convolution kernels. Each random convolution kernel performs a sliding dot product with the input sequence to generate a feature map, extracting two features: the proportion of positive values ​​and the maximum value. The expression for calculating the sliding dot product of the convolution kernel and the sequence is:

[0125]

[0126] In the formula, X i To apply convolution computation from position i, ω is the random convolution kernel, and p is the dilation value. In the second stage, the ridge regression module is used to perform regression analysis on the intermediate representation after the random convolution kernel transformation to obtain the predicted results of satellite signal quality.

[0127] Specifically, in practical applications, we can use this trained quality prediction model to predict satellite signal quality in real time. For example, during sudden weather changes, we can acquire the latest sky images and historical satellite signal-to-noise ratio (SNR) data, and convert them into a target comprehensive feature vector. This vector is then input into the quality prediction model to predict the current satellite signal quality. Based on the prediction results, we can make targeted adjustments and optimizations to ensure the normal operation of the system and the stability of communication quality. In summary, training a quality prediction model using comprehensive feature vectors and historical satellite SNR data can more accurately predict satellite signal quality. By combining image features and numerical features, and employing techniques such as random convolution kernel transformation and ridge regression, this method has significant advantages in improving prediction accuracy and generalization ability.

[0128] Another embodiment of the present invention provides a satellite signal quality prediction apparatus, comprising:

[0129] The acquisition module is used to acquire target sky images and corresponding satellite signal-to-noise ratio data.

[0130] The processing module is used to extract features from the target sky image to obtain target image features, and to normalize the satellite signal-to-noise ratio data to obtain target numerical data features.

[0131] The processing module is also used to concatenate the target image features with the target numerical data features to obtain a target comprehensive feature vector.

[0132] The processing module is also used to input the target integrated feature vector into the quality prediction model obtained by the satellite signal quality prediction model construction method to obtain the satellite signal quality prediction result.

[0133] Another embodiment of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described satellite signal quality prediction method.

[0134] It should be noted that this device can be a computer device such as a server or mobile terminal.

[0135] Figure 7An internal structural diagram of a computer device in one embodiment is shown. The computer device includes a processor, memory, network interface, input device, and display screen connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program that, when executed by the processor, enables the processor to implement a satellite signal quality prediction method. The internal memory may also store a computer program that, when executed by the processor, enables the processor to execute the satellite signal quality prediction method. The display screen of the computer device can be a liquid crystal display (LCD) or an e-ink display. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0136] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the satellite signal quality prediction method described above.

[0137] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0138] It should be noted that, in this document, terms such as "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 limitation, 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.

[0139] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

[0140] While the present invention has been disclosed above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and all such changes and modifications will fall within the scope of protection of the present invention.

Claims

1. A method for constructing a satellite signal quality prediction model, characterized in that, include: Acquire sky images and corresponding historical satellite signal-to-noise ratio data; Feature extraction is performed on the sky image to obtain sky image features, and the historical signal-to-noise ratio data of the satellite is normalized to obtain numerical data features; The image feature data and the numerical data features are concatenated to obtain a comprehensive feature vector; The original quality prediction model is trained and optimized based on the comprehensive feature vector and the historical signal-to-noise ratio data of the satellite to obtain a quality prediction model. The quality prediction model is used to predict the satellite signal quality and includes a random convolution kernel transformation module and a ridge regression module.

2. The method for constructing a satellite signal quality prediction model according to claim 1, characterized in that, The step of extracting features from the sky image to obtain sky image features includes: The sky image is input into the sky image feature extraction model to obtain the sky image features. The sky image feature extraction model is based on an adversarial autocoding system.

3. The method for constructing a satellite signal quality prediction model according to claim 2, characterized in that, The method for constructing the sky image feature extraction model includes: Obtain historical sky images and corresponding historical feature images. The original sky image feature extraction model is trained and optimized based on the historical sky image and the historical feature image to obtain the sky image feature extraction model.

4. The method for constructing a satellite signal quality prediction model according to claim 3, characterized in that, The adversarial autocoding system includes an encoder, a decoder, and a discriminator; the step of training and optimizing the original sky image feature extraction model based on the historical sky image and the historical feature image to obtain the sky image feature extraction model includes: The historical sky image is input into the encoder to obtain a temporary feature vector; The temporary feature vector is input into the discriminator to obtain the target feature vector; The temporary feature vector and the target feature vector are input into the decoder to obtain a temporary sky feature image. The original sky image feature extraction model is optimized based on the temporary sky feature image and the historical feature image to obtain the sky image feature extraction model.

5. The method for constructing a satellite signal quality prediction model according to claim 2, characterized in that, The process of training and optimizing the original quality prediction model based on the comprehensive feature vector and the historical signal-to-noise ratio data of the satellite to obtain the quality prediction model includes: The comprehensive feature vector is input into the random convolution kernel conversion module to obtain intermediate feature data; The intermediate feature data is input into the ridge regression module to obtain temporary signal quality data; The original quality prediction model is optimized based on the temporary signal quality data and the satellite's historical signal-to-noise ratio data to obtain the quality prediction model.

6. The method for constructing a satellite signal quality prediction model according to claim 5, characterized in that, The step of optimizing the original quality prediction model based on the temporary signal quality data and the satellite's historical signal-to-noise ratio data to obtain the quality prediction model includes: Based on the temporary signal quality data and the satellite historical signal-to-noise ratio data, the loss of the original quality prediction model is calculated to obtain the loss function output. The model parameters of the original quality prediction model are adjusted according to the output of the loss function until the input of the loss function meets the preset conditions. The original quality prediction model after parameter adjustment is then used as the quality prediction model.

7. The method for constructing a satellite signal quality prediction model according to claim 1, characterized in that, The historical signal-to-noise ratio (SNR) data of the satellite includes multiple sub-SNR data. The normalization process of the historical SNR data of the satellite to obtain numerical data characteristics includes: Based on all the aforementioned sub-signal-to-noise ratio data, obtain the maximum and minimum values ​​of the satellite's historical signal-to-noise ratio data; Based on the maximum and minimum values, the historical signal-to-noise ratio data of the satellite is processed using Equation 1 to obtain the numerical data characteristics; Wherein, Equation 1 is: ； in, The numerical data feature corresponding to the Nth sub-signal-to-noise ratio data. For the Nth sub-signal-to-noise ratio data, This refers to the maximum value of the historical signal-to-noise ratio data of the satellite. This is the minimum value of the historical signal-to-noise ratio data of the satellite.

8. A satellite signal quality prediction model construction device, characterized in that, include: The acquisition unit is used to acquire sky images and corresponding historical signal-to-noise ratio data of satellites; The processing unit is used to extract features from the sky image to obtain sky image features, and to normalize the historical signal-to-noise ratio data of the satellite to obtain numerical data features. The processing unit is also used to concatenate the image feature data with the numerical data features to obtain a comprehensive feature vector; The processing unit is also used to train and optimize the original quality prediction model based on the comprehensive feature vector and the satellite historical signal-to-noise ratio data to obtain a quality prediction model. The quality prediction model is used to predict satellite signal quality, and the quality prediction model includes a random convolution kernel transformation module and a ridge regression module.

9. A method for predicting satellite signal quality, characterized in that, include: Acquire target sky images and corresponding satellite signal-to-noise ratio data; Feature extraction is performed on the target sky image to obtain target image features, and the satellite signal-to-noise ratio data is normalized to obtain target numerical data features; The target image features are concatenated with the target numerical data features to obtain the target comprehensive feature vector; The target integrated feature vector is input into the quality prediction model obtained by the satellite signal quality prediction model construction method as described in any one of claims 1 to 7 to obtain the satellite signal quality prediction result.

10. A satellite signal quality prediction method and apparatus, characterized in that, include: The acquisition module is used to acquire target sky images and corresponding satellite signal-to-noise ratio data; The processing module is used to extract features from the target sky image to obtain target image features, and to normalize the satellite signal-to-noise ratio data to obtain target numerical data features. The processing module is also used to concatenate the target image features with the target numerical data features to obtain a target comprehensive feature vector; The processing module is further configured to input the target integrated feature vector into the quality prediction model obtained by the satellite signal quality prediction model construction method as described in any one of claims 1 to 7, and obtain the satellite signal quality prediction result.

Images