CN-122001453-A - Multi-mode satellite communication link intelligent selection method and system based on deep learning

Abstract

The embodiment of the invention provides a multi-mode satellite communication link intelligent selection method and system based on deep learning, which relate to the technical field of deep learning, wherein an in-orbit satellite communication node receives a link detection pulse sequence which is sent by a ground gateway station and contains different modulation and coding strategy identifiers, analyzes the link detection pulse sequence to obtain an atmospheric attenuation disturbance parameter and an ionosphere scintillation effect parameter, inputs the atmospheric attenuation disturbance parameter and the ionosphere scintillation effect parameter into a deep learning feature extraction network to generate a multi-dimensional link state feature tensor, combines a service type priority code, distributes candidate link weight coefficients through a attention mechanism, constructs a sequencing list, sequentially selects a main communication link and a standby communication link according to the weight coefficients, and generates a link switching instruction. The invention improves the accuracy and reliability of satellite communication link selection and ensures the communication quality.

Inventors

• PENG WENBIN
• HUANG JINGYING
• LIU JIANRUI

Assignees

• 邦盛时空(广西)科技股份公司

Dates

Publication Date

20260508

Application Date

20260326

Claims (10)

1. 1. An intelligent selection method of multimode satellite communication links based on deep learning is characterized by comprising the following steps: A plurality of satellite communication nodes running in orbit respectively receive a link detection pulse sequence sent by a ground gateway station, wherein the link detection pulse sequence comprises detection pulse units carrying different modulation and coding strategy identifiers; performing impulse response characteristic analysis processing on the link detection pulse sequence to obtain an atmospheric attenuation disturbance parameter and an ionosphere scintillation effect parameter which are experienced by each detection pulse unit in a propagation path; inputting the atmospheric attenuation disturbance parameter and the ionosphere scintillation effect parameter into a deep learning feature extraction network which is pre-deployed on a satellite communication node, and generating a multi-dimensional link state feature tensor, wherein the multi-dimensional link state feature tensor comprises a frequency domain fading feature component and a time domain phase deviation feature component; According to the multidimensional link state characteristic tensor, combining the service type priority codes synchronously transmitted by the ground gateway station, distributing the weight coefficient of each candidate link through an attention mechanism, and constructing a candidate link sequencing list containing the weight coefficient; And selecting a candidate link with the highest weight coefficient as a main communication link based on the weight coefficient sequence of the candidate link ordered list, marking the candidate link with the highest weight coefficient as a standby communication link, and generating a link switching instruction containing the main communication link identifier and the standby communication link identifier.
2. 2. The intelligent selection method of a multimode satellite communication link based on deep learning according to claim 1, wherein the step of performing impulse response feature analysis processing on the link detection pulse sequence to obtain an atmospheric attenuation disturbance parameter and an ionospheric scintillation effect parameter undergone by each detection pulse unit in a propagation path specifically comprises the following steps: Receiving a link detection pulse sequence sent by a ground gateway station according to a preset pulse emission time interval, wherein the link detection pulse sequence comprises a first detection pulse unit, a second detection pulse unit and a third detection pulse unit which are arranged continuously, each detection pulse unit carries the same modulation and coding strategy identifier, and each detection pulse unit carries a node identifier of a satellite communication node when being received at the satellite communication node side; extracting signal amplitude attenuation and signal phase rotation of the first detection pulse unit in a propagation path, and taking the signal amplitude attenuation and the signal phase rotation as an original pulse response parameter set corresponding to the first detection pulse unit; performing frequency domain transformation on the original impulse response parameter set to generate a frequency domain response spectral line distribution diagram corresponding to the first detection impulse unit, wherein the frequency domain response spectral line distribution diagram comprises amplitude attenuation spectral values and phase offset spectral values corresponding to a plurality of frequency points; Performing curve fitting treatment on amplitude attenuation spectrum values in the frequency domain response spectrum line distribution diagram to obtain an atmospheric absorption attenuation curve which changes along with frequency, and extracting attenuation values at a plurality of preset characteristic frequency points from the atmospheric absorption attenuation curve to form an atmospheric attenuation disturbance parameter vector consisting of a plurality of attenuation values; Performing differential operation on phase shift spectrum values in the frequency domain response spectrum line distribution diagram to obtain a phase difference sequence between adjacent frequency points, calculating a phase scintillation index caused by an ionosphere according to the fluctuation amplitude of the phase difference sequence, and taking the phase scintillation index as a scalar ionosphere scintillation effect parameter; Repeating the processing procedure of the first detection pulse unit, and sequentially executing impulse response characteristic analysis processing on the second detection pulse unit and the third detection pulse unit to respectively obtain an atmospheric attenuation disturbance parameter and an ionospheric scintillation effect parameter corresponding to the second detection pulse unit and an atmospheric attenuation disturbance parameter and an ionospheric scintillation effect parameter corresponding to the third detection pulse unit; Performing element weighted average processing on the atmospheric attenuation disturbance parameter vector corresponding to the first detection pulse unit and the atmospheric attenuation disturbance parameter vector corresponding to the second detection pulse unit to generate a first weighted average atmospheric attenuation disturbance parameter vector, and performing element weighted average processing on the first weighted average atmospheric attenuation disturbance parameter vector and the atmospheric attenuation disturbance parameter vector corresponding to the third detection pulse unit to generate a final atmospheric attenuation disturbance parameter vector, wherein the final atmospheric attenuation disturbance parameter vector is used for representing the comprehensive degree of atmospheric attenuation influence of a plurality of detection pulse units in a propagation path; And carrying out mean value processing on the scalar ionospheric scintillation effect parameter corresponding to the first detection pulse unit and the scalar ionospheric scintillation effect parameter corresponding to the second detection pulse unit to generate a first mean value ionospheric scintillation effect parameter, and carrying out mean value processing on the first mean value ionospheric scintillation effect parameter and the scalar ionospheric scintillation effect parameter corresponding to the third detection pulse unit to generate a final ionospheric scintillation effect parameter, wherein the final ionospheric scintillation effect parameter is used for representing the comprehensive degree of ionospheric scintillation influence of a plurality of detection pulse units in a propagation path.
3. 3. The intelligent selection method of a deep learning-based multimode satellite communication link according to claim 1, wherein the inputting the atmospheric attenuation disturbance parameter and the ionospheric scintillation effect parameter into a deep learning feature extraction network pre-deployed on a satellite communication node generates a multidimensional link state feature tensor, the multidimensional link state feature tensor including a frequency domain fading feature component and a time domain phase shift feature component, specifically comprising the steps of: Inputting the atmospheric attenuation disturbance parameter vector and the ionospheric scintillation effect parameter into an input feature splicing layer of the deep learning feature extraction network, and sequentially splicing the atmospheric attenuation disturbance parameter vector and the ionospheric scintillation effect parameter in the input feature splicing layer according to feature dimensions to generate an initial spliced feature vector; Transmitting the initial spliced feature vector to a first convolution feature extraction module of the deep learning feature extraction network, wherein the first convolution feature extraction module comprises a plurality of one-dimensional convolution kernels which are arranged in parallel, each one-dimensional convolution kernel has different convolution kernel lengths, and the initial spliced feature vector is subjected to parallel convolution operation through the plurality of one-dimensional convolution kernels which are arranged in parallel to generate a first local feature map, a second local feature map and a third local feature map which correspond to different convolution kernel lengths respectively; Inputting the first local feature map, the second local feature map and the third local feature map into a feature map fusion sub-layer of the first convolution feature extraction module, and performing element-by-element addition fusion operation on the first local feature map, the second local feature map and the third local feature map in the feature map fusion sub-layer to generate a fusion local feature map; Transmitting the fused local feature map to a second convolution feature extraction module of the deep learning feature extraction network, wherein the second convolution feature extraction module comprises a first depth separable convolution layer and a second depth separable convolution layer, performing space convolution operation of depth dimension on the fused local feature map through the first depth separable convolution layer to generate a first depth convolution feature map, and performing point-by-point convolution operation on the first depth convolution feature map through the second depth separable convolution layer to generate a second depth convolution feature map; Transmitting the second depth convolution feature map to a feature recalibration module of the depth learning feature extraction network, wherein the feature recalibration module comprises a global average pooling layer, a first full-connection layer and a second full-connection layer, global average pooling processing of space dimensions is carried out on the second depth convolution feature map through the global average pooling layer, a channel description vector is generated, dimension reduction processing is carried out on the channel description vector through the first full-connection layer, a dimension reduction channel vector is generated, dimension increase processing is carried out on the dimension reduction channel vector through the second full-connection layer, and a channel weight vector is generated; carrying out channel-by-channel multiplication operation on the channel weight vector and the second depth convolution feature map to generate a recalibration feature map subjected to channel attention weighting; Transmitting the recalibration feature map to a feature separation output layer of the deep learning feature extraction network, performing channel dimension segmentation operation on the recalibration feature map in the feature separation output layer, segmenting the recalibration feature map into a first feature subset and a second feature subset according to a preset channel allocation proportion, taking the first feature subset as a frequency domain fading feature component and taking the second feature subset as a time domain phase deviation feature component; And carrying out dimension alignment processing on the frequency domain fading characteristic components and the time domain phase deviation characteristic components, ensuring that the characteristic dimension quantity of the frequency domain fading characteristic components is consistent with the characteristic dimension quantity of the time domain phase deviation characteristic components, and combining the frequency domain fading characteristic components and the time domain phase deviation characteristic components after dimension alignment together to form the multidimensional link state characteristic tensor.
4. 4. The intelligent selection method of multimode satellite communication links based on deep learning according to claim 1, wherein the step of constructing a candidate link ordered list containing weight coefficients by allocating the weight coefficients of each candidate link through an attention mechanism according to the multidimensional link state characteristic tensor and combining the service type priority codes synchronously transmitted by the ground gateway station specifically comprises the following steps: Receiving a service type priority code synchronously transmitted by a ground gateway station through a special signaling channel, wherein the service type priority code comprises a first priority value corresponding to a real-time voice service, a second priority value corresponding to a broadband data service and a third priority value corresponding to an instruction control service; The multi-dimensional link state characteristic tensor is input into a query vector generation sub-module of the attention mechanism module, and linear transformation processing is carried out on the multi-dimensional link state characteristic tensor through the query vector generation sub-module to generate a link state query vector; Inputting the service type priority code into a key value vector generation sub-module of the attention mechanism module, and performing embedded representation learning processing on the service type priority code through the key value vector generation sub-module to generate a service priority key value vector; Inputting the link state query vector and the service priority key value vector into an attention score calculation unit of the attention mechanism module, and calculating the dot product similarity of the link state query vector and the service priority key value vector in the attention score calculation unit to obtain an initial attention score; Inputting the initial attention score into a score normalization processing unit of the attention mechanism module, scaling the initial attention score through the score normalization processing unit to generate a scaled attention score, and performing probability distribution transformation on the scaled attention score to generate normalized attention weight distribution; Acquiring a link identifier list of all candidate links which can be detected by a satellite communication node, wherein the link identifier list comprises a first link identifier corresponding to a first candidate link, a second link identifier corresponding to a second candidate link and a third link identifier corresponding to a third candidate link; Respectively distributing the normalized attention weight distribution to the first candidate link, the second candidate link and the third candidate link according to the sequence of the link identifier list to obtain a first weight coefficient corresponding to the first candidate link, a second weight coefficient corresponding to the second candidate link and a third weight coefficient corresponding to the third candidate link; Respectively carrying out association binding on the first weight coefficient, the second weight coefficient and the third weight coefficient and the corresponding first link identifier, second link identifier and third link identifier to generate an initial candidate link association list comprising a first association item, a second association item and a third association item; And comparing the values of the first weight coefficient, the second weight coefficient and the third weight coefficient in the initial candidate link association list, and rearranging the first association item, the second association item and the third association item according to the comparison result from high to low according to the weight coefficient to generate a candidate link sorting list containing the weight coefficient.
5. 5. The intelligent selection method for multimode satellite communication links based on deep learning according to claim 1, wherein the selecting the candidate link with the highest weight coefficient as the primary communication link based on the weight coefficient sequence of the ordered list of candidate links, and marking the candidate link with the highest weight coefficient as the backup communication link, and generating the link switching instruction including the primary communication link identifier and the backup communication link identifier, specifically comprises the following steps: analyzing the candidate link sorting list, extracting a first association item corresponding to the highest weight coefficient positioned at the first position from the arrangement sequence of the candidate link sorting list, and reading a first link identifier associated and bound with the highest weight coefficient from the first association item; Determining the first candidate link identified by the read first link identifier as a main communication link, writing the first link identifier into a main link identifier storage area, and generating a main link identifier to be confirmed; Extracting a second association entry corresponding to a next-highest weight coefficient positioned at a next-highest level from the candidate link sorting list, and reading a second link identifier associated and bound with the next-highest weight coefficient from the second association entry; Determining the second candidate link identified by the read second link identifier as a standby communication link, writing the second link identifier into a standby link identifier storage area, and generating a standby link identifier to be confirmed; inputting the main link identifier to be confirmed and the standby link identifier to be confirmed into a link state confirmation module, respectively sending short-time detection signaling to the first candidate link and the second candidate link through the link state confirmation module, and receiving first link state response information returned by the first candidate link and second link state response information returned by the second candidate link; Analyzing the first link state response information, extracting a current available bandwidth parameter and a current bit error rate parameter of the first candidate link, comparing the current available bandwidth parameter with a preset available bandwidth threshold value, comparing the current bit error rate parameter with a preset bit error rate threshold value, and if the current available bandwidth parameter is greater than or equal to the preset available bandwidth threshold value and the current bit error rate parameter is less than or equal to the preset bit error rate threshold value, confirming that the to-be-confirmed active link identifier is valid; Analyzing the second link state response information, extracting the current available bandwidth parameter and the current bit error rate parameter of the second candidate link, comparing the current available bandwidth parameter of the second candidate link with a preset available bandwidth threshold value, comparing the current bit error rate parameter of the second candidate link with the preset bit error rate threshold value, and if the current available bandwidth parameter of the second candidate link is greater than or equal to the preset available bandwidth threshold value and the current bit error rate parameter of the second candidate link is less than or equal to the preset bit error rate threshold value, confirming that the standby link identifier to be confirmed is valid; Taking the valid main link identifier to be confirmed as a formal main link identifier, taking the valid standby link identifier to be confirmed as a formal standby link identifier, and carrying out combined encapsulation on the formal main link identifier and the formal standby link identifier to generate a link switching instruction; And sending the generated link switching instruction to a radio frequency front end switching unit through a control channel of a satellite communication node, triggering the radio frequency front end switching unit to switch a current communication link to a main communication link identified by the formal main link identifier, and configuring a standby communication link identified by the formal standby link identifier into a hot standby state.
6. 6. The deep learning based multimode satellite communication link intelligent selection method of claim 1, further comprising a link quality dynamic monitoring step performed after generating a link switch instruction comprising a primary communication link identifier and a backup communication link identifier, in particular comprising the steps of: after switching the current communication link to the main communication link, continuously receiving a service data stream sent by a ground gateway station through the main communication link, wherein the service data stream comprises continuously arranged data frame units; Carrying out real-time demodulation and decoding processing on the data frame units in the service data stream, and extracting frame synchronization header information and a forward error correction coding verification result carried by each data frame unit; Calculating time interval jitter parameters between adjacent data frame units according to the frame synchronization header information, and taking the time interval jitter parameters as delay jitter characteristic values of a main communication link; counting the number of data frame units with decoding errors in a preset counting time window according to the forward error correction coding checking result, calculating the ratio of the number of the data frame units with decoding errors to the number of total data frame units received in the counting time window, and generating real-time frame error rate parameters of a main communication link; Inputting the delay jitter characteristic value and the real-time frame error rate parameter of the main communication link into a link quality evaluation model pre-deployed on a satellite communication node, and carrying out comprehensive mapping processing on the delay jitter characteristic value and the real-time frame error rate parameter through the link quality evaluation model to generate a current quality grading value of the main communication link; comparing the current quality grading value with a pre-stored primary communication link initial quality grading value, and calculating the descending amplitude percentage of the current quality grading value relative to the primary communication link initial quality grading value; And if the descending amplitude percentage exceeds a preset descending amplitude threshold value, triggering a standby link activation flow, extracting a standby communication link identifier from a link switching instruction, sending the link switching activation instruction containing the standby communication link identifier to a radio frequency front-end switching unit through a control channel of a satellite communication node, and triggering the radio frequency front-end switching unit to switch the current communication link from a main communication link to a standby communication link identified by the standby communication link identifier.
7. 7. The intelligent selection method of multimode satellite communication link based on deep learning according to claim 6, wherein the step of inputting the delay jitter feature value and the real-time frame error rate parameter of the active communication link into a link quality evaluation model pre-deployed on a satellite communication node, and performing comprehensive mapping processing on the delay jitter feature value and the real-time frame error rate parameter through the link quality evaluation model to generate the current quality score value of the active communication link specifically comprises the following steps: Performing characteristic dimension alignment operation on the delay jitter characteristic value and the real-time frame error rate parameter of the main communication link, and respectively converting the delay jitter characteristic value and the real-time frame error rate parameter into a first input characteristic vector and a second input characteristic vector with the same characteristic dimension quantity; respectively carrying out normalization processing on the first input feature vector and the second input feature vector, mapping feature values in the first input feature vector and the second input feature vector to the same numerical range, and generating a normalized first input feature vector and a normalized second input feature vector; Inputting the normalized first input feature vector and the normalized second input feature vector into an input feature merging layer of the link quality evaluation model, and performing vector splicing operation on the normalized first input feature vector and the normalized second input feature vector in the input feature merging layer to generate a merged input feature vector; Transmitting the combined input feature vector to a first full-connection hidden layer of the link quality assessment model, wherein the first full-connection hidden layer comprises a first number of neuron nodes, and performing nonlinear transformation processing on the combined input feature vector through the first full-connection hidden layer to generate a first hidden layer output vector; Transmitting the first hidden layer output vector to a second full-connection hidden layer of the link quality evaluation model, wherein the second full-connection hidden layer comprises a second number of neuron nodes, the second number is smaller than the first number, and performing dimension reduction nonlinear transformation on the first hidden layer output vector through the second full-connection hidden layer to generate a second hidden layer output vector; transmitting the second hidden layer output vector to a feature activation function layer of the link quality evaluation model, and performing activation function mapping processing on the second hidden layer output vector through the feature activation function layer to generate an activation feature vector containing a plurality of activation values within a preset range; Transmitting the activation feature vector to an output layer neuron of the link quality evaluation model, and performing weighted summation operation on the activation feature vector by the output layer neuron to generate an original quality scoring value; inputting the original quality grading value into a grading correction module of the link quality evaluation model, and performing linear scaling processing on the original quality grading value according to a preset grading correction coefficient by the grading correction module to generate a corrected current quality grading value; And taking the corrected current quality score value as the final output of the link quality evaluation model, wherein the corrected current quality score value is used for quantitatively representing the comprehensive transmission quality level of the main communication link at the current moment.
8. 8. The deep learning based multimode satellite communication link intelligent selection method of claim 1, further comprising a candidate link dynamic update step performed prior to generating a link switch instruction comprising a primary communication link identifier and a backup communication link identifier, comprising the steps of: Continuously scanning satellite beacon signals in a surrounding space range through a wide area receiving antenna of a satellite communication node, and receiving a plurality of beacon frames broadcast and transmitted by an in-orbit satellite, wherein each beacon frame carries an orbit position parameter and a beam coverage parameter of a satellite transmitting the beacon frame; Analyzing each received beacon frame, extracting an orbit position parameter and a beam coverage parameter of a satellite transmitting the beacon frame from the beacon frame, and calculating a visible angle range and a signal arrival angle of the satellite communication node relative to the satellite transmitting the beacon frame according to the orbit position parameter and the beam coverage parameter; Comparing the calculated visible angle range and the signal arrival angle of the satellite communication node relative to each satellite transmitting the beacon frame with a preset minimum visible angle threshold value and a preset maximum signal arrival angle threshold value respectively, screening out satellites meeting the condition that the visible angle range is greater than or equal to the preset minimum visible angle threshold value and the signal arrival angle is less than or equal to the preset maximum signal arrival angle threshold value, and taking the satellites as candidate satellite sets; Acquiring a communication frequency band list and a polarization mode list supported by each candidate satellite in the candidate satellite set, and further screening satellites with communication frequency bands and polarization modes matched with the satellite communication nodes from the candidate satellite set according to the communication frequency bands and the polarization modes supported by the satellite communication nodes as available satellite sets; Predicting the ephemeris track of each available satellite in a preset time period in the future according to the orbit position parameters of each available satellite in the available satellite set, and generating satellite motion track prediction data containing an ephemeris track coordinate sequence; Comparing the satellite motion trail prediction data with ephemeris trail of satellites corresponding to candidate links in a candidate link sorting list which is currently in use, identifying a satellite to be invalidated which is about to exceed the visible range of the satellite communication node, and removing the candidate link corresponding to the satellite to be invalidated from the candidate link sorting list; And taking the satellite newly added into the available satellite set as a newly added candidate satellite, carrying out impulse response characteristic analysis processing on a link detection pulse sequence sent by the newly added candidate satellite to obtain an atmospheric attenuation disturbance parameter and an ionospheric scintillation effect parameter corresponding to the newly added candidate satellite, generating a multidimensional link state characteristic tensor corresponding to the newly added candidate satellite through a deep learning characteristic extraction network, and supplementing the multidimensional link state characteristic tensor corresponding to the newly added candidate satellite into the construction process of a candidate link sequencing list.
9. 9. The multi-mode satellite communication link intelligent selection system based on deep learning is characterized by comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are in communication with each other through the communication bus, the memory is used for storing a computer program, and the processor is used for realizing the multi-mode satellite communication link intelligent selection method based on deep learning according to any one of claims 1-8 when executing the computer program.
10. 10. A computer readable storage medium having stored thereon a computer program, wherein the computer program when executed by a processor implements the deep learning based multimode satellite communication link intelligent selection method steps of any one of claims 1-8.

Description

Multi-mode satellite communication link intelligent selection method and system based on deep learning Technical Field The invention relates to the technical field of deep learning, in particular to a multimode satellite communication link intelligent selection method and system based on deep learning. Background In the field of satellite communications, with increasing diversification and complexity of service requirements, how to efficiently and reliably select an appropriate communication link becomes a key issue. Currently, there are a number of in-orbit satellite communication nodes in a satellite communication network that require frequent data transmissions with ground gateway stations. However, the state of the satellite communication link is affected by a number of factors, such as atmospheric attenuation and ionospheric scintillation effects. The energy of the signal is gradually weakened in the propagation process by the atmospheric attenuation, so that the signal quality is reduced, and the phase and amplitude of the signal are rapidly changed by the ionosphere scintillation effect, so that the stability of communication is affected. Most of the existing satellite communication link selection methods are based on simple signal intensity measurement or preset fixed rules, and the influence of dynamic factors such as atmospheric attenuation and ionosphere scintillation effect on the link state is not fully considered in the method, so that the actual quality of the link cannot be accurately estimated. Moreover, when the requirements of different service types are met, flexible link selection cannot be performed according to the priority of the service, so that the communication quality of important service cannot be ensured easily, and the overall performance and reliability of the satellite communication system are affected. Disclosure of Invention Accordingly, an objective of the embodiments of the present invention is to provide a method and a system for intelligently selecting multimode satellite communication links based on deep learning. According to an aspect of the embodiment of the present invention, there is provided a multimode satellite communication link intelligent selection method based on deep learning, the method including: A plurality of satellite communication nodes running in orbit respectively receive a link detection pulse sequence sent by a ground gateway station, wherein the link detection pulse sequence comprises detection pulse units carrying different modulation and coding strategy identifiers; performing impulse response characteristic analysis processing on the link detection pulse sequence to obtain an atmospheric attenuation disturbance parameter and an ionosphere scintillation effect parameter which are experienced by each detection pulse unit in a propagation path; inputting the atmospheric attenuation disturbance parameter and the ionosphere scintillation effect parameter into a deep learning feature extraction network which is pre-deployed on a satellite communication node, and generating a multi-dimensional link state feature tensor, wherein the multi-dimensional link state feature tensor comprises a frequency domain fading feature component and a time domain phase deviation feature component; According to the multidimensional link state characteristic tensor, combining the service type priority codes synchronously transmitted by the ground gateway station, distributing the weight coefficient of each candidate link through an attention mechanism, and constructing a candidate link sequencing list containing the weight coefficient; And selecting a candidate link with the highest weight coefficient as a main communication link based on the weight coefficient sequence of the candidate link ordered list, marking the candidate link with the highest weight coefficient as a standby communication link, and generating a link switching instruction containing the main communication link identifier and the standby communication link identifier. According to another aspect of the embodiment of the invention, a multimode satellite communication link intelligent selection system based on deep learning is provided, which comprises a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory are in communication with each other through the communication bus, the memory is used for storing a computer program, and the processor is used for realizing any one of the steps of the multimode satellite communication link intelligent selection method based on deep learning when executing the computer program. According to another aspect of the embodiments of the present invention, there is provided a readable storage medium having stored thereon a computer program which, when executed by a processor, can perform the steps of the above-described deep learning based multimode satellite communication link intellig