Phased array antenna multi-beam automatic optimization method, device, equipment and storage medium

Abstract

The present invention discloses a method, device, equipment and storage medium for automatic optimization of multiple beams of phased array antennas, the method comprising: constructing a beam optimization model based on sparse coding and deep neural network; optimizing the constructed beam optimization model by using reinforcement learning strategy, so that the beam optimization model can autonomously adjust the beam control strategy according to the feedback of the actual space environment and working state; establishing a multi-task learning framework, so that the beam optimization model can simultaneously learn the mutual influence and coupling effect between multiple beams to achieve a global optimal solution; deploying the trained beam optimization model to a satellite-borne satellite, so that the trained beam optimization model outputs the optimal beam control code according to the working parameters of the current phased array antenna of the satellite-borne satellite, so that the satellite-borne satellite controls the operation of the phased array antenna based on the optimal beam control code. The present invention can adaptively optimize the multiple beams of the phased array antenna in real time.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a method, device, equipment and storage medium for automatic multi-beam optimization of a phased array antenna. Background Art

[0002] Phased array antennas have many working modes, many beams, and high pointing accuracy requirements. In addition, they need to store beam control codes after ground testing to be input into the beam control system. The beam control codes contain feed component control information that controls beam scanning, such as the feed amplitude attenuation control amount of the transmitting and receiving components, the feed phase shift control amount, and the delay control amount of the delay amplifier component. In order to complete the multi-beam calibration of phased array antennas and obtain the beam control codes of each beam, a suitable optimization method needs to be adopted. Therefore, how to effectively develop a fully automated optimization method for the multi-beam system of phased array antennas to ensure that the beam control codes of each beam are accurately determined and applied in various complex space environments, thereby achieving fast and accurate beam pointing and switching, and meeting the requirements of space communication and detection missions for high pointing accuracy and multi-target tracking has become an increasingly urgent problem to be solved. Summary of the invention

[0003] The embodiments of the present invention provide a phased array antenna multi-beam automatic optimization method, device, equipment and storage medium, which can adaptively optimize the phased array antenna multi-beam in real time, realize fast and accurate beam pointing and switching, meet the requirements of space communication and detection tasks for high pointing accuracy and multi-target tracking, and ensure that it always performs at its best in a complex and changeable space environment.

[0004] An embodiment of the present invention provides a phased array antenna multi-beam automatic optimization method, comprising:

[0005] A beam optimization model based on sparse coding and deep neural network is constructed; the beam optimization model is used to predict the optimal feeding parameter combination under different beam pointing directions, the input layer of the beam optimization model is used to receive feature vectors, and the output layer of the beam optimization model is used to predict the corresponding beam control code value; the beam optimization model is trained based on the working data samples of the phased array antenna obtained in advance;

[0006] The constructed beam optimization model is optimized by using a reinforcement learning strategy, so that the beam optimization model can autonomously adjust the beam control strategy according to feedback from the actual space environment and working status;

[0007] Establishing a multi-task learning framework to allow the beam optimization model to simultaneously learn the mutual influence and coupling effects between multiple beams to achieve a global optimal solution;

[0008] The trained beam optimization model is deployed to the onboard satellite so that the trained beam optimization model outputs the optimal beam control code according to the current operating parameters of the onboard satellite's phased array antenna, thereby enabling the onboard satellite to control the operation of the phased array antenna based on the optimal beam control code.

[0009] As an improvement of the above solution, the construction of a beam optimization model based on sparse coding and deep neural network includes:

[0010] Initialize the beam optimization model based on sparse coding and deep neural network; the model architecture of the initialized beam optimization model includes input layer, hidden layer and output layer; the input layer is used to receive feature vectors; the first layer of the hidden layer is a sparse coding layer, the coding dimension is set to 1 / 4 of the length of the original feature vector, the activation function uses a hard threshold function to maintain the sparsity of the input, and the subsequent hidden layers are 3 fully connected layers, the number of nodes in each layer is 512, 256, and 128 respectively, and the activation function is ReLU; the output layer is used to predict the beam control code value under the corresponding beam pointing; the optimizer of the initialized beam optimization model uses the Adam optimizer, and the learning rate is set to 0.001; the loss function is set to:

[0011] N beams is the total number of beams, is the weight of each control code loss item;

[0012] The acquired working data samples of the phased array antenna are used for feature extraction; the working data samples include: beam pointing scanning angle and feeding parameters, and the feeding parameters include: injection signal power, transmission power, feeding amplitude Ai and feeding phase

[0013] Apply deep learning feature extraction technology to extract features from the working data samples to obtain corresponding feature vectors: key features include: electrical characteristic curves of the transmitting and receiving components, beam shape parameters and spatial distribution relationships;

[0014] Pair the feature vectors with the actual measured beam steering code values ​​and divide them into a training set and a validation set;

[0015] Divide the training set into multiple small batches and perform mini-batch gradient descent. In each iteration, randomly extract a batch of data from the training set and feed it into the model for forward propagation.

[0016] Calculate the prediction control code output by the model on the current batch training set and calculate the loss function value;

[0017] According to the gradient of the loss function, the gradient of the model parameters is calculated through the back propagation algorithm;

[0018] Use the Adam optimizer to update model parameters based on the gradient;

[0019] After each predetermined number of iterations, the validation set is used to evaluate the model performance; if the validation set performance no longer improves or reaches a preset stopping condition, the training of the beam optimization model is terminated.

[0020] As an improvement of the above solution, the deep learning feature extraction technology is applied to extract features from the working data sample to obtain a corresponding feature vector, including:

[0021] For each transmitting and receiving component, the interpolation method is used to fit the feed amplitude attenuation curve Ai(θ, φ) and the feed phase shift curve φi(θ, φ) under different beam pointing directions, and the dynamic change rate of the two over time is calculated to obtain the electrical characteristic change curve;

[0022] For each beam pointing scan angle, the beam gain G is calculated based on the transmitted power j , thus obtaining the beam shape parameters:

[0023] P transmitj is the transmitted power, P injected is the injected signal power;

[0024] By using three-dimensional drawing software, a beam coverage map is drawn according to the beam pointing data to clarify the overlapping area of ​​adjacent beams, thereby obtaining the spatial distribution relationship.

[0025] As an improvement of the above solution, the constructed beam optimization model is optimized by using a reinforcement learning strategy, so that the beam optimization model can autonomously adjust the beam control strategy according to feedback from the actual space environment and working status, including:

[0026] Define a reinforcement learning environment, where the state of reinforcement learning is defined as follows: Among them, θ is the azimuth angle of the current beam, indicating the direction of the beam in the horizontal plane; φ is the pitch angle of the current beam, indicating the direction of the beam in the vertical plane; is the feeding amplitude of the i-th transmitting and receiving component, reflecting the signal strength of the antenna at this component; T is the ambient temperature, which affects the electrical properties of the antenna material and the stability of the transmitting and receiving components; N is the spatial noise level, which affects the communication quality and beam detection accuracy; the reinforcement learning action is defined in the following way: Where ΔA i is the feed amplitude adjustment amount of the i-th transmitting and receiving component, which is used to correct amplitude mismatch or optimize beam pointing; is the feed phase adjustment of the i-th transmitting and receiving component, which is used to adjust the beam pointing or improve the beam shape; Δt k is the delay control value of the kth delay amplifier component (Delaycontrol, which is used to accurately control the time difference between the signal reaching each sending and receiving component, and then adjust the beam pointing; the reward function of reinforcement learning is defined in the following way: R = f(G, HPBW, SLL, P reflect ,P transmit ,N total ) Where G is the beam gain, which reflects the ability of the antenna to concentrate energy into the main lobe; HPBW is the half-power beam width, which characterizes the degree of beam focusing; SLL is the sidelobe level, which measures the impact of the beam sidelobe on the main lobe; P reflect is the reflected power, which represents the energy scattered back by the antenna; P transmit is the transmitted power, i.e. the power effectively transmitted to the specified direction; N total is the total noise power, reflecting the system signal-to-noise ratio; the specific formula of the reward function is set as: Among them, w G , w HPBW , w SLL , is the weight coefficient used to balance the contribution of different performance indicators to the reward function;

[0027] Create a table Q-table whose size is the state space size (|S|) multiplied by the action space size (|A|), and all elements are initially set to 0; Q-table is used to store the expected cumulative reward for performing each action A in each state S;

[0028] According to the current state S, select an action A from the Q-table according to the ε-greedy strategy;

[0029] Execute the selected action on the phased array antenna, observe the new state S′ and obtain the immediate reward R;

[0030] Update the corresponding items in the Q-table according to the Q-learning update rule set below:

[0031] Among them, α is the learning rate, which controls the impact of new experience on the existing Q-value, α = 0.1; γ is the discount factor, which determines the importance of future rewards, γ = 0.9.

[0032] As an improvement to the above solution, a multi-task learning framework is established to allow the beam optimization model to simultaneously learn the mutual influence and coupling effects between multiple beams to achieve a global optimal solution, including:

[0033] In the model structure of the beam optimization model, a shared hidden layer is set for multiple beams, the shared hidden layer includes a sparse coding layer and the first two fully connected layers, and the shared hidden layer is used to learn a common beamforming rule;

[0034] Set up an independent output layer for each beam to predict the respective beam control code;

[0035] When predicting a new beam control code, the state of other beams is considered through the attention mechanism to achieve a global optimal solution; the attention mechanism implementation process includes:

[0036] For the control code prediction of the jth beam, an attention vector a is introduced j To weight and fuse the status information of other beams: Among them, N beams is the total number of beams, indicating the number of beams that can be independently controlled in the system; h i is the hidden layer output vector corresponding to the i-th beam, which contains the feature representation of the beam learned inside the model; w is a learnable attention weight vector, which is used to calculate the attention score of each beam hidden layer output vector. After normalization by the softmax function, it forms a weighted combination of other beam information; softmax(·) is a normalization function that converts the dot product result of the weight vector and the hidden layer output vector into a probability distribution to ensure that the weighted sum is 1;

[0037] The weighted attention vector a j and the hidden layer output vector h of the current beam j i Concatenate and pass it as a new input to the independent output layer to predict the control code of beam j: They are the predicted values ​​of the feed amplitude adjustment, feed phase adjustment and delay control of the jth beam respectively. OutputLayer represents the last layer of the neural network of the model, which is responsible for converting the concatenated vector into the control code prediction;

[0038] During training, the shared underlying feature layer and independent output layer parameters are updated simultaneously in the following way: Using a multi-task learning strategy, the control code prediction of multiple beams is treated as different subtasks to jointly optimize a loss function:

[0039] , where L is the total loss function, which is used to measure the performance of the model on all beam control code prediction tasks; MSE is the mean square error function, which is used to calculate the difference between the model prediction value and the true value; are respectively the feeding amplitude adjustment amount, feeding phase adjustment amount and delay control amount of the j-th beam predicted by the model; are respectively the actual feeding amplitude adjustment amount, feeding phase adjustment amount and delay control amount of the j-th beam.

[0040] As an improvement of the above solution, the trained beam optimization model is deployed to the onboard satellite so that the trained beam optimization model outputs an optimal beam control code according to the current operating parameters of the phased array antenna of the onboard satellite, including:

[0041] Obtain the current status information of beam j and other beam status information; the status information includes: beam pointing scanning angle and feeding parameters, and the feeding parameters include: injection signal power, transmission power, feeding amplitude Ai and feeding phase

[0042] The state information of all other beams is used as the input of the beam optimization model, and features are extracted through the shared underlying feature layer to obtain the hidden layer output vector h corresponding to each other beam. i (i≠j);

[0043] For each other beam, the hidden layer output vector h i , use the attention weight wi vector set during the training process to perform dot product and normalize it through the softmax function to get the attention score α i :α i =softmax(w T h i );

[0044] According to the attention score, the hidden layer output vectors of other beams are weighted and summed to generate the attention vector a of beam j j : N beams is the total number of beams;

[0045] The hidden layer output vector h corresponding to the current state of beam j i And the calculated attention vector a j Concatenate to form a new input vector:

[0046] The concatenated input vector The control code adjustment amount of beam j is obtained by sending it to the independent output layer OutputLayer of the beam optimization model and calculating it through the neural network:

[0047] According to the predicted control code adjustment amount, adjust the actual feeding amplitude A of beam j j and Phase A j ←A j +△A j

[0048]

[0049] According to the delay control amount Δt j adjusting the delay of the delay amplifier component associated with beam j;

[0050] Finally, the overall performance of all beams is optimized by the multi-task learning strategy pre-set during the training process of the trained beam optimization model.

[0051] Another embodiment of the present invention provides a phased array antenna multi-beam automatic optimization device, including:

[0052] A construction module is used to construct a beam optimization model based on sparse coding and deep neural network; the beam optimization model is used to predict the optimal feeding parameter combination under different beam pointing directions, the input layer of the beam optimization model is used to receive feature vectors, and the output layer of the beam optimization model is used to predict the corresponding beam control code value; the beam optimization model is trained based on the working data samples of the phased array antenna obtained in advance;

[0053] An optimization module, used to optimize the constructed beam optimization model by using a reinforcement learning strategy, so that the beam optimization model can autonomously adjust the beam control strategy according to feedback from the actual space environment and working status;

[0054] Establishing a module for establishing a multi-task learning framework so that the beam optimization model can simultaneously learn the mutual influence and coupling effects between multiple beams to achieve a global optimal solution;

[0055] The application module is used to deploy the trained beam optimization model to the onboard satellite, so that the trained beam optimization model outputs the optimal beam control code according to the current operating parameters of the phased array antenna of the onboard satellite, so that the onboard satellite controls the operation of the phased array antenna based on the optimal beam control code.

[0056] Another embodiment of the present invention provides a phased array antenna multi-beam automatic optimization system, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, the phased array antenna multi-beam automatic optimization method described in the above-mentioned embodiment of the invention is implemented.

[0057] Another embodiment of the present invention provides a storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the phased array antenna multi-beam automatic optimization device method described in the above-mentioned embodiment of the invention.

[0058] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0059] The embodiment of the present invention first constructs a beam optimization model based on sparse coding and deep neural network. The model receives a vector containing the working state characteristics of the phased array antenna as input, predicts the optimal feed parameter combination under different beam pointing, and outputs the corresponding beam control code value. The model is trained based on the pre-acquired phased array antenna working data samples to ensure that it has the ability to adapt to actual scenarios. Then, the constructed beam optimization model is optimized using a reinforcement learning strategy, so that the model can autonomously adjust the beam control strategy according to the feedback of the actual space environment and working state, and realize dynamic adaptability and self-learning ability. In addition, by establishing a multi-task learning framework, the beam optimization model simultaneously learns the mutual influence and coupling effects between multiple beams, ensuring that the model takes into account the global optimal solution of the overall system while optimizing the performance of a single beam. Finally, the trained beam optimization model is deployed to the onboard satellite, so that it outputs the optimal beam control code in real time according to the current working parameters of the phased array antenna, and guides the satellite to accurately control the phased array antenna based on this control code, so as to achieve fast and accurate beam pointing and switching. From the above analysis, it can be seen that the embodiments of the present invention can adaptively optimize the operation of the multi-beam phased array antenna in real time, ensure that the best performance is always maintained in a complex and changeable space environment, and effectively meet the requirements of space communication and detection tasks for high pointing accuracy and multi-target tracking. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a flow chart of a method for automatic multi-beam optimization of a phased array antenna provided by one embodiment of the present invention;

[0061] Figure 2 It is a flow chart of a phased array antenna multi-beam automatic optimization device provided by one embodiment of the present invention;

[0062] Figure 3 It is a flow chart of a phased array antenna multi-beam automatic optimization system provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] See also Figure 1, is a flow chart of a phased array antenna multi-beam automatic optimization method provided by an embodiment of the present invention. The phased array antenna multi-beam automatic optimization method comprises steps S10 to S13:

[0065] S10, constructing a beam optimization model based on sparse coding and deep neural network; the beam optimization model is used to predict the optimal feeding parameter combination under different beam pointing directions, the input layer of the beam optimization model is used to receive feature vectors, and the output layer of the beam optimization model is used to predict the corresponding beam control code value; the beam optimization model is trained based on the working data samples of the phased array antenna obtained in advance;

[0066] S11, optimizing the constructed beam optimization model by using a reinforcement learning strategy, so that the beam optimization model can autonomously adjust the beam control strategy according to feedback from the actual space environment and working status;

[0067] S102, establishing a multi-task learning framework to allow the beam optimization model to simultaneously learn the mutual influence and coupling effects between multiple beams to achieve a global optimal solution;

[0068] S13, deploying the trained beam optimization model to the onboard satellite, so that the trained beam optimization model outputs the optimal beam control code according to the current operating parameters of the phased array antenna of the onboard satellite, so that the onboard satellite controls the operation of the phased array antenna based on the optimal beam control code.

[0069] The embodiment of the present invention first constructs a beam optimization model based on sparse coding and deep neural network. The model receives a vector containing the working state characteristics of the phased array antenna as input, predicts the optimal feed parameter combination under different beam pointing, and outputs the corresponding beam control code value. The model is trained based on the pre-acquired phased array antenna working data samples to ensure that it has the ability to adapt to actual scenarios. Then, the constructed beam optimization model is optimized using a reinforcement learning strategy, so that the model can autonomously adjust the beam control strategy according to the feedback of the actual space environment and working state, and realize dynamic adaptability and self-learning ability. In addition, by establishing a multi-task learning framework, the beam optimization model simultaneously learns the mutual influence and coupling effects between multiple beams, ensuring that the model takes into account the global optimal solution of the overall system while optimizing the performance of a single beam. Finally, the trained beam optimization model is deployed to the onboard satellite, so that it outputs the optimal beam control code in real time according to the current working parameters of the phased array antenna, and guides the satellite to accurately control the phased array antenna based on this control code, so as to achieve fast and accurate beam pointing and switching. From the above analysis, it can be seen that the embodiments of the present invention can adaptively optimize the operation of the multi-beam phased array antenna in real time, ensure that the best performance is always maintained in a complex and changeable space environment, and effectively meet the requirements of space communication and detection tasks for high pointing accuracy and multi-target tracking.

[0070] As an improvement of the above solution, the construction of a beam optimization model based on sparse coding and deep neural network includes:

[0071] Initialize the beam optimization model based on sparse coding and deep neural network; the model architecture of the initialized beam optimization model includes input layer, hidden layer and output layer; the input layer is used to receive feature vectors; the first layer of the hidden layer is a sparse coding layer, the coding dimension is set to 1 / 4 of the length of the original feature vector, the activation function uses a hard threshold function to maintain the sparsity of the input, and the subsequent hidden layers are 3 fully connected layers, the number of nodes in each layer is 512, 256, and 128 respectively, and the activation function is ReLU; the output layer is used to predict the beam control code value under the corresponding beam pointing; the optimizer of the initialized beam optimization model uses the Adam optimizer, and the learning rate is set to 0.001; the loss function is set to:

[0072]

[0073] N beams is the total number of beams, is the weight of each control code loss item;

[0074] The acquired working data samples of the phased array antenna are used for feature extraction; the working data samples include: beam pointing scanning angle and feeding parameters, and the feeding parameters include: injection signal power, transmission power, feeding amplitude Ai and feeding phase

[0075] Apply deep learning feature extraction technology to extract features from the working data samples to obtain corresponding feature vectors: key features include: electrical characteristic curves of the transmitting and receiving components, beam shape parameters and spatial distribution relationships;

[0076] Pair the feature vectors with the actual measured beam steering code values ​​and divide them into a training set and a validation set;

[0077] Divide the training set into multiple small batches and perform mini-batch gradient descent. In each iteration, randomly extract a batch of data from the training set and feed it into the model for forward propagation.

[0078] Calculate the prediction control code output by the model on the current batch training set and calculate the loss function value;

[0079] According to the gradient of the loss function, the gradient of the model parameters is calculated through the back propagation algorithm;

[0080] Use the Adam optimizer to update model parameters based on the gradient;

[0081] After each predetermined number of iterations, the validation set is used to evaluate the model performance; if the validation set performance no longer improves or reaches a preset stopping condition, the training of the beam optimization model is terminated.

[0082] As an improvement of the above solution, the deep learning feature extraction technology is applied to extract features from the working data sample to obtain a corresponding feature vector, including:

[0083] For each transmitting and receiving component, the interpolation method is used to fit the feed amplitude attenuation curve Ai(θ, φ) and the feed phase shift curve φi(θ, φ) under different beam pointing directions, and the dynamic change rate of the two over time is calculated to obtain the electrical characteristic change curve;

[0084] For each beam pointing scan angle, the beam gain G is calculated based on the transmitted power j , thus obtaining the beam shape parameters:

[0085] P transmitj is the transmitted power, P injected is the injected signal power;

[0086] By using three-dimensional drawing software, a beam coverage map is drawn according to the beam pointing data to clarify the overlapping area of ​​adjacent beams, thereby obtaining the spatial distribution relationship.

[0087] As an improvement of the above solution, the constructed beam optimization model is optimized by using a reinforcement learning strategy, so that the beam optimization model can autonomously adjust the beam control strategy according to feedback from the actual space environment and working status, including:

[0088] Define a reinforcement learning environment, where the state of reinforcement learning is defined as follows: Among them, θ is the azimuth angle of the current beam, indicating the direction of the beam in the horizontal plane; φ is the pitch angle of the current beam, indicating the direction of the beam in the vertical plane; is the feeding amplitude of the i-th transmitting and receiving component, reflecting the signal strength of the antenna at this component; T is the ambient temperature, which affects the electrical properties of the antenna material and the stability of the transmitting and receiving components; N is the spatial noise level, which affects the communication quality and beam detection accuracy; the reinforcement learning action is defined in the following way: Where ΔA i is the feed amplitude adjustment amount of the i-th transmitting and receiving component, which is used to correct amplitude mismatch or optimize beam pointing; is the feed phase adjustment of the i-th transmitting and receiving component, which is used to adjust the beam pointing or improve the beam shape; Δt k is the delay control of the kth delay amplifier component (Delaycontrol, which is used to accurately control the time difference between the signal reaching each sending and receiving component, and then adjust the beam pointing; the reward function of reinforcement learning is defined in the following way: R = f(G, HPBW, SLL, P reflect , P transmit , N total ), where G is the beam gain, which reflects the ability of the antenna to concentrate energy into the main lobe; HPBW is the half-power beam width, which characterizes the degree of beam focusing; SLL is the sidelobe level, which measures the impact of the beam sidelobe on the main lobe; P reflect is the reflected power, which represents the energy scattered back by the antenna; P transmit is the transmitted power, i.e. the power effectively transmitted to the specified direction; N total is the total noise power, reflecting the system signal-to-noise ratio; the specific formula of the reward function is set as: Among them, w G ,w HPBW ,w SLL , is the weight coefficient used to balance the contribution of different performance indicators to the reward function;

[0089] Create a table Q-table whose size is the state space size (|S|) multiplied by the action space size (|A|), and all elements are initially set to 0; Q-table is used to store the expected cumulative reward for performing each action A in each state s;

[0090] According to the current state s, select an action A from the Q-table according to the ε-greedy strategy;

[0091] Execute the selected action on the phased array antenna, observe the new state S′ and obtain the immediate reward R;

[0092] Update the corresponding items in the Q-table according to the Q-learning update rule set below:

[0093] Among them, α is the learning rate, which controls the impact of new experience on the existing Q-value, α = 0.1; γ is the discount factor, which determines the importance of future rewards, γ = 0.9.

[0094] As an improvement to the above solution, a multi-task learning framework is established to allow the beam optimization model to simultaneously learn the mutual influence and coupling effects between multiple beams to achieve a global optimal solution, including:

[0095] In the model structure of the beam optimization model, a shared hidden layer is set for multiple beams, the shared hidden layer includes a sparse coding layer and the first two fully connected layers, and the shared hidden layer is used to learn a common beamforming rule;

[0096] Set up an independent output layer for each beam to predict the respective beam control code;

[0097] When predicting a new beam control code, the state of other beams is considered through the attention mechanism to achieve a global optimal solution; the attention mechanism implementation process includes:

[0098] For the control code prediction of the jth beam, an attention vector a is introduced j To weight and fuse the status information of other beams: Among them, N beams is the total number of beams, indicating the number of beams that can be independently controlled in the system; h i is the hidden layer output vector corresponding to the i-th beam, which contains the feature representation of the beam learned inside the model; w is a learnable attention weight vector, which is used to calculate the attention score of each beam hidden layer output vector. After normalization by the softmax function, it forms a weighted combination of other beam information; softmax(·) is a normalization function that converts the dot product result of the weight vector and the hidden layer output vector into a probability distribution to ensure that the weighted sum is 1;

[0099] The weighted attention vector a j and the hidden layer output vector h of the current beam j j Concatenate and pass it as a new input to the independent output layer to predict the control code of beam j: They are the predicted values ​​of the feed amplitude adjustment, feed phase adjustment and delay control of the jth beam respectively. OutputLayer represents the last layer of the neural network of the model, which is responsible for converting the concatenated vector into the control code prediction;

[0100] During training, the shared underlying feature layer and independent output layer parameters are updated simultaneously in the following way: Using a multi-task learning strategy, the control code prediction of multiple beams is treated as different subtasks to jointly optimize a loss function:

[0101]

[0102] , where L is the total loss function, which is used to measure the performance of the model on all beam control code prediction tasks; MSE is the mean square error function, which is used to calculate the difference between the model prediction value and the true value;

[0103] are respectively the feeding amplitude adjustment amount, feeding phase adjustment amount and delay control amount of the j-th beam predicted by the model; are respectively the actual feeding amplitude adjustment amount, feeding phase adjustment amount and delay control amount of the j-th beam.

[0104] As an improvement of the above solution, the trained beam optimization model is deployed to the onboard satellite so that the trained beam optimization model outputs an optimal beam control code according to the current operating parameters of the phased array antenna of the onboard satellite, including:

[0105] Obtain the current status information of beam j and other beam status information; the status information includes: beam pointing scanning angle and feeding parameters, and the feeding parameters include: injection signal power, transmission power, feeding amplitude Ai and feeding phase

[0106] The state information of all other beams is used as the input of the beam optimization model, and features are extracted through the shared underlying feature layer to obtain the hidden layer output vector h corresponding to each other beam. i (i≠j);

[0107] For each other beam, the hidden layer output vector h i , use the attention weight wi vector set during the training process to perform dot product and normalize it through the softmax function to get the attention score α i :α i =softmax(w T h i );

[0108] According to the attention score, the hidden layer output vectors of other beams are weighted and summed to generate the attention vector a of beam j j : N beams is the total number of beams;

[0109] The hidden layer output vector h corresponding to the current state of beam j i And the calculated attention vector a j Concatenate to form a new input vector:

[0110] The concatenated input vector The control code adjustment amount of beam j is obtained by sending it to the independent output layer OutputLayer of the beam optimization model and calculating it through the neural network:

[0111] According to the predicted control code adjustment amount, adjust the actual feeding amplitude A of beam j j and Phase A j ←A j +△A j

[0112]

[0113] According to the delay control amount Δt j adjusting the delay of the delay amplifier component associated with beam j;

[0114] Finally, the overall performance of all beams is optimized by the multi-task learning strategy pre-set during the training process of the trained beam optimization model.

[0115] See also Figure 2 , is a schematic diagram of the structure of a phased array antenna multi-beam automatic optimization device provided by an embodiment of the present invention. The phased array antenna multi-beam automatic optimization device comprises:

[0116] A construction module 10 is used to construct a beam optimization model based on sparse coding and deep neural network; the beam optimization model is used to predict the optimal feeding parameter combination under different beam pointing directions, the input layer of the beam optimization model is used to receive feature vectors, and the output layer of the beam optimization model is used to predict the corresponding beam control code value; the beam optimization model is trained based on the working data samples of the phased array antenna obtained in advance;

[0117] An optimization module 11 is used to optimize the constructed beam optimization model by using a reinforcement learning strategy, so that the beam optimization model can autonomously adjust the beam control strategy according to feedback from the actual space environment and working status;

[0118] Establishing module 12, for establishing a multi-task learning framework, so that the beam optimization model can simultaneously learn the mutual influence and coupling effects between multiple beams to achieve a global optimal solution;

[0119] The application module 13 is used to deploy the trained beam optimization model to the onboard satellite, so that the trained beam optimization model outputs the optimal beam control code according to the current operating parameters of the phased array antenna of the onboard satellite, so that the onboard satellite controls the operation of the phased array antenna based on the optimal beam control code.

[0120] The embodiment of the present invention first constructs a beam optimization model based on sparse coding and deep neural network. The model receives a vector containing the working state characteristics of the phased array antenna as input, predicts the optimal feed parameter combination under different beam pointing, and outputs the corresponding beam control code value. The model is trained based on the pre-acquired phased array antenna working data samples to ensure that it has the ability to adapt to actual scenarios. Then, the constructed beam optimization model is optimized using a reinforcement learning strategy, so that the model can autonomously adjust the beam control strategy according to the feedback of the actual space environment and working state, and realize dynamic adaptability and self-learning ability. In addition, by establishing a multi-task learning framework, the beam optimization model simultaneously learns the mutual influence and coupling effects between multiple beams, ensuring that the model takes into account the global optimal solution of the overall system while optimizing the performance of a single beam. Finally, the trained beam optimization model is deployed to the onboard satellite, so that it outputs the optimal beam control code in real time according to the current working parameters of the phased array antenna, and guides the satellite to accurately control the phased array antenna based on this control code, so as to achieve fast and accurate beam pointing and switching. From the above analysis, it can be seen that the embodiments of the present invention can adaptively optimize the operation of the multi-beam phased array antenna in real time, ensure that the best performance is always maintained in a complex and changeable space environment, and effectively meet the requirements of space communication and detection tasks for high pointing accuracy and multi-target tracking.

[0121] See also Figure 3 , is a schematic diagram of a phased array antenna multi-beam automatic optimization system provided in one embodiment of the present invention. The phased array antenna multi-beam automatic optimization system of this embodiment includes: a processor 100, a memory 101, and a computer program stored in the memory 101 and executable on the processor 100, such as a phased array antenna multi-beam automatic optimization program. When the processor 100 executes the computer program, the steps in each of the above-mentioned phased array antenna multi-beam automatic optimization method embodiments are implemented. Alternatively, when the processor 100 executes the computer program, the functions of each module / unit in the above-mentioned device embodiments are implemented.

[0122] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program in the phased array antenna multi-beam automatic optimization system.

[0123] The phased array antenna multi-beam automatic optimization system may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the schematic diagram is merely an example of a phased array antenna multi-beam automatic optimization system and does not constitute a limitation on the phased array antenna multi-beam automatic optimization system. The system may include more or fewer components than shown in the diagram, or may combine certain components, or different components. For example, the phased array antenna multi-beam automatic optimization system may also include input and output devices, network access devices, buses, and the like.

[0124] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the phased array antenna multi-beam automatic optimization system, and uses various interfaces and lines to connect various parts of the entire phased array antenna multi-beam automatic optimization system.

[0125] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the phased array antenna multi-beam automatic optimization system by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0126] Wherein, if the module / unit integrated in the phased array antenna multi-beam automatic optimization system is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media does not include electrical carrier signals and telecommunication signals.

[0127] It should be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the accompanying drawings of the device embodiments provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0128] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A phased array antenna multi-beam automatic optimization method, characterized in that: include: A beam optimization model based on sparse coding and deep neural network is constructed; the beam optimization model is used to predict the optimal feeding parameter combination under different beam pointing directions, the input layer of the beam optimization model is used to receive feature vectors, and the output layer of the beam optimization model is used to predict the corresponding beam control code value; the beam optimization model is trained based on the working data samples of the phased array antenna obtained in advance; The constructed beam optimization model is optimized by using a reinforcement learning strategy, so that the beam optimization model can autonomously adjust the beam control strategy according to feedback from the actual space environment and working status; Establish a multi-task learning framework to allow the beam optimization model to simultaneously learn the mutual influence and coupling effects between multiple beams to achieve a global optimal solution; The trained beam optimization model is deployed to the onboard satellite so that the trained beam optimization model outputs the optimal beam control code according to the current operating parameters of the onboard satellite's phased array antenna, thereby enabling the onboard satellite to control the operation of the phased array antenna based on the optimal beam control code.

2. The phased array antenna multi-beam automatic optimization method according to claim 1, characterized in that: The construction of a beam optimization model based on sparse coding and deep neural network includes: Initialize the beam optimization model based on sparse coding and deep neural network; the model architecture of the initialized beam optimization model includes input layer, hidden layer and output layer; the input layer is used to receive feature vectors; the first layer of the hidden layer is a sparse coding layer, the coding dimension is set to 1 / 4 of the length of the original feature vector, the activation function uses a hard threshold function to maintain the sparsity of the input, and the subsequent hidden layers are 3 fully connected layers, the number of nodes in each layer is 512, 256, and 128 respectively, and the activation function is ReLU; the output layer is used to predict the beam control code value under the corresponding beam pointing; the optimizer of the initialized beam optimization model uses the Adam optimizer, and the learning rate is set to 0.001; the loss function is set to: N beams is the total number of beams, w ΔA′ w Δt is the weight of each control code loss item; The acquired working data samples of the phased array antenna are used for feature extraction; the working data samples include: beam pointing scanning angle and feeding parameters, and the feeding parameters include: injection signal power, transmission power, feeding amplitude Ai and feeding phase Apply deep learning feature extraction technology to extract features from the working data samples to obtain corresponding feature vectors: key features include: electrical characteristic curves of the transmitting and receiving components, beam shape parameters and spatial distribution relationships; Pair the feature vectors with the actual measured beam steering code values ​​and divide them into a training set and a validation set; Divide the training set into multiple small batches and perform mini-batch gradient descent. In each iteration, randomly extract a batch of data from the training set and feed it into the model for forward propagation. Calculate the prediction control code output by the model on the current batch training set and calculate the loss function value; According to the gradient of the loss function, the gradient of the model parameters is calculated through the back propagation algorithm; Use the Adam optimizer to update model parameters based on the gradient; After each predetermined number of iterations, the validation set is used to evaluate the model performance; if the validation set performance no longer improves or reaches a preset stopping condition, the training of the beam optimization model is terminated.

3. The phased array antenna multi-beam automatic optimization method according to claim 2, characterized in that: The applying of deep learning feature extraction technology to extract features from the working data sample to obtain a corresponding feature vector includes: For each transmitting and receiving component, the interpolation method is used to fit the feed amplitude attenuation curve Ai(θ, φ) and the feed phase shift curve φi(θ, φ) under different beam pointing directions, and the dynamic change rate of the two over time is calculated to obtain the electrical characteristic change curve; For each beam pointing scan angle, the beam gain G is calculated based on the transmitted power j , thus obtaining the beam shape parameters: is the transmitted power, P injected is the injected signal power; By using three-dimensional drawing software, a beam coverage map is drawn according to the beam pointing data to clarify the overlapping area of ​​adjacent beams, thereby obtaining the spatial distribution relationship.

4. The phased array antenna multi-beam automatic optimization method according to claim 3, characterized in that: The method of optimizing the constructed beam optimization model by using a reinforcement learning strategy so that the beam optimization model can autonomously adjust the beam control strategy according to feedback from the actual space environment and working status includes: Define a reinforcement learning environment, where the state of reinforcement learning is defined as follows: Among them, θ is the azimuth angle of the current beam, indicating the direction of the beam in the horizontal plane; φ is the pitch angle of the current beam, indicating the direction of the beam in the vertical plane; is the feeding amplitude of the i-th transmitting and receiving component, reflecting the signal strength of the antenna at this component; T is the ambient temperature, which affects the electrical properties of the antenna material and the stability of the transmitting and receiving components; N is the spatial noise level, which affects the communication quality and beam detection accuracy; the reinforcement learning action is defined in the following way: Among them, ΔA i is the feed amplitude adjustment amount of the i-th transmitting and receiving component, which is used to correct amplitude mismatch or optimize beam pointing; is the feed phase adjustment of the i-th transmitting and receiving component, which is used to adjust the beam pointing or improve the beam shape; Δt k is the delay control of the kth delay amplifier component (Delaycontrol, which is used to accurately control the time difference between the signal reaching each sending and receiving component, and then adjust the beam pointing; the reward function of reinforcement learning is defined in the following way: R = f(G, HPBW, SLL, P reflect , P transmit , N total ), where G is the beam gain, which reflects the ability of the antenna to concentrate energy into the main lobe; HPBW is the half-power beam width, which characterizes the degree of beam focusing; SLL is the sidelobe level, which measures the impact of the beam sidelobe on the main lobe; P reflect is the reflected power, which represents the energy scattered back by the antenna; P transmit is the transmitted power, i.e. the power effectively transmitted to the specified direction; N total is the total noise power, reflecting the system signal-to-noise ratio; the specific formula of the reward function is set as: Among them, w G , w HPBW , w SLL , is the weight coefficient used to balance the contribution of different performance indicators to the reward function; Create a table Q-table whose size is the state space size (|S|) multiplied by the action space size (|A|), and all elements are initially set to 0; Q-table is used to store the expected cumulative reward for performing each action A in each state S; According to the current state S, select an action A from the Q-table according to the ε-greedy strategy; Execute the selected action on the phased array antenna, observe the new state S′ and obtain the immediate reward R; Update the corresponding items in the Q-table according to the Q-learning update rule set below: Among them, α is the learning rate, which controls the impact of new experience on the existing Q-value, α = 0.1; γ is the discount factor, which determines the importance of future rewards, γ = 0.

9.

5. The phased array antenna multi-beam automatic optimization method according to claim 4, characterized in that: The multi-task learning framework is established to allow the beam optimization model to simultaneously learn the mutual influence and coupling effects between multiple beams to achieve a global optimal solution, including: In the model structure of the beam optimization model, a shared hidden layer is set for multiple beams, the shared hidden layer includes a sparse coding layer and the first two fully connected layers, and the shared hidden layer is used to learn a common beamforming rule; Set up an independent output layer for each beam to predict the respective beam control code; When predicting a new beam control code, the state of other beams is considered through the attention mechanism to achieve a global optimal solution; the attention mechanism implementation process includes: For the control code prediction of the jth beam, an attention vector a is introduced j To weight and fuse the status information of other beams: Among them, N beams : is the total number of beams, indicating the number of beams that can be independently controlled in the system; h i : is the hidden layer output vector corresponding to the jth beam, which contains the feature representation of the beam learned inside the model; w is a learnable attention weight vector, which is used to calculate the attention score of each beam hidden layer output vector. After normalization by the softmax function, it forms a weighted combination of other beam information; softmax(·) is a normalization function that converts the dot product result of the weight vector and the hidden layer output vector into a probability distribution to ensure that the weighted sum is 1; The weighted attention vector a j and the hidden layer output vector h of the current beam j j Concatenate and pass it as a new input to the independent output layer to predict the control code of beam j: ΔA j , Δt j They are the predicted values ​​of the feed amplitude adjustment, feed phase adjustment and delay control of the jth beam respectively. OutputLayer represents the last layer of the neural network of the model, which is responsible for converting the concatenated vector into the control code prediction; During training, the shared underlying feature layer and independent output layer parameters are updated simultaneously in the following way: Using a multi-task learning strategy, the control code prediction of multiple beams is treated as different subtasks to jointly optimize a loss function: Among them, L is the total loss function, which is used to measure the performance of the model on all beam control code prediction tasks; MSE is the mean square error function, which is used to calculate the difference between the model prediction value and the true value; ΔA j，pred , Δt j，pred are respectively the feed amplitude adjustment, feed phase adjustment and delay control of the jth beam predicted by the model; ΔA j，true , Δt j，true are respectively the actual feeding amplitude adjustment amount, feeding phase adjustment amount and delay control amount of the j-th beam.

6. The phased array antenna multi-beam automatic optimization method according to claim 5, characterized in that: The step of deploying the trained beam optimization model to the onboard satellite so that the trained beam optimization model outputs an optimal beam control code according to the operating parameters of the current phased array antenna of the onboard satellite includes: Obtain the current status information of beam j and other beam status information; the status information includes: beam pointing scanning angle and feeding parameters, and the feeding parameters include: injection signal power, transmission power, feeding amplitude Ai and feeding phase The state information of all other beams is used as the input of the beam optimization model, and features are extracted through the shared underlying feature layer to obtain the hidden layer output vector h corresponding to each other beam. i (i≠j); For each other beam, the hidden layer output vector h i , use the attention weight w vector set during training to perform dot product and normalize it through the softmax function to get the attention score α i :α i =softmax(w T h i ); According to the attention score, the hidden layer output vectors of other beams are weighted and summed to generate the attention vector a of beam j j : N beams : is the total number of beams; The hidden layer output vector h corresponding to the current state of beam j i And the calculated attention vector a j Concatenate to form a new input vector: The concatenated input vector The control code adjustment amount of beam j is obtained by sending it to the independent output layer OutputLayer of the beam optimization model and calculating it through the neural network: According to the predicted control code adjustment amount, adjust the actual feeding amplitude A of beam j j and Phase A j ←A j +ΔA j According to the delay control amount Δt j adjusting the delay of the delay amplifier component associated with beam j; Finally, the overall performance of all beams is optimized by the multi-task learning strategy pre-set during the training process of the trained beam optimization model.

7. A phased array antenna multi-beam automatic optimization device, characterized in that: include: A construction module is used to construct a beam optimization model based on sparse coding and deep neural network; the beam optimization model is used to predict the optimal feeding parameter combination under different beam pointing directions, the input layer of the beam optimization model is used to receive feature vectors, and the output layer of the beam optimization model is used to predict the corresponding beam control code value; the beam optimization model is trained based on the working data samples of the phased array antenna obtained in advance; An optimization module, used to optimize the constructed beam optimization model by using a reinforcement learning strategy, so that the beam optimization model can autonomously adjust the beam control strategy according to feedback from the actual space environment and working status; Establishing a module for establishing a multi-task learning framework so that the beam optimization model can simultaneously learn the mutual influence and coupling effects between multiple beams to achieve a global optimal solution; The application module is used to deploy the trained beam optimization model to the onboard satellite, so that the trained beam optimization model outputs the optimal beam control code according to the current operating parameters of the phased array antenna of the onboard satellite, so that the onboard satellite controls the operation of the phased array antenna based on the optimal beam control code.

8. A phased array antenna multi-beam automatic optimization system, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the method for automatic multi-beam optimization of a phased array antenna as claimed in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the phased array antenna multi-beam automatic optimization method according to any one of claims 1 to 6.

Images