Optimal switching beam pair searching method and device

Abstract

The invention discloses an optimal switching beam pair searching method and device. The method comprises the following steps: acquiring a target beam pair information graph of a plurality of beam pairs between a base station and a target unmanned aerial vehicle in a coverage range of the base station at a first moment; the target beam pair information graph is analyzed by using a search range prediction model, a corresponding target optimal beam pair search range is obtained, and the search range prediction model is constructed based on an improved depth residual network; and an optimal switching beam pair of the base station and the unmanned aerial vehicle at a second moment is determined from the target optimal beam pair search range by using a heuristic algorithm, and the second moment is the next moment of the first moment. According to the method and the device, the technical problem that the communication between the base station and the unmanned aerial vehicle is delayed or interrupted due to the fact that the optimal switching beam pair is difficult to quickly find in a plurality of candidate beam pairs of the communication between the base station and the unmanned aerial vehicle in related technologies is solved.

Description

Technical Field

[0001] This application relates to the field of wireless communication technologies, and in particular, to an optimal handover beam pair search method and apparatus. Background Art

[0002] In low-altitude communication scenarios such as drone inspections and live broadcasts, the pictures need to be transmitted in real time, which poses high requirements on the communication rate and stability of the air network. However, due to the relatively fast flight speed of drones, the base station needs to update the beam in real time to ensure high-speed and stable communication with the drones and avoid communication delays or interruptions caused by high latency. Therefore, how to quickly switch the optimal beam during the communication between the base station and the drones is a difficult problem that urgently needs to be solved.

[0003] Currently, the path loss of the deployed frequency band of the base station is relatively high, and millimeter wave band sites have been deployed. In the post-5G era, AAU needs to rely on a very large-scale antenna array to obtain beamforming gain to combat path loss. The more antennas there are, the more and narrower the beams are, and more time is required to search for the optimal beam pair to maintain high-speed and low-latency stable communication. Therefore, the existing solutions cannot meet the requirements of fast beam pair switching between the base station and the drones. Summary of the Invention

[0004] Embodiments of this application provide an optimal handover beam pair search method and apparatus, which can at least solve the technical problem that it is difficult to quickly find the optimal handover beam pair among multiple candidate beam pairs for the communication between the base station and the drones in the related art, resulting in communication delays or interruptions between the base station and the drones.

[0005] According to one aspect of the embodiments of this application, an optimal handover beam pair search method is provided, including: obtaining a target beam pair information map of multiple beam pairs between a base station and a target drone within the coverage range of the base station at a first moment; analyzing the target beam pair information map by using a pre-trained search range prediction model to obtain a corresponding target optimal beam pair search range, where the search range prediction model is constructed based on an improved deep residual network, and the improved deep residual network at least includes: a shallow feature extraction module, a deep feature extraction module based on a residual nesting structure, and a prediction output module; determining an optimal handover beam pair between the base station and the drone at a second moment from the target optimal beam pair search range by using a heuristic algorithm, where the second moment is the next moment of the first moment.

[0006] Optionally, obtaining a target beam pair information map of multiple beam pairs between a base station and a target unmanned aerial vehicle (UAV) within the coverage range of the base station at a first moment includes: obtaining communication quality parameters of each of the multiple beam pairs between the base station and the target UAV at the first moment, where the communication quality parameters include at least one of the following: spectral efficiency, signal-to-interference-plus-noise ratio (SINR); obtaining first communication state information of the base station at the first moment and second communication state information of the target UAV at the first moment, where the first communication state information at least includes: first beam state information of each of the multiple beams of the base station, base station state information, and the second communication state information at least includes: second beam state information of each of the multiple beams of the target UAV, UAV state information, and at least one of the first beam state information and the second beam state information includes: beam azimuth angle, beam elevation angle, beam width, beam gain; for each beam pair, taking the first beam state information, second beam state information, communication quality parameters of the two beams within the beam pair, base station state information, and UAV state information as multiple features of the beam pair, forming a feature vector of the beam pair from the multiple features, and performing normalization processing on the feature vector; arranging the multiple beams of the base station and the multiple beams of the target UAV in ascending order of beam numbers as multiple matrix rows and multiple matrix columns, taking the normalized feature vectors of each beam pair as matrix elements, and constructing a target beam pair information map.

[0007] Optionally, the training process of the search range prediction model includes: constructing a deep learning model including a shallow feature extraction module, a deep feature extraction module, and a prediction output module; obtaining a training sample set and a sample label set, where the training sample set includes multiple beam pair information maps between a base station and a UAV as training samples, and the sample label set includes the true optimal beam pair search range corresponding to each beam pair information map as a sample label; using the training sample set and the sample label set to perform iterative training on the deep learning model to obtain a search range prediction model.

[0008] Optionally, the shallow feature extraction module includes at least one convolutional layer, and the shallow feature extraction module is used to extract low-level image features of the beam pair information map; the deep feature extraction module includes multiple residual blocks connected by short skip connections, each residual block includes multiple convolutional layers, batch normalization layers, and activation function layers connected by residual connections, and the deep feature extraction module is used to extract high-level image features based on the low-level image features; the prediction output module includes two fully connected layers and a Dropout layer, and the prediction output module is used to convert the high-level image features into a predicted search range.

[0009] Optionally, an iterative training is performed on the deep learning model using the training sample set and the sample label set to obtain a search range prediction model, including: for each training batch in the iterative training process, inputting each training sample in the training batch into the deep learning model to obtain the search range of each predicted optimal beam pair output by the deep learning model, constructing an objective loss function using the search range of the predicted optimal beam pair and the corresponding sample label, and adjusting the model parameters of the deep learning model according to the objective loss function until the model parameters converge to obtain the search range prediction model.

[0010] Optionally, the objective loss function includes: an angle loss function, a search radius loss function, and a physical constraint loss function, and the expression of the objective loss function is: , where represents the angle loss function, represents the search radius loss function, represents the physical constraint loss function, , , respectively represent the adjustment coefficients of the angle loss function, the search radius loss function, and the physical constraint loss function, where: the angle loss function has the expression: , where N represents the total number of training samples in the training batch, represents the true beam azimuth angle of the i-th training sample, represents the predicted beam azimuth angle of the i-th training sample, represents the true beam elevation angle of the i-th training sample, represents the predicted beam elevation angle of the i-th training sample, represents a preset angle threshold, respectively represent weight coefficients; the search radius loss function has the expression: , where N represents the total number of training samples in the training batch, represents the true search radius of the i-th training sample, represents the predicted search radius of the i-th training sample, represents the preset maximum allowable search range, represents the adjustment coefficient; the physical constraint loss function has the expression: , where represents the beam direction gradient constraint loss function, and the beam direction gradient constraint loss function is obtained by calculating the base station beam pattern function, denotes the energy coverage constraint loss function, and when the energy coverage rate of the predicted optimal beam pair search range corresponding to the training sample is less than the preset energy threshold, the value of the energy coverage constraint loss function increases, while when the energy coverage rate of the predicted optimal beam pair search range corresponding to the training sample is not less than the energy threshold, the value of the energy coverage constraint loss function decreases.

[0011] Optionally, the target optimal beam pair search range is determined by the central beam pair and the search radius. Among them, using a heuristic algorithm to determine the optimal handover beam pair between the base station and the UAV at the second moment within the target optimal beam pair search range includes: initializing the temperature and the cooling coefficient ; in the first round of search, using the central beam pair as the current optimal beam pair in the current search round, determining multiple first neighboring beam pairs adjacent to the current optimal beam pair within the target optimal beam pair search range, and calculating the spectral efficiency change rate of each first neighboring beam pair compared to the current optimal beam pair according to the following formula: , where represents the current optimal beam pair, represents the first neighboring beam pair, and ; when the spectral efficiency change rate of the first target neighboring beam pair compared to the current optimal beam pair is higher than the preset first threshold, calculate the transition probability of the first target neighboring beam according to the following formula: , where represents the predicted search radius, represents the maximum allowable search radius, represents the adjustment coefficient; when the transition probability of the first target neighboring beam is higher than the preset second threshold, take the first target neighboring beam as the optimal beam pair in the next round of search, and update the temperature of the current search round according to the following formula: ; in the second round and subsequent rounds of search, take the neighboring beam pair switched in the previous round of search as the current optimal beam pair in the current search round, re-determine multiple second neighboring beam pairs adjacent to the current optimal beam pair within the target optimal beam pair search range, and determine the spectral efficiency change rate of each second neighboring beam pair compared to the current optimal beam pair; when the spectral efficiency change rate of the second target neighboring beam pair compared to the current optimal beam pair is higher than the first threshold, calculate the transition probability of the second target neighboring beam according to the following formula: , where represents the spectral efficiency change rate of the second target neighboring beam, represents the spectral efficiency change rate of the current optimal beam pair in the current search round, represents the temperature of the k-th round, and , , where K represents the maximum number of iterative searches; when the transition probability of the second target neighboring beam is higher than the second threshold, the second target neighboring beam is used as the optimal beam pair in the next round of the search process, and the temperature of the current search round is updated; repeat the above steps, and use the optimal beam pair of the target search round with a temperature lower than the preset third threshold as the optimal handover beam pair between the base station and the UAV at the second moment.

[0012] According to another aspect of the embodiments of the present application, there is also provided an optimal handover beam pair search device, including: an acquisition module, configured to acquire a target beam pair information map of multiple beam pairs between a base station and a target UAV within the coverage range of the base station at a first moment; a prediction module, configured to analyze the target beam pair information map by using a pre-trained search range prediction model to obtain a corresponding target optimal beam pair search range, where the search range prediction model is constructed based on an improved deep residual network, and the improved deep residual network at least includes: a shallow feature extraction module, a deep feature extraction module based on a residual nesting structure, and a prediction output module; a search module, configured to determine the optimal handover beam pair between the base station and the UAV at a second moment from the target optimal beam pair search range by using a heuristic algorithm, where the second moment is the next moment of the first moment.

[0013] According to another aspect of the embodiments of the present application, there is also provided a computer program product, including: a computer program, where when the computer program is executed by a processor, the above-mentioned optimal handover beam pair search method is implemented.

[0014] According to another aspect of the embodiments of the present application, there is also provided an electronic device, including: a memory and a processor, where a computer program is stored in the memory, and the processor is configured to execute the above-mentioned optimal handover beam pair search method through the computer program.

[0015] In the embodiments of the present application, a target beam pair information map of multiple beam pairs between a base station and a target UAV within the coverage range of the base station is acquired at a first moment, and the target beam pair information map is analyzed by using a search range prediction model constructed based on an improved deep residual network to obtain a corresponding target optimal beam pair search range, thereby greatly reducing the search space and improving the search efficiency; a heuristic algorithm is used to determine the optimal handover beam pair between the base station and the UAV at a second moment from the target optimal beam pair search range, ensuring continuous high-quality and low-latency communication, thereby realizing efficient and stable communication, and further solving the technical problem that due to the difficulty of quickly finding the optimal handover beam pair among multiple candidate beam pairs for communication between the base station and the UAV in the related art, the communication between the base station and the UAV is delayed or interrupted. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings described herein are used to provide a further understanding of the present application and form a part of the present application. The schematic embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation to the present application. In the drawings:

[0017] Figure 1 is a schematic diagram of the beam pair gain in an optional line-of-sight communication mode according to an embodiment of the present application;

[0018] Figure 2 is a schematic diagram of the beam pair gain in an optional non-line-of-sight communication mode according to an embodiment of the present application;

[0019] Figure 3 is a schematic flowchart of an optional optimal handover beam pair search method according to an embodiment of the present application;

[0020] Figure 4 is a schematic diagram of an optional optimal beam pair search range according to an embodiment of the present application;

[0021] Figure 5 is a schematic diagram of a comparison of the number of searches according to an embodiment of the present application;

[0022] Figure 6 is a schematic structural diagram of an optional optimal handover beam pair search device according to an embodiment of the present application;

[0023] Figure 7 is a schematic diagram of an optional electronic device according to an embodiment of the present application. Detailed implementation manners

[0024] In order to enable those skilled in the art to better understand the solution of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0025] It should be noted that the terms "first", "second", etc. in the description, claims and drawings of this application are used to distinguish similar objects and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments of this application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units need not be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0026] To better understand the embodiments of this application, the following is a translation and explanation of some nouns or terms that appear in the description process of the embodiments of this application:

[0027] Line of Sight (LOS) communication: It refers to the transmission of signals between the transmitting antenna and the receiving antenna within the distance where "they can see each other", that is, the radio wave directly propagates from the transmitting point to the receiving point (generally including the reflected wave on the ground). Generally, the distance of line-of-sight propagation is 20 - 50 Km.

[0028] Not Line of Sight (NLOS) communication: It refers to the non-direct point-to-point communication between the receiver and the transmitter. The most direct explanation is that the line of sight between the two points is blocked and they cannot see each other.

[0029] Spectral Efficiency: Abbreviated as spectral efficiency, also known as system capacity and bandwidth utilization rate. It is defined as the ratio of the effective information rate transmitted by the system to the communication channel bandwidth, that is, the number of bits that can be transmitted per second on the transmission channel per unit bandwidth, representing the utilization rate of the spectrum resources by the system.

[0030] Signal to Interference plus Noise Ratio (SINR): Abbreviated as signal-to-dry ratio, SINR is the ratio of the intensity of the received useful signal to the intensity of the received dry signal (including noise and interference). It is used to measure the index of the signal intensity relative to the dry source intensity. Therefore, the larger the signal-to-dry ratio, the better the communication quality. Thus, improving the signal-to-dry ratio is one of the main tasks to improve communication quality.

[0031] Simulated Annealing (SA) algorithm: It is a probabilistic global optimization heuristic algorithm inspired by the annealing process of solid materials. In the physical annealing process, the material is first heated to a high temperature to make the internal particles in a highly disordered state, and then slowly cooled. As the temperature drops, the particles tend to be arranged in an orderly manner, and finally reach the lowest energy (i.e., the ground state) at room temperature. The simulated annealing algorithm abstracts this physical process into a strategy for solving optimization problems.

[0032] Deep Residual Networks (ResNet): By introducing residual learning and a special network structure, it solves the gradient message problem in traditional deep neural networks and realizes an efficient and scalable deep model. The deep residual network includes: an initial convolutional layer, a residual blocks group, a global average pooling layer, and a fully connected layer. Among them, the residual blocks group contains multiple residual blocks, which are the basic building blocks of ResNet. Through the residual blocks, ResNet effectively solves the gradient vanishing problem and can train extremely deep networks.

[0033] Embodiment 1

[0034] The line-of-sight communication method is mainly used between the base station and the drone. However, there are a large number of high-rise residential areas in dense urban areas, and coupled with the different heights of the airspace stations, in some scenarios, the non-line-of-sight communication method will be used between the base station and the drone. Figure 1 、 Figure 2 They are respectively the gain diagrams of the base station and the drone using different beams for communication in the line-of-sight scenario and the non-line-of-sight scenario. Among them, the lighter the color, the higher the communication rate. Therefore, it is not difficult to see that in Figure 1 the shown line-of-sight scenario, there are fewer light-colored areas, and there is only one beam pair with the maximum gain. Although other beam pairs near this beam pair can also communicate, there is a large gap in the communication rate; while in Figure 2 the shown non-line-of-sight scenario, there are relatively more light-colored areas and there are multiple beam pairs with similar communication rates, that is, there are multiple high-rate alternative beams.

[0035] Therefore, during low-altitude communication, there are many candidate beams for communication between the base station and the drone, and the beam search range is large. However, the traditional beam switching method is difficult to quickly find the optimal beam pair for switching from a large search range, resulting in stuttering or even interruption in the communication process between the base station and the drone.

[0036] To solve the above problems, an embodiment of the present application provides an optimal handover beam pair search method to solve the problem of low beam search efficiency caused by a large beam search range. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0037] Figure 3 is a schematic flowchart of an optimal handover beam pair search method provided according to an embodiment of the present application, as Figure 3 shown, the method includes the following steps:

[0038] Step S302, obtain the target beam pair information map of multiple beam pairs between the base station and the target unmanned aerial vehicle within the coverage range of the base station at the first moment.

[0039] Step S304, analyze the target beam pair information map by using a pre-trained search range prediction model to obtain the corresponding target optimal beam pair search range, wherein the search range prediction model is constructed based on an improved deep residual network, and the improved deep residual network at least includes: a shallow feature extraction module, a deep feature extraction module based on a residual nesting structure, and a prediction output module.

[0040] Step S306, use a heuristic algorithm to determine the optimal handover beam pair between the base station and the unmanned aerial vehicle at the second moment from the target optimal beam pair search range, where the second moment is the next moment of the first moment.

[0041] Based on the solution defined in the above steps S302 to S306, it can be known that in the embodiment of the present application, the target beam pair information map is analyzed by using the search range prediction model constructed based on the improved deep residual network to obtain the target optimal beam pair search range where the optimal handover beam pair between the base station and the unmanned aerial vehicle is located, greatly reducing the search space and improving the search efficiency. Subsequently, a heuristic algorithm is used to further accurately determine the optimal handover beam pair within the predicted search range, ensuring continuous high-quality and low-latency communication, thereby achieving the technical effect of efficient and stable communication, solving the challenges faced by traditional beam switching methods in the high-speed movement of unmanned aerial vehicles, especially in the non-line-of-sight communication environment in dense urban areas, and significantly improving the communication rate and stability, meeting the timeliness requirements of emergency response and resource scheduling.

[0042] The following describes each step of the optimal handover beam pair search method in combination with a specific implementation process.

[0043] As an alternative implementation, in the technical solution provided in the above step S302, the target beam pair information graph can be obtained according to the following steps, including:

[0044] First step: Obtain the communication quality parameters of multiple beam pairs between the base station and the target UAV at a first moment respectively.

[0045] Among them, the communication quality parameters include at least one of the following: spectral efficiency, gain information, signal-to-interference-plus-noise ratio, etc.

[0046] Second step: Obtain the first communication state information of the base station at the first moment and the second communication state information of the target UAV at the first moment.

[0047] Among them, the first communication state information at least includes: the first beam state information of each of the multiple beams of the base station, such as the beam azimuth angle (determined by the mechanical azimuth angle and the electronic azimuth angle), the beam downward tilt angle (determined by the mechanical downward tilt angle and the electronic downward tilt angle), the beam horizontal width, the beam vertical width, the beam gain, the beam activation time, etc. In addition, the first communication state information also includes: state information such as the base station height. These information jointly describe the spatial pointing, energy characteristics and life cycle of the base station beam.

[0048] The second communication state information at least includes: the second beam state information of each of the multiple beams of the target UAV, such as the beam azimuth angle, the beam downward tilt angle, etc. In addition, the second communication state information also includes: the latitude, longitude and altitude coordinates of the target UAV, the three-dimensional velocity components, the three-dimensional acceleration components, the pitch angle, the roll angle, the residence time of the target UAV at this base station, the residence time of the target UAV under each transmitting beam of this base station, etc. These information reflect the precise position, motion state of the target UAV and its communication persistence with the base station.

[0049] Third step: For each beam pair, take the first beam state information, the second beam state information, the communication quality parameters of the two beams in the beam pair, as well as the base station state information and the UAV state information of each of the two beams in the beam pair as multiple features of the beam pair, form a feature vector of the beam pair from the multiple features, and perform normalization processing on the feature vector.

[0050] Among them, the purpose of the normalization processing is to eliminate the influence of the feature scale and make different features on the same magnitude. The normalization processing usually includes: mean removal and scaling, that is, subtracting the mean value of each feature from it and dividing by its standard deviation to ensure that the mean value of all features is 0 and the variance is 1. After such processing, even if the original value ranges of the features are different, they can fairly participate in the subsequent machine learning process.

[0051] Step 4: Arrange the multiple beams of the base station and the multiple beams of the target UAV in ascending order of beam numbers as multiple matrix rows and multiple matrix columns, and use the normalized eigenvectors of each beam pair as matrix elements to construct the target beam pair information map.

[0052] It can be understood that the beam numbers of the base station beams and the beam numbers of the target UAV beams are arranged in ascending order and used as the horizontal and vertical coordinates to construct a two-dimensional grid, and the normalized eigenvectors are mapped to the corresponding grid points as the information of the grid points, so as to obtain the target beam pair information map.

[0053] Among them, the above-mentioned target beam pair information map can be in the following form:

[0054] ,

[0055] In the formula, represents the number of beams of the base station, represents the number of beams of the target UAV, represents the th eigenvector (also known as "beam pair information block") of the beam pair composed of the th base station beam and the

[0056] Further, input the obtained target beam pair information map into a pre-trained search range prediction model for analysis to obtain the corresponding target optimal beam pair search range.

[0057] As an optional implementation manner, the training process of the above-mentioned search range prediction model may include:

[0058] Step 1: Construct a deep learning model including a shallow feature extraction module, a deep feature extraction module, and a prediction output module;

[0059] Step 2: Obtain a training sample set and a sample label set. Among them, the training sample set includes multiple beam pair information maps between the base station and the UAV as training samples, and the sample label set includes the true optimal beam pair search range corresponding to each beam pair information map as a sample label;

[0060] Step 3: Use the training sample set and the sample label set to iteratively train the deep learning model to obtain the search range prediction model.

[0061] In the search range prediction model, the shallow feature extraction module usually includes several convolutional layers (for example, the shallow feature extraction module includes a convolutional layer with a convolutional kernel size of 7*7 and a stride of 2), which are used to extract the low-level features and local features of the beam pair information map.

[0062] The deep feature extraction module usually includes multiple residual blocks connected by short skip connections. Each residual block includes multiple convolutional layers, batch normalization layers, and activation function layers (such as ReLU) connected by residual connections, which are used to effectively extract high-level and abstract features based on the low-level features of the image. Among them, the output of each residual block is to add the input signal "residual" to the convolved signal, which avoids the vanishing gradient and degradation problems when the network depth increases, enabling the model to learn more complex feature representations.

[0063] The prediction output module predicts the prediction search range of the optimal handover beam pair between the base station and the UAV based on the high-level image features and low-frequency image features output by the above two modules. This module generally includes two fully connected layers and a Dropout layer, and the connection relationship is the first fully connected layer, the Dropout layer, and the second fully connected layer. Among them, the first fully connected layer is used to convert the high-level image features into low-dimensional prediction outputs, while the second fully connected layer converts the features output by the Dropout layer into prediction outputs to obtain the prediction search range.

[0064] In the specific training process, the deep learning model can be iteratively trained using the training sample set and the sample label set in the following way to obtain the search range prediction model: For each training batch in the iterative training process, input the training samples of the training batch into the deep learning model to obtain the predicted optimal beam pair search ranges output by the deep learning model. Use the predicted optimal beam pair search ranges and the corresponding sample labels to construct the objective loss function, and adjust the model parameters of the deep learning model according to the objective loss function until the model parameters converge to obtain the search range prediction model.

[0065] Among them, the above objective loss function includes: an angular loss function, a search radius loss function, and a physical constraint loss function. Therefore, the expression of the above objective loss function is: , where represents the angular loss function, represents the search radius loss function, represents the physical constraint loss function, , , respectively represent the adjustment coefficients of the angular loss function, the search radius loss function, and the physical constraint loss function. Among them:

[0066] The above angular loss function has the following expression:

[0067] ,

[0068] where N represents the total number of training samples in the training batch, Represents the true beam azimuth angle of the i-th training sample, Represents the predicted beam azimuth angle of the i-th training sample, Represents the true beam elevation angle of the i-th training sample, Represents the predicted beam elevation angle of the i-th training sample, Represents a preset angle threshold, Respectively represent weight coefficients.

[0069] Search radius loss function The asymmetric penalty function is used to encourage the predicted radius to be slightly larger than the true value to avoid missed detection. Therefore, the search radius loss function The expression can be written as:

[0070] ,

[0071] In the formula, N represents the total number of training samples in a training batch, Represents the true search radius of the i-th training sample, Represents the predicted search radius of the i-th training sample, Represents the preset maximum allowable search range, Represents a regulation coefficient.

[0072] In addition, the physical constraint loss function Introduces the beam direction gradient constraint function and the energy coverage constraint function. Therefore, . By introducing two physical constraints, prior knowledge such as the electromagnetic wave propagation law and antenna characteristics is embedded in the neural network, improving the rationality and generalization ability of the model. Specifically:

[0073] Regarding the beam direction gradient constraint loss function , it can be directly obtained by taking the derivative of the base station beam pattern function . Among them, the base station beam pattern function describes the intensity distribution of the base station transmitting signal in different directions in space, and its derivative (gradient) reflects the rate of change of this intensity distribution with direction. Therefore, in the embodiment of the present application, by taking the derivative of the base station beam pattern function in a certain direction (such as azimuth and elevation), the rate of change of the beam intensity in this direction can be obtained to set a reasonable gradient threshold as the beam direction constraint. Subsequently, if the gradient value of a certain beam pair exceeds the set threshold, it means that the directivity change of this beam pair is too fast, so this beam pair is not conducive to stable communication; on the contrary, if the gradient value of the beam pair is small, it means that the beam direction change of this beam pair is gentle and the communication stability is better.

[0074] Regarding the energy coverage constraint loss function , its goal is to ensure that the predicted search radius contains sufficient energy to avoid missing detection of beams. Therefore, when the energy coverage rate of the search range of the predicted optimal beam pair corresponding to the training sample is less than the preset energy threshold , the energy coverage constraint loss function increases; while when the energy coverage rate of the search range of the predicted optimal beam pair corresponding to the training sample is not less than the preset energy threshold , the energy coverage constraint loss function decreases. Among them, the expression of the energy coverage rate of the search range of the predicted optimal beam pair corresponding to the training sample can be written as:

[0075] .

[0076] Therefore, the embodiment of the present application uses the search range prediction model obtained by the above training to perform multi-level and multi-perspective feature extraction and analysis on the target beam information map, and accurately predicts the search range of the target optimal beam pair for beam switching at the next moment. This prediction process combines the advantages of image recognition and the powerful function of deep learning, can efficiently screen the predicted search range in a complex environment, significantly reduces the need to traverse all possible beam pairs, thereby greatly reducing the search time for the optimal switching beam pair between the search base station and the drone, and improving the stability and efficiency of communication.

[0077] Finally, since the search range of the target optimal beam pair obtained by model prediction is determined by the central beam pair and the search radius, and the search range of the target optimal beam pair includes multiple candidate beam pairs, as Figure 4 shown. Therefore, the embodiment of the present application can use a heuristic algorithm to iteratively search for the optimal switching beam pair between the base station and the drone at the second moment within the search range of the target optimal beam pair. Among them, the heuristic algorithm includes: particle swarm optimization algorithm, simulated annealing algorithm.

[0078] Next, taking the simulated annealing algorithm as an example, the above search process will be described.

[0079] (I) Initialization: Set the temperature and the cooling coefficient , where the initial temperature should be set high enough to accept more inferior solutions at the initial stage of the search, increasing the randomness of the search, and the setting of the initial temperature usually depends on the scale and complexity of the problem.

[0080] (II) In the first round of the search process, first use the central beam pair (i.e., Figure 4The intersection point within it) is used as the current optimal beam pair in the current search round. Determine multiple first neighboring beam pairs adjacent to the current optimal beam pair within the target optimal beam pair search range, and calculate the spectral efficiency change rate of each first neighboring beam pair compared to the current optimal beam pair according to the following formula:

[0081] ,

[0082] In the formula, represents the current optimal beam pair, represents the first neighboring beam pair, and .

[0083] When the spectral efficiency change rate of the first target neighboring beam pair within multiple first neighboring beams compared to the current optimal beam pair is higher than the preset first threshold, the transition probability of the first target neighboring beam can be calculated according to the following formula:

[0084] ,

[0085] In the formula, represents the predicted search radius, represents the maximum allowable search radius, represents the adjustment coefficient.

[0086] When the transition probability of the first target neighboring beam is higher than the preset second threshold, the first target neighboring beam is used as the optimal beam pair in the next round of search process, and the temperature is updated according to the following formula: .

[0087] (3) In the second round and each subsequent round of search process, use the neighboring beam pair switched in the previous round of search process as the current optimal beam pair in the current search round. Redetermine multiple second neighboring beam pairs adjacent to the current optimal beam pair within the target optimal beam pair search range, and determine the spectral efficiency change rate of each second neighboring beam pair compared to the current optimal beam pair.

[0088] When the spectral efficiency change rate of the second target neighboring beam pair among multiple second neighboring beam pairs compared to the current optimal beam pair is higher than the first threshold, calculate the transition probability of the second target neighboring beam according to the following formula:

[0089] ,

[0090] In the formula, represents the spectral efficiency change rate of the second target neighboring beam, represents the spectral efficiency change rate of the current optimal beam pair in the current search round, represents the temperature of the k-th round, and , , where K represents the maximum number of iterative searches.

[0091] When the transition probability of the second target adjacent beam is higher than the second threshold value, the second target adjacent beam is used as the optimal beam pair in the next round of search process, and the temperature of the current search round is updated according to the following formula: .

[0092] (4) Repeat the above step (3), and use the optimal beam pair of the target search round with a temperature lower than the preset third threshold value (such as approaching 0, that is, the algorithm stops accepting inferior solutions) as the optimal handover beam pair between the base station and the UAV at the second moment.

[0093] Therefore, through the above heuristic search method, the embodiment of the present application can dynamically adjust the search strategy in a changing communication environment to ensure that it always focuses on the most potential beam pairs instead of blindly searching the entire beam space. This search scheme can adapt to the communication requirements in different scenarios while maintaining high efficiency.

[0094] Compared with the traditional exhaustive search method, the above heuristic search method can quickly locate the optimal handover beam pair with a significant reduction in the number of searches, thereby significantly reducing the delay of the communication system and improving the communication efficiency. For example, if the base station has 128 antennas and 2 radio frequency links and adopts a fully connected structure; and the UAV under the coverage of the base station has 16 antennas and 2 links and also adopts a fully connected structure. Then, when using the traditional exhaustive search method and the heuristic search algorithm to search for the optimal handover beam pair between the base station and the UAV respectively, the comparison results of the required number of searches are as Figure 5 shown. It is not difficult to see that the number of searches required to obtain the optimal handover beam pair using the traditional exhaustive search method is 4194304, and the number of searches required to obtain the optimal handover beam pair using the heuristic search algorithm provided by the embodiment of the present application is 32 times. Therefore, the complexity of the heuristic search algorithm provided by the embodiment of the present application is reduced by 99.999%.

[0095] Embodiment 2

[0096] According to the embodiment of the present application, there is also provided an optimal handover beam pair search device for implementing the optimal handover beam pair search method in Embodiment 1, as Figure 6 shown. The optimal handover beam pair search device at least includes: an acquisition module 62, a prediction module 64, and a search module 66, where:

[0097] The acquisition module 62 is configured to acquire the target beam pair information map of multiple beam pairs between the base station and the target UAV within the coverage range of the base station at the first moment;

[0098] A prediction module 64, configured to analyze the target beam pair information graph by using a pre-trained search range prediction model, so as to obtain a corresponding target optimal beam pair search range, where the search range prediction model is constructed based on an improved deep residual network, and the improved deep residual network at least includes: a shallow feature extraction module, a deep feature extraction module based on a residual nesting structure, and a prediction output module;

[0099] A search module 66, configured to determine an optimal handover beam pair between the base station and the UAV at the second moment from the target optimal beam pair search range by using a heuristic algorithm, where the second moment is the next moment of the first moment.

[0100] The functions of the various modules of the optimal handover beam pair search device are described below in combination with a specific implementation process.

[0101] As an optional implementation manner, the acquisition module 62 may acquire the target beam pair information graph according to the following steps, including:

[0102] The first step: Acquire the communication quality parameters of each of the multiple beam pairs between the base station and the target UAV at the first moment.

[0103] Wherein, the communication quality parameters include at least one of the following: spectral efficiency, gain information, signal-to-interference-plus-noise ratio, etc.

[0104] The second step: Acquire the first communication state information of the base station at the first moment and the second communication state information of the target UAV at the first moment.

[0105] Wherein, the first communication state information at least includes: the first beam state information of each of the multiple beams of the base station, such as beam azimuth angle (determined by mechanical azimuth angle and electronic azimuth angle), beam downward tilt angle (determined by mechanical downward tilt angle and electronic downward tilt angle), beam horizontal width, beam vertical width, beam gain, beam activation time, etc. In addition, the first communication state information also includes: state information such as the height of the base station. These information jointly describe the spatial pointing, energy characteristics and life cycle of the base station beam.

[0106] The second communication state information at least includes: the second beam state information of each of the multiple beams of the target UAV, such as beam azimuth angle, beam downward tilt angle, etc. In addition, the second communication state information also includes: the latitude, longitude and altitude coordinates of the target UAV, three-dimensional velocity components, three-dimensional acceleration components, pitch angle, roll angle, the residence time of the target UAV at the base station, the residence time of the target UAV under each transmitting beam of the base station, etc. These information reflect the precise position, motion state of the target UAV and its communication persistence with the base station.

[0107] Step 3: For each beam pair, the first beam state information, the second beam state information, the communication quality parameters of the beam pair, the base station status information and the drone status information of each of the two beams in the beam pair are used as multiple features of the beam pair. The feature vector of the beam pair is composed of multiple features and the feature vector is standardized.

[0108] Step 4: Arrange the multiple beams of the base station and the multiple beams of the target UAV in ascending order of beam numbers as multiple matrix rows and multiple matrix columns, use the standardized eigenvectors of each beam pair as matrix elements, and construct the target beam pair information graph.

[0109] Furthermore, the prediction module 64 may use a pre-trained search range prediction model to analyze the target beam pair information graph to obtain a corresponding target optimal beam pair search range.

[0110] Optionally, the training process of the search range prediction model may include:

[0111] Step 1: Build a deep learning model including a shallow feature extraction module, a deep feature extraction module and a prediction output module;

[0112] Step 2: Obtain a training sample set and a sample label set, wherein the training sample set includes multiple beam pair information graphs between the base station and the UAV as training samples, and the sample label set includes the true optimal beam pair search range corresponding to each beam pair information graph as a sample label;

[0113] Step 3: Use the training sample set and sample label set to iteratively train the deep learning model to obtain the search range prediction model.

[0114] In the search range prediction model, the shallow feature extraction module usually includes several convolutional layers for extracting low-level features and local features of the beam pair information graph. The deep feature extraction module usually includes multiple residual blocks connected by short skip connections. Each residual block includes multiple convolutional layers, batch normalization layers, and activation function layers (such as ReLU) connected by residual connections, which are used to effectively extract high-level and abstract features based on the low-level features of the image. Among them, the output of each residual block is obtained by adding the input signal "residual" to the convolved signal, which avoids the problems of gradient disappearance and degradation when the network depth increases, enabling the model to learn more complex feature representations. The prediction output module predicts the predicted search range of the optimal handover beam pair between the base station and the UAV based on the high-level features and low-frequency features of the image output by the above two modules. This module generally includes two fully connected layers and a Dropout layer, and the connection relationship is the first fully connected layer, the Dropout layer, and the second fully connected layer. Among them, the first fully connected layer is used to convert the high-level features of the image into low-dimensional prediction outputs, and the second fully connected layer converts the features output by the Dropout layer into prediction outputs to obtain the predicted search range.

[0115] In the specific training process, the deep learning model can be iteratively trained using the training sample set and the sample label set in the following way to obtain the search range prediction model: For each training batch in the iterative training process, each training sample in the training batch is input into the deep learning model to obtain the predicted optimal beam pair search ranges output by the deep learning model. The target loss function is constructed using the predicted optimal beam pair search ranges and the corresponding sample labels, and the model parameters of the deep learning model are adjusted according to the target loss function until the model parameters converge to obtain the search range prediction model.

[0116] Among them, the above target loss function includes: an angle loss function, a search radius loss function, and a physical constraint loss function. Therefore, the expression of the above target loss function is: , where represents the angle loss function, represents the search radius loss function, represents the physical constraint loss function, , , respectively represent the adjustment coefficients of the angle loss function, the search radius loss function, and the physical constraint loss function. Among them:

[0117] The above angle loss function has the following expression:

[0118] ,

[0119] Wherein, N represents the total number of training samples in a training batch, represents the true beam azimuth angle of the i-th training sample, represents the predicted beam azimuth angle of the i-th training sample, represents the true beam elevation angle of the i-th training sample, represents the predicted beam elevation angle of the i-th training sample, represents a preset angle threshold, respectively represent weight coefficients.

[0120] Search radius loss function The asymmetric penalty function is used to encourage the predicted radius to be slightly larger than the true value to avoid missing detections. Therefore, the search radius loss function can be expressed as:

[0121] ,

[0122] Wherein, N represents the total number of training samples in a training batch, represents the true search radius of the i-th training sample, represents the predicted search radius of the i-th training sample, represents the preset maximum allowable search range, represents a regulation coefficient.

[0123] In addition, in the physical constraint loss function the beam direction gradient constraint function and the energy coverage constraint function are introduced. Therefore, . By introducing the two physical constraints, the prior knowledge such as the electromagnetic wave propagation law and the antenna characteristics is embedded into the neural network, improving the rationality and generalization ability of the model. Specifically:

[0124] Regarding the beam direction gradient constraint loss function , it can be directly obtained by taking the derivative of the beam direction pattern function of the base station. Among them, the base station beam direction pattern function describes the intensity distribution of the signals transmitted by the base station in different directions in space, and its derivative (gradient) reflects the rate of change of this intensity distribution with the direction. Therefore, in the embodiment of the present application, by taking the derivative of the beam direction pattern function of the base station in a certain direction (such as azimuth and elevation), the rate of change of the beam intensity in this direction can be obtained to set a reasonable gradient threshold as the beam direction constraint. Subsequently, if the gradient value of a certain beam pair exceeds the set threshold, it means that the directivity change of this beam pair is too fast, so this beam pair is not conducive to stable communication; on the contrary, if the gradient value of the beam pair is small, it indicates that the beam direction change of this beam pair is gentle and the communication stability is better.

[0125] Regarding the energy coverage constraint loss function , whose goal is to ensure that the predicted search radius contains sufficient energy to avoid missing detected beams. Therefore, when the energy coverage rate of the search range of the predicted optimal beam pair corresponding to the training sample is less than the preset energy threshold , the energy coverage constraint loss function increases; while when the energy coverage rate of the search range of the predicted optimal beam pair corresponding to the training sample is not less than the preset energy threshold , the energy coverage constraint loss function decreases. Among them, the expression of the energy coverage rate of the search range of the predicted optimal beam pair corresponding to the training sample can be written as:

[0126] .

[0127] Finally, since the search range of the target optimal beam pair obtained by model prediction is determined by the central beam pair and the search radius, and the search range of the target optimal beam pair includes multiple candidate beam pairs. Therefore, the search module 66 can use a heuristic algorithm to iteratively search for the optimal handover beam pair between the base station and the UAV at the second moment within the search range of the target optimal beam pair. Among them, the heuristic algorithms include: particle swarm optimization algorithm, simulated annealing algorithm.

[0128] Taking the simulated annealing algorithm as an example below, the above search process will be described.

[0129] (I) Initialization: Set the temperature and the cooling coefficient , where the initial temperature should be set high enough to accept more inferior solutions at the beginning of the search, increasing the randomness of the search, and the setting of the initial temperature usually depends on the scale and complexity of the problem.

[0130] (II) In the first round of the search process, first use the central beam pair (i.e., the intersection point within Figure 4 ) as the current optimal beam pair in the current search round, determine multiple first neighboring beam pairs adjacent to the current optimal beam pair within the search range of the target optimal beam pair, and calculate the spectral efficiency change rate of each first neighboring beam pair compared to the current optimal beam pair according to the following formula:

[0131] ,

[0132] In the formula, represents the current optimal beam pair, represents the first neighboring beam pair, and .

[0133] When the spectral efficiency change rate of the first target neighboring beam pair within multiple first neighboring beams is higher than a preset first threshold value compared to the current optimal beam pair, the transition probability of the first target neighboring beam can be calculated according to the following formula:

[0134] ,

[0135] In the formula, represents the predicted search radius, represents the maximum allowable search radius, represents the adjustment coefficient.

[0136] When the transition probability of the first target neighboring beam is higher than a preset second threshold value, the first target neighboring beam is used as the optimal beam pair in the next round of search process, and the temperature is updated according to the following formula: .

[0137] (3) In the second round and each subsequent round of search process, the neighboring beam pair switched in the previous round of search process is used as the current optimal beam pair in the current search round. Multiple second neighboring beam pairs adjacent to the current optimal beam pair are re-determined within the target optimal beam pair search range, and the spectral efficiency change rate of each second neighboring beam pair compared to the current optimal beam pair is determined.

[0138] When the spectral efficiency change rate of the second target neighboring beam pair among multiple second neighboring beam pairs is higher than the first threshold value compared to the current optimal beam pair, the transition probability of the second target neighboring beam is calculated according to the following formula:

[0139] ,

[0140] In the formula, represents the spectral efficiency change rate of the second target neighboring beam, represents the spectral efficiency change rate of the current optimal beam pair in the current search round, represents the temperature of the k-th round, and , , K represents the maximum number of iterative searches.

[0141] When the transition probability of the second target neighboring beam is higher than the second threshold value, the second target neighboring beam is used as the optimal beam pair in the next round of search process, and the temperature of the current search round is updated according to the following formula: .

[0142] (4) Repeat the above step (3), and use the optimal beam pair of the target search round with a temperature lower than a preset third threshold value (such as approaching 0, that is, the algorithm stops accepting inferior solutions) as the optimal switching beam pair between the base station and the UAV at the second moment.

[0143] Therefore, through the above heuristic search method, the search module 66 can dynamically adjust the search strategy in a changing communication environment to ensure that it always focuses on the most promising beam pairs instead of blindly searching the entire beam space. This search scheme can adapt to the communication requirements in different scenarios while maintaining high efficiency.

[0144] It should be noted that each module in the optimal handover beam pair search device in the embodiments of the present application corresponds one by one to each implementation step of the optimal handover beam pair search method in Embodiment 1. Since a detailed description has been given in Embodiment 1, details not shown in this embodiment can be referred to Embodiment 1 and will not be elaborated here.

[0145] Embodiment 3

[0146] According to an embodiment of the present application, there is also provided a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the optimal handover beam pair search method in Embodiment 1.

[0147] According to an embodiment of the present application, there is also provided a non-volatile storage medium, which includes a stored computer program. The device where the non-volatile storage medium is located executes the optimal handover beam pair search method in Embodiment 1 by running the computer program.

[0148] According to an embodiment of the present application, there is also provided a processor, which is used to run a computer program. When the computer program runs, it executes the optimal handover beam pair search method in Embodiment 1.

[0149] According to an embodiment of the present application, there is also provided an electronic device, which includes: a memory and a processor. A computer program is stored in the memory, and the processor is configured to execute the optimal handover beam pair search method in Embodiment 1 through the computer program.

[0150] Specifically, when the computer program runs, it executes the following steps: obtaining a target beam pair information map of multiple beam pairs between a base station and a target unmanned aerial vehicle within the coverage range of the base station at a first moment; analyzing the target beam pair information map by using a pre-trained search range prediction model to obtain a corresponding target optimal beam pair search range, where the search range prediction model is constructed based on an improved deep residual network, and the improved deep residual network at least includes: a shallow feature extraction module, a deep feature extraction module based on a residual nesting structure, and a prediction output module; using a heuristic algorithm to determine the optimal handover beam pair between the base station and the unmanned aerial vehicle at a second moment from the target optimal beam pair search range, where the second moment is the next moment of the first moment.

[0151] As an alternative embodiment, the above-mentioned electronic device may exist in the form of a mobile terminal, a computer terminal, or a similar computing device. Figure 7 A hardware structure block diagram of an electronic device for implementing an optimal handover beam pair search method is shown. As Figure 7 shown, the electronic device 70 may include one or more (shown as 702a, 702b,..., 702n in the figure) processors 702 (the processor 702 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 704 for storing data, and a transmission device 706 for communication functions. In addition, it may further include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply, and / or a camera. Those of ordinary skill in the art can understand that Figure 7 the structure shown is only illustrative and does not limit the structure of the above-mentioned electronic device. For example, the electronic device 70 may further include more or fewer components than Figure 7 shown, or have a different configuration from Figure 7 shown.

[0152] It should be noted that the above one or more processors 702 and / or other data processing circuits are generally referred to as "data processing circuits" in this document. The data processing circuit may be embodied in software, hardware, firmware, or any combination thereof, in whole or in part. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any one of the other elements in the electronic device 70. As involved in the embodiments of the present application, the data processing circuit is used for processor control (such as the selection of a variable resistor terminal path connected to an interface).

[0153] The memory 704 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the optimal handover beam pair search method in the embodiments of the present application. The processor 702 executes various functional applications and data processing by running the software programs and modules stored in the memory 704, that is, implements the vulnerability detection method of the above-mentioned application program. The memory 704 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, a flash memory, or other non-volatile solid-state memories. In some instances, the memory 704 may further include a memory remotely set relative to the processor 702, and these remote memories may be connected to the electronic device 70 through a network. Examples of the above network include, but are not limited to, the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0154] The transmission device 706 is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wireless network provided by the communication provider of the electronic device 70. In one example, the transmission device 706 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 706 can be a Radio Frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0155] The display can be, for example, a touch-screen liquid crystal display (LCD), which enables users to interact with the user interface of the electronic device 70.

[0156] The above-mentioned embodiment numbers are only for description and do not represent the advantages or disadvantages of the embodiments.

[0157] In the above embodiments of the present application, the descriptions of each embodiment have their own emphases. For the parts not detailed in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0158] In several embodiments provided in the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of units can be a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of units or modules can be in electrical or other forms.

[0159] The units described as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they can be located in one place or distributed to multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0160] In addition, in each embodiment of the present application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0161] When the integrated unit 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 such an understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods in various embodiments of this application. The aforementioned storage medium includes: various media such as USB flash drives, read-only memories (ROMs), random access memories (RAMs), mobile hard disks, magnetic disks, or optical discs that can store program codes.

[0162] The above are only the preferred embodiments of this application. It should be noted that for those of ordinary skill in the art in this technical field, without departing from the principle of this application, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of this application.

Claims

1. An optimal switching beam pair search method, characterized in that, Including: Obtaining a target beam pair information map of multiple beam pairs between a base station and a target unmanned aerial vehicle (UAV) within the coverage range of the base station at a first moment; Analyzing the target beam pair information map by using a pre-trained search range prediction model to obtain a corresponding target optimal beam pair search range, wherein the search range prediction model is constructed based on an improved deep residual network, and the improved deep residual network at least includes: a shallow feature extraction module, a deep feature extraction module based on a residual nesting structure, and a prediction output module; Using a heuristic algorithm to determine an optimal handover beam pair between the base station and the UAV at a second moment from the target optimal beam pair search range, wherein the second moment is the next moment of the first moment.

2. The method according to claim 1, characterized in that, Obtaining a target beam pair information map of multiple beam pairs between a base station and a target UAV within the coverage range of the base station at a first moment includes: Obtaining communication quality parameters of multiple beam pairs between the base station and the target UAV at the first moment respectively, wherein the communication quality parameters include at least one of the following: spectral efficiency, signal-to-interference-plus-noise ratio; Obtaining first communication state information of the base station at the first moment and second communication state information of the target UAV at the first moment, wherein the first communication state information at least includes: first beam state information of multiple beams of the base station respectively, base station state information, and the second communication state information at least includes: second beam state information of multiple beams of the target UAV respectively, UAV state information, and at least one of the first beam state information and the second beam state information includes: beam azimuth angle, beam elevation angle, beam width, beam gain; For each of the beam pairs, using the first beam state information, the second beam state information of the two beams within the beam pair, the communication quality parameter of the beam pair, and the base station state information and the UAV state information as multiple features of the beam pair, forming a feature vector of the beam pair from the multiple features, and performing normalization processing on the feature vector; Arranging multiple beams of the base station and multiple beams of the target UAV in ascending order of beam numbers as multiple matrix rows and multiple matrix columns, and using the normalized feature vectors of each of the beam pairs as matrix elements to construct the target beam pair information map.

3. The method according to claim 1, wherein The training process of the search range prediction model includes: Constructing a deep learning model including the shallow feature extraction module, the deep feature extraction module, and the prediction output module; Obtaining a training sample set and a sample label set, wherein the training sample set includes multiple beam pair information maps between a base station and a UAV as training samples, and the sample label set includes the true optimal beam pair search range corresponding to each of the beam pair information maps as a sample label; Iteratively training the deep learning model by using the training sample set and the sample label set to obtain the search range prediction model.

4. The method according to claim 3, wherein The shallow feature extraction module includes at least one convolutional layer, and the shallow feature extraction module is used to extract the low-level image features of the beam pair information map; The deep feature extraction module includes a plurality of residual blocks connected by short skip connections. Each residual block includes a plurality of convolutional layers, batch normalization layers, and activation function layers connected by residual connections, and the deep feature extraction module is used to extract high-level image features based on the low-level image features; The prediction output module includes two fully connected layers and a Dropout layer, and the prediction output module is used to convert the high-level image features into a predicted search range.

5. The method according to claim 3, wherein Using the training sample set and the sample label set to iteratively train the deep learning model to obtain the search range prediction model, including: For each training batch in the iterative training process, input the training samples of the training batch into the deep learning model to obtain the predicted optimal beam pair search ranges output by the deep learning model. Use the predicted optimal beam pair search ranges and the corresponding sample labels to construct an objective loss function, and adjust the model parameters of the deep learning model according to the objective loss function until the model parameters converge to obtain the search range prediction model, where the objective loss function includes: an angle loss function, a search radius loss function, and a physical constraint function.

6. The method according to claim 5, wherein The objective loss function includes: an angle loss function, a search radius loss function, and a physical constraint loss function, and the expression of the objective loss function is: ， In the formula, represents the angle loss function, represents the search radius loss function, represents the physical constraint loss function, , , respectively represent the adjustment coefficients of the angle loss function, the search radius loss function, and the physical constraint loss function, where: The angle loss function has the following expression: ， where N represents the total number of training samples within a training batch, represents the true beam azimuth angle of the i-th training sample, represents the predicted beam azimuth angle of the i-th training sample, represents the true beam elevation angle of the i-th training sample, represents the predicted beam elevation angle of the i-th training sample, represents a preset angle threshold, respectively represent weight coefficients; The search radius loss function has the following expression: ， Where N represents the total number of training samples within a training batch, represents the true search radius of the i-th training sample, represents the predicted search radius of the i-th training sample, represents the preset maximum allowable search range, represents the adjustment coefficient; The physical constraint loss function has the following expression: ， In the formula, represents the beam direction gradient constraint loss function, and the beam direction gradient constraint loss function is obtained by calculating the base station beam pattern function, represents the energy coverage constraint loss function, and when the energy coverage rate within the search range of the predicted optimal beam pair corresponding to the training sample is less than a preset energy threshold, the energy coverage constraint loss function increases in value, while when the energy coverage rate within the search range of the predicted optimal beam pair corresponding to the training sample is not less than the energy threshold, the energy coverage constraint loss function decreases in value.

7. The method according to claim 1, characterized in that, The target optimal beam pair search range is determined by the central beam pair and the search radius. Among them, using a heuristic algorithm to determine the optimal handover beam pair between the base station and the UAV at the second moment within the target optimal beam pair search range, including: Initial temperature and cooling coefficient ; In the first round of search, use the central beam pair as the current optimal beam pair in the current search round. Determine a plurality of first neighboring beam pairs adjacent to the current optimal beam pair within the target optimal beam pair search range, and calculate the spectral efficiency change rate of each first neighboring beam pair compared to the current optimal beam pair according to the following formula: ， In the formula, represents the current optimal beam pair, represents the first neighboring beam pair, and ; when the spectral efficiency change rate of the first target neighboring beam pair compared to the current optimal beam pair is higher than a preset first threshold, calculate the transition probability of the first target neighboring beam according to the following formula: ， wherein, represents the search radius; represents the maximum allowable search radius; represents the adjustment coefficient; when the transition probability of the first target adjacent beam is higher than a preset second threshold, the first target adjacent beam is used as the optimal beam pair in the next round of search process, and the temperature of the current search round is updated according to the following formula: ; In the second round and subsequent rounds of search, use the neighboring beam pair switched in the previous round of search as the current optimal beam pair in the current search round. Re-determine a plurality of second neighboring beam pairs adjacent to the current optimal beam pair within the target optimal beam pair search range, and determine the spectral efficiency change rate of each second neighboring beam pair compared to the current optimal beam pair; when the spectral efficiency change rate of the second target neighboring beam pair compared to the current optimal beam pair is higher than the first threshold, calculate the transition probability of the second target neighboring beam according to the following formula: ， wherein, represents the spectral efficiency change rate of the second target neighboring beam, represents the spectral efficiency change rate of the current optimal beam pair in the current search round, represents the temperature of the k-th round, and , , where K represents the maximum iterative search times; when the transition probability of the second target neighboring beam is higher than the second threshold, the second target neighboring beam is used as the optimal beam pair in the next search process, and the temperature of the current search round is updated; the above steps are repeatedly executed, and the optimal beam pair of the target search round with a temperature lower than the preset third threshold is used as the optimal handover beam pair between the base station and the UAV at the second moment.

8. An optimal handover beam pair search device, characterized in that Including: An acquisition module for acquiring the target beam pair information map of multiple beam pairs between the base station and the target UAV within the coverage range of the base station at the first moment; A prediction module, configured to analyze the target beam pair information graph by using a pre-trained search range prediction model to obtain a corresponding target optimal beam pair search range, where the search range prediction model is constructed based on an improved deep residual network, and the improved deep residual network at least includes: a shallow feature extraction module, a deep feature extraction module based on a residual nesting structure, and a prediction output module; A search module, configured to determine an optimal handover beam pair between the base station and the UAV at a second moment from the target optimal beam pair search range by using a heuristic algorithm, where the second moment is the next moment of the first moment.

9. A computer program product, characterized in that, Comprising: A computer program, where when the computer program is executed by a processor, it implements the optimal handover beam pair search method according to any one of claims 1 to 7.

10. An electronic device, characterized in that, Comprising: A memory and a processor, where a computer program is stored in the memory, and the processor is configured to execute the optimal handover beam pair search method according to any one of claims 1 to 7 through the computer program.

Images