Unmanned aerial vehicle flight communication control system based on single Beidou positioning

Abstract

The invention discloses an unmanned aerial vehicle flight communication control system based on single Beidou positioning. The system comprises a data acquisition module which acquires Beidou position data of an unmanned aerial vehicle through Beidou positioning; the strategy integration module is used for dividing a flight three-dimensional space into voxels based on the Beidou position data, counting the occurrence probability of the unmanned aerial vehicle in each voxel, constructing a space-time probability distribution diagram and determining an optimal coding strategy; the transmission optimization module is used for predicting the optimal prediction position of the unmanned aerial vehicle in the next five seconds by using an improved LSTM prediction model, comparing the optimal prediction position with an actual position to calculate a difference residual error, and then encoding according to an optimal encoding strategy according to a voxel region where the difference residual error is located to obtain optimal track information; and the flight control module obtains an optimal track based on the optimal track information and performs flight communication control of the unmanned aerial vehicle based on the optimal track, the data volume compression ratio can reach 67% or above, and the communication bandwidth requirement and the transmission delay are greatly reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of UAV control, and particularly to a UAV flight communication control system based on single Beidou positioning. Background Art

[0002] Currently, UAV flight control systems generally rely on multi-mode positioning modules (such as GPS / Beidou dual-mode) to ensure positioning accuracy, resulting in high hardware costs and increased system complexity. At present, in terms of UAV flight communication control, traditional methods directly transmit original track coordinate data (usually 10 bytes / point), which has problems of large data redundancy and high bandwidth occupancy. Especially in high-frequency position update scenarios (such as 10Hz sampling rate), the communication load increases sharply. At the same time, trajectory prediction mostly uses the standard LSTM model, and its 32-bit floating-point weight parameters result in a large model volume (usually hundreds of KB), consuming a large amount of computing resources and being difficult to be efficiently deployed on airborne embedded devices. Therefore, a UAV flight communication control system based on single Beidou positioning is proposed herein. Summary of the Invention

[0003] In order to overcome the above defects of the prior art and achieve the above object, the present invention proposes the following technical solutions: A UAV flight communication control system based on single Beidou positioning, comprising: A data acquisition module: acquiring Beidou position data of the UAV through Beidou positioning; A strategy integration module: based on the Beidou position data, dividing the flight three-dimensional space into voxels, statistically calculating the probability of the UAV appearing in each voxel, constructing a spatio-temporal probability distribution map, and determining an optimal coding strategy; A transmission optimization module: using an improved LSTM prediction model to predict the best predicted position of the UAV in the next 5 seconds, calculating the differential residual by comparing with the actual position, and then encoding according to the optimal coding strategy according to the voxel area where the differential residual is located to obtain optimal track information; Wherein, the improved LSTM prediction model includes an input layer, a binary LSTM layer, and an output layer; A flight control module: obtaining an optimal trajectory based on the optimal track information and performing UAV flight communication control based on the optimal trajectory.

[0004] The process of dividing the flight three-dimensional space into voxels is as follows: Based on the Beidou position data, dividing the three-dimensional space of the UAV flight into grids at intervals of 1m in the horizontal direction and into layers at intervals of 0.5m in the vertical direction to form a three-dimensional voxel structure.

[0005] The process of obtaining the spatio-temporal probability distribution map is as follows: Based on the collected Beidou positioning data of the UAV, the number of UAV appearances is statistically counted for each voxel, and the probability of UAV appearance within different voxels is obtained based on the number of UAV appearances; Traverse all Beidou position data, statistically count the probability of UAV appearance in each voxel, construct a probability distribution map, and use a three-dimensional graphics visualization tool to construct a spatio-temporal probability distribution map according to the probability value of each voxel.

[0006] The process of determining the optimal coding strategy is as follows: Based on the probability of UAV appearance within the voxel size, the voxels are divided into three categories, including: A preset probability threshold , ; When , this voxel is classified as a high-probability area; When , it is a medium-probability area; When , it is a low-probability area; Set different coding methods for different regions, including: Let the predicted appearance position of the UAV in the voxel be , and the initial position be , and the differential residual formula is expressed as ; The high-probability area uses 4-byte differential residual coding; The medium-probability area uses 3-byte differential residual coding; The low-probability area uses 2-byte coding.

[0007] The process of constructing the improved LSTM prediction model is as follows: Both the input layer and the input layer are set with 3 neurons; The binary LSTM layer uses 1-bit weights and activation values, and the weight matrix of the traditional LSTM is binary processed using the sign function sgn(y), where y is the independent variable of the LSTM layer; For each neuron j in the LSTM layer, it receives data from the input layer and the state information of the previous hidden layer , where t is the time; Let the binary weight matrices be the weight matrix connecting the input layer and the current neuron and the weight matrix connecting the previous hidden layer and the current neuron , and the bias vector is , and the calculation process of the binary LSTM layer neuron is:

[0008] Among them, is the activation function, is the output value after being processed by the activation function, is an intermediate variable, is the input position data. Through such binary weight calculation, the neuron can update its own state and complete the binary LSTM layer calculation task.

[0009] The process of obtaining the optimal flight path information is as follows: Taking 0.1 second as the time interval, the position information at the current moment and the previous moment is input into the improved LSTM prediction model according to the time series, and the predicted position points at multiple moments within the next 5 seconds are output to form a predicted trajectory; Based on the position information at different moments and arranged in sequence, the initial trajectory of the UAV is formed. The predicted trajectory is compared with the initial trajectory, and for the position points corresponding to each time interval, the differential residual is obtained; Determine a voxel division rule, and according to the optimal coding strategy, set a new probability threshold for coding; The coding data obtained through the optimal coding strategy is combined and arranged in chronological order to form the optimal flight path information.

[0010] The voxel division rule is as follows: Suppose in a three-dimensional space, the side lengths of the voxel in the x, y, and z directions are respectively , , , for the differential residual , the index of the voxel where it is located is obtained and expressed as: = , = , = ; Among them, represents the floor operation.

[0011] The process of setting a new probability threshold for coding is as follows: Based on the preset probability threshold , ; When , this voxel is divided into a high-probability area; When , it is a medium-probability area; When , it is a low-probability area; The high-probability region uses 4-byte differential residual coding, the medium-probability region uses 3-byte differential residual coding, and the low-probability region uses 2-byte coding.

[0012] The present invention has the following beneficial effects: In the present invention, first, based on the spatio-temporal probability distribution map, 4-byte differential residual coding is used for the high-probability region (to ensure accuracy), 3-byte coding is used for the medium-probability region (to balance accuracy and data volume), and 2-byte abbreviated coding is used for the low-probability region (only recording the general profile). Compared with the traditional absolute coordinate transmission of 10 bytes / point, the data volume compression ratio can reach 67% (from 500 bytes to 165 bytes for 50 points), significantly reducing the communication bandwidth requirement and transmission delay (from 200 ms to within 80 ms). Secondly, the binary LSTM model predicts the future 5-second flight path. Only the deviation between the actual path and the predicted value is transmitted through differential residual coding, further reducing redundant data. Especially when the UAV is flying smoothly (such as in straight-line cruising), the proportion of the high-probability region increases, and the compression effect is more significant. The binary LSTM model is trained through historical data to control the future 5-second flight path prediction error within 0.3 m (the error of the traditional model is about 1 m). Combining with the high-precision region adaptation of differential residual coding, the deviation between the reconstructed path and the actual position is less than 0.5 m, meeting the precise control requirements of inspection, logistics and other scenarios. Finally, only the position change amount is transmitted through differential residual coding, which is insensitive to the random noise in the Beidou signal. Even in weak signal areas (such as canyons and between buildings), the control stability can still be maintained through the precise coding of the high-probability region. The compressed data volume is reduced, the power consumption of the communication module is reduced by about 30%, and the endurance time of the UAV is increased by more than 15%, meeting the requirements of long-endurance missions. Description of the Drawings

[0013] Figure 1 It is a system block diagram of the UAV flight communication control system based on single Beidou positioning proposed by the present invention. Detailed Embodiments

[0014] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0015] Embodiment: As Figure 1 shown, the UAV flight communication control system based on single Beidou positioning proposed by the present invention includes: Data acquisition module: Collect Beidou position data of the UAV through Beidou positioning; Use single Beidou positioning to continuously collect the initial Beidou position data of the UAV at a frequency of 10 Hz; Specifically, the initial Beidou position data covers data of various flight scenarios, including when flying in a straight line, maintaining a fixed flight speed and direction, and recording positions at different times; in the turning scenario, set different turning radii and angles to obtain corresponding position information; during ascent and descent, change the altitude at different rates while collecting position information, and these position information constitute the Beidou position data; The collected initial Beidou position data has different value ranges in the x, y, and z dimensions. Through the min-max normalization method, it is mapped to the interval [0, 1]. After normalization, the Beidou position data is obtained , where t is the corresponding time, and x, y, and z represent coordinate values respectively, serving as the input of the transmission optimization module.

[0016] Strategy integration module: Based on the Beidou position data, divide the three-dimensional flight space into voxels, count the probability of the UAV appearing in each voxel, construct a spatio-temporal probability distribution map, and determine the optimal coding strategy; The process of constructing the spatio-temporal probability distribution map is as follows: Based on the Beidou position data, the three-dimensional space in which the UAV flies is carefully divided. In the horizontal direction, the grid is divided at intervals of 1 m, and in the vertical direction, the layers are divided at intervals of 0.5 m to form a regular three-dimensional voxel structure V; Specifically, this division accuracy ensures the accuracy of the UAV position description while taking into account the computational amount and storage amount; According to the collected Beidou positioning data of the UAV, count the number of times the UAV appears in each voxel one by one. Based on the number of times the UAV appears, obtain the probability of the UAV appearing in different voxels. Let the total amount of collected data be N, and the number of times the UAV appears in voxel V be n, then the probability of the UAV appearing in this voxel ; Traverse all the Beidou position data, count the probability values of each voxel, construct a probability distribution map, and use a three-dimensional graphics visualization tool to construct a spatio-temporal probability distribution map according to the calculated probability values of each voxel; Specifically, in the spatio-temporal probability distribution map, different probability values can be represented by different colors or transparencies through a three-dimensional graphics visualization tool, intuitively showing the probability distribution of the UAV in the three-dimensional space, which is convenient for subsequent analysis and coding strategy formulation; The process of determining the optimal coding strategy is as follows: Based on the probability of the UAV appearing in the voxel size, divide the voxels into three categories, including: Preset probability threshold , ; When , the voxel is divided into a high - probability region, indicating that the UAV frequently appears in this region; When , it is a medium - probability region; When , it is a low - probability region, that is, the possibility of the UAV appearing in this region is extremely small; Set different coding methods for different regions, including: Set the predicted appearance position of the UAV in the voxel as , and the initial position as , and the differential residual formula is expressed as ; High - probability region: Since the UAV has a high probability of appearing in this region and requires high precision for position information, 4 - byte differential residual coding is adopted. By encoding , data redundancy is effectively reduced. This coding method is encoded based on the difference between the predicted position and the actual position, and is expressed as ; Among them, represents the coding function of the high - probability region. Through a specific coding rule (adopting 4 - byte differential residual coding), is encoded to ensure high precision while effectively reducing data redundancy; Medium - probability region: 3 - byte differential residual coding is adopted, that is, on the premise of ensuring a certain precision, the data transmission volume is reduced, and the precision and data volume are balanced. Encoding is based on the differential residual, and the coding function is denoted as ; Among them, represents the coding function of the medium - probability region (adopting 3 - byte differential residual coding). Through this coding method, a trade - off is made between precision and data volume, and under the condition of meeting a certain precision requirement, the space required for data transmission and storage is reduced as much as possible.

[0017] Low - probability region: 2 - byte abbreviated coding method is adopted, mainly recording the approximate position range. Because the possibility of the UAV appearing in this region is small and high - precision coding is not required, let the abbreviated coding function be ; Among them, represents the coding function of the low - probability region (adopting 2 - byte differential residual coding). The coding function directly abbreviates the actual position to reduce the data volume.

[0018] Transmission Optimization Module: Use the improved LSTM prediction model to predict the best predicted position of the UAV in the next 5 seconds, calculate the differential residual by comparing with the actual position, and then encode according to the optimal coding strategy in the voxel region where the differential residual is located to obtain the optimal flight path information; The construction process of the improved LSTM prediction model is as follows: The position data after normalization is sequentially input into an LSTM model containing an input layer, an LSTM layer, and an output layer according to the time sequence; Adjust the input layer: The input layer is set with 3 neurons, corresponding to the coordinate values of the x, y, and z dimensions respectively; This design can realize the preliminary access of position information and provide basic data for the subsequent processing of the model; The input layer inputs the normalized position data into the model to provide an initial input value for the next layer of calculation; Adjust and improve the LSTM layer to construct a binary LSTM layer. The implementation process is as follows: Use an LSTM layer with 1-bit weights and activation values; Perform binary processing on the weight matrix of the traditional LSTM using the sign function sgn(y), where y is the independent variable of the LSTM layer, that is, the elements in the traditional LSTM weight matrix; When then ; When then ; For each neuron j in the LSTM layer , M is the number of neurons, receiving data from the input layer and the state information of the hidden layer at the previous moment , t is the time; In the traditional LSTM, the calculation process involves complex weight matrix multiplication and accumulation operations. In the binary LSTM layer, let the binary weight matrices be (connecting the input layer to the current neuron) and (connecting the hidden layer at the previous moment to the current neuron), and the bias vector is , then the calculation process of the neurons in the binary LSTM layer is:

[0019] Among them, is the activation function, is the output value after being processed by the activation function, is an intermediate variable, For the input position data, through such binary weight calculation, the neuron can update its own state, thus completing the calculation task of this layer; Output layer adjustment: The output layer sets 3 neurons, corresponding to predicting the coordinate values of the drone in the x, y, and z dimensions at future moments; To ensure that the output value can directly and accurately reflect the coordinate change situation, and avoid the complex calculations and possible information loss caused by non-linear transformation, the output layer uses a linear activation function; Let the neuron output vector of the output layer be , after being processed by the linear activation function, the output value directly corresponds to the predicted coordinate change amount, that is Corresponding to the predicted change in the x dimension, Corresponding to the predicted change in the y dimension, Corresponding to the predicted change in the z dimension, t is the moment, to more quickly realize the output function of the coordinate prediction value; Model training and evaluation: Divide the preprocessed Beidou positioning data according to the ratio of 8:2, use the constructed model to predict the test set data, calculate the mean squared error MAE. If the mean squared error value is less than the preset threshold of 0.01 and the value is within the acceptable range (less than 0.1), then the model performance is considered to meet the standard, otherwise it needs to be further optimized; If the model performance does not meet the requirements, then adjust the number j of neurons in the binary LSTM layer. Increasing the number of neurons can improve the model's expression ability, and decreasing the number of neurons will have the opposite effect. Find the optimal number through multiple experiments, and finally complete the model training and evaluation; Example: The weight parameters of the traditional LSTM model are usually 32-bit or 64-bit floating-point values, with large storage and calculation overheads. While the improved LSTM prediction model uses 1-bit weights, limiting the weight values to +1 or -1; Suppose the weight matrix scale of a certain layer of the traditional LSTM is 100×100, and each weight parameter is 32 bits (4 bytes), then the storage of this matrix requires 100×100×4 = 40000 bytes; After binaryization, the same-scale weight matrix only requires 100×100×1 = 10000 bits, that is, 1250 bytes. The parameter storage amount is greatly reduced. In the actual model, the number of parameters is less than 5KB. Compared with the traditional AI algorithm, the model volume is reduced to ; In actual tests, for the drone trajectory prediction task, a certain number of test data samples (100 groups) were collected. When the traditional LSTM model predicted the drone's future 5-second trajectory, the average prediction error was about 1m; The improved LSTM prediction model controls the flight path prediction error of the UAV within 0.3 m in the next 5 seconds, and the prediction is significantly improved; The process of obtaining the optimal flight path information is as follows: Using the trained improved LSTM prediction model (improved LSTM prediction model), accurately predict the flight path of the UAV in the next 5 seconds; Let seconds be the time interval, and input the position information at the current moment and the previous moment into the improved LSTM prediction model (improved LSTM model) according to the time series. The neurons inside the model calculate through binary weights and output the predicted trajectory of multiple moments within the next 5 seconds; That is, when, the total output is predicted position points, and these position points form the predicted trajectory , where N is the number of predicted position points of the predicted trajectory, is the predicted position point, which is represented as in the three-dimensional space; Specifically, is the predicted position point of the UAV in the three-dimensional space at the future moment output by the improved LSTM prediction model; Collect the initial trajectory L, continuously collect the position information of the UAV at different moments through the Beidou positioning function, and arrange them in sequence to form the initial trajectory of the UAV, where is the initial position point; Compare the predicted trajectory with the initial trajectory , and for each position point corresponding to the time interval, obtain the differential residual ; where is a three-dimensional vector, representing the precise deviation value between the best predicted position and the actual position at each moment; In order to determine the voxel region where each differential residual is located, first clarify the voxel division rule; Suppose in the three-dimensional space, the side lengths of the voxel in the x, y, and z directions are , , , for the differential residual , the index of the voxel where it is located is obtained and expressed as: = , = , = ; Among them, Denotes the floor operation; According to the optimal coding strategy determined previously, set a new probability threshold and ; If the voxel where a certain differential residual is located belongs to the high-probability region (assuming the probability threshold of the high-probability region is , that is, the probability of the UAV appearing in this voxel ), then use the 4-byte coding method (coding function ) to encode it; If , located in the medium-probability region, is the probability threshold of the low-probability region, use the 3-byte coding method (coding function ) ; If is located in the low-probability region, use the 2-byte coding method (coding function ) ; Specifically, in this way, the UAV flight path information is efficiently compressed and encoded, greatly reducing the data transmission volume compared with the traditional absolute coordinate transmission of 10 bytes per time; For example, the traditional method requires 10N bytes to transmit N position points, while after this coding compression, the data transmission volume can be greatly reduced according to the coding conditions of different regions, effectively improving the data transmission efficiency; Combining the encoded data of the high, medium, and low probability regions in chronological order constitutes the optimal flight path information; Example: Suppose a UAV is performing a mission in a certain flight area, and this area is divided into three-dimensional voxels (1m interval in the horizontal direction and 0.5m interval in the vertical direction); Count the number of times the UAV appears in each voxel to obtain the probability distribution of the following typical voxels: High-probability region: Voxel A (coordinate range: x = 10 - 11m, y = 20 - 21m, z = 5 - 5.5m), the number of appearances is 5,000 times, probability (p = 5000 / 10000 = 0.5 (close to the threshold α = 0.5, classified into the high-probability area); Medium-probability region: Voxel B (x = 15 - 16m, y = 18 - 19m, z = 3 - 3.5m), the number of appearances is 1,500 times, probability (p = 0.15) (between β = 0.1 and α = 0.5, classified into the medium-probability area); Low-probability region: Voxel C (x = 25 - 26m, y = 5 - 6m, z = 8 - 8.5m), the number of appearances is 500 times, probability (p = 0.05) (lower than β = 0.1, classified into the low-probability area); Predict the future 5-second flight path of the UAV (with an interval of 0.1 second, a total of 50 points) using an improved LSTM model. The prediction errors (differential residuals) of 3 typical points are as follows: Point 1 (located in voxel A, high-probability area): Predicted position point m; Initial position point m; Differential residual m, encoded using a 4-byte differential residual encoding function for encoding; Point 2 (located in voxel B, medium-probability area): Predicted position point m; Initial position point m, Differential residual m, encoded using a 3-byte differential residual encoding function for encoding; Point 3 (located in voxel C, low-probability area): Initial position point m, encoded using 2-byte abbreviated encoding for encoding, only recording the approximate area coordinates (such as x = 25 - 26m, y = 5 - 6m, z = 8 - 9m); Data compression effect: Traditional absolute coordinate transmission requires 10 bytes per point, and a total of 500 bytes for 50 points; After optimization, the high-probability area (20 points) is 20 × 4 = 80 bytes, the medium-probability area (25 points): 25 × 3 = 75 bytes, and the low-probability area (assuming 5 points): 5 × 2 = 10 bytes, The total data volume is 165 bytes, and the transmission optimization (compression ratio) is 67%.

[0020] Flight control module: Obtain the optimal trajectory based on the optimal flight path information, and perform UAV flight communication control based on the optimal trajectory; After receiving the compressed and encoded flight path information, the flight control module parses the encoded data of each area in chronological order. For the high / medium-probability area data: Restore the differential residual through the decoding function , and combine it with the predicted trajectory to calculate the optimal trajectory ; Finally, reconstruct the complete optimal trajectory , where is the corrected optimal initial position point; Compare the reconstructed optimal trajectory with the initial trajectory Compare and calculate the position deviation ; Generate a control command based on the deviation value: If exceeds the allowable error threshold (0.5 m), trigger the PID controller to adjust the flight parameters: The horizontal deviation (represented by x and y in the position points) is corrected by adjusting the motor speed difference or the servo angle to correct the heading; The vertical deviation (represented by z in the position points) is controlled by adjusting the throttle output to control the ascent and descent speed; If the deviation is within the threshold, maintain the current flight state.

[0021] In the application, several formulas involved are calculated by taking their numerical values after dimensionless. The establishment of the formulas is obtained by collecting a large amount of data for software simulation to get a formula closest to the real situation. Some coefficients or weights in the formulas are set by those skilled in the art according to the actual situation, so no more details will be elaborated here.

[0022] The above embodiments can be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution.

[0023] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A drone flight communication control system based on single Beidou positioning, characterized in that It includes: Data acquisition module: Collect Beidou position data of the UAV through Beidou positioning. Strategy integration module: Based on the Beidou position data, divide the three-dimensional flight space into voxels, count the occurrence probability of UAVs in each voxel, construct a spatio-temporal probability distribution map, and determine the optimal coding strategy. Transmission optimization module: Use the improved LSTM prediction model to predict the best predicted position of the UAV in the next 5 seconds, calculate the differential residual by comparing with the actual position, and then encode according to the optimal coding strategy according to the voxel area where the differential residual is located to obtain the optimal flight path information. Among them, the improved LSTM prediction model includes an input layer, a binary LSTM layer, and an output layer. Flight control module: Obtain the optimal trajectory based on the optimal flight path information, and perform UAV flight communication control based on the optimal trajectory.

2. The UAV flight communication control system based on single Beidou positioning according to claim 1, characterized in that The process of dividing the three-dimensional flight space into voxels is as follows: Based on the Beidou position data, divide the three-dimensional space where the UAV flies into grids at an interval of 1m in the horizontal direction and into layers at an interval of 0.5m in the vertical direction to form a three-dimensional voxel structure.

3. The UAV flight communication control system based on single Beidou positioning according to claim 2, characterized in that, The process of obtaining the spatio-temporal probability distribution map is as follows: According to the collected Beidou positioning data of the UAV, count the number of UAV appearances in each voxel one by one, and obtain the occurrence probability of UAVs in different voxels based on the number of UAV appearances. Traverse all Beidou position data, count the occurrence probability of UAVs in each voxel, construct a probability distribution map, and use a three-dimensional graphics visualization tool to construct a spatio-temporal probability distribution map according to the probability value of each voxel.

4. The UAV flight communication control system based on single Beidou positioning according to claim 3, wherein The process of determining the optimal coding strategy is as follows: According to the probability of the appearance of the drone within the voxel size, the voxels are divided into three categories, including: Predetermined probability threshold , ; When the voxel is divided into a high-probability region; When it is the medium probability region; When it is a low-probability area; Set different coding methods for different regions, including: Let the predicted appearance position of the UAV in the voxel be , and the initial position be . The differential residual formula is expressed as ; The high-probability region uses 4-byte differential residual coding. The medium-probability region uses 3-byte differential residual coding. The low-probability region uses 2-byte coding.

5. The drone flight communication control system based on single Beidou positioning according to claim 1, characterized in that, The process of constructing the improved LSTM prediction model is as follows: Both the input layer and the input layer are set with 3 neurons. The binary LSTM layer uses 1-bit weights and activation values, and performs binary processing on the weight matrix of the traditional LSTM using the sign function sgn(y), where y is the independent variable of the LSTM layer. For each neuron j in the LSTM layer, it receives data from the input layer and the state information of the hidden layer at the previous moment , where t is the moment; Let the weight matrices after binarization be the weight matrix connecting the input layer and the current neuron and the weight matrix connecting the hidden layer at the previous moment and the current neuron , and the bias vector is . The calculation process of the neurons in the binarized LSTM layer is as follows: ； Among them, is the activation function, is the output value after being processed by the activation function, is an intermediate variable, is the input position data. Through such binary weight calculation, the neuron can update its own state and complete the binary LSTM layer calculation task.

6. The UAV flight communication control system based on single Beidou positioning according to claim 5, wherein, The process of obtaining the optimal flight path information is as follows: Taking 0.1 second as the time interval, input the position information at the current moment and the previous moment into the improved LSTM prediction model in time series, and output the predicted position points at multiple moments within the next 5 seconds to form a predicted trajectory. Based on the position information at different moments and arranging them in sequence, form the initial trajectory of the UAV, compare the predicted trajectory with the initial trajectory, and obtain the differential residual for each position point corresponding to the time interval. Determine a voxel division rule, and set a new probability threshold for coding according to the optimal coding strategy. Combine and arrange the coding data obtained through the optimal coding strategy in chronological order to form the optimal flight path information.

7. The UAV flight communication control system based on single Beidou positioning according to claim 6, wherein The voxel division rule is as follows: Suppose in a three-dimensional space, the side lengths of a voxel in the x, y, and z directions are respectively , , . For the differential residual , the index of the voxel where it is located is obtained and expressed as: = , = , = ; Among them, represents the floor operation.

8. The UAV flight communication control system based on single Beidou positioning according to claim 6, characterized in that The process of setting a new probability threshold for coding is as follows: Based on a preset probability threshold , ; When occurs, the voxel is divided into a high-probability region; When it is the medium probability region; When it is a low-probability area; The high-probability region uses 4-byte differential residual coding, the medium-probability region uses 3-byte differential residual coding, and the low-probability region uses 2-byte coding.

Images