Unmanned aerial vehicle flight communication control system based on single-beidou positioning

Abstract

The application discloses a UAV flight communication control system based on single Beidou positioning, comprising: a data acquisition module: collecting Beidou position data of the UAV through Beidou positioning; a strategy integration module: based on the Beidou position data, dividing a flight three-dimensional space into voxels, counting the probability of the UAV appearing in each voxel, constructing a space-time 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, comparing the actual position to calculate the differential residual, and then coding according to the voxel area where the differential residual is located according to the optimal coding strategy to obtain optimal track information; a flight control module: obtaining an optimal trajectory based on the optimal track information, and controlling the UAV flight communication based on the optimal trajectory, so that the data compression ratio can reach more than 67%, and the communication bandwidth demand and transmission delay are greatly reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) control, and in particular to a UAV flight communication control system based on single Beidou positioning. Background Art

[0002] Current 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;

[0003] Currently, in the field of UAV flight communication control, the traditional method directly transmits the original track coordinate data (usually requiring 10 bytes / point), which has the 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, track prediction mostly uses the standard LSTM model, whose 32-bit floating-point weight parameters lead to a large model size (usually hundreds of KB), high computing resource consumption, and difficult to efficiently deploy on airborne embedded devices. Therefore, a UAV flight communication control system based on single Beidou positioning is proposed here. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art and to achieve the above-mentioned objectives, the present invention proposes the following technical solutions:

[0005] The UAV flight communication control system based on single Beidou positioning includes:

[0006] Data collection module: collects Beidou location data of drones through Beidou positioning;

[0007] Strategy Integration Module: Based on BeiDou location data, the three-dimensional flight space is divided into voxels, the probability of drone appearance in each voxel is calculated, a spatiotemporal probability distribution map is constructed, and the optimal encoding strategy is determined;

[0008] Transmission optimization module: Use the improved LSTM prediction model to predict the best predicted position of the drone in the next 5 seconds, compare it with the actual position to calculate the difference residual, and then encode it according to the optimal encoding strategy based on the voxel area where the difference residual is located to obtain the optimal track information;

[0009] The improved LSTM prediction model includes an input layer, a binary LSTM layer, and an output layer;

[0010] Flight control module: obtains the optimal trajectory based on the optimal track information, and performs UAV flight communication control based on the optimal trajectory.

[0011] The process of dividing the flight three-dimensional space into voxels is as follows:

[0012] Based on Beidou position data, the three-dimensional space where the drone flies is divided into grids with an interval of 1m in the horizontal direction and divided into layers with an interval of 0.5m in the vertical direction to form a three-dimensional voxel structure.

[0013] The process of obtaining the spatiotemporal probability distribution map is as follows:

[0014] Based on the collected Beidou positioning data of drones, the number of drone appearances is counted voxel by voxel, and the probability of drone appearance in different voxels is obtained based on the number of drone appearances;

[0015] Traverse all Beidou location data, count the probability of drone appearance in each voxel, build a probability distribution map, and use 3D graphics visualization tools to build a spatiotemporal probability distribution map based on the probability value of each voxel.

[0016] The optimal coding strategy determination process is as follows:

[0017] Based on the probability of drone appearance within the voxel Size,classifies voxels into three categories, including:

[0018] Preset probability threshold , ;

[0019] when When , the voxel is divided into a high probability area;

[0020] when When , it is the medium probability area;

[0021] when When , it is a low probability area;

[0022] Set different encoding methods for different regional codes, including:

[0023] Assume that the predicted appearance position of the drone in the voxel is , the initial position is , the difference residual formula is expressed as ;

[0024] The high probability area is encoded using 4-byte differential residual;

[0025] The medium probability region is encoded using a 3-byte differential residual;

[0026] The low probability area is encoded in a 2-byte manner.

[0027] The improved LSTM prediction model construction process is as follows:

[0028] The input layer and the output layer are both provided with 3 neurons;

[0029] The binary LSTM layer uses 1-bit weights and activation values, and uses the symbolic function sgn(y) to binaryize the weight matrix of the traditional LSTM, where y is the independent variable of the LSTM layer.

[0030] 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. , t is the time;

[0031] Assume that the binary weight matrix is ​​the weight matrix connecting the input layer and the current neuron And connect the previous hidden layer with the current neuron , the bias vector is , the calculation process of the binary LSTM layer neurons is:

[0032]

[0033] in, 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 and complete the binary LSTM layer calculation task.

[0034] The optimal track information acquisition process is as follows:

[0035] At 0.1 second intervals, the current and previous position information is input into the improved LSTM prediction model in time series, and the predicted position points at multiple moments in the next 5 seconds are output to form a predicted trajectory.

[0036] The initial trajectory of the UAV is constructed by arranging the position information at different times in sequence. The predicted trajectory is compared with the initial trajectory, and the difference residual is obtained for the position point corresponding to each time interval.

[0037] Determine a voxel partitioning rule and set a new probability threshold for encoding based on the optimal encoding strategy;

[0038] The coded data obtained through the optimal coding strategy are combined and arranged in chronological order to form the optimal track information.

[0039] The voxel partitioning rule is:

[0040] Assume that in three-dimensional space, the side lengths of the voxel in the x, y, and z directions are , , , for the difference residual , the index of the voxel where it is located Get the representation: = , = , = ;

[0041] in, Indicates a floor operation.

[0042] The process of setting a new probability threshold for encoding is:

[0043] Based on a preset probability threshold , ;

[0044] when When , the voxel is divided into a high probability area;

[0045] when When , it is the medium probability area;

[0046] when When , it is a low probability area;

[0047] The high probability region adopts 4-byte differential residual coding, the medium probability region adopts 3-byte differential residual coding, and the low probability region adopts 2-byte coding.

[0048] The present invention has the following beneficial effects:

[0049] In this invention, based on the spatiotemporal probability distribution map, 4-byte differential residual coding is used for high-probability areas (to ensure accuracy), 3-byte coding is used for medium-probability areas (to balance accuracy and data volume), and 2-byte simplified coding is used for low-probability areas (to only record the general outline). Compared with the traditional 10-byte / point absolute coordinate transmission, the data compression ratio can reach more than 67% (50 points are reduced from 500 bytes to 165 bytes), significantly reducing communication bandwidth requirements and transmission delay (from 200ms to within 80ms).

[0050] Secondly, the binary LSTM model predicts the trajectory for the next 5 seconds. Using differential residual coding, it only transmits the deviation between the actual trajectory and the predicted value, further reducing redundant data. This is especially true when the drone is in stable flight (such as a straight cruise), as the proportion of high-probability areas increases, resulting in a more significant compression effect. The binary LSTM model, trained on historical data, keeps the prediction error for the next 5 seconds within 0.3 meters (compared to the approximately 1 meter error for traditional models). Combined with the high-precision regional adaptation of differential residual coding, the deviation between the reconstructed trajectory and the actual position is less than 0.5 meters, meeting the precise control requirements of inspection, logistics, and other scenarios.

[0051] Finally, differential residual coding is used to transmit only the position change, which is insensitive to random noise in Beidou signals. Even in areas with weak signals (such as canyons and between buildings), control stability can still be maintained through precise coding in high-probability areas. The amount of compressed data is reduced, the power consumption of the communication module is reduced by about 30%, and the flight time of the UAV is increased by more than 15%, adapting to the needs of long-duration missions. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is the system block diagram of the UAV flight communication control system based on single Beidou positioning proposed in this invention. DETAILED DESCRIPTION

[0053] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0054] Example: Figure 1 As shown, the UAV flight communication control system based on single Beidou positioning proposed by the present invention includes:

[0055] Data collection module: collects Beidou location data of drones through Beidou positioning;

[0056] Using single BeiDou positioning, the initial BeiDou position data of the drone is continuously collected at a frequency of 10Hz;

[0057] Specifically, the initial Beidou position data covers data from various flight scenarios, including maintaining a fixed flight speed and direction during straight-line flight and recording positions at different times; setting different turning radii and angles in turning scenarios to obtain corresponding position information; and changing altitude at different rates during ascent and descent while collecting position information. This position information constitutes the Beidou position data.

[0058] The initial Beidou position data collected has different value ranges in the x, y, and z dimensions. Through the minimum-maximum normalization method, it is mapped to the [0, 1] interval. After normalization, the Beidou position data is obtained. , where t is the corresponding time, x, y, and z represent the coordinate values, which serve as the input of the transmission optimization module.

[0059] Strategy Integration Module: Based on BeiDou location data, the three-dimensional flight space is divided into voxels, the probability of drone appearance in each voxel is calculated, a spatiotemporal probability distribution map is constructed, and the optimal encoding strategy is determined;

[0060] The process of constructing the spatiotemporal probability distribution map is as follows:

[0061] Based on Beidou position data, the three-dimensional space where the drone flies is divided into grids with a horizontal interval of 1m and vertical intervals of 0.5m, forming a regular three-dimensional voxel structure V.

[0062] Specifically, this division accuracy ensures the accuracy of the UAV position description while taking into account the amount of computation and storage;

[0063] According to the collected Beidou positioning data of drones, the number of drone appearances is counted voxel by voxel, and the probability of drone appearance in different voxels is obtained based on the number of drone appearances. Assuming that the total amount of collected data is N and the number of drone appearances in voxel V is n, the probability of drone appearance in the voxel is ;

[0064] Traverse all Beidou position data, count the probability value of each voxel, build a probability distribution map, and use 3D graphics visualization tools to build a spatiotemporal probability distribution map based on the calculated probability value of each voxel;

[0065] Specifically, in the spatiotemporal probability distribution graph, different colors or transparencies can be used through 3D graphic visualization tools to represent different probability values, intuitively displaying the probability distribution of drones in 3D space, facilitating subsequent analysis and coding strategy formulation;

[0066] The optimal coding strategy determination process is:

[0067] Based on the probability of drone appearance within the voxel Size,classifies voxels into three categories, including:

[0068] Preset probability threshold , ;

[0069] when When , the voxel is divided into a high-probability area, indicating that drones appear frequently in this area;

[0070] when When , it is the medium probability area;

[0071] when When , it is a low probability area, that is, the possibility of drones appearing in this area is extremely small;

[0072] Set different encoding methods for different regional codes, including:

[0073] Assume that the predicted appearance position of the drone in the voxel is , the initial position is , the difference residual formula is expressed as ;

[0074] High probability area: Since the probability of drones appearing in this area is high, the position information accuracy is required to be high. 4-byte differential residual coding is used. Encoding effectively reduces data redundancy. This encoding method is based on the difference between the predicted position and the actual position, which is expressed as ;

[0075] in, The coding function representing the high probability area is encoded by a specific coding rule (using 4-byte differential residual coding). Perform encoding processing to ensure high accuracy while effectively reducing data redundancy;

[0076] Medium probability area: 3-byte differential residual coding is used, that is, under the premise of ensuring a certain accuracy, the data transmission volume is reduced, the accuracy and data volume are balanced, and the coding is based on the differential residual. The coding function is recorded as ;

[0077] in, The coding function representing the medium probability region (using 3-byte differential residual coding) uses this coding method to strike a balance between accuracy and data volume, minimizing the space required for data transmission and storage while meeting certain accuracy requirements.

[0078] Low probability area: 2-byte simplified encoding is used to record the approximate location range. Since the possibility of drones appearing in this area is small, high-precision encoding is not required. The simplified encoding function is set as ;

[0079] in, The encoding function representing the low probability area (using 2-byte differential residual encoding) is directly related to the actual position Perform simplification to reduce the amount of data.

[0080] Transmission optimization module: Use the improved LSTM prediction model to predict the best predicted position of the drone in the next 5 seconds, compare it with the actual position to calculate the difference residual, and then encode it according to the optimal encoding strategy based on the voxel area where the difference residual is located to obtain the optimal track information;

[0081] The improved LSTM prediction model construction process is as follows:

[0082] The normalized position data , which are sequentially input into an LSTM model consisting of an input layer, an LSTM layer, and an output layer;

[0083] Adjust the input layer:

[0084] The input layer has three neurons, corresponding to the coordinate values ​​of the x, y, and z dimensions respectively;

[0085] This design enables preliminary access to location information and provides basic data for subsequent model processing;

[0086] The input layer takes the normalized position data Input into the model to provide initial input values ​​for the calculation of the next layer;

[0087] Adjust and improve the LSTM layer and build a binary LSTM layer. The implementation process is as follows:

[0088] LSTM layers using 1-bit weights and activations;

[0089] The traditional LSTM weight matrix is ​​binarized using the symbolic function sgn(y), where y is the independent variable of the LSTM layer, i.e., the element in the traditional LSTM weight matrix.

[0090] when hour, ;

[0091] when hour, ;

[0092] 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;

[0093] In traditional LSTM, the calculation process involves complex weight matrix multiplication and accumulation operations. In the binary LSTM layer, the binary weight matrices are (connecting the input layer to the current neuron) and (Connecting the previous hidden layer with the current neuron), the bias vector is , then the calculation process of the binary LSTM layer neurons is:

[0094]

[0095] in, 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, thereby completing the calculation task of this layer;

[0096] Output layer adjustment:

[0097] The output layer has three neurons, corresponding to the predicted coordinate values ​​of the drone in the x, y, and z dimensions at the next moment.

[0098] To ensure that the output value can directly and accurately reflect the coordinate changes and avoid the complex calculations and possible information loss caused by nonlinear transformations, the output layer uses a linear activation function.

[0099] Assume that the neuron output vector of the output layer is , after being processed by the linear activation function, the output value directly corresponds to the predicted coordinate change, that is, Corresponding to the predicted change in x dimension, Corresponding to the predicted change in y dimension, Corresponding to the predicted change in the z dimension, t is the moment, which enables the output function of the coordinate prediction value to be realized more quickly;

[0100] Model training evaluation:

[0101] The pre-processed BeiDou positioning data is divided into 8:2 ratios, and the constructed model is used to predict the test set data. The mean square error (MAE) is calculated. If the mean square error value is less than the preset threshold of 0.01 and is within the acceptable range (less than 0.1), the model performance is considered to meet the standard. Otherwise, further optimization is required.

[0102] If the model performance does not meet the requirements, adjust the number of neurons j in the binary LSTM layer. Increasing the number of neurons can improve the model's expressiveness, while reducing the number of neurons will have the opposite effect. Through multiple experiments, find the optimal number and finally complete the model training evaluation.

[0103] Example:

[0104] Traditional LSTM model weight parameters are usually 32-bit or 64-bit floating-point values, which have high storage and computational overhead. However, the improved LSTM prediction model uses 1-bit weights, limiting the weight values ​​to +1 or -1.

[0105] Assuming that the weight matrix size of a traditional LSTM layer is 100×100 and each weight parameter is 32 bits (4 bytes), the matrix storage requires 100×100×4=40,000 bytes;

[0106] After binarization, the same-sized weight matrix only requires 100×100×1=10,000 bits, or 1,250 bytes, which greatly reduces the parameter storage capacity. The number of parameters in the actual model is less than 5KB. Compared with traditional AI algorithms, the model size is reduced to its ;

[0107] In actual testing, a certain number of test data samples (100 groups) were collected for the drone trajectory prediction task. When the traditional LSTM model predicted the drone's trajectory for the next 5 seconds, the average prediction error was about 1 meter.

[0108] The improved LSTM prediction model controls the drone's trajectory prediction error for the next 5 seconds to within 0.3m, significantly improving the prediction;

[0109] The process of obtaining the optimal track information is as follows:

[0110] Using the trained improved LSTM prediction model (Improved LSTM Prediction Model), the drone's trajectory for the next 5 seconds is accurately predicted;

[0111] Assume Seconds is the time interval, the current time and the moment before The location information is input into the improved LSTM prediction model (improved LSTM model) in time series. The neurons in the model are calculated through binary weights to output the predicted location points and trajectories at multiple moments in the next 5 seconds.

[0112] Right now, The total output is predicted location points, which constitute the predicted trajectory , where N is the number of predicted position points of the predicted trajectory, To predict the location point, it is represented in three-dimensional space as ;

[0113] Specifically, It is the predicted position of the drone in three-dimensional space at the future moment output by the improved LSTM prediction model;

[0114] Collect the initial trajectory L, and continuously collect the position information of the UAV at different times through the Beidou positioning function, and arrange them in sequence to form the initial trajectory of the UAV ,in, is the initial position point;

[0115] The predicted trajectory With the initial trajectory For comparison, for each location point corresponding to each time interval, obtain the difference residual ;in, It is a three-dimensional vector, which represents the exact deviation between the best predicted position and the actual position at each moment;

[0116] In order to determine the voxel region where each differential residual is located, the voxel division rule is first clarified;

[0117] The voxel is set in a three-dimensional space, and the edge length of the voxel in the x, y and z directions is respectively , , For the differential residual , the index of the voxel where the differential residual is located is obtained as follows: = floor (x / 8) * 8 + floor (y / 8) * 8 + floor (z / 8) , = floor (x / 16) * 16 + floor (y / 16) * 16 + floor (z / 16) , = floor (x / 32) * 32 + floor (y / 32) * 32 + floor (z / 32) ;

[0118] wherein, denotes a floor operation;

[0119] According to the optimal encoding strategy determined in the foregoing, new probability threshold values and are set;

[0120] If the voxel where a certain differential residual is located belongs to a high-probability region (assuming that the probability threshold value of the high-probability region is , that is, the probability of the presence of the unmanned aerial vehicle in the voxel is ), the differential residual is encoded using a 4-byte encoding mode (encoding function ) ;

[0121] If , the voxel is located in a medium-probability region, and the probability threshold value of a low-probability region is used, a 3-byte encoding mode (encoding function ) is used;

[0122] If , the voxel is located in a low-probability region, a 2-byte encoding mode (encoding function ) is used;

[0123] Specifically, in this way, the unmanned aerial vehicle track information is efficiently compressed and encoded, and compared with the traditional 10-byte / once absolute coordinate transmission, the data transmission amount is greatly reduced;

[0124] For example, the traditional mode requires 10N bytes to transmit N position points, and after the encoding and compression, the transmission data amount can be greatly reduced according to different region encoding conditions, and the data transmission efficiency is effectively improved;

[0125] The encoding data of the high-probability region, the medium-probability region and the low-probability region are combined and arranged in time sequence, thereby forming the optimal track information;

[0126] Example:

[0127] Assume that the drone is performing a mission in a certain flight area, which is divided into three-dimensional voxels (1m interval in the horizontal direction and 0.5m interval in the vertical direction);

[0128] By counting the number of drones appearing in each voxel, we can get the following probability distribution of typical voxels:

[0129] High probability region: voxel A (coordinate range: x = 10-11m, y = 20-21m, z = 5-5.5m), 5,000 occurrences, probability (p = 5000 / 10000 = 0.5 (close to the threshold α = 0.5, classified as a high probability region);

[0130] Medium probability region: voxel B (x=15-16m, y=18-19m, z=3-3.5m), 1,500 occurrences, probability (p=0.15) (between β=0.1 and α=0.5, classified as medium probability region);

[0131] Low probability region: voxel C (x=25-26m, y=5-6m, z=8-8.5m), 500 occurrences, probability (p=0.05) (lower than β=0.1, classified as low probability region);

[0132] The improved LSTM model is used to predict the drone's trajectory for the next 5 seconds (with an interval of 0.1 seconds and a total of 50 points). The prediction errors (differential residuals) of three typical points are as follows:

[0133] Point 1 (located at voxel A, high probability area):

[0134] Predicted location point m;

[0135] Initial position point m;

[0136] differential residuals m, using 4-byte differential residual coding function Encoding;

[0137] Point 2 (located at voxel B, medium probability area):

[0138] Predicted location point m;

[0139] Initial position point m,

[0140] differential residuals m, using 3-byte differential residual coding function Encoding;

[0141] Point 3 (located at voxel C, low probability area):

[0142] Initial position point m, using 2-byte abbreviated encoding Encode and record only the approximate area coordinates (e.g. x=25-26m, y=5-6m, z=8-9m);

[0143] Data compression effect:

[0144] Traditional absolute coordinate transmission requires 10 bytes per point, and 50 points require a total of 500 bytes.

[0145] 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.

[0146] The total data volume is 165 bytes, and the transmission optimization (compression ratio) is 67%.

[0147] Flight control module: obtains the optimal trajectory based on the optimal track information, and performs UAV flight communication control based on the optimal trajectory;

[0148] After receiving the compressed and encoded track information, the flight control module parses the encoded data of each area in chronological order, and the data of the high / medium probability area is:

[0149] Restore the difference residual through the decoding function , and combined with the predicted trajectory Calculate the best trajectory ;

[0150] Finally, the complete optimal trajectory is reconstructed ,in, is the best initial position point after correction;

[0151] The reconstructed optimal trajectory With the initial trajectory Compare and calculate position deviation ;

[0152] Generate control instructions based on the deviation value: If When the error threshold (0.5m) is exceeded, the PID controller is triggered to adjust the flight parameters:

[0153] Horizontal deviation (represented by x and y in the position point) is used to correct the heading by adjusting the motor speed difference or the servo angle;

[0154] The vertical deviation (z in the position point) is different, and the lifting speed is controlled by adjusting the throttle output;

[0155] If the deviation is within the threshold, maintain the current flight state.

[0156] In the application, several formulas involved are calculated by taking their numerical values ​​after removing the dimensions, and the formulas are established by collecting a large amount of data and performing software simulation to obtain a formula for the most recent real situation. Some coefficients or weights in the formulas are set by technical personnel in this field according to actual conditions, so they will not be elaborated here.

[0157] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution.

[0158] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. UAV flight communication control system based on single Beidou positioning, characterized by: include: Data collection module: collects Beidou location data of drones through Beidou positioning; Strategy Integration Module: Based on BeiDou location data, the three-dimensional flight space is divided into voxels, the probability of drone appearance in each voxel is calculated, a spatiotemporal probability distribution map is constructed, and the optimal encoding strategy is determined; The optimal coding strategy determination process is as follows: Based on the probability of drone appearance within the voxel Size,classifies voxels into three categories, including: Preset probability threshold , ; when When , the voxel is divided into a high probability area; when When , it is the medium probability area; when When , it is a low probability area; Set different encoding methods for different regional codes, including: Assume that the predicted appearance position of the drone in the voxel is , the initial position is , the difference residual formula is expressed as ; The high probability area is encoded using 4-byte differential residual; The medium probability region is encoded using a 3-byte differential residual; The low probability area adopts a 2-byte encoding method; Transmission optimization module: Use the improved LSTM prediction model to predict the best predicted position of the drone in the next 5 seconds, compare it with the actual position to calculate the difference residual, and then encode it according to the optimal encoding strategy based on the voxel area where the difference residual is located to obtain the optimal track information; The improved LSTM prediction model includes an input layer, a binary LSTM layer, and an output layer; The improved LSTM prediction model construction process is as follows: The input layer and the output layer are both provided with 3 neurons; The binary LSTM layer uses 1-bit weights and activation values, and uses the symbolic function sgn(y) to binaryize the weight matrix of the traditional LSTM, 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. , t is the time; Assume that the binary weight matrix is ​​the weight matrix connecting the input layer and the current neuron And connect the previous hidden layer with the current neuron , the bias vector is , the calculation process of the binary LSTM layer neurons is: ； in, 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 and complete the binary LSTM layer calculation task; The optimal track information acquisition process is as follows: At 0.1 second intervals, the current and previous position information is input into the improved LSTM prediction model in time series, and the predicted position points at multiple moments in the next 5 seconds are output to form a predicted trajectory. The initial trajectory of the UAV is constructed by arranging the position information at different times in sequence. The predicted trajectory is compared with the initial trajectory, and the difference residual is obtained for the position point corresponding to each time interval. Determine a voxel partitioning rule and set a new probability threshold for encoding based on the optimal encoding strategy; The coded data obtained by the optimal coding strategy are combined and arranged in chronological order to form the optimal track information; Flight control module: obtains the optimal trajectory based on the optimal track information, and performs 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 is characterized in that: The process of dividing the flight three-dimensional space into voxels is as follows: Based on Beidou position data, the three-dimensional space where the drone flies is divided into grids with an interval of 1m in the horizontal direction and divided into layers with 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 is characterized in that: The process of obtaining the spatiotemporal probability distribution map is as follows: Based on the collected Beidou positioning data of drones, the number of drone appearances is counted voxel by voxel, and the probability of drone appearance in different voxels is obtained based on the number of drone appearances; Traverse all Beidou location data, count the probability of drone appearance in each voxel, build a probability distribution map, and use 3D graphics visualization tools to build a spatiotemporal probability distribution map based on the probability value of each voxel.

4. The UAV flight communication control system based on single Beidou positioning according to claim 3 is characterized in that: The voxel partitioning rule is: Assume that in three-dimensional space, the side lengths of the voxel in the x, y, and z directions are , , , for the difference residual , the index of the voxel where it is located Get the representation: , , ; in, Indicates a floor operation.

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

Images