Spatial Parameter Optimization Methods, Apparatus and Electronic Equipment

Abstract

This application discloses a method, apparatus, and electronic device for optimizing airspace parameters. The method includes: acquiring communication status data of a UAV during flight; processing the communication status data using a prediction model to obtain a combination of airspace parameters for the UAV, wherein the prediction model is trained using samples from a preset rule base and includes a composite loss function, comprising a first loss function for quantifying the numerical difference between the predicted and actual values, a second loss function for weighting samples with prediction difficulty greater than a preset threshold, and a third loss function for controlling gradient fluctuations; and adjusting the initial control parameters of the UAV based on the airspace parameter combination. This application solves the technical problems of sluggish handover, frequent ping-pong effects, and uneven resource allocation caused by network handover mechanisms in related technologies that rely on fixed parameters in airspace movement scenarios.

Description

Technical Field

[0001] This application relates to the field of wireless communication technology, and more specifically, to a spatial parameter optimization method, apparatus, and electronic device. Background Technology

[0002] With the deep integration of 5G networks and drone technology, low-altitude networks are becoming increasingly important as a key infrastructure supporting future high-dynamic, high-density communication. However, the high mobility of drones, their frequent and unpredictable trajectory changes, and the unique dynamic and complex interference in the airspace pose significant challenges to the network switching mechanisms in related technologies.

[0003] Currently, airspace base stations use the handover mode of terrestrial networks, relying on a set of pre-set and unchanging parameters for UAV handover. This fixed-parameter-based handover mechanism has significant drawbacks in dynamic and ever-changing low-altitude scenarios. Base stations cannot adaptively optimize and adjust the handover parameters according to the UAV's real-time location, high-speed movement, and complex low-altitude environment, severely limiting the overall performance of low-altitude networks in scenarios with high dynamics, high density, and high reliability requirements.

[0004] There is currently no effective solution to the above problems. Summary of the Invention

[0005] This application provides a method, apparatus, and electronic device for optimizing airspace parameters, in order to at least solve the technical problems of sluggish handover, frequent ping-pong effects, and uneven resource allocation caused by the reliance on fixed parameters in network handover mechanisms in related technologies during airspace mobility scenarios.

[0006] According to one aspect of the embodiments of this application, a method for optimizing airspace parameters is provided, comprising: acquiring communication status data of a UAV during flight; processing the communication status data using a prediction model to obtain a combination of airspace parameters of the UAV, wherein the prediction model is obtained by training samples obtained from a preset rule base, the preset rule base including historical communication status data of the UAV and corresponding historical airspace parameter combinations, the historical airspace parameter combinations being airspace parameters optimized by using a neighborhood search algorithm on the initial airspace parameter combinations corresponding to the historical communication status data, the prediction model further comprising a composite loss function, the composite loss function including a first loss function for quantifying the numerical difference between the predicted value and the true value of the prediction model, a second loss function for weighting samples whose prediction difficulty is greater than a preset threshold, and a third loss function for controlling gradient fluctuations; and adjusting the initial control parameters of the UAV according to the airspace parameter combination.

[0007] Optionally, the preset rule base is determined by: acquiring historical communication status data of the UAV, wherein the historical communication status data includes the UAV's flight speed, reference signal received power, reference signal received quality, and base station spacing in multiple historical flight scenarios; determining the initial airspace parameter combination corresponding to the historical communication status data, wherein the initial airspace parameter combination includes the event bias value, hysteresis value, and trigger time corresponding to the historical communication status data; optimizing the initial airspace parameter combination using a neighborhood search algorithm to obtain the historical airspace parameter combination; and determining the preset rule base based on the historical communication status data and the historical airspace parameter combination.

[0008] Optionally, a neighborhood search algorithm is used to optimize the initial airspace parameter combination, including: obtaining a first parameter combination corresponding to the target communication state data from a preset rule base, wherein the target communication state data is the communication state data of the UAV on a preset flight path; determining a first switching score corresponding to the first parameter combination, wherein the first switching score is used to reflect the network performance of the first parameter combination; determining the median of the three-dimensional interval of the first parameter combination as the starting point, and determining a second parameter combination at 26 neighboring points corresponding to the starting point; determining a second switching score corresponding to the second parameter combination, wherein the second switching score is used to reflect the network performance of the second parameter combination; determining a target parameter combination corresponding to the target communication state data based on the first switching score and the second switching score, and optimizing the initial airspace parameter combination based on the target parameter combination, wherein the target parameter combination is the parameter combination with the highest score among the first switching score and the second switching score.

[0009] Optionally, determining the first handover score corresponding to the first parameter combination includes: determining a first uplink rate, a first ping-pong handover rate, and a first handover delay of the UAV on a preset flight path based on the forward parameter combination in the first parameter combination, wherein the forward parameter combination is a parameter combination corresponding to the communication status data of the UAV when flying forward on the preset flight path; determining a second uplink rate, a second ping-pong handover rate, and a second handover delay of the UAV on the preset flight path based on the reverse parameter combination in the first parameter combination, wherein the reverse parameter combination is a parameter combination corresponding to the communication status data of the UAV when flying backward on the preset flight path; determining a forward handover score corresponding to the forward parameter combination based on the first uplink rate, the first ping-pong handover rate, and the first handover delay, and determining a reverse handover score corresponding to the reverse parameter combination based on the second uplink rate, the second ping-pong handover rate, and the second handover delay; and determining the first handover score based on the forward handover score and the reverse handover score.

[0010] Optionally, determining the second parameter combination on the 26 neighboring points corresponding to the starting point includes: determining the median of the three-dimensional interval of the first parameter combination; and at the target neighboring point, increasing or decreasing the median by a preset step size to obtain the second parameter combination corresponding to the target neighboring point, wherein the target neighboring point is any one of the 26 neighboring points.

[0011] Optionally, the communication state data is processed by a prediction model to obtain a combination of UAV spatial parameters, including: extracting features from the communication state data through a first convolutional layer in the prediction model to obtain a first feature; processing the first feature through multiple residual blocks in the prediction model to obtain a second feature, wherein each residual block includes multiple residual units, each residual unit includes a second convolutional layer and a residual connection structure, the second convolutional layer is connected to an activation function to alleviate the gradient vanishing phenomenon of the first feature during processing; and processing the second feature through a fully connected layer in the prediction model to obtain a combination of spatial parameters.

[0012] Optionally, the composite loss function is determined as follows: The combination of prediction parameters of the prediction model during training is obtained, and the hyperparameters for quantifying the prediction error of the prediction model are determined. A first loss function is determined based on the combination of prediction parameters, the combination of historical spatial parameters, and the hyperparameters. The prediction confidence, class weight factor, and focus factor required for training the target sample are determined, and a second loss function is determined based on the prediction confidence, class weight factor, and focus factor. Here, the target sample is a sample whose prediction difficulty is greater than a preset threshold; the class weight factor is used to quantify the weight of the target sample during training; and the focus factor is used to enhance the focusing ability on the target sample. The regularization coefficient and loss gradient required for training the prediction model are determined, and a third loss function is determined based on the regularization coefficient and loss gradient. Finally, a composite loss function is determined based on the first, second, and third loss functions.

[0013] Optionally, the method further includes: training the prediction model based on historical communication status data, and determining the loss value of each training sample in the historical communication status data through a composite loss function; comparing the loss value with a preset threshold, and determining the training sample corresponding to the loss value being greater than or equal to the preset threshold as the target sample, and determining the training sample corresponding to the loss value being less than the preset threshold as the normal sample; and performing the next round of training on the prediction model based on the target sample and the normal sample, wherein the weight value assigned to the target sample in the composite loss function is higher than the weight value assigned to the normal sample in the composite loss function.

[0014] According to another aspect of the embodiments of this application, an airspace parameter optimization device is also provided, comprising: an acquisition module for acquiring communication status data of a UAV during flight; a processing module for processing the communication status data through a prediction model to obtain a combination of airspace parameters of the UAV, wherein the prediction model is obtained by training samples obtained from a preset rule base, the preset rule base including historical communication status data of the UAV and corresponding historical airspace parameter combinations, the historical airspace parameter combinations being airspace parameters optimized by using a neighborhood search algorithm on the initial airspace parameter combinations corresponding to the historical communication status data, the prediction model further including a composite loss function, the composite loss function including a first loss function for quantifying the numerical difference between the predicted value and the true value of the prediction model, a second loss function for weighting samples whose prediction difficulty is greater than a preset threshold, and a third loss function for controlling gradient fluctuations; and an optimization module for adjusting the initial control parameters of the UAV according to the airspace parameter combination.

[0015] According to another aspect of the embodiments of this application, an electronic device is also provided, including: a memory and a processor, wherein the memory is used to store program instructions; and the processor is connected to the memory and used to execute the above-described spatial parameter optimization method.

[0016] According to another aspect of the embodiments of this application, a non-volatile storage medium is also provided, the non-volatile storage medium including a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned spatial parameter optimization method by running the computer program.

[0017] According to another aspect of the embodiments of this application, a computer program product is also provided, including computer instructions that, when executed by a processor, implement the above-described spatial parameter optimization method.

[0018] In this embodiment, communication status data of the UAV during flight is acquired; the communication status data is processed by a prediction model to obtain the UAV's airspace parameter combination. The prediction model is trained by obtaining samples from a preset rule base, which includes historical communication status data of the UAV and corresponding historical airspace parameter combinations. The historical airspace parameter combinations are airspace parameters optimized by using a neighborhood search algorithm to optimize the initial airspace parameter combinations corresponding to the historical communication status data. The prediction model also includes a composite loss function, which includes a first loss function for quantifying the numerical difference between the predicted value and the true value of the prediction model, a second loss function for weighting samples whose prediction difficulty is greater than a preset threshold, and a third loss function for controlling gradient fluctuations. The initial control parameters of the UAV are adjusted according to the airspace parameter combination, achieving the purpose of adaptively adjusting the UAV network switching parameters. This improves the airspace switching efficiency and ensures the quality and stability of wireless communication, thereby solving the technical problems of sluggish switching, frequent ping-pong effects, and uneven resource allocation caused by the reliance on fixed parameters in network switching mechanisms in related technologies during airspace movement scenarios. Attached Figure Description

[0019] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0020] Figure 1 This is a hardware structure diagram of a computer terminal for implementing a spatial parameter optimization method according to an embodiment of this application;

[0021] Figure 2 This is a flowchart of a spatial parameter optimization method according to an embodiment of this application;

[0022] Figure 3 This is a schematic diagram comparing the effects of different parameter verification methods according to embodiments of this application;

[0023] Figure 4 This is a schematic diagram of the processing logic of a deep neural network in a prediction model according to an embodiment of this application;

[0024] Figure 5 This is a schematic diagram of the network structure of a ResNet model according to an embodiment of this application;

[0025] Figure 6 This is a schematic diagram illustrating the change in the training loss value of a ResNet model according to an embodiment of this application;

[0026] Figure 7This is a schematic diagram illustrating the change in training accuracy of a ResNet model according to an embodiment of this application;

[0027] Figure 8 This is a structural diagram of a spatial parameter optimization device according to an embodiment of this application. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0029] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0030] First, some nouns or terms that appear in the explanation of the embodiments of this application shall be interpreted as follows:

[0031] RSRP (Reference Signal Received Power): In LTE (Long Term Evolution) and 5G networks, RSRP is a measurement metric used to describe the strength of a reference signal from the serving cell or neighboring cells. It is one of the important parameters for evaluating signal coverage and quality in wireless communication. The magnitude of the RSRP value directly affects the communication stability between the UE (User Equipment) and the base station.

[0032] RSRQ (Reference Signal Received Quality): A metric that measures RSRP (Reference Signal Received Rate), reflecting the average quality of the serving cell reference signal received by the UE. The RSRQ value is related to the RSRP value and the total received power, and is used to assess channel quality, especially under interference conditions.

[0033] SINR (Signal to Interference plus Noise Ratio): In wireless communication systems, SINR is a key parameter for measuring the quality of the received signal. It represents the ratio of signal power to the total power of interference and noise. A high SINR value indicates good signal quality and a stable communication link, while a low SINR value may indicate the presence of interference or a weak signal, resulting in degraded communication quality.

[0034] Hysteresis: In wireless communication, hysteresis refers to a threshold difference set by the system during network handover to prevent ping-pong handovers caused by slight fluctuations in signal strength. A UE will only initiate a cell handover request when the signal strength of a neighboring cell exceeds the signal strength of the current serving cell plus the hysteresis threshold. Setting the hysteresis value is crucial for improving network handover stability and reducing unnecessary handovers.

[0035] Time to Trigger (TTT): This refers to the time the network waits before actually executing a handover operation after detecting that the handover conditions have been met. This waiting time is to ensure the accuracy of the handover decision and avoid unnecessary handovers caused by momentary signal fluctuations. Properly setting the TTT helps reduce the probability of ping-pong handovers, while also reducing resource consumption and communication latency during the handover process.

[0036] To address the poor performance of spatial parameter switching mechanisms in related technologies, this application provides a spatial parameter optimization method that can be implemented in... Figure 1 The computer terminal shown is described below.

[0037] The spatial parameter optimization method embodiments provided in this application can be executed on mobile terminals, computer terminals, or similar computing devices. Figure 1 A hardware block diagram of a computer terminal for implementing a spatial parameter optimization method is shown. Figure 1As shown, the computer terminal 10 may include one or more processors (shown as 102a, 102b, ..., 102n in the figure) (the processor may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 104 for storing data, and a transmission module 106 for communication functions connected via wired and / or wireless networks. In addition, it may also include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, and a BUS bus. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0038] It should be noted that the aforementioned one or more processors and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10. As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0039] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the spatial parameter optimization method in this embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned spatial parameter optimization method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0040] The transmission module 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission module 106 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission module 106 may be a radio frequency (RF) module, used for wireless communication with the Internet.

[0041] The display can be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the computer terminal 10.

[0042] It should be noted here that, in some optional embodiments, the above... Figure 1 The computer terminal shown may include hardware elements (including circuitry), software elements (including computer code stored on a computer-readable medium), or a combination of both hardware and software elements. It should be noted that... Figure 1 This is only one instance of a specific particular instance, and is intended to illustrate the types of components that may exist in the aforementioned computer terminal.

[0043] In the above operating environment, this application provides an embodiment of a spatial parameter optimization method. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0044] Figure 2 This is a flowchart of a spatial parameter optimization method according to an embodiment of this application, such as... Figure 2 As shown, the method includes the following steps:

[0045] Step S202: Obtain communication status data of the UAV during flight.

[0046] In step S202 above, the first step is to select a UAV flight path within the coverage area of ​​the airspace base station and collect multiple communication status data during the UAV's flight in real time, including but not limited to the UAV's speed, received reference signal power (RSRP), signal-to-noise ratio (SINR), and distance between the UAV and the base station (base station spacing). This communication status data can be acquired through a built-in communication module or external sensors and uploaded in real time to the network control center or the server where the prediction model is located, enabling the prediction model to generate the most suitable combination of airspace parameters based on the current flight conditions and network environment.

[0047] Step S204: The communication status data is processed by the prediction model to obtain the airspace parameter combination of the UAV. The prediction model is obtained by training samples obtained from a preset rule base. The preset rule base includes the historical communication status data of the UAV and the corresponding historical airspace parameter combination. The historical airspace parameter combination is the airspace parameter after optimizing the initial airspace parameter combination corresponding to the historical communication status data using a neighborhood search algorithm. The prediction model also includes a composite loss function, which includes a first loss function for quantifying the numerical difference between the predicted value and the true value of the prediction model, a second loss function for weighting samples whose prediction difficulty is greater than a preset threshold, and a third loss function for controlling gradient fluctuations.

[0048] In step S204 above, the communication status data is input into the deployed prediction model (the ResNet-34 prediction model is used in this application) for inference, which can output the optimal airspace parameter combination for the corresponding flight scenario, including A3 event bias, hysteresis value and trigger time (TTT). These will be used to guide the handover behavior between the UAV and the network, so as to minimize communication interruption, reduce the frequency of ping-pong handover, and improve the overall communication quality and network performance.

[0049] It is worth noting that the prediction model is a deep neural network based on the ResNet architecture. It can generate prediction parameters suitable for the current communication scenario by learning historical communication state data from a heuristic hierarchical rule base (i.e., a pre-defined rule base) and corresponding optimized spatial parameters. In practical applications, the prediction model also uses a composite loss function for optimization. The composite loss function consists of three sub-losses: a first loss function (such as Huber loss or mean squared error MSE) used to quantify the difference between the prediction model's predicted value and the true value; a second loss function (such as FocalLoss) used to enhance attention to difficult-to-predict samples; and a third loss function (such as gradient norm regularization) used to control gradient fluctuations.

[0050] Step S206: Adjust the initial control parameters of the UAV based on the airspace parameter combination.

[0051] In step S206 above, the optimal airspace parameter combination output by the prediction model can be injected into the handover strategy table of the current communication controller through the network control interface or the wireless resource management module, so as to dynamically update the handover parameters of the cells (including the serving cell and the corresponding target cell) involved in the handover process of the UAV, so as to achieve efficient and stable network handover.

[0052] In this embodiment of the application, step S208 is also included: after the switching strategy is executed, the communication status data of the UAV is continuously monitored to determine the UAV's transmission rate, switching latency and whether abnormal states such as ping-pong switching occur, for subsequent updates and evaluations of the prediction model.

[0053] In this embodiment of the application, step S210 is also included, in which the input features (communication status data) recorded during the application process and the actual switching effect (output spatial parameter combination) are periodically combined to form a feedback data packet and sent back to the central server or training module for expanding the sample set, fine-tuning the model, and realizing adaptive optimization of the prediction model.

[0054] Through steps S202 to S210, the goal of adaptively adjusting the UAV network handover parameters is achieved, thereby improving airspace handover efficiency and ensuring the quality and stability of wireless communication. This solves the technical problems of sluggish handover, frequent ping-pong effects, and uneven resource allocation caused by network handover mechanisms relying on fixed parameters in airspace mobile scenarios. The following is a detailed explanation.

[0055] In this embodiment, the construction and optimization of the heuristic hierarchical rule base (i.e., the aforementioned preset rule base, hereinafter referred to as the rule base) is the core step in realizing intelligent adaptive adjustment of airspace parameters, providing powerful training data for the ResNet prediction model. The preset rule base can be determined as follows: Acquire historical communication state data of the UAV, including the UAV's flight speed, reference signal received power, reference signal received quality, and base station spacing in multiple historical flight scenarios; determine the initial airspace parameter combination corresponding to the historical communication state data, including the event bias value, hysteresis value, and trigger time corresponding to the historical communication state data; optimize the initial airspace parameter combination using a neighborhood search algorithm to obtain the historical airspace parameter combination; and determine the preset rule base based on the historical communication state data and the historical airspace parameter combination. The specific process analysis is as follows:

[0056] 1. Collect and integrate historical communication status data experienced by UAVs during past flights.

[0057] This historical communication status data includes, but is not limited to, the drone's flight speed, RSRP, RSRQ, and base station spacing. These provide intuitive feedback on the drone's communication performance and network environment, and serve as input parameters for the rule base.

[0058] 1) Drone speed: Low (L) <7 m / s, Medium (M) 7-13 m / s, High (H) >13 m / s;

[0059] 2) RSRP: Weak (W) < -95 dBm, Medium (M) -95 to -85 dBm, Strong (S) > -85 dBm;

[0060] 3) SINR: Poor (P) <0 dB, Good (G) 0-10 dB, Excellent (E) >10 dB;

[0061] 4) Base station spacing: Near (S) <3 km, Medium (M) 3-8 km, Far (L) >8 km;

[0062] The above four types of parameters together constitute the perceptual basis of the rule base, forming... For a discretized input scenario, the corresponding output parameters are as follows:

[0063] 1) A3 event bias: Low (L) 2 dB, Medium (M) 3 dB, High (H) 4 dB;

[0064] 2) Hysteresis: Low (L) 2 dB, Medium (M) 3 dB, High (H) 4 dB;

[0065] 3) Trigger time: Short (S) 40-160 ms, Medium (M) 256-640 ms, Long (L) 1024-1280 ms.

[0066] Specifically, the rule base contains 81 scenario rules. Each scenario rule is a combination of initial spatial parameters corresponding to different input parameters. Representative rules are shown in Table 1.

[0067] Table 1. Examples of Representative Rules

[0068]

[0069] 2. Optimize the initial spatial parameter combination.

[0070] Furthermore, to verify the accuracy of the rule base, the scene rules within it need to be tested and corrected. Since traditional parameter traversal testing methods suffer from high test counts and low efficiency, this application introduces a three-dimensional 26-neighborhood search algorithm to optimize the initial spatial parameter combinations determined in the rule base. Based on a heuristic hierarchical rule base, this neighborhood search algorithm determines the TTT interval according to UAV speed, RSRP, SINR, and base station spacing. It uses the median of the three-dimensional interval composed of TTT, A3 bias, and hysteresis as the starting point to expand testing to the surrounding 26 neighbors. The parameter combination with the best handover performance is selected as the new starting point. Iteration continues until the handover performance at the starting point is superior to the parameter combinations of the surrounding 26 neighbors. The parameter combination at this point is then taken as the optimal spatial parameter combination under the current input parameters and updated in the rule base for subsequent training of the ResNet prediction model.

[0071] Figure 3 This paper demonstrates the effectiveness of different parameter verification methods simulated on a Monte Carlo simulation platform, including complete traversal, 26-neighbor asymptotic, 6-neighbor asymptotic, improved 6-neighbor asymptotic, hierarchical traversal, hierarchical 26-neighbor, and hierarchical improved 26-neighbor testing methods. While the traditional complete traversal testing method achieves a 100% search success rate, its 117 test cases (each requiring multiple tests and averaging) result in high testing costs. In contrast, the hierarchical 26-neighbor search algorithm used in this application requires only an average of 29.2 test cases, achieving an optimal performance of 97.84% while reducing the number of searches by 75%.

[0072] 3. Improve the heuristic hierarchical rule base.

[0073] Finally, by combining the optimized historical airspace parameter combinations with corresponding historical communication status data, a comprehensive heuristic hierarchical rule base is constructed. It should be noted that through the dynamic update mechanism of the rule base, the system can continuously absorb new data and adjust optimization strategies, ensuring the accuracy and effectiveness of the rule base. In this way, the preset rule base can provide the most suitable switching parameters when the UAV faces different flight scenarios and network conditions, achieving intelligent switching with high throughput, low ping-pong switching, and low overhead, providing strong technical support for UAV communication.

[0074] In the verification process of the aforementioned rule base, a handover score model was introduced to evaluate the handover effect of airspace parameter combinations. This model calculates the handover score for each airspace parameter combination by real-time monitoring of the drone's uplink transmission rate, ping-pong handover count, and handover latency. The handover score model consists of the following three parts:

[0075] (1) Uplink speed: used to reflect the communication quality of the UAV (corresponding weight) ), 90th percentile of historical communication status data The target uplink rate is calculated using the following formula:

[0076]

[0077] (2) Ping-pong switching rate: used to measure switching stability (corresponding weight) Exponential decay is used to avoid excessive differences between the first and second orders. The formula is as follows:

[0078]

[0079] (3) Switching delay: Affects real-time communication performance (corresponding weight) The maximum allowable delay is set to 200ms, as shown in the following formula:

[0080]

[0081] Based on the above rate term Table tennis and delay items Switching score corresponding to spatial parameter combination It can be represented as:

[0082]

[0083] Based on the aforementioned handover score model, a neighborhood search algorithm can be used to optimize the initial airspace parameter combination, including: obtaining a first parameter combination corresponding to the target communication state data from a preset rule base, wherein the target communication state data is the communication state data of the UAV on a preset flight path; determining a first handover score corresponding to the first parameter combination, wherein the first handover score is used to reflect the network performance of the first parameter combination; determining the median of the three-dimensional interval of the first parameter combination as the starting point, and determining a second parameter combination at 26 neighboring points corresponding to the starting point; determining a second handover score corresponding to the second parameter combination, wherein the second handover score is used to reflect the network performance of the second parameter combination; determining a target parameter combination corresponding to the target communication state data based on the first handover score and the second handover score, and optimizing the initial airspace parameter combination based on the target parameter combination, wherein the target parameter combination is the parameter combination with the highest score between the first handover score and the second handover score.

[0084] The determination of the first handover score corresponding to the first parameter combination includes: determining the first uplink rate, first ping-pong handover rate, and first handover delay of the UAV on a preset flight path based on the positive parameter combination in the first parameter combination, wherein the positive parameter combination is the parameter combination corresponding to the communication status data of the UAV when flying forward on the preset flight path; determining the second uplink rate, second ping-pong handover rate, and second handover delay of the UAV on the preset flight path based on the negative parameter combination in the first parameter combination, wherein the negative parameter combination is the parameter combination corresponding to the communication status data of the UAV when flying backward on the preset flight path; determining the positive handover score corresponding to the positive parameter combination based on the first uplink rate, first ping-pong handover rate, and first handover delay, and determining the negative handover score corresponding to the negative parameter combination based on the second uplink rate, second ping-pong handover rate, and second handover delay; and determining the first handover score based on the positive handover score and the negative handover score. The specific process analysis is as follows:

[0085] S1: First, determine the UAV's preset flight path, such as flying from serving cell A to serving cell B, and continuously monitor the cell ID in the MQTT messages returned by the UAV until the cell ID changes (A→B). Record this location as the start point of the handover zone. After the UAV continues to fly to a stable signal area, ensure it stably connects to cell B, then turn around and fly back to cell A until the cell ID changes again (B→A). Record this location as the end point of the handover zone. After completing the bidirectional handover zone boundary calibration, establish the formal test path benchmark, i.e., the determined preset flight path.

[0086] S2: When the drone flies from cell A to near the handover zone of cell B, record the drone's current flight speed, RSRP, SINR, and base station spacing in real time, i.e., the target communication status data during forward flight. Match the current input parameters according to the rule base to obtain the initial range of the three-dimensional output parameters, i.e., the forward parameter combination in the first parameter combination. Subsequently, adjust the three-dimensional parameters of cell A synchronously via the base station API to the median value of the A3 event bias. Hysteresis is taken as the median value TTT takes the midpoint of the interval Perform an A→B handover operation. After the handover is completed, fly stably for 5 seconds. Record the first uplink communication rate, the first number of ping-pong handovers, and the first handover delay during the forward flight of the UAV. Calculate the corresponding parameter combinations based on the handover score model. , , The positive switching score.

[0087] S3: When the drone turns back from cell B to the vicinity of the handover zone of cell A, the drone's current flight speed, RSRP, SINR, and base station spacing are recorded in real time, i.e., the target communication status data during reverse flight. Based on the rule base, the current input parameters are matched to obtain the initial range of the three-dimensional output parameters, i.e., the reverse parameter combination in the first parameter combination. Subsequently, the three-dimensional parameters of cell B are synchronously adjusted via the base station API to the median value of the A3 event bias. Hysteresis is taken as the median value TTT takes the midpoint of the interval Perform a B→A handover operation, and after the handover is completed, fly stably for 5 seconds. Record the second uplink communication rate, the second ping-pong handover count, and the second handover delay during the reverse flight process of the UAV, and calculate the corresponding parameter combination based on the handover score model. , , The reverse switching score.

[0088] S4: Repeat steps S2 and S3, calculate the average of the two forward and reverse switching scores respectively, and use the average as the baseline switching score for the first parameter combination. That is, the first switching score mentioned above.

[0089] S5: To further optimize the first parameter combination, it also includes: determining the median of the three-dimensional interval of the first parameter combination; increasing or decreasing the median by a preset step size at the target neighborhood point to obtain the second parameter combination corresponding to the target neighborhood point, wherein the target neighborhood point is any one of the 26 neighborhood points.

[0090] Specifically, the A3 offset, hysteresis, and TTT of cells A and B are adjusted via the base station API to a step size of +1 or -1 from the median value within the range (i.e., a preset step size), resulting in a new parameter combination (i.e., a second parameter combination). The second handover score is then calculated across all 26 neighborhoods based on the handover score model. For example, the median A3 event offset is 3dB, the median hysteresis is 3dB, and the median TTT is 640ms. If the preset step size is 1dB, the parameters for A3 event offset and hysteresis at neighboring points may change to 2dB, 4dB, or remain constant at 3dB; for TTT, if the step size is 160ms, the parameters may change to 480ms, 800ms, or remain constant at 640ms.

[0091] S6: Taking A→B handover as an example: If the second handover score of all neighboring points is lower than the benchmark first handover score... Then additional parameter combinations are executed ( , , If the results of two verification tests are consistent, then it is determined that () , , The optimal spatial parameter combination is determined by the following: If a neighboring point has a higher score, the search continues in that direction, and untested parameter combinations from the 26 neighbors of that point are tested and their corresponding second switching scores are recorded. After comparing the scores of all neighboring points, the direction with the highest score is selected to continue exploring until performance degrades. The same logic is used to optimize the B→A switching path.

[0092] S7: Repeat step S6 until a combination of spatial parameters is found ( , , If the switching score is higher than 26 neighboring points in 3 independent tests, the combination is determined to be the optimal 3D spatial parameter combination in the current scene, and the corresponding initial spatial parameter combination is entered into the rule base to update it.

[0093] Through the aforementioned three-dimensional 26-neighborhood search process, a rule base containing input features and corresponding measured optimal spatial parameter combinations can be determined. This dynamic optimization mechanism, combining quantitative evaluation of network performance with systematic testing of the neighboring parameter space, provides UAVs with an intelligent and adaptive parameter adjustment strategy during network switching, effectively improving the efficiency and stability of network switching.

[0094] In this embodiment of the application, in order to further improve the generalization ability of the ResNet prediction model under different base station environments, the nonlinear relationship between the neural network learning input features and the optimal combination of spatial parameters is utilized, so that it can predict the optimal three-dimensional continuous spatial parameters for any given continuous four-dimensional input parameters.

[0095] Optionally, the communication state data is processed by a prediction model to obtain the spatial parameter combination of the UAV. This includes: extracting features from the communication state data using a first convolutional layer in the prediction model to obtain a first feature; processing the first feature using multiple residual blocks in the prediction model to obtain a second feature, wherein each residual block includes multiple residual units, each residual unit includes a second convolutional layer and a residual connection structure, and the second convolutional layer is connected to an activation function to alleviate the gradient vanishing phenomenon during the processing of the first feature; and processing the second feature using a fully connected layer in the prediction model to obtain the spatial parameter combination. The model architecture and processing flow of the ResNet prediction model are described in detail below.

[0096] I. Data Construction

[0097] Based on a heuristic hierarchical rule base, a dataset containing the optimal combination of input and output is constructed, where each sample consists of the following:

[0098] 1) Input feature set: includes four continuous variables, namely UAV flight speed, RSRP, SINR and distance to base station. The input features are all actual values ​​and are not discretized.

[0099] 2) Output parameter set: The optimal combination of spatial parameters for the corresponding scenario, including A3 event bias, switching hysteresis value and switching trigger time, all of which are continuous real values.

[0100] II. Model Structure

[0101] A ResNet-based convolutional neural network model is employed. The input is a normalized four-dimensional continuous feature vector, representing the communication state data. This vector is passed through multiple one-dimensional convolutional layers and activation function layers to extract nonlinear features between the input features. Residual connection structures are used to enhance feature representation and training stability. The output layer, through a fully connected mapping, outputs a three-dimensional continuous vector, representing the optimal combination of spatial parameters. The following explains the deep neural network within the ResNet prediction model.

[0102] In deep neural networks, as the number of layers increases, the input features are learned directly. To complex output parameters Mapping can easily lead to gradient vanishing and network degradation. In this embodiment, residual blocks are introduced to address this problem. Figure 4 As shown, input features The residual function is obtained after convolution, activation, and other operations. and with The output is obtained by adding elements one by one. This skip connection structure greatly simplifies the learning objective, allowing gradients to flow more efficiently through the identity mapping path during backpropagation, effectively mitigating the gradient vanishing problem. Simultaneously, the network focuses on learning the residuals. This allows for more effective separation and utilization of useful features in different layers, avoiding network degradation.

[0103] like Figure 5 As shown, ResNet includes five network structures: 18, 34, 50, 101, and 152. In this embodiment, the ResNet-34 architecture is adopted, consisting of convolutional layers, residual blocks, and fully connected layers. The input is the preprocessed image-like format of the current UAV's communication status data (UAV speed, RSRP, SINR, and base station spacing). The specific processing procedure is as follows: The original data image is first input into a 7×7 convolutional layer with a stride of 2 and padding of 3 for initial feature extraction to obtain the first feature. Subsequently, the first feature output from the convolution is sequentially input into four residual block groups for processing, resulting in residual feature maps (i.e., the second feature). The first group of residual blocks contains 3 residual units, the second group contains 4, the third group contains 6, and the fourth group contains 3. Each residual unit contains two 3×3 convolutional layers with a stride of 1 and padding of 1, as well as a residual connection. Each convolutional layer is followed by batch normalization and the ReLU activation function. In the first unit of the second, third, and fourth residual blocks, the feature map size is halved while the number of channels is doubled by setting the stride of the first convolutional layer to 2. After processing by these four residual block groups, the resulting second feature is input into a fully connected layer, ultimately outputting a three-dimensional continuous vector, which represents the optimal combination of spatial parameters to be predicted.

[0104] III. Composite Loss Function

[0105] Optionally, the composite loss function is determined as follows: The combination of prediction parameters of the prediction model during training is obtained, and the hyperparameters for quantifying the prediction error of the prediction model are determined. A first loss function is determined based on the combination of prediction parameters, the combination of historical spatial parameters, and the hyperparameters. The prediction confidence, class weight factor, and focus factor required for training the target sample are determined, and a second loss function is determined based on the prediction confidence, class weight factor, and focus factor. Here, the target sample is a sample whose prediction difficulty is greater than a preset threshold; the class weight factor is used to quantify the weight of the target sample during training; and the focus factor is used to enhance the focusing ability on the target sample. The regularization coefficient and loss gradient required for training the prediction model are determined, and a third loss function is determined based on the regularization coefficient and loss gradient. Finally, a composite loss function is determined based on the first, second, and third loss functions.

[0106] In this embodiment, the composite loss function introduces multiple sub-loss terms to collaboratively address problems in traditional training such as instability, sample imbalance, large boundary prediction errors, and poor multi-objective fitting ability. These include the following three sub-terms:

[0107] (1) Basic prediction error term: The error loss function, i.e., the first loss function, such as Huber loss or MSE loss, is used to measure the numerical difference between the model's predicted output and the actual output value. The specific expression is as follows:

[0108]

[0109] In the formula, This represents the mean square error function. Indicates the first The actual output value of the sample is obtained directly from the corresponding historical spatial parameter combination from the rule base; Indicates the first The model prediction value for each sample.

[0110]

[0111] In the formula, Represents the Huber loss function; The hyperparameters representing the prediction error of the quantization prediction model serve as a turning point, determining the boundary between squared error and linear error; the error is less than a threshold. The error is the squared error, and the error is greater than the threshold. The error is linear, thus enhancing tolerance to outliers while maintaining accuracy. Weights can be increased in the early stages of training to improve learning efficiency, and then gradually decreased in the later stages to control the risk of overfitting.

[0112] (2) Focusing on the loss term: The Focal Loss mechanism is introduced to enhance the model's ability to focus on difficult-to-predict samples. The specific expression is as follows:

[0113]

[0114] In the formula, This represents the focusing loss function, also known as the second loss function; This represents the confidence level of the ResNet model's prediction of the current target. For category weighting factors, This is the focusing factor.

[0115] In this embodiment of the application, the focus loss function is... Focusing factor in To enhance the training focus on samples with a high probability of prediction errors; Class balance factor The loss value of the sample can be dynamically assigned based on the loss value during the training process:

[0116]

[0117] In the formula, Indicates the first The loss value for each sample. This represents the cumulative loss value for j samples.

[0118] (3) Gradient control and regularization: A gradient norm term is introduced to limit drastic gradient fluctuations during training, thereby improving training convergence and numerical stability. The specific expression is as follows:

[0119]

[0120] In the formula, Represents the third loss function; This represents the regularization coefficient, also known as the weight coefficient, which controls the proportion of the gradient regularization term in the total loss and is usually a hyperparameter. For the third loss function Regarding model parameters The gradient is the update direction and magnitude corresponding to each parameter in the model.

[0121] Finally, the above sub-items are combined into a composite loss function using a weighted strategy. The weights of each sub-item are dynamically adjusted based on the error descent trend, gradient fluctuation range, or output stability during the training process. The specific expression is as follows:

[0122]

[0123] In the formula, , , The weights for each loss function can be set through cross-validation or a dynamic adjustment mechanism.

[0124] IV. Model Training and Optimization

[0125] In this embodiment, the training process of the prediction model employs a dynamic sample weighting and iterative optimization strategy to enhance the model's performance in handling complex and edge scenarios. This includes: training the prediction model based on historical communication state data and determining the loss value of each training sample in the historical communication state data using a composite loss function; comparing the loss value with a preset threshold and identifying the training sample whose loss value is greater than or equal to the preset threshold as the target sample, and identifying the training sample whose loss value is less than the preset threshold as the normal sample; and performing the next round of training on the prediction model based on the target sample and the normal sample, wherein the target sample is assigned a higher weight value in the composite loss function than the normal sample.

[0126] Specifically, a ResNet network is trained based on the aforementioned composite loss function and optimized over multiple epochs to obtain a ResNet model that exhibits high fitting accuracy to input features, strong boundary prediction capabilities, and stable training. To enhance the model's ability to fit boundary samples and rare input combinations, a sample difficulty guidance mechanism is introduced:

[0127] S1: Iterate through all training samples in the current epoch, calculate the loss value of each sample using a composite loss function; select the sample with the largest loss value. A set of samples is used to identify the most difficult target samples to learn from, where samples with a loss value greater than or equal to a preset threshold are identified and their indices are recorded in the set. In the middle, samples with loss values ​​less than a preset threshold are classified as normal samples, and their indices are recorded in a set. middle.

[0128] S2: Add the above target samples to the next round of mini-batch training to form an enhanced training sample set, which is the sample set for the next round of model training. From the current normal sample set and difficult sample set composition:

[0129]

[0130] S3: For difficult sample sets The target sample is given a higher weight in the loss function:

[0131]

[0132] In the formula, It is a weighting factor for enhancing the influence of difficult samples.

[0133] By employing a composite loss function to evaluate training samples and dynamically dividing them into target and normal samples based on the loss values, and then adjusting sample weights for targeted training, this demonstrates an efficient optimization strategy in deep learning model training. This strategy not only improves the model's prediction accuracy when handling difficult samples but also ensures the stability and generalization ability of the model training, providing a solid technical foundation for intelligent prediction of UAV communication parameters. Through continuous iteration of this training process, the model can gradually reduce prediction errors, especially in parameter prediction under highly dynamic and complex environments, thereby achieving adaptive optimization of intelligent parameter switching and effectively improving the stability and efficiency of UAV network communication.

[0134] Assuming an initial learning rate of 0.001 and 100 training epochs, the model training performance is as follows: Figure 6 and Figure 7 As shown:

[0135] Figure 6 The diagram illustrates the trend of loss values ​​over 100 training epochs. As the number of training epochs increases, all four loss curves corresponding to the ResNet prediction model (total loss, base loss (Huber), Focal Loss, and gradient regularization) show a significant decreasing trend, reaching a stable convergence state after approximately 70 epochs. The total loss decreases from approximately 0.78 initially to around 0.14, indicating that as the number of iterations increases, the difference between the ResNet prediction model's predictions and actual results gradually decreases, demonstrating its ability to effectively learn the inherent patterns of parameter switching.

[0136] Figure 7 The training accuracy trend over 100 training rounds is shown. As the number of training rounds increases, the training accuracy of the ResNet prediction model steadily rises from approximately 46% initially to 88%, indicating that the ResNet prediction model's ability to predict combinations of spatial parameters gradually improves with training and possesses good generalization performance.

[0137] After completing offline model training and obtaining the optimal combination of spatial parameters, the system can be integrated and deployed to edge nodes of wireless communication systems, UAV platforms, or base station control modules to achieve, for example... Figure 2 The diagram illustrates intelligent prediction and real-time scheduling of spatial parameter combinations in a dynamic communication environment.

[0138] In this embodiment, a heuristic hierarchical rule base is constructed and combined with a three-dimensional 26-neighborhood search algorithm to achieve real-time dynamic adjustment of network handover parameters in high-speed UAV movement scenarios. Specifically, the rule base takes multi-dimensional environmental information such as UAV flight speed, RSRP, SINR, and base station spacing as input, and dynamically generates optimal handover control parameters through expert experience and environmental perception. The three-dimensional 26-neighborhood search algorithm can accurately verify and optimize the parameters in the rule base while significantly reducing the number of tests, thus significantly improving testing efficiency and parameter optimization accuracy. In addition, a ResNet prediction model based on composite loss function optimization is introduced, which can predict the optimal combination of spatial parameters for continuous input features, greatly enhancing the system's generalization ability and prediction accuracy. This effectively overcomes the limitations of traditional fixed-parameter handover mechanisms when facing highly dynamic UAV flight, significantly reduces ping-pong handover phenomena, and improves network handover stability and communication quality.

[0139] In summary, this approach not only effectively improves the communication efficiency and user experience of drones in low-altitude network environments, but also provides a feasible technical solution for future high-dynamic, high-density communication scenarios, which has significant theoretical implications and broad application prospects.

[0140] According to embodiments of this application, a spatial parameter optimization apparatus is provided. It should be noted that the spatial parameter optimization apparatus of this application can be used to execute the spatial parameter optimization method provided in the embodiments of this application. The spatial parameter optimization apparatus provided in the embodiments of this application will be described below.

[0141] Figure 8 This is a structural diagram of a spatial parameter optimization device provided according to an embodiment of this application. Figure 8 As shown, the device includes:

[0142] The acquisition module 80 is used to acquire communication status data of the UAV during flight.

[0143] The processing module 82 is used to process the communication status data through the prediction model to obtain the airspace parameter combination of the UAV. The prediction model is obtained by training with samples obtained from a preset rule base. The preset rule base includes the historical communication status data of the UAV and the corresponding historical airspace parameter combination. The historical airspace parameter combination is the airspace parameter after optimizing the initial airspace parameter combination corresponding to the historical communication status data using a neighborhood search algorithm. The prediction model also includes a composite loss function, which includes a first loss function for quantifying the numerical difference between the predicted value and the true value of the prediction model, a second loss function for weighting samples whose prediction difficulty is greater than a preset threshold, and a third loss function for controlling gradient fluctuations.

[0144] The optimization module 84 is used to adjust the initial control parameters of the UAV based on the combination of airspace parameters.

[0145] Through the acquisition module, processing module, and optimization module in the aforementioned airspace parameter optimization device, the goal of adaptively adjusting the UAV network switching parameters is achieved. This results in improving airspace switching efficiency and ensuring the quality and stability of wireless communication. Furthermore, it solves the technical problems of sluggish switching, frequent ping-pong effects, and uneven resource allocation caused by the reliance on fixed parameters in network switching mechanisms in related technologies during airspace mobility scenarios.

[0146] In the airspace parameter optimization device provided in this application embodiment, the processing module is further configured to acquire historical communication status data of the UAV, wherein the historical communication status data includes the UAV's flight speed, reference signal received power, reference signal received quality, and base station spacing in multiple historical flight scenarios; determine an initial airspace parameter combination corresponding to the historical communication status data, wherein the initial airspace parameter combination includes the event bias value, hysteresis value, and trigger time corresponding to the historical communication status data; optimize the initial airspace parameter combination using a neighborhood search algorithm to obtain a historical airspace parameter combination; and determine a preset rule base based on the historical communication status data and the historical airspace parameter combination.

[0147] In the airspace parameter optimization device provided in this application embodiment, the processing module is further configured to obtain a first parameter combination corresponding to target communication state data from a preset rule base, wherein the target communication state data is the communication state data of the UAV on a preset flight path; determine a first switching score corresponding to the first parameter combination, wherein the first switching score is used to reflect the network performance of the first parameter combination; determine the median of the three-dimensional interval of the first parameter combination as the starting point, and determine a second parameter combination at 26 neighboring points corresponding to the starting point; determine a second switching score corresponding to the second parameter combination, wherein the second switching score is used to reflect the network performance of the second parameter combination; determine a target parameter combination corresponding to the target communication state data based on the first switching score and the second switching score, and optimize the initial airspace parameter combination based on the target parameter combination, wherein the target parameter combination is the parameter combination with the highest score among the first switching score and the second switching score.

[0148] In the airspace parameter optimization device provided in this application embodiment, the processing module is further configured to determine, based on the positive parameter combination in the first parameter combination, a first uplink rate, a first ping-pong switching rate, and a first switching delay of the UAV on a preset flight path, wherein the positive parameter combination is a parameter combination corresponding to the communication status data of the UAV when flying forward on the preset flight path; determine, based on the negative parameter combination in the first parameter combination, a second uplink rate, a second ping-pong switching rate, and a second switching delay of the UAV on the preset flight path, wherein the negative parameter combination is a parameter combination corresponding to the communication status data of the UAV when flying backward on the preset flight path; determine, based on the first uplink rate, the first ping-pong switching rate, and the first switching delay, a positive switching score corresponding to the positive parameter combination, and determine, based on the second uplink rate, the second ping-pong switching rate, and the second switching delay, a negative switching score corresponding to the negative parameter combination; and determine, based on the positive switching score and the negative switching score, a first switching score.

[0149] In the spatial parameter optimization device provided in this application embodiment, the processing module is further used to determine the median of the three-dimensional interval of the first parameter combination; at the target neighborhood point, the median is increased or decreased by a preset step size to obtain the second parameter combination corresponding to the target neighborhood point, wherein the target neighborhood point is any one of the 26 neighborhood points.

[0150] In the spatial parameter optimization device provided in this application embodiment, the processing module is further configured to extract features from the communication state data through the first convolutional layer in the prediction model to obtain a first feature; process the first feature through multiple residual blocks in the prediction model to obtain a second feature, wherein the residual block includes multiple residual units, each residual unit includes a second convolutional layer and a residual connection structure, the second convolutional layer is connected to an activation function to alleviate the gradient vanishing phenomenon of the first feature during processing; and process the second feature through a fully connected layer in the prediction model to obtain a spatial parameter combination.

[0151] In the spatial parameter optimization apparatus provided in this application embodiment, the processing module is further configured to obtain the prediction parameter combination of the prediction model during the training process, and determine the hyperparameters that quantify the prediction error of the prediction model; determine a first loss function based on the prediction parameter combination, historical spatial parameter combination, and hyperparameters; determine the prediction confidence, class weight factor, and focusing factor required for training the target sample, and determine a second loss function based on the prediction confidence, class weight factor, and focusing factor, wherein the target sample is a sample whose prediction difficulty is greater than a preset threshold, the class weight factor is used to quantify the weight of the target sample during the training process, and the focusing factor is used to enhance the focusing ability of the target sample; determine the regularization coefficient and loss gradient required for training the prediction model, and determine a third loss function based on the regularization coefficient and loss gradient; and determine a composite loss function based on the first loss function, the second loss function, and the third loss function.

[0152] In the spatial parameter optimization device provided in this application embodiment, the processing module is further configured to train the prediction model based on historical communication state data, and determine the loss value of each training sample in the historical communication state data through a composite loss function; compare the loss value with a preset threshold, and determine the training sample corresponding to the loss value being greater than or equal to the preset threshold as the target sample, and determine the training sample corresponding to the loss value being less than the preset threshold as the normal sample; and perform the next round of training on the prediction model based on the target sample and the normal sample, wherein the weight value assigned to the target sample in the composite loss function is higher than the weight value assigned to the normal sample in the composite loss function.

[0153] This application also provides an electronic device, including: a memory and a processor, wherein the memory is used to store program instructions; and the processor is connected to the memory and used to execute the above-described spatial parameter optimization method.

[0154] It should be noted that the aforementioned electronic equipment is used to perform Figure 2 The spatial parameter optimization method shown above is also applicable to this electronic device, and will not be repeated here.

[0155] This application also provides a non-volatile storage medium, which includes a stored computer program, wherein the device containing the non-volatile storage medium executes the above-described spatial parameter optimization method by running the computer program.

[0156] It should be noted that the aforementioned non-volatile storage media is used for execution. Figure 2 The spatial parameter optimization method shown above is also applicable to this non-volatile storage medium, and will not be repeated here.

[0157] This application also provides a computer program product, including computer instructions that, when executed by a processor, implement the above-described spatial parameter optimization method.

[0158] It should be noted that the above-mentioned computer program product is used to execute... Figure 2 The spatial parameter optimization method shown above is also applicable to this computer program product, and will not be repeated here.

[0159] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0160] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0161] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For instance, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual coupling, direct coupling, or communication connection may be through some interfaces; the indirect coupling or communication connection between units or modules may be electrical or other forms.

[0162] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0163] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0164] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard drive, magnetic disk, or optical disk.

[0165] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for optimizing spatial parameters, characterized in that, The method comprises: acquiring communication state data of a UAV in flight; processing the communication state data by a prediction model to obtain airspace parameter combinations of the UAV, wherein the prediction model is trained by samples obtained from a preset rule library, the preset rule library comprising historical communication state data of the UAV and corresponding historical airspace parameter combinations, the historical airspace parameter combinations being airspace parameters obtained by optimizing initial airspace parameter combinations corresponding to the historical communication state data by a neighborhood search algorithm, the prediction model further comprising a composite loss function, the composite loss function comprising a first loss function for quantifying a numerical difference between predicted values and true values of the prediction model, a second loss function for weighting samples with a prediction difficulty greater than a preset threshold, and a third loss function for controlling gradient fluctuations; optimizing the initial airspace parameter combinations by the neighborhood search algorithm, comprising: obtaining a first parameter combination corresponding to target communication state data from the preset rule library, wherein the target communication state data is communication state data of the UAV on a preset flight path; determining a first switching score corresponding to the first parameter combination, wherein the first switching score is used to reflect network performance of the first parameter combination; determining a median value of a three-dimensional interval of the first parameter combination as a starting point, and determining second parameter combinations on 26 neighborhood points corresponding to the starting point, wherein the three-dimensional interval is composed of an event bias value, a hysteresis value and a trigger time in the first parameter combination; determining a second switching score corresponding to the second parameter combination, wherein the second switching score is used to reflect network performance of the second parameter combination; determining a target parameter combination corresponding to the target communication state data according to the first switching score and the second switching score, and optimizing the initial airspace parameter combination according to the target parameter combination, wherein the target parameter combination is the parameter combination with the highest score in the first switching score and the second switching score; adjusting initial control parameters of the UAV according to the airspace parameter combinations.

2. The method of claim 1, wherein, The preset rule library is determined by: acquiring historical communication state data of the UAV, wherein the historical communication state data comprises flight speeds, reference signal received powers, reference signal received qualities and base station distances of the UAV in multiple historical flight scenarios; determining initial airspace parameter combinations corresponding to the historical communication state data, wherein the initial airspace parameter combinations comprise event bias values, hysteresis values and trigger times corresponding to the historical communication state data; optimizing the initial airspace parameter combinations by a neighborhood search algorithm to obtain the historical airspace parameter combinations; determining the preset rule library according to the historical communication state data and the historical airspace parameter combinations.

3. The method of claim 1, wherein, determining a first switching score corresponding to the first parameter combination, comprising: determine a first uplink rate, a first ping-pong switching rate and a first switching time delay of the UAV on the preset flight path according to a forward parameter combination in the first parameter combination, wherein the forward parameter combination corresponds to the communication state data when the UAV flies forward on the preset flight path; determine a second uplink rate, a second ping-pong switching rate and a second switching time delay of the UAV on the preset flight path according to a reverse parameter combination in the first parameter combination, wherein the reverse parameter combination corresponds to the communication state data when the UAV flies reverse on the preset flight path; determine a forward switching score corresponding to the forward parameter combination according to the first uplink rate, the first ping-pong switching rate and the first switching time delay, and determine a reverse switching score corresponding to the reverse parameter combination according to the second uplink rate, the second ping-pong switching rate and the second switching time delay; determine the first switching score according to the forward switching score and the reverse switching score.

4. The method of claim 1, wherein, determine a second parameter combination corresponding to each of the 26 neighboring points of the starting point, including: determine a median value of a three-dimensional interval of the first parameter combination; add or subtract a preset step length to the median value at a target neighboring point to obtain a second parameter combination corresponding to the target neighboring point, wherein the target neighboring point is any one of the 26 neighboring points.

5. The method of claim 1, wherein, processing the communication state data through a prediction model to obtain the airspace parameter combination of the UAV, including: extracting features of the communication state data through a first convolutional layer in the prediction model to obtain first features; processing the first features through a plurality of residual blocks in the prediction model to obtain second features, wherein each of the residual blocks includes a plurality of residual units, and each of the residual units includes a second convolutional layer and a residual connection structure, and the second convolutional layer is connected with an activation function and is used to alleviate the gradient vanishing phenomenon of the first features in the processing process; processing the second features through a fully connected layer in the prediction model to obtain the airspace parameter combination.

6. The method of claim 1, wherein, The composite loss function is determined by: obtaining a predicted parameter combination of the prediction model in a training process, and determining a hyperparameter for quantifying a prediction error of the prediction model, and determining the first loss function according to the predicted parameter combination, the historical airspace parameter combination and the hyperparameter; determining a prediction confidence, a category weight factor and a focusing factor required when training a target sample, and determining the second loss function according to the prediction confidence, the category weight factor and the focusing factor, wherein the target sample is a sample with a prediction difficulty greater than a preset threshold, the category weight factor is used to quantify the weight of the target sample in the training process, and the focusing factor is used to enhance the focusing ability on the target sample; determining a regularization coefficient and a loss gradient required when training the prediction model, and determining the third loss function according to the regularization coefficient and the loss gradient; The composite loss function is determined according to the first loss function, the second loss function and the third loss function.

7. The method of claim 6, wherein, The method further comprises: The prediction model is trained according to the historical communication state data, and a loss value of each training sample in the historical communication state data is determined through the composite loss function; The loss value and a preset threshold value are compared, and a training sample corresponding to the loss value greater than or equal to the preset threshold value is determined as a target sample, and a training sample corresponding to the loss value less than the preset threshold value is determined as a normal sample; The prediction model is trained according to the target sample and the normal sample in the next round, wherein a weight value of the target sample in the composite loss function is higher than a weight value of the normal sample in the composite loss function.

8. An apparatus for optimizing spatial parameters, characterized by Comprise: The communication state data of the unmanned aerial vehicle in the flight process is acquired; The communication state data is processed through a prediction model to obtain the airspace parameter combination of the unmanned aerial vehicle, wherein the prediction model is trained by acquiring samples from a preset rule library, the preset rule library comprises historical communication state data and corresponding historical airspace parameter combination of the unmanned aerial vehicle, the historical airspace parameter combination is an airspace parameter obtained by optimizing an initial airspace parameter corresponding to the historical communication state data by using a neighborhood search algorithm, the prediction model further comprises a composite loss function, the composite loss function comprises a first loss function for quantifying a numerical difference between a predicted value and an actual value of the prediction model, a second loss function for weighting a sample with a prediction difficulty greater than a preset threshold, and a third loss function for controlling gradient fluctuation; the initial airspace parameter combination is optimized by using a neighborhood search algorithm, comprising: acquiring a first parameter combination corresponding to target communication state data from the preset rule library, wherein the target communication state data is communication state data of the unmanned aerial vehicle on a preset flight path; determining a first switching score corresponding to the first parameter combination, wherein the first switching score is used to reflect network performance of the first parameter combination; determining a median value of a three-dimensional interval of the first parameter combination as a starting point, and determining second parameter combinations on 26 neighborhood points corresponding to the starting point, wherein the three-dimensional interval is composed of an event bias value, a hysteresis value and a trigger time in the first parameter combination; determining a second switching score corresponding to the second parameter combination, wherein the second switching score is used to reflect network performance of the second parameter combination; determining a target parameter combination corresponding to the target communication state data according to the first switching score and the second switching score, and optimizing the initial airspace parameter combination according to the target parameter combination, wherein the target parameter combination is a parameter combination with the highest score in the first switching score and the second switching score; The initial control parameter of the unmanned aerial vehicle is adjusted according to the airspace parameter combination.

9. An electronic device, comprising: Comprise: A memory and a processor, wherein the memory is configured to store program instructions; The processor is connected with the memory and is configured to execute the spatial parameter optimization method according to any one of claims 1 to 7.

10. A non-volatile storage medium, comprising: The non-volatile storage medium comprises a stored computer program, wherein a device in which the non-volatile storage medium is located executes the spatial parameter optimization method according to any one of claims 1 to 7 by running the computer program.

11. A computer program product comprising computer instructions, characterized in that, The computer instructions are executed by the processor to implement the spatial parameter optimization method according to any one of claims 1 to 7.

Images