Low-altitude flight communication network resource allocation method based on large model algorithm

Abstract

The application relates to the technical field of resource allocation, and specifically discloses a low-altitude flight communication network resource allocation method based on a large model algorithm. The method acquires a UAV task instruction, performs intention analysis and demand prediction based on an LLM model to capture communication network resource demand characteristics, and further combines network state data and environment data, fuses network state characteristics, environment state characteristics and resource demand characteristics to construct a comprehensive state expression of a UAV task execution scene, and then makes intelligent resource allocation decisions to generate and execute network resource allocation instructions. In this way, intelligent allocation of low-altitude flight communication network resources is realized, resource utilization efficiency and task execution capability are improved, the accuracy and real-time performance of resource allocation are effectively ensured, and the intelligent level of resource allocation is further improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of resource allocation, and more specifically, to a low-altitude flight communication network resource allocation method based on a large model algorithm. BACKGROUND

[0002] Under the background of rapid development of low-altitude flight communication network, unmanned aerial vehicle task resource allocation has become an important research direction in the field of communication. Efficient allocation of communication resources is of great significance to ensure the real-time, reliability and flexibility of unmanned aerial vehicle task execution. Traditional communication resource allocation methods mainly rely on fixed quality of service (QoS) parameter inputs such as bandwidth, delay and reliability requirements. These methods have obvious limitations when faced with complex and variable unmanned aerial vehicle task scenarios, and lack the ability to respond to dynamic changes in the task execution phase and environmental conditions (such as weather, interference, network congestion), and cannot adjust resource allocation strategies in real time to meet task requirements.

[0003] Specifically, traditional resource allocation methods are mainly based on QoS parameter hard coding mechanism, which realizes network resource scheduling by manually configuring fixed thresholds such as bandwidth, delay and reliability. Such methods are limited by pre-set fixed thresholds and are difficult to adapt to dynamic changes in unmanned aerial vehicle task requirements, resulting in low resource allocation efficiency and inability to meet the diversified needs of complex task scenarios. That is, due to the high time-varying nature of low-altitude unmanned aerial vehicle task requirements (such as sudden high-definition video backhaul, emergency obstacle avoidance instructions), when faced with dynamic environmental changes, existing technologies lack the ability to dynamically reallocate resources, which may result in "over-provisioning" or "under-provisioning" of resources, for example, the sudden high bandwidth demand of a reconnaissance task when a suspicious target is found, which may cause critical data transmission to be interrupted due to network congestion.

[0004] Therefore, an optimized low-altitude flight communication network resource allocation method based on a large model algorithm is expected. SUMMARY

[0005] To solve the above technical problems, the present application is proposed. The embodiments of the present application provide a low-altitude flight communication network resource allocation method based on a large model algorithm, which acquires a UAV task instruction, and performs intention analysis and demand prediction based on an LLM model to capture communication network resource demand characteristics. Meanwhile, network state data and environment data are further combined, and network state characteristics, environment state characteristics and resource demand characteristics are fused to construct a comprehensive state expression of a UAV task execution scene, and then intelligent resource allocation decisions are made based on this to generate and execute network resource allocation instructions. In this way, intelligent allocation of low-altitude flight communication network resources is achieved, resource utilization efficiency and task execution capability are improved, the accuracy and real-time performance of resource allocation can be effectively ensured, and the intelligent level of resource allocation is further improved.

[0006] According to an aspect of the present application, a low-altitude flight communication network resource allocation method based on a large model algorithm is provided, which includes:

[0007] acquiring a UAV task instruction;

[0008] performing intention analysis and demand prediction based on an LLM model on the UAV task instruction to obtain a future resource demand structured coding vector;

[0009] acquiring network state data and environment data;

[0010] structurally processing the network state data and the environment data using a network state embedding coding matrix and an environment data embedding coding matrix to obtain a network state structured coding vector and an environment state structured coding vector;

[0011] fusing the future resource demand structured coding vector, the network state structured coding vector and the environment state structured coding vector to obtain a comprehensive state feature vector;

[0012] inputting the comprehensive state feature vector into an intelligent resource allocation decision module to obtain a network resource allocation instruction, which is transmitted to a network controller for execution.

[0013] Compared with the prior art, the low-altitude flight communication network resource allocation method based on the large model algorithm provided by the application can capture the communication network resource demand characteristics by obtaining the unmanned aerial vehicle task instruction and performing intention analysis and demand prediction based on the LLM model, and further combine network state data and environment data to fuse the network state characteristics, environment state characteristics and resource demand characteristics to construct the comprehensive state expression of the unmanned aerial vehicle task execution scene, and then make intelligent resource allocation decision based on this to generate and execute network resource allocation instruction. In this way, intelligent allocation of low-altitude flight communication network resources is realized, resource utilization efficiency and task execution capability are improved, the accuracy and real-time performance of resource allocation can be effectively ensured, and the intelligent level of resource allocation is further improved. BRIEF DESCRIPTION OF DRAWINGS

[0014] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description of embodiments of the present application taken in conjunction with the accompanying drawings. The accompanying drawings are provided to aid in the understanding of the present application and constitute a part of the specification, together with the written description, to explain the present application and, together with the written description, serve to explain the present application. The drawings do not limit the present application. In the drawings, the same reference numerals generally represent the same components or steps.

[0015] Figure 1 A flowchart of the low-altitude flight communication network resource allocation method based on the large model algorithm according to the embodiments of the present application.

[0016] Figure 2 A data flow diagram of the low-altitude flight communication network resource allocation method based on the large model algorithm according to the embodiments of the present application.

[0017] Figure 3 A flowchart of sub-step S2 of the low-altitude flight communication network resource allocation method based on the large model algorithm according to the embodiments of the present application.

[0018] Figure 4 A flowchart of sub-step S22 of the low-altitude flight communication network resource allocation method based on the large model algorithm according to the embodiments of the present application.

[0019] Figure 5 A flowchart of sub-step S222 of the low-altitude flight communication network resource allocation method based on the large model algorithm according to the embodiments of the present application. DETAILED DESCRIPTION

[0020] As shown in the present application and claims, unless the context clearly indicates otherwise, the words "a," "an," "the," and / or "this" do not necessarily refer to the singularly identified item, but can comprise (1) the singularly identified item or (2) a plurality of the identified items. In general, the terms "comprise" and "contain" are intended to be open-ended, meaning that the methods and devices described can include other steps or elements not expressly listed or implied by such terms in addition to those enumerated specifically.

[0021] Although the present application makes various references to certain modules in the system according to embodiments of the present application, however, any number of different modules can be used and run on the user terminal and / or server. The modules are merely illustrative, and different aspects of the system and method can use different modules.

[0022] Flowcharts are used in the present application to illustrate the operations performed by the system according to embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in sequence. On the contrary, various steps can be processed in reverse order or simultaneously, as needed. Meanwhile, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0023] In the following, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It is obvious that the described embodiments are only a part of the embodiments of the present application, and not all embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein.

[0024] It is worth noting that in the present application, all actions of obtaining data are carried out in compliance with the corresponding data protection regulations and policies of the country where the device is located, and with the authorization given by the corresponding device owner.

[0025] In view of the technical problems described in the foregoing background, the present application proposes a mixed solid waste resource processing method, which obtains a UAV task instruction, and performs intent analysis and demand prediction based on an LLM model to capture communication network resource demand characteristics. Meanwhile, network state data and environmental data are further combined, and network state characteristics, environmental state characteristics, and resource demand characteristics are fused to construct a comprehensive state expression of the UAV task execution scene, and then intelligent resource allocation decisions are made based on this to generate and execute network resource allocation instructions. In this way, intelligent allocation of low-altitude flight communication network resources is achieved, resource utilization efficiency and task execution capability are improved, and the accuracy and real-time performance of resource allocation can be effectively ensured, further improving the intelligent level of resource allocation.

[0026] Figure 1 Flowchart of the low-altitude flight communication network resource allocation method based on large model algorithm according to embodiments of the present application. Figure 2A data flow diagram of a low-altitude flight communication network resource allocation method based on a large model algorithm according to an embodiment of the present application. As shown in Figure 1 and Figure 2 The low-altitude flight communication network resource allocation method based on a large model algorithm includes the following steps: S1, obtaining a UAV task instruction; S2, performing intent analysis and demand prediction based on an LLM model on the UAV task instruction to obtain a future resource demand structured coding vector; S3, obtaining network state data and environment data; S4, using a network state embedding coding matrix and an environment data embedding coding matrix to perform structured processing on the network state data and the environment data to obtain a network state structured coding vector and an environment state structured coding vector; S5, fusing the future resource demand structured coding vector, the network state structured coding vector, and the environment state structured coding vector to obtain a comprehensive state feature vector; S6, inputting the comprehensive state feature vector into an intelligent resource allocation decision module to obtain a network resource allocation instruction, which is transmitted to a network controller for execution.

[0027] In the low-altitude flight communication network resource allocation method based on a large model algorithm described above, the step S1, obtaining a UAV task instruction. It should be understood that, due to the high dynamic nature of the UAV task scenario, for example, sudden high-definition video backhaul or emergency obstacle avoidance instructions will cause sharp fluctuations in bandwidth demand, and relying only on preset QoS parameters cannot respond to such changes in time. Therefore, the present application captures the original input of the UAV task instruction (such as task type, target parameter, operation sequence, etc.) in real time, converts it into an initial demand signal that can be understood by the system, provides basic data support for subsequent resource demand prediction and dynamic adjustment, so that the system can real-time perceive the dynamic changes in the task execution phase (such as sudden high-bandwidth demand or priority transition of emergency obstacle avoidance instructions), avoid resource allocation lag or failure caused by relying on fixed thresholds. For example, when the UAV triggers high-definition video backhaul due to the discovery of a suspicious target, the immediate acquisition of the task instruction can trigger the priority reconstruction of the resource allocation process, thereby reducing the risk of interruption of critical data transmission, while reducing the resource redundancy occupation of non-emergency tasks, improving the overall efficiency and reliability of the network.

[0028] In the process of implementation, in order to realize the accurate task instruction acquisition, it is necessary to establish an efficient mechanism to capture these information in real time. The mechanism first needs to design a reliable communication link to ensure that the unmanned aerial vehicle and the ground control station can stably exchange data. This usually involves the use of advanced wireless communication technology, such as 5G network, so as to maintain high-quality data transmission even in complex and variable environment. In this way, no matter where the unmanned aerial vehicle is located or what weather conditions it faces, the task instructions it sends out can be received in time and accurately. In addition, considering the high dynamic nature of unmanned aerial vehicle task scenarios, such as the demand for high-definition video backhaul in emergency situations or emergency obstacle avoidance instructions, this communication link also needs to have fast response capability, which can adapt to environmental changes in a short time and quickly adjust its own state to meet the new task requirements.

[0029] Further, in the process of implementation, the unmanned aerial vehicle may receive multiple different types of task instructions at the same time. For example, on the one hand, it needs to continuously monitor a certain area and upload monitoring data regularly, and on the other hand, it may suddenly encounter obstacles and need to take immediate measures to avoid obstacles. In this case, the instruction acquisition mechanism not only needs to be able to distinguish between different types of instructions, but also needs to sort these instructions according to the preset priority rules to ensure that high-priority tasks can be processed first. For example, when the unmanned aerial vehicle detects obstacles in front, the related obstacle avoidance instructions should be immediately identified and trigger the corresponding action sequence, such as adjusting the flight path or changing the flight speed, etc. At the same time, for those non-emergency task instructions, they can continue to be executed or temporarily shelved under the premise of ensuring safety.

[0030] In addition to the task instructions obtained directly from the unmanned aerial vehicle, the system's database may also store a large amount of data records about similar past tasks. These historical data are of great significance to understand the context of the current task instructions. For example, by studying the instructions generated during the execution of similar tasks in the past, it can be found that there are certain task patterns or rules. Using this information, the system can help more accurately predict the types of tasks that may occur in the future and their specific requirements. For example, if the analysis results show that emergency obstacle avoidance events frequently occur in a certain area during a certain period of time, the route can be planned to avoid this area in the future, thereby reducing the probability of such events. Such practices not only help to improve the safety and efficiency of task execution, but also provide an important reference for subsequent resource allocation decisions.

[0031] It is worth noting that in the implementation of the UAV task instruction acquisition process, special attention should be paid to how to effectively integrate information from different sources. In addition to the historical data mentioned above, it also includes real-time collected environmental data (such as weather forecasts, air traffic conditions) and network status data (such as current available bandwidth, network delay). All these information together constitute the complete background information required for the UAV to perform the task. Therefore, building a platform that can automatically collect, organize and analyze these diverse information not only helps the system better understand the specific requirements of each individual task, but also can dig out some common trends or characteristics from it, and then guide future task planning and resource allocation strategies. For example, if the system monitors that the current network congestion is serious, which may affect the upcoming high-definition video backhaul task, then it can make corresponding adjustments in advance, such as optimizing routing selection or reserving more bandwidth resources, to ensure the smooth transmission of critical data.

[0032] In the low-altitude flight communication network resource allocation method based on large model algorithm described above, the step S2, the intention analysis and demand prediction of the UAV task instruction based on the LLM model are performed to obtain the future resource demand structured coding vector. Specifically, in the present application, the intention analysis and demand prediction of the UAV task instruction by the LLM model (Large Language Model) can deeply understand the task semantics and mine the implicit dynamic resource demand pattern, convert the task target into a quantifiable future resource demand structured coding vector, combine the historical data with the real-time context to predict the resource demand inflection point that may occur in the task execution process, so as to realize the real-time matching of network resource supply and dynamic demand in the complex and variable low-altitude scene, effectively reduce the transmission interruption risk caused by resource estimation deviation, and improve the network resource utilization rate and the task execution success rate. Among them, Figure 3 The flow chart of the sub-step S2 of the low-altitude flight communication network resource allocation method based on large model algorithm according to the embodiment of the present application is shown as follows. Figure 3 As shown in the figure, the step S2 includes the following steps: S21, after the tokenization processing of the UAV task instruction, input the text encoder to obtain the sequence distribution of the UAV task word granularity semantic embedding coding vector; S22, the sequence distribution of the UAV task word granularity semantic embedding coding vector is encoded by the UAV task word granularity context semantic correlation to obtain the UAV task semantic coding vector; S23, the UAV task semantic coding vector is input into the demand predictor based on the LLM model to obtain the future resource demand structured coding vector.

[0033] Specifically, the step S21 inputs the unmanned aerial vehicle task instruction after word segmentation processing to obtain a sequence distribution of unmanned aerial vehicle task word granularity semantic embedding coding vectors. Specifically, since the unmanned aerial vehicle task instruction often describes complex requirements in natural language form, its semantic connotation has the characteristics of multi-granularity, strong correlation and dynamic context dependence, therefore, the present application first disassembles the continuous unmanned aerial vehicle task instruction into word granularity units with independent semantics through word segmentation processing, and inputs the text encoder to generate a sequence distribution of unmanned aerial vehicle task word granularity semantic embedding coding vectors at the word level, which can convert unstructured text instructions into calculable mathematical representations while retaining the grammatical relationships and semantic hierarchies between words in the instructions, thereby providing a fine-grained and semantically consistent basic input for the fusion of network state and environmental data. In a specific example of the present application, the text encoder is a word embedding coding model based on the BERT model. It should be understood that the BERT model is a pre-trained language representation model that can capture complex semantic relationships and context dependencies between words through unsupervised learning of large-scale corpora, generating high-quality word-level semantic embeddings. On this basis, the present application uses the BERT model to perform deep semantic understanding of each word unit in the unmanned aerial vehicle task instruction, converts key information (such as task type, target parameter, operation sequence, etc.) in the task description from unstructured text to vector representation in a high-dimensional semantic space, and obtains a sequence distribution of unmanned aerial vehicle task word granularity semantic embedding coding vectors to facilitate subsequent task intent understanding and demand prediction.

[0034] Specifically, the step S22 performs unmanned aerial vehicle task word granularity context semantic association coding on the sequence distribution of unmanned aerial vehicle task word granularity semantic embedding coding vectors to obtain unmanned aerial vehicle task semantic coding vectors. It should be understood that since the semantics of the unmanned aerial vehicle task instruction has dynamic temporal and multi-dimensional correlation, the sequence distribution of unmanned aerial vehicle task word granularity semantic embedding coding vectors at a single word granularity cannot capture the spatio-temporal evolution relationship between key semantic units in the task scenario. Therefore, the present application further performs context semantic association coding on the sequence distribution of unmanned aerial vehicle task word granularity semantic embedding coding vectors based on spatio-temporal collaborative constraints to enhance the accuracy of task intent analysis, dynamically analyze the semantic dependency between words in the task instruction, and map discrete word-level semantics to continuous unmanned aerial vehicle task semantic coding vectors, providing fine-grained semantic input with spatio-temporal reasoning capability for the resource allocation module. Wherein, Figure 4 The flow chart of the sub-step S22 of the low-altitude flight communication network resource allocation method based on the large model algorithm according to the embodiment of the present application. As Figure 4As shown, step S22 includes the following steps: S221, inputting the sequence distribution of the UAV task word-granular semantic embedding encoding vector into a sequence encoder based on a recurrent neural network to obtain the sequence distribution of the UAV task semantic transmission encoding vector; S222, calculating the message transmission spatiotemporal coordination constraint factor of each UAV task semantic transmission encoding vector in the sequence distribution of the UAV task semantic transmission encoding vector; S223, dynamically adjusting and aggregating each UAV task semantic transmission encoding vector based on the message transmission spatiotemporal coordination constraint factor to obtain the UAV task semantic encoding vector.

[0035] More specifically, step S221 is expressed by the formula:

[0036]

[0037]

[0038] in, This represents the sequence distribution of the semantic embedding encoding vectors at the word granularity of the UAV task. , , and These represent the 1st, 2nd, and 3rd words in the sequence distribution of the semantic embedding encoding vector at the word granularity of the UAV task. The and the first Each UAV task term-level semantic embedding encoding vector , , and These represent the 1st, 2nd, and 3rd elements in the sequence distribution of the semantic transfer encoding vector for UAV missions. The and the first A single UAV mission semantic transfer encoding vector This represents a recurrent neural network.

[0039] In other words, by using an RNN model to progressively pass through and accumulate features from the sequence distribution of the UAV task word-level semantic embedding encoding vectors, deep implicit patterns of task requirements in the temporal context are extracted, generating a comprehensive feature expression with temporal context awareness, namely the sequence distribution of the UAV task semantic transmission encoding vectors. This dynamic feature extraction method provides an interpretable and traceable contextual information foundation for subsequent intelligent resource allocation decisions.

[0040] Figure 5 This is a flowchart of sub-step S222 of the low-altitude flight communication network resource allocation method based on the large model algorithm according to an embodiment of this application. Figure 5As shown, the step S222 includes steps of: S2221, calculating a time sequence confidence constraint factor of each unmanned aerial vehicle task semantic transfer encoding vector in the sequence distribution of the unmanned aerial vehicle task semantic transfer encoding vector; S2222, calculating a space confidence constraint factor of each unmanned aerial vehicle task semantic transfer encoding vector in the sequence distribution of the unmanned aerial vehicle task semantic transfer encoding vector; S2223, constructing a message passing space-time collaborative constraint factor of the each unmanned aerial vehicle task semantic transfer encoding vector based on the space confidence constraint factor and the time sequence confidence constraint factor of the each unmanned aerial vehicle task semantic transfer encoding vector.

[0041] In one specific example of the present application, the step S2221 is expressed by a formula as follows:

[0042]

[0043] wherein, and respectively represent different weight parameters, and respectively represent different weight matrices, represents a hyperbolic tangent function, represents matrix multiplication, represents an unmanned aerial vehicle task semantic time sequence context transfer feature importance score conversion vector, represents an unmanned aerial vehicle task semantic transfer time sequence context confidence factor, represents a softmax normalization function, represents a time sequence confidence constraint factor.

[0044] That is, by weighting and fusing the original unmanned aerial vehicle task word granularity semantic embedding encoding vector and the context transferred unmanned aerial vehicle task semantic transfer encoding vector, and using the nonlinear mapping capability of the neural network, the correlation features between the two are captured to dynamically evaluate the confidence weight of each unmanned aerial vehicle task word granularity semantic embedding encoding vector in the context. Specifically, based on the generated time sequence confidence constraint factor, noise suppression and information enhancement can be achieved, that is, by reducing the word granularity weight of low confidence, the interference of the global resource allocation is reduced, while the key features of high confidence word granularity are amplified, providing high-quality feature input for subsequent resource demand prediction, thereby significantly optimizing the screening efficiency and decision reliability of information.

[0045] In one specific example of the present application, the step S2222 is expressed by a formula as follows:

[0046]

[0047] wherein, a spatial confidence score of denotes an exponential function operation with base e, denotes a spatial confidence constraint factor of

[0048] That is, through the structured information embedding, the topology awareness ability of resource allocation is enhanced. Specifically, the spatial confidence constraint factor guides the model to prioritize the resource demand of high-structure-value nodes and suppress the redundant requests of low-value nodes in subsequent decision-making by quantifying the importance of node spatial roles. In this structured way, the dynamic adaptation ability of the model is optimized, so that the resource allocation strategy is coordinated with the node spatial function to avoid local congestion or global inefficiency caused by ignoring structural heterogeneity. Based on the spatial confidence constraint factor generated in this way, the topology rationality and propagation efficiency of network resource allocation are significantly improved. In addition, through the priority allocation of high-confidence nodes, efficient information transmission on the critical path is promoted, while the noise propagation of non-critical nodes is suppressed, so that the subsequent resource allocation decision can generate more accurate network scheduling instructions based on the reliable structural features, and finally realize efficient and robust resource management of low-altitude flight communication network under complex dynamic topology.

[0049] In one specific example of the present application, the step S2223 includes: inputting the spatial confidence constraint factor and the temporal confidence constraint factor of the UAV task semantic transmission encoding vector after weighted fusion to a sigmoid function for normalization processing to obtain a message transmission space-time coordination constraint factor of the UAV task semantic transmission encoding vector, which is expressed by the formula:

[0050]

[0051] wherein, and respectively represent different constraint weight parameters, denotes a sigmoid function, denotes a message transmission space-time coordination constraint factor of

[0052] That is, through the space-time coordination weight fusion, the confidence weight of the UAV task semantic transmission encoding vector in the spatial position and the temporal context can be effectively balanced, so as to more comprehensively evaluate the comprehensive importance of each UAV task semantic transmission encoding vector in the complex dynamic environment. In addition, through the normalization processing of the sigmoid function, the message transmission space-time coordination constraint factor of the fused UAV task semantic transmission encoding vector is further mapped to the interval [0, 1], realizing numerical stability and providing standardized weight input for subsequent message transmission encoding. ​

[0053] More specifically, in one preferred example of the present application, the step S223 comprises: first, performing multi-dimensional constraint manifold-based spatiotemporal coupling compensation optimization on the message passing spatiotemporal coordination constraint factor of the respective UAV task semantic passing coding vector to obtain the optimized message passing spatiotemporal coordination constraint factor of the respective UAV task semantic passing coding vector, which is expressed in a formula as:

[0054]

[0055]

[0056]

[0057]

[0058] wherein, represents a spatiotemporal confidence balance factor between the temporal confidence constraint factor and the spatial confidence constraint factor, represents a spatiotemporal constraint curvature compensation factor between the temporal confidence constraint factor and the spatial confidence constraint factor represents a sine function, represents a message passing spatiotemporal constraint coupling compensation factor, represents a corresponding optimized message passing spatiotemporal coordination constraint factor.

[0059] That is, through geometric constraint optimization, the distortion effect of spatiotemporal dimension coupling is eliminated, and the synergistic efficiency of the temporal confidence constraint factor and the spatial confidence constraint factor in the fusion space is improved. Specifically, the multi-dimensional constraint manifold compensation mechanism forces the spatiotemporal coupling to converge to a flat space form by dynamically adjusting the geometric properties of the temporal confidence constraint factor and the spatial confidence constraint factor. In this way, the model adaptively balances the spatiotemporal curvature difference, so that the optimized message passing spatiotemporal coordination constraint factor after fusion maintains a stable attractor structure in the flat space, avoiding resource allocation misjudgment caused by geometric distortion. In this way, through multi-dimensional constraint manifold compensation optimization, the model realizes geometric distortion suppression, and the compensation mechanism can real-time correct the curvature deviation of the message passing spatiotemporal coordination constraint factor, avoiding the invalidation of the fusion weight caused by geometric distortion. Secondly, by forcing the spatiotemporal coupling to converge to a flat space, the complementarity of the temporal confidence constraint factor and the spatial confidence constraint factor is strengthened, and the logical consistency of the resource allocation strategy is improved. The optimized message passing spatiotemporal coordination constraint factor generated based on this has stronger fault tolerance for network state mutations, and provides a weight basis with geometric stability for subsequent message passing coding.

[0060] Then, based on the optimization message passing spatiotemporal coordination constraint factor of the respective unmanned aerial vehicle task semantic transfer encoding vector, the respective unmanned aerial vehicle task semantic transfer encoding vector is modulated in message passing structure to obtain a sequence distribution of unmanned aerial vehicle task semantic transfer structural modulation encoding vectors; the position sum of the sequence distribution of the unmanned aerial vehicle task semantic transfer structural modulation encoding vectors is calculated to obtain the unmanned aerial vehicle task semantic encoding vector, which is expressed by a formula as:

[0061]

[0062] wherein, represents the unmanned aerial vehicle task semantic encoding vector.

[0063] That is, through the structured information propagation and aggregation optimization, a comprehensive semantic representation with spatiotemporal consistency is generated. Specifically, the message passing structural modulation realizes the information screening based on the spatiotemporal constraint driving through the weighting strategy, ensuring the reliability of the resource allocation decision. Through this position sum, the sequence distribution of the unmanned aerial vehicle task semantic transfer structural modulation encoding vectors is further compressed into the unmanned aerial vehicle task semantic encoding vector, capturing the global contextual semantic features of the unmanned aerial vehicle task, and focusing on the key task semantics through the adaptive weight constraint, which helps to improve the accuracy of task understanding and network demand prediction.

[0064] Specifically, the step S23 inputs the unmanned aerial vehicle task semantic encoding vector into a demand predictor based on an LLM model to obtain a future resource demand structured encoding vector. Specifically, the LLM model is a natural language processing (NLP) model based on deep learning, which pre-trains large-scale text data through self-supervised learning, learns the statistical rules and semantic information of language, and performs supervised fine-tuning through a small amount of labeled data to adapt to specific tasks. In this application, through the context reasoning of the unmanned aerial vehicle task semantic encoding vector by the demand predictor based on the LLM model, the intention semantics implied in the task instruction can be deeply analyzed, and relying on the powerful context reasoning and prediction ability of the LLM model, the future resource demand structured encoding vector is dynamically generated, the discrete task demand is converted into the continuous unmanned aerial vehicle task semantic encoding vector, thereby providing a forward-looking data basis for subsequent resource allocation decisions, enhancing the matching accuracy of resource allocation and task objectives, and ultimately improving the real-time response capability and resource utilization efficiency of the low-altitude network in complex scenarios.

[0065] In the low-altitude flight communication network resource allocation method based on the large model algorithm, the step S3, the network state data and the environment data are acquired. It should be understood that, since the low-altitude flight communication network faces dynamic changing task demand and complex environmental interference, and the traditional static resource allocation method cannot realize real-time perception of network load, channel quality and external environment change, leading to disconnection between resource supply and task demand. Therefore, the application realizes real-time acquisition of network state data (such as bandwidth occupancy, node load, channel quality) and environment data (such as meteorological information, geographical obstacles, interference source distribution), and dynamically evaluates the current network carrying capacity and potential risk factors. In this way, real-time environmental disturbance and network performance indicators can be included in the resource allocation decision model, so that the model can predict the resource demand trend based on global scene features, and realize accurate adaptation of resource supply to dynamic task demand.

[0066] Specifically, in terms of network state data, real-time monitoring of bandwidth occupancy becomes an important means to understand the current network load. Through sensors deployed on key nodes in the network, the bandwidth usage of each node can be continuously collected and uploaded to the central processing unit for analysis. This approach not only helps to identify the bottleneck position in the network, but also provides a basis for subsequent adjustment of resource allocation strategies. For example, when an abnormal increase in bandwidth occupancy is found in a certain area, it indicates that there may be a large amount of data transmission demand in that area, and at this time, existing resources may need to be redistributed to avoid task interruption or delay due to network congestion.

[0067] At the same time, node load is also an important parameter to measure the health status of the network. By monitoring the CPU usage, memory occupation and other key indicators of each network node, potential overload risks can be detected in a timely manner. Once it is detected that the load of a node approaches its maximum bearing capacity, the system can take appropriate measures, such as dispersing part of the task to other nodes with lower load, or optimizing routing selection to alleviate the pressure on high-load nodes. In addition, machine learning algorithms can be used to predict the trend of node load in the future period of time, so as to make resource scheduling preparations in advance and ensure that the network always maintains an efficient operation state.

[0068] Channel quality, as one of the key factors affecting data transmission efficiency, also needs to be fully focused. Using advanced signal processing technology, the signal strength, bit error rate and other parameters of each communication link can be monitored in real time. Based on these data, the system can quickly judge which channels are in good condition and which may have quality problems. For the latter, communication effects can be improved by switching to backup channels or increasing signal transmission power. At the same time, with the help of historical data analysis, fluctuations in channel quality in different time periods can also be predicted, so as to develop more scientific and reasonable resource allocation plans.

[0069] In addition to network status data, the acquisition of environmental data is also indispensable. Considering the complex natural conditions and human interference factors in low-altitude flight environments, a complete environmental monitoring system must be established. First, real-time updating of meteorological information is of great significance to ensure the safe flight of unmanned aerial vehicles. By using the latest weather forecast data provided by weather stations, combined with satellite remote sensing images and feedback from other meteorological observation equipment, a detailed meteorological map can be constructed to show the current regional wind speed, rainfall, temperature changes, etc. According to this information, not only can the unmanned aerial vehicle choose the best flight route to avoid adverse weather areas, but also can help predict the network performance decline problem caused by climate change and take measures in advance.

[0070] The existence of geographical obstacles also has a significant impact on the flight path planning and communication link of unmanned aerial vehicles. Therefore, it is necessary to rely on high-precision map services and advanced sensing technologies such as LiDAR to accurately map the topographic features. In this way, when planning the flight route of the unmanned aerial vehicle, it can effectively avoid obstacles such as high mountains and buildings, reducing the risk of flight accidents. At the same time, geographical obstacles may also block radio signals, causing local communication blind areas. Therefore, understanding the distribution of obstacles also helps to optimize the base station layout and ensure the maximization of network coverage.

[0071] The distribution of interference sources is another environmental factor that needs to be focused on. With the increasing number of wireless communication devices, electromagnetic interference is becoming increasingly serious, which poses a great challenge to low-altitude flight communication networks. In order to effectively identify and locate the interference source, spectrum monitoring equipment can be used to scan the entire working frequency band and record the location and strength of all abnormal signals. Based on these data, not only can the specific location of the interference source be tracked, but also the specific impact of the interference on network performance can be further analyzed. On this basis, targeted measures such as adjusting the frequency setting or increasing the anti-interference coding can be taken to reduce the negative impact of interference.

[0072] In the low-altitude flight communication network resource allocation method based on the large model algorithm, the step S4 is to use the network state embedding coding matrix and the environment data embedding coding matrix to structure the network state data and the environment data to obtain a network state structured coding vector and an environment state structured coding vector. It should be understood that by using the network state embedding coding matrix and the environment data embedding coding matrix, the original network state data and the environment data are mapped to a high-dimensional semantic space to generate a network state structured coding vector and an environment state structured coding vector with dynamic perception capability. In this way, discrete and heterogeneous network performance indicators and environment disturbance parameters can be converted into continuous vector representations to facilitate the model to capture the hidden spatio-temporal correlation and task coupling relationship between the data, thereby dynamically identifying the spatio-temporal evolution law of the network bottleneck and the potential conflict of the task demand, and finally supporting the intelligent decision module to realize joint optimization and adaptive allocation of multi-dimensional resource constraints in a complex scenario. In an embodiment of the present application, the network state embedding coding matrix and the environment data embedding coding matrix are trained by a pre-trained word vector model such as Word2vec.

[0073] In the low-altitude flight communication network resource allocation method based on the large model algorithm, the step S5 is to fuse the future resource demand structured coding vector, the network state structured coding vector and the environment state structured coding vector to obtain a comprehensive state feature vector. In one specific example of the present application, the step S5 includes cascading fusion of the future resource demand structured coding vector, the network state structured coding vector and the environment state structured coding vector to obtain the comprehensive state feature vector. It should be understood that by cascading fusion of the future resource demand structured coding vector, the network state structured coding vector and the environment state structured coding vector, a collaborative representation framework of multi-dimensional information can be constructed. On the one hand, the cascading mechanism can retain the time evolution sequence of the task demand, the topological features of the network state and the spatial heterogeneity of the environment condition, avoiding feature dilution of single-dimensional information in the fusion process. On the other hand, by forming high-order semantic association through cross-modal cascading, the comprehensive state feature vector has context reasoning capability. Through this cascading manner, the model can dynamically perceive the chain reaction among the three, thereby generating a resource allocation strategy with strong robustness to realize network load balancing, key task zero interruption and Pareto optimization of resource utilization in a complex scenario.

[0074] In the low-altitude flight communication network resource allocation method based on the large model algorithm described above, the step S6 inputs the comprehensive state feature vector into an intelligent resource allocation decision module to obtain a network resource allocation instruction, which is passed to the network controller for execution. In one specific example of the present application, step S6 includes inputting the comprehensive state feature vector into a decoder-based intelligent resource allocation decision module to obtain a network resource allocation instruction. Specifically, through the autoregressive generation mechanism of the decoder architecture, the present application can decouple the spatiotemporal coupling relationship between task demand, network state and environmental changes layer by layer, and use the semantic understanding ability of the pre-trained large model to map multi-modal features into hierarchical resource allocation strategies (such as channel allocation, power control, bandwidth adjustment, etc.), thereby realizing fine management and efficient use of resources, and constructing a continuous decision-making closed loop in a dynamic environment. In this way, while capturing sudden task demands in real time, the spatial distribution characteristics of network congestion nodes and meteorological interference areas are simultaneously perceived, so that the intelligent resource allocation decision module can generate a flexible allocation scheme with anti-interference capability based on the global resource topology, significantly improving the context awareness accuracy of resource scheduling, and helping to eliminate the decision-making blind spots caused by artificial feature engineering in traditional methods, while ensuring low latency constraints to achieve dynamic optimization of network resource allocation.

[0075] After the network resource allocation instructions are generated, these instructions need to be passed to the network controller to perform the corresponding resource adjustment and management operations. It is worth noting that during the execution of the instructions, various unexpected situations or challenges may be encountered. For example, due to the dynamic changes in the network environment, the originally planned path may suddenly become unavailable, or some critical devices fail to respond normally to the instructions. To address these issues, the system should have certain fault tolerance mechanisms. On the one hand, backup paths or redundant devices can be set up to enhance the robustness of the overall system; on the other hand, when abnormal situations are detected, an alarm mechanism is triggered in a timely manner and detailed information is reported to the administrator, so that remedial measures can be taken as soon as possible. In addition, a self-repairing function can also be introduced to allow the system to automatically solve some common problems within a certain range, reducing the need for human intervention.

[0076] In summary, the low-altitude flight communication network resource allocation method based on the large model algorithm based on the embodiments of the present application is illustrated, which acquires the unmanned aerial vehicle task instruction, and performs intention analysis and demand prediction based on the LLM model to capture the communication network resource demand characteristics. At the same time, further combined with network state data and environment data, the network state characteristics, environment state characteristics and resource demand characteristics are fused to construct the comprehensive state expression of the unmanned aerial vehicle task execution scene, and then intelligent resource allocation decision is made on this basis to generate and execute network resource allocation instructions. In this way, intelligent allocation of low-altitude flight communication network resources is realized, resource utilization efficiency and task execution capability are improved, the accuracy and real-time of resource allocation can be effectively ensured, and the intelligent level of resource allocation is further improved.

[0077] The basic principles of the application are described above in conjunction with specific embodiments, but it should be noted that the advantages, advantages, effects, etc. mentioned in the application are only examples and not limitations, and these advantages, advantages, effects, etc. cannot be considered as the must-have of each embodiment of the application. In addition, the specific details of the above embodiments are only for the purpose of example and for the purpose of understanding, and the above details do not limit the application to the must-use specific details.

[0078] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described or recorded in a certain embodiment can be referred to the related description of other embodiments. In the several embodiments provided by the application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the system embodiments described above are only schematic, for example, the unit division is only a logical function division, and there can be other division ways in actual implementation. The units described as separated components can or can not be physically separated, and the components displayed as units can be or can not be physical units, that is, they can be located in one place or distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0079] It is obvious for those skilled in the art that the application is not limited to the details of the above exemplary embodiments, and the application can be implemented in other specific forms without departing from the spirit or essential characteristics of the application. Therefore, the embodiments should be regarded as exemplary and non-limiting, and the scope of the application is defined by the appended claims rather than the above description, and all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the application. Any reference signs in the claims should not be regarded as limiting the claims to which they relate.

[0080] Furthermore, the word "comprising" does not exclude other elements or steps, and the singular does not exclude the plural and vice-versa, unless the context clearly requires these exclusions. The conjunction "or" is used to link items in a list or a set of alternatives, and is not disjunctive, unless the context clearly requires it to be disjunctive. The conjunction "and" is used to link items in a list or a set of alternatives, and is not conjunctive, unless the context clearly requires it to be conjunctive. The prefix "re-" when used in the context of a method or process means that additional steps can be performed in addition to those described in the method or process, unless the context clearly indicates otherwise.

[0081] Finally, it should be noted that the description has been given for illustrative and descriptive purposes only and is not intended to limit the technical solutions of the present application. Furthermore, the above embodiments are merely used to illustrate the technical solutions of the present application but not to limit the present application, and although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalent replaced without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A resource allocation method for low-altitude flight communication networks based on a large model algorithm, characterized in that, include: Obtain drone mission instructions; The UAV mission instructions are subjected to intent parsing and demand prediction based on an LLM model to obtain a structured encoding vector of future resource demands. Acquire network status data and environmental data; The network state data and the environment data are structured using a network state embedding coding matrix and an environment data embedding coding matrix to obtain network state structured coding vectors and environment state structured coding vectors. The structured encoding vectors of future resource demand, network state, and environment state are fused to obtain a comprehensive state feature vector. The comprehensive state feature vector is input into the intelligent resource allocation decision module to obtain a network resource allocation instruction, which is then transmitted to the network controller for execution. The UAV mission commands are subjected to intent parsing and demand prediction based on an LLM model to obtain a structured encoded vector of future resource demands, including: After the UAV mission command is segmented, it is input into a text encoder to obtain the sequence distribution of the UAV mission word-granular semantic embedding encoding vector; The sequence distribution of the UAV task word-granular semantic embedding encoding vector is subjected to UAV task word-granular context semantic association encoding to obtain the UAV task semantic encoding vector; The UAV mission semantic encoding vector is input into an LLM-based demand predictor to obtain the future resource demand structured encoding vector. Perform UAV task word-granularity semantic embedding encoding on the sequence distribution of the UAV task word-granularity semantic embedding encoding vector to obtain the UAV task semantic encoding vector, including: The sequence distribution of the UAV mission word-granular semantic embedding encoding vector is input into a sequence encoder based on a recurrent neural network to obtain the sequence distribution of the UAV mission semantic transfer encoding vector. Calculate the temporal confidence constraint factor of each UAV mission semantic transfer encoding vector in the sequence distribution of the UAV mission semantic transfer encoding vector; Calculate the spatial confidence constraint factor of each UAV mission semantic transfer encoding vector in the sequence distribution of the UAV mission semantic transfer encoding vector; Based on the spatial confidence constraint factor and temporal confidence constraint factor of each UAV mission semantic transfer encoding vector, a message transfer spatiotemporal coordination constraint factor for each UAV mission semantic transfer encoding vector in the sequence distribution of the UAV mission semantic transfer encoding vector is constructed. Based on the aforementioned spatiotemporal coordination constraint factor for message passing, the semantic transmission encoding vectors of each UAV mission are dynamically adjusted and aggregated to obtain the semantic encoding vector of the UAV mission.

2. The low-altitude flight communication network resource allocation method based on the large model algorithm according to claim 1, characterized in that, The text encoder is a word embedding encoding model based on the BERT model.

3. The low-altitude flight communication network resource allocation method based on the large model algorithm according to claim 2, characterized in that, Based on the spatial confidence constraint factor and temporal confidence constraint factor of each UAV mission semantic transfer encoding vector, a message transfer spatiotemporal coordination constraint factor is constructed for each UAV mission semantic transfer encoding vector, including: The spatial confidence constraint factor and temporal confidence constraint factor of the UAV mission semantic transfer encoding vector are weighted and fused, and then input into the sigmoid function for normalization to obtain the message transfer spatiotemporal coordination constraint factor of the UAV mission semantic transfer encoding vector.

4. The low-altitude flight communication network resource allocation method based on the large model algorithm according to claim 3, characterized in that, Based on the aforementioned spatiotemporal coordination constraint factor for message passing, the semantic transmission encoding vectors of each UAV mission are dynamically adjusted and aggregated to obtain the UAV mission semantic encoding vector, including: The spatiotemporal coordination constraint factor of message passing for each UAV mission semantic transfer encoding vector is optimized by performing spatiotemporal coupling compensation based on multidimensional constraint manifold to obtain the optimized spatiotemporal coordination constraint factor of message passing for each UAV mission semantic transfer encoding vector. Based on the optimized message passing spatiotemporal coordination constraint factor of each UAV mission semantic passing coding vector, message passing structure modulation is performed on each UAV mission semantic passing coding vector to obtain the sequence distribution of UAV mission semantic passing structure modulation coding vector. The positional summation of the sequence distribution of the UAV mission semantic transfer structured modulation and coding vector is calculated to obtain the UAV mission semantic coding vector.

5. The low-altitude flight communication network resource allocation method based on the large model algorithm according to claim 4, characterized in that, The future resource demand structured encoding vector, the network state structured encoding vector, and the environment state structured encoding vector are fused to obtain a comprehensive state feature vector, including: The structured encoding vectors of future resource demand, network state, and environment state are concatenated and fused to obtain the comprehensive state feature vector.

6. The low-altitude flight communication network resource allocation method based on the large model algorithm according to claim 5, characterized in that, The comprehensive state feature vector is input into the intelligent resource allocation decision module to obtain a network resource allocation instruction, which is then passed to the network controller for execution, including: The comprehensive state feature vector is input into the decoder-based intelligent resource allocation decision module to obtain network resource allocation instructions.

Images