5g-a low-altitude communication link quality optimization and adaptive adjustment method based on ai prediction

Abstract

The application relates to the technical field of low-altitude communication and discloses a 5G-A low-altitude communication link quality optimization and adaptive adjustment method based on AI prediction, which comprises the following steps: performing cross-domain conflict resolution on the link state of a terminal in a low-altitude communication scene and multidimensional environment data to obtain conflict-free fusion data; performing high-dimensional situation extraction on the conflict-free fusion data to obtain deep state representation; intelligently deducing the space-time evolution mode of the link state based on the deep state representation and the historical link quality of the terminal to obtain a link quality prediction trajectory; deriving preventive constraint of potential degradation modes of the terminal to obtain a constraint condition set; encoding a cooperative strategy for the terminal to obtain a cooperative instruction; and globally stably deploying the link state based on the cooperative instruction to obtain an optimized link state; and the application can improve the efficiency of the 5G-A low-altitude communication link quality optimization and adaptive adjustment based on AI prediction.

Description

Technical Field

[0001] This invention relates to the field of low-altitude communication technology, and in particular to a method for optimizing and adaptively adjusting the quality of 5G-A low-altitude communication links based on AI prediction. Background Technology

[0002] In current practices of link quality management in low-altitude communication networks for 5G-A, there is a widespread reliance on simple aggregation of multi-dimensional monitoring data and threshold comparison mechanisms. These methods struggle to effectively handle spatiotemporal conflicts and logical inconsistencies among heterogeneous data from multiple sources, including terminals, airspace environments, and wireless links. This results in inherent contradictions and noise in the underlying data used for state assessment, limiting the reliability of subsequent analyses. Furthermore, traditional feature extraction methods often focus on single-dimensional or shallow statistics, failing to construct a unified, high-dimensional representation that profoundly reflects the system's internal operational status from complex cross-domain interactions. This leaves state perception at a fragmented and superficial level.

[0003] Furthermore, most existing link quality predictions are based on statistical extrapolation of historical data or simple fitting of isolated models. They fail to effectively model and deduce the spatiotemporal evolution patterns and causal relationships of link states in complex low-altitude scenarios, resulting in predictions lacking foresight and robustness, and making it difficult to accurately capture potential rapid degradation trends. Optimization adjustments based on such predictions are mostly delayed, passive, and single-point resource reconfigurations, lacking the ability to proactively generate preventative collaborative strategies at the system level based on forward-looking prediction results. They also cannot achieve globally stable deployment and closed-loop verification of configurations across multiple terminals. This poses a significant challenge to the continuous optimization and robust maintenance of overall network link quality in dynamic environments. Therefore, improving the efficiency of data report generation has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method for optimizing and adaptively adjusting the quality of 5G-A low-altitude communication links based on AI prediction, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a method for optimizing and adaptively adjusting the quality of 5G-A low-altitude communication links based on AI prediction, comprising:

[0006] S01. Perform cross-domain conflict resolution on the link status and multi-dimensional environmental data of the terminal in the low-altitude communication scenario to obtain conflict-free fused data of the terminal.

[0007] S02. Perform high-dimensional situation extraction on the conflict-free fused data to obtain the deep state representation of the conflict-free fused data;

[0008] S03. Based on the deep state representation and the historical link quality of the terminal, perform intelligent graph deduction on the spatiotemporal evolution pattern of the link state to obtain the link quality prediction trajectory of the terminal.

[0009] S04. Based on the link quality prediction trajectory, perform preventive constraint derivation on the potential degradation mode of the terminal to obtain the constraint condition set of the terminal.

[0010] S05. Based on the set of constraints, perform cooperative strategy encoding on the terminal to obtain the cooperative instructions of the terminal;

[0011] S06. Based on the cooperative instructions, the link state is globally stabilized to obtain the optimized link state of the terminal.

[0012] In a preferred embodiment, the step of performing cross-domain conflict resolution on the link status and multi-dimensional environmental data of the terminal in the low-altitude communication scenario to obtain conflict-free fused data of the terminal includes:

[0013] In low-altitude communication scenarios, the link status of the terminal and multi-dimensional environmental data are dynamically aligned heterogeneously to obtain the spatiotemporal aligned data sequence of the terminal.

[0014] Conflict identification is performed on the spatiotemporally aligned data sequence to obtain the set of conflicting elements in the spatiotemporally aligned data sequence;

[0015] By removing the elements in the spatiotemporally aligned data sequence that correspond to the set of conflicting elements, the conflict-free fused data of the terminal is obtained.

[0016] In a preferred embodiment, the step of performing high-dimensional situation extraction on the conflict-free fused data to obtain a deep state representation of the conflict-free fused data includes:

[0017] Cross-domain causal influence factors are decoupled from the ternary coupling relationship of terminal, environment, and link in the conflict-free fused data to obtain the set of influence elements of the conflict-free fused data;

[0018] Based on the set of influencing elements, a structured deduction of the link state is performed to obtain a probability distribution map of the link state;

[0019] The probability distribution map is subjected to topological manifold diffusion to obtain a stability description of the conflict-free fused data;

[0020] Based on the stability description, the key modes of the comprehensive communication situation in the low-altitude communication scenario are reorganized to obtain the deep state characterization of the conflict-free fused data.

[0021] The process of decoupling the ternary coupling relationship between the terminal, environment, and link in the conflict-free fused data through cross-domain causal influence factors yields a set of causal influence elements for the conflict-free fused data, including:

[0022] Cross-domain phase synchronization analysis is performed on the ternary coupling relationship of terminal, environment, and link in the conflict-free fused data to obtain the phase oscillation sequence of the conflict-free fused data;

[0023] Based on the phase oscillation sequence, the influence intensity value of the cross-domain factor in the ternary coupling relationship is calculated, wherein the formula for calculating the influence intensity value is:

[0024] ;

[0025] In the formula, cross-domain factors Cross-domain factors The influence intensity value, The total length of the trans-domain phase oscillation sequence. The preset delay time for causal analysis, For the current moment, Transdomain factors in the phase oscillation sequence The instantaneous phase sequence, It is a sine trigonometric function. For the cross-domain factors The instantaneous phase value, Transdomain factors in the phase oscillation sequence The instantaneous phase value, It is an exponential function. The sequence of first derivatives of the instantaneous phase sequence. Trans-domain factors in the phase oscillation sequence The sequence of second derivatives of the instantaneous phase, This is the absolute value operator;

[0026] Based on the influence intensity value, the conflict-free fused data is subjected to sparsity filtering to obtain the causal influence element set of the conflict-free fused data.

[0027] In a preferred embodiment, the step of intelligently extrapolating the spatiotemporal evolution pattern of the link state based on the deep state representation and the historical link quality of the terminal to obtain the predicted link quality trajectory of the terminal includes:

[0028] A topological analysis of the deep state representation and the historical link quality of the terminal is performed to obtain the spatiotemporal entanglement relationship of the terminal;

[0029] Based on the spatiotemporal entanglement relationship, the transition conditions of the link state are extracted to obtain the constraint boundary set of the link state;

[0030] Based on the constraint boundary set, the spatiotemporal topological relationship of the link state is synthesized to obtain the spatiotemporal evolution pattern map of the link state;

[0031] Based on the spatiotemporal evolution pattern map, simulated path exploration is performed on the deep state representation to obtain a set of candidate state transition paths for the deep state representation;

[0032] The candidate state transition path set is refined into a dominant path to obtain the link quality prediction trajectory of the terminal.

[0033] The step of performing topological analysis on the deep state representation and the historical link quality of the terminal to obtain the spatiotemporal entanglement relationship of the terminal includes:

[0034] Based on the historical link quality of the terminal, a multi-scale feature hierarchy is constructed for the deep state representation to obtain a multi-scale knowledge representation of the deep state representation.

[0035] Based on the multi-scale knowledge representation, the co-occurrence patterns of cross-domain state variables in the deep state representation are correlated and structured to obtain the higher-order correlation tensor of the multi-scale knowledge representation.

[0036] The state evolution patterns of the higher-order correlation tensor are summarized to obtain the set of transition patterns of the intrinsic states in the higher-order correlation tensor.

[0037] Cross-scale fusion of the dynamic interaction dependencies of the transfer mode set yields the spatiotemporal entanglement relationship of the terminal.

[0038] In a preferred embodiment, the step of performing preventative constraint derivation on the potential degradation modes of the terminal based on the link quality prediction trajectory to obtain the constraint set of the terminal includes:

[0039] Dynamic vulnerability insight is performed on the predicted link quality trajectory to obtain potential vulnerable segments of the predicted link quality trajectory;

[0040] The degradation mode characteristics of the potentially vulnerable sections are obtained by performing degradation mode deconstruction on the potentially vulnerable sections.

[0041] The degradation mode features are subjected to inverse optimization conditions to construct a set of resilience enhancement features.

[0042] Based on the resilience enhancement feature set, the dynamic boundary of the communication state in the terminal is constrained and deduced in reverse to obtain the constraint condition set of the terminal.

[0043] The reverse optimization of the degradation mode features to construct the resilience enhancement feature set of the degradation mode features includes:

[0044] Key factors are extracted from the degradation mode features to obtain the set of dominant driving factors for the degradation mode features;

[0045] Based on a pre-defined multi-faceted reverse intervention scheme, the dynamic evolution path of the dominant driving factor set is simulated to obtain the evolution path sequence of the dominant driving factor set.

[0046] Based on the evolutionary path sequence, the resilience gain value of the multi-faceted reverse intervention scheme is calculated, wherein the formula for calculating the resilience gain value is:

[0047] ;

[0048] In the formula, This is the toughness gain value. The total time step of the evolutionary path sequence. The first in the evolutionary path sequence The variance of each quality indicator The first in the evolutionary path sequence The average value of each quality indicator The preset zero-prevention constant is used. The first in the preset evolutionary path sequence Sensitivity weighting coefficients for each quality indicator In time step No. The quality indicator and the first The preset dynamic correlation coefficient between the quality indicators. For the first The quality indicator and the first The preset ideal correlation coefficient between the quality indicators This represents the total number of quality indicators in the evolutionary path sequence. It is an exponential function;

[0049] Based on the resilience gain value, the Pareto front screening is performed on the multivariate reverse intervention schemes, and multidimensional feature extraction is performed on the screened intervention schemes to obtain the resilience enhancement feature set of the degradation mode features.

[0050] In a preferred embodiment, the step of encoding the cooperative strategy for the terminal based on the constraint set to obtain the cooperative instructions for the terminal includes:

[0051] The decision architecture of the terminal is obtained by performing multi-objective collaborative logic analysis on the set of constraints.

[0052] Based on the aforementioned decision-making architecture, the resource supply and demand relationship of the terminal is dynamically coupled and optimized to obtain a resource allocation scheme for the terminal.

[0053] Based on the resource allocation scheme, the migration path and control elements of the link status are aggregated together to obtain the collaborative parameter set of the terminal.

[0054] The set of collaborative parameters is encapsulated into collaborative instructions for the terminal.

[0055] In a preferred embodiment, the step of globally stabilizing the link state based on the cooperative instructions to obtain the optimized link state of the terminal includes:

[0056] The timing dependency of the collaborative instructions is parsed to obtain the timing execution plan of the terminal;

[0057] Based on the timing execution plan, the configuration parameters of the terminal are coordinated and deployed to obtain the synchronization link status of the terminal;

[0058] A steady-state assessment of the synchronization link status is performed to obtain an assessment report of the synchronization link status;

[0059] Based on the evaluation report, the link configuration of the terminal is adaptively corrected to obtain the optimized link status of the terminal.

[0060] Compared with the prior art, the present invention has the following beneficial effects:

[0061] 1. This invention effectively integrates and enhances the internal consistency and representational capabilities of multi-source heterogeneous data through cross-domain conflict resolution and high-dimensional situational depth extraction, thereby constructing a deep state description capable of accurately depicting complex coupled dynamics at low altitudes. This foundation allows subsequent deductions of the spatiotemporal evolution patterns of link quality to be based on a more realistic and fundamental understanding of the situation, significantly improving the accuracy, continuity, and foresight of predicted trajectories, and providing a reliable basis for proactive management.

[0062] 2. Based on accurate prediction, this invention achieves a paradigm shift from passive response to preventative collaborative optimization, enabling proactive derivation of constraints and generation of system-level collaborative strategies. Through global steady-state deployment and closed-loop verification mechanisms, it ensures dynamic optimal allocation of resources among multiple terminals and robust convergence of configuration parameters, thereby continuously maintaining and proactively optimizing link quality in highly dynamic low-altitude environments, significantly enhancing the overall service robustness and adaptive maintenance capability of the entire communication network. Attached Figure Description

[0063] Figure 1 A flowchart illustrating an AI-predictive-based method for optimizing and adaptively adjusting the quality of 5G-A low-altitude communication links, as provided in an embodiment of the present invention.

[0064] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0065] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0066] This application provides a method for optimizing and adaptively adjusting the quality of 5G-A low-altitude communication links based on AI prediction. The execution entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cluster of cloud servers. The server can be an independent server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0067] Reference Figure 1 The diagram shown is a flowchart illustrating a method for optimizing and adaptively adjusting the quality of 5G-A low-altitude communication links based on AI prediction, according to an embodiment of the present invention. In this embodiment, the method for optimizing and adaptively adjusting the quality of 5G-A low-altitude communication links based on AI prediction includes:

[0068] S1. Perform cross-domain conflict resolution on the link status and multi-dimensional environmental data of the terminal in the low-altitude communication scenario to obtain conflict-free fused data of the terminal.

[0069] In this embodiment of the invention, the step of performing cross-domain conflict resolution on the link status and multi-dimensional environmental data of the terminal in a low-altitude communication scenario to obtain conflict-free fused data of the terminal includes:

[0070] In low-altitude communication scenarios, the link status of the terminal and multi-dimensional environmental data are dynamically aligned heterogeneously to obtain the spatiotemporal aligned data sequence of the terminal.

[0071] Conflict identification is performed on the spatiotemporally aligned data sequence to obtain the set of conflicting elements in the spatiotemporally aligned data sequence;

[0072] By removing the elements in the spatiotemporally aligned data sequence that correspond to the set of conflicting elements, the conflict-free fused data of the terminal is obtained.

[0073] When dynamically aligning heterogeneous data such as the link status of terminals and multidimensional environmental data in low-altitude communication scenarios, all input data sources are synchronized based on a unified time reference and spatial coordinate system. Each data source outputs raw data with its own timestamp and spatial location information. First, all these timestamps are corrected to an absolute time axis based on the system's time synchronization server, eliminating time deviations caused by different device acquisition cycles and transmission delays. Then, the spatial coordinates of all data records are transformed to a unified three-dimensional Cartesian coordinate system with a specific ground base station as the origin. Coordinate mapping is completed based on the latitude, longitude, and altitude data reported by the terminal in real time, combined with the geographical information from environmental sensing devices. Each data point processed above has a precise correspondence in both the time and spatial dimensions. These data points, arranged chronologically and spatially aligned, are categorized and integrated by source to generate the spatiotemporally aligned data sequence of the terminal.

[0074] When identifying conflicts in the spatiotemporally aligned data sequence, logical consistency comparisons are performed on data from different sources but at the same spatiotemporal point in the sequence according to predefined domain consistency rules. These rules explicitly define the reasonable correlation range between different physical quantities, such as the corresponding interval between specific signal strength and bit error rate, the matching relationship between known terminal motion state and Doppler frequency shift, and the reasonable impact threshold of environmental meteorological conditions on signal attenuation in a specific frequency band. The identification process traverses each synchronization point in the spatiotemporally aligned data sequence, checking whether the parameter value combinations provided by all data sources at that time fully satisfy all preset consistency rules. The parameter records of specific data sources that cause rule violations at specific points in time are identified, and the index information of all these identified data records is summarized, thereby obtaining the conflict element set of the spatiotemporally aligned data sequence.

[0075] When removing elements from the spatiotemporally aligned data sequence that correspond to the conflict element set, each data record in the spatiotemporally aligned data sequence is traversed, and its existence in the index information recorded in the conflict element set is checked. Data records verified to exist in the conflict element set are removed from the sequence. Data records not marked as conflicting are retained intact. During this process, to ensure data continuity in subsequent processing, for gaps in certain parameters caused by data removal at a specific time, linear interpolation is performed using the same parameter value from the nearest valid time before and after that time to fill the gap. After the above processes of removing conflicting records and interpolation filling, the retained and reconstructed data points constitute a new data set that is spatiotemporally continuous and satisfies internal consistency logic in all dimensions. This set is the conflict-free fused data of the terminal.

[0076] The beneficial effects are as follows: by performing strict time and space benchmark synchronization operations, the raw data from the terminal link and multi-dimensional environment are transformed into a unified spatiotemporal coordinate system, generating a highly consistent spatiotemporally aligned data sequence, thereby eliminating the misalignment problem caused by differences in acquisition and transmission of multi-source heterogeneous data. Furthermore, by utilizing domain rules to perform consistency verification on all synchronized data points, all conflicting data elements that violate physical or logical relationships are accurately identified and located, forming a conflict element set, effectively filtering out internal contradictions and noise. Based on this conflict element set, targeted elimination and neighbor interpolation are performed on the original sequence, ultimately constructing internally self-consistent and spatiotemporally continuous conflict-free fused data. This lays a solid and reliable data quality foundation for subsequent deep situational awareness extraction and accurate prediction, directly improving the input reliability and process stability of the entire link quality optimization process.

[0077] S2. Perform high-dimensional situation extraction on the conflict-free fused data to obtain the deep state representation of the conflict-free fused data;

[0078] In this embodiment of the invention, the step of performing high-dimensional situation extraction on the conflict-free fused data to obtain a deep state representation of the conflict-free fused data includes:

[0079] Cross-domain causal influence factors are decoupled from the ternary coupling relationship of terminal, environment, and link in the conflict-free fused data to obtain the set of influence elements of the conflict-free fused data;

[0080] Based on the set of influencing elements, a structured deduction of the link state is performed to obtain a probability distribution map of the link state;

[0081] The probability distribution map is subjected to topological manifold diffusion to obtain a stability description of the conflict-free fused data;

[0082] Based on the stability description, the key modes of the comprehensive communication situation in the low-altitude communication scenario are reorganized to obtain the deep state characterization of the conflict-free fused data.

[0083] The process of decoupling the ternary coupling relationship between the terminal, environment, and link in the conflict-free fused data through cross-domain causal influence factors yields a set of causal influence elements for the conflict-free fused data, including:

[0084] Cross-domain phase synchronization analysis is performed on the ternary coupling relationship of terminal, environment, and link in the conflict-free fused data to obtain the phase oscillation sequence of the conflict-free fused data;

[0085] Based on the phase oscillation sequence, the influence intensity value of the cross-domain factor in the ternary coupling relationship is calculated, wherein the formula for calculating the influence intensity value is:

[0086] ;

[0087] In the formula, cross-domain factors Cross-domain factors The influence intensity value, The total length of the trans-domain phase oscillation sequence. The preset delay time for causal analysis, For the current moment, Trans-domain factors in the phase oscillation sequence The instantaneous phase sequence, It is a sine trigonometric function. For the cross-domain factors The instantaneous phase value, Trans-domain factors in the phase oscillation sequence The instantaneous phase value, It is an exponential function. The sequence of first derivatives of the instantaneous phase sequence. Trans-domain factors in the phase oscillation sequence The sequence of second derivatives of the instantaneous phase, This is the absolute value operator;

[0088] Based on the influence intensity value, the conflict-free fused data is subjected to sparsity filtering to obtain the causal influence element set of the conflict-free fused data.

[0089] When performing cross-domain phase synchronization analysis on the ternary coupling relationship of the terminal environment link in the conflict-free fused data, the periodic fluctuation characteristics of the time series of each dimension are extracted. For the continuous data sequence of each dimension, the corresponding instantaneous phase value is calculated by the analytical signal method. This method transforms the original sequence into a pair of orthogonal components and solves for the phase angle at each moment accordingly. The instantaneous phase values ​​obtained by the above processing of all dimensions are arranged in chronological order to form a set of parallel phase change sequences. This set of sequences is the phase oscillation sequence of the conflict-free fused data.

[0090] When calculating the influence intensity of cross-domain factors in a ternary coupling relationship based on the phase oscillation sequence, two specific dimensional sequences are selected for analysis. During the calculation, the phase value of one dimension at a specific time interval in advance is differed from the phase value of the other dimension at the current moment. The sine of this phase difference is calculated to obtain an instantaneous influence measure. Simultaneously, the phase change rate of the preceding dimension at the in advance time interval and the phase change acceleration of the following dimension at the current moment are calculated, and the absolute value of their difference is obtained. This absolute value is used as input for an exponential decay term to weight the aforementioned instantaneous influence measure. This calculation is repeated for each time point throughout the entire time series, and the weighted results at all time points are arithmetically averaged. The final scalar value obtained is the influence intensity value of the cross-domain factor.

[0091] When performing sparsity filtering on the conflict-free fused data based on the influence strength values, a specific strength threshold is set. The influence strength values ​​between all calculated cross-domain factors are compared with this threshold. Only specific cross-domain factor pairs whose influence strength values ​​exceed the threshold are retained, and the causal dominant dimension element in these factor pairs is recorded as an independent entry. All these independent entries that meet the strength requirements are aggregated to form a new set, which is the causal influence element set of the conflict-free fused data.

[0092] When performing a structured deduction of the link state based on the set of influencing elements, a state transition structure is constructed according to the causal relationships revealed by the set of influencing elements. This structure uses different performance levels of the link state as nodes and the causal paths defined by the set of influencing elements as directed connections between nodes. Each directed connection is assigned an initial transition probability weight, which originates from the strength information of the corresponding causal relationship in the set of influencing elements. By simulating the state transmission process over time on this structure, the weights of all possible paths are iteratively updated until the weight distribution of the entire structure reaches stability. Finally, the stable weight distribution is expressed graphically, where the node position represents the state and the thickness of the connecting lines represents the magnitude of the transition probability; this graph is the probability distribution diagram of the link state.

[0093] When performing topological manifold diffusion on the probability distribution graph, each state node in the graph is considered an initial energy source, and its initial energy value is determined by the node's centrality in the probability distribution graph. The diffusion rule is defined as follows: energy can flow from high-energy nodes to adjacent low-energy nodes along connecting lines in the graph, with the flow rate proportional to the weight of the connecting line. Based on this rule, multiple rounds of iterative energy transfer calculations are performed among all nodes until the energy distribution of the entire graph structure no longer changes significantly. The stable energy value held by each node at this point is recorded. This series of energy values ​​and the global distribution situation they reflect together constitute the stability description of the conflict-free fused data.

[0094] When performing key mode reorganization of the comprehensive communication situation in low-altitude communication scenarios based on the aforementioned stability description, peak points with energy values ​​significantly higher than the surrounding areas are first identified in the stability description. These peak points correspond to core states that the system tends to maintain stability. Simultaneously, channel regions connecting different core states with drastic energy gradient changes are identified. The original data features corresponding to these core states and key channel regions are extracted, including the specific parameter combinations of their respective terminal environment links. These features are then integrated and abstracted to form a new feature set that can simultaneously characterize the system's stable core and key transition paths. This new feature set is the deep state representation of the conflict-free fused data.

[0095] All parameters in the formula are derived from the cross-domain phase oscillation sequence generated in the preceding steps. This sequence is obtained by processing the original data sequences of each dimension of the terminal environment link using analytical signal methods. The total sequence length is the total number of data points contained in the phase oscillation sequence. The causal analysis delay duration is a fixed time interval value pre-set based on domain prior knowledge. The current moment refers to a specific time point index during the sequence processing calculation. The instantaneous phase value of the cross-domain factor is directly taken from the value of the corresponding dimension of the factor in the phase oscillation sequence at a specific moment. The first derivative sequence of the instantaneous phase sequence is obtained by calculating the difference between each data point in the phase sequence and its previous adjacent data point and dividing by the time interval. The second derivative sequence of the instantaneous phase is obtained by again calculating the difference between adjacent points on the first derivative sequence.

[0096] The formula quantifies the intensity of the directional causal influence of one cross-domain factor on another. The calculation process iterates through every time point of the phase oscillation sequence. At each time point, the difference between the phase value of the causal factor at a specific delay and the phase value of the consequent factor at the current time is calculated. A sine function is derived from this difference to obtain the instantaneous directional driving force. Simultaneously, the absolute value of the difference between the rate of phase change of the causal factor at the delay and the acceleration of phase change of the consequent factor at the current time is calculated. This absolute value is input into an exponential function to obtain the dynamic coordination damping weight. The instantaneous directional driving force is multiplied by the dynamic coordination damping weight to obtain the contribution at that moment. Finally, the contributions at all time points are summed and divided by the total sequence length to obtain the average influence intensity value. This calculation mechanism simultaneously evaluates the traction direction reflected by the phase difference and the motion coordination reflected by the rate of change matching.

[0097] The formula's trend shows that the influence intensity value changes synchronously with the phase pull effect and the degree of dynamic coordination. When the lagging phase of the causal factor exerts a stable pull on the current phase of the result factor, and the rate of phase change and acceleration of both are highly matched, the exponential damping term approaches its maximum value. This makes the calculated result approach the maximum potential driving force output by the sine function, thus yielding a higher influence intensity value. When the phase difference is in the region with no pull effect, or when the rate of change of the causal factor and the acceleration of change of the result factor differ significantly, the dynamic coordination damping weight decays rapidly, thus significantly reducing the contribution at that moment. By averaging the contribution over the entire time series, the final influence intensity value output by this formula can effectively highlight causal relationships with continuous coordinated phase driving relationships while suppressing interference signals caused by random fluctuations or false synchronization.

[0098] The beneficial effect is that by obtaining the oscillation sequence through phase synchronization analysis of the three-element coupling data of the terminal environment link, and calculating the directional influence intensity between cross-domain factors based on the matching degree of temporal phase difference and rate of change, core causal elements exceeding the intensity threshold are screened to form a set of causal influence elements. Based on this set of elements, the link state is subjected to structured deduction and weighted iteration with causal path as the skeleton, generating a probability distribution map expressed by nodes and weighted connections, which is then transformed into a stability description of the system's steady-state distribution through an energy diffusion process on the topological manifold. Finally, based on this description, the features of the system's core steady state and key conversion channels are extracted, integrated, and abstracted to construct a deep state representation that can uniformly characterize the inherent stable structure and evolution path of complex situations in low-altitude communication scenarios, providing a high-dimensional, stable, and semantically rich feature foundation for subsequent accurate prediction.

[0099] S3. Based on the deep state representation and the historical link quality of the terminal, perform intelligent graph deduction on the spatiotemporal evolution pattern of the link state to obtain the link quality prediction trajectory of the terminal.

[0100] In this embodiment of the invention, the step of intelligently extrapolating the spatiotemporal evolution pattern of the link state based on the deep state representation and the historical link quality of the terminal to obtain the predicted link quality trajectory of the terminal includes:

[0101] A topological analysis of the deep state representation and the historical link quality of the terminal is performed to obtain the spatiotemporal entanglement relationship of the terminal;

[0102] Based on the spatiotemporal entanglement relationship, the transition conditions of the link state are extracted to obtain the constraint boundary set of the link state;

[0103] Based on the constraint boundary set, the spatiotemporal topological relationship of the link state is synthesized to obtain the spatiotemporal evolution pattern map of the link state;

[0104] Based on the spatiotemporal evolution pattern map, simulated path exploration is performed on the deep state representation to obtain a set of candidate state transition paths for the deep state representation;

[0105] The candidate state transition path set is refined into a dominant path to obtain the link quality prediction trajectory of the terminal.

[0106] The step of performing topological analysis on the deep state representation and the historical link quality of the terminal to obtain the spatiotemporal entanglement relationship of the terminal includes:

[0107] Based on the historical link quality of the terminal, a multi-scale feature hierarchy is constructed for the deep state representation to obtain a multi-scale knowledge representation of the deep state representation.

[0108] Based on the multi-scale knowledge representation, the co-occurrence patterns of cross-domain state variables in the deep state representation are correlated and structured to obtain the higher-order correlation tensor of the multi-scale knowledge representation.

[0109] The state evolution patterns of the higher-order correlation tensor are summarized to obtain the set of transition patterns of the intrinsic states in the higher-order correlation tensor.

[0110] Cross-scale fusion of the dynamic interaction dependencies of the transfer mode set yields the spatiotemporal entanglement relationship of the terminal.

[0111] Based on the historical link quality of the terminal, when constructing the multi-scale feature hierarchy of the deep state representation, a fixed-time-window sliding segmentation method is used to process the historical link quality data. The historical link quality data is segmented according to multiple time scales from short to long, forming a set of subsequences with different time granularities. These subsequences of different scales are then aligned and fused with the features of the corresponding time segments in the deep state representation. The fusion method involves concatenating the statistical features of the subsequences with the feature vectors of the deep state representation. After processing and concatenating all time scales, a feature set containing multi-level time information is obtained; this feature set is the multi-scale knowledge representation of the deep state representation.

[0112] Based on the multi-scale knowledge representation, when performing a structured mapping of the co-occurrence patterns of cross-domain state variables in the deep state representation, feature dimensions belonging to different domains of the terminal environment link in the multi-scale knowledge representation are identified. The frequency of specific combinations of values ​​occurring in pairs of these dimensions is statistically analyzed at different time scales, and these frequency values ​​are arranged into a multi-dimensional array according to dimension pairs and time scales. Each dimension of this array corresponds to a state variable domain, and each element in the array represents the co-occurrence strength of a specific variable combination at a specific scale. This multi-dimensional array is flattened and reorganized according to its dimensional order to form a dense multi-dimensional data structure with a unified index structure. This data structure is the high-order correlation tensor of the multi-scale knowledge representation.

[0113] When summarizing the state evolution patterns of the higher-order correlation tensor, the higher-order correlation tensor is regarded as an encoding of state transition relations. By decomposing the various dimensional patterns of the tensor, common state combinations implicit in the tensor are extracted. For each common state combination, other related state combinations in the tensor are found, and the strength of this association is recorded. Each common state combination, its related state combinations, and the association strength are organized into a transition rule, and all such transition rules constitute a set. This set of transition rules describes the possible transition relationships between states, which is the set of transition patterns of the intrinsic states in the higher-order correlation tensor.

[0114] When performing cross-scale fusion of the dynamic interaction dependencies of the aforementioned transition pattern set, the transition pattern sets summarized at different time scales are merged. For transition rules describing the same state combination, their association strength at different scales is weighted and averaged, with the weight determined by the length of the corresponding time scale. For conflicting transition rules, the rule with the highest strength after weighted averaging is retained. After merging and weighting, a unified set of transition rules that integrates multi-scale information is obtained. The rules in this set not only describe state transitions but also reflect the confidence and temporal characteristics of the transitions through their strength values. This unified set of transition rules represents the spatiotemporal entanglement relationship of the terminal.

[0115] Based on the spatiotemporal entanglement relationship, when extracting transition conditions for the link state, the prerequisite state combination and the result state combination of each transition rule in the spatiotemporal entanglement relationship are analyzed. The value boundaries of each state variable in the prerequisite state combination are extracted, and these boundary conditions serve as constraints required for the transition to occur. Simultaneously, the possible value ranges of each state variable in the result state combination are extracted, serving as constraints for the state after the transition. The prerequisite and result constraints of all transition rules are summarized separately, and after removing duplicates, they are merged into two sets, corresponding to the state boundary conditions before and after the transition, respectively. These two sets together constitute the constraint boundary set of the link state.

[0116] Based on the constraint boundary set, when synthesizing the spatiotemporal topological relationships of the link states, a directed graph structure is constructed using different value levels of the link states as nodes and the allowed transition relationships defined by the constraint boundary set as edges. Each edge in the graph connects a state node satisfying the prerequisite constraints to a state node satisfying the result constraints, and the direction of the edge indicates the transition direction. Each edge is assigned a weight, which is determined by the strength value of the corresponding transition rule in the spatiotemporal entanglement relationship. This weighted directed graph structure fully characterizes the evolutionary possibilities of the link states in the spatiotemporal dimension, thus forming the spatiotemporal evolution pattern map of the link states.

[0117] Based on the spatiotemporal evolution pattern graph, when simulating path exploration for the deep state representation, starting from the current state node corresponding to the deep state representation, multi-step forward simulation is performed in the spatiotemporal evolution pattern graph according to the direction of the edges. In each simulation step, among all outgoing edges from the current node, an edge is randomly selected according to the edge weight and probability to move to the next node, and the sequence of nodes traversed is recorded. This process is repeated multiple times, generating a large number of possible node sequences starting from the current state. All generated node sequences are collected, and each sequence represents a possible future state evolution path. The set of these paths is the candidate state transition path set for the deep state representation.

[0118] When refining the candidate state transition path set into dominant paths, the frequency of each node in all candidate state transition paths is statistically analyzed. The node sequences with the highest frequency are selected as representative paths. For paths of different lengths, scores are assigned based on the cumulative weight of nodes in the path and the smoothness of the path, and the highest-scoring paths are selected. These selected representative paths are then averaged; specifically, for each future time step, the average of the state nodes of all representative paths at that time step is taken as the predicted state at that time step. These predicted states are then connected in chronological order to form a continuous state trajectory, which is the link quality prediction trajectory of the terminal.

[0119] The beneficial effects are as follows: by constructing multi-scale feature hierarchies of deep state representation and historical link quality, a unified knowledge representation integrating multi-granular temporal information is formed. Based on this, a high-order correlation tensor is generated by performing a structured mapping of the co-occurrence patterns of cross-domain state variables. On this basis, an intrinsic state transition pattern set is summarized, and by fusing their dynamic interaction dependencies across scales, a spatiotemporal entanglement relationship that accurately characterizes the spatiotemporal control law between states is obtained. Based on this relationship, a constraint boundary set of link state transitions is extracted, and a spatiotemporal evolution pattern map representing the possibility of state evolution is synthesized. Using this as a navigation tool, multi-path simulation exploration of deep state representation is conducted to obtain a set of candidate state transition paths. Finally, by refining the dominant paths and synthesizing trajectories from the candidate path set, a predictive trajectory that can accurately reflect the continuous change trend of future link quality is generated, providing a reliable and continuous decision-making basis for forward-looking resource optimization, significantly improving the spatiotemporal coherence and practical guiding value of the prediction.

[0120] S4. Based on the link quality prediction trajectory, perform preventive constraint derivation on the potential degradation mode of the terminal to obtain the constraint condition set of the terminal;

[0121] In this embodiment of the invention, the step of performing preventative constraint derivation on the potential degradation mode of the terminal based on the link quality prediction trajectory to obtain the constraint set of the terminal includes:

[0122] Dynamic vulnerability insight is performed on the predicted link quality trajectory to obtain potential vulnerable segments of the predicted link quality trajectory;

[0123] The degradation mode characteristics of the potentially vulnerable sections are obtained by performing degradation mode deconstruction on the potentially vulnerable sections.

[0124] The degradation mode features are subjected to inverse optimization conditions to construct a set of resilience enhancement features.

[0125] Based on the resilience enhancement feature set, the dynamic boundary of the communication state in the terminal is constrained and deduced in reverse to obtain the constraint condition set of the terminal.

[0126] The reverse optimization of the degradation mode features to construct the resilience enhancement feature set of the degradation mode features includes:

[0127] Key factors are extracted from the degradation mode features to obtain the set of dominant driving factors for the degradation mode features;

[0128] Based on a pre-defined multi-faceted reverse intervention scheme, the dynamic evolution path of the dominant driving factor set is simulated to obtain the evolution path sequence of the dominant driving factor set.

[0129] Based on the evolutionary path sequence, the resilience gain value of the multi-faceted reverse intervention scheme is calculated, wherein the formula for calculating the resilience gain value is:

[0130] ;

[0131] In the formula, This is the toughness gain value. The total time step of the evolutionary path sequence. The first in the evolutionary path sequence The variance of each quality indicator, The first in the evolutionary path sequence The average value of each quality indicator The preset zero-prevention constant is used. The first in the preset evolutionary path sequence Sensitivity weighting coefficients for each quality indicator In time step No. The quality indicator and the first The preset dynamic correlation coefficient between the quality indicators. For the first The quality indicator and the first The preset ideal correlation coefficient between the quality indicators The total number of quality indicators in the evolutionary path sequence. It is an exponential function;

[0132] Based on the resilience gain value, the Pareto front screening is performed on the multivariate reverse intervention schemes, and multidimensional feature extraction is performed on the screened intervention schemes to obtain the resilience enhancement feature set of the degradation mode features.

[0133] When performing dynamic vulnerability insight on the predicted link quality trajectory, the deviation of the link quality index from a preset quality threshold at each time point in the trajectory is analyzed. Sections on the trajectory where the quality index is below the set threshold are continuously monitored, and the rate and duration of quality degradation within these sections are calculated. Continuous time sections where the degradation rate exceeds a critical value and the duration reaches a certain length are identified; these identified continuous time sections are the potentially vulnerable sections of the predicted link quality trajectory.

[0134] When performing degradation mode deconstruction on the potentially vulnerable segments, complete data records of terminal status, environmental parameters, and link performance at all time points within each segment are extracted. The changing trends and collaborative patterns of various parameters in these data records are analyzed to identify common parameter change patterns that lead to continuous quality degradation. These identified common parameter change patterns are categorized and characterized according to the physical and protocol layers they affect, forming a set of feature entries describing the causes and manifestations of degradation. This set of feature entries constitutes the degradation mode features of the potentially vulnerable segments.

[0135] When extracting key factors from the degradation mode features, the contribution of each feature item in the degradation mode features to the link quality degradation is evaluated. The contribution is quantified by calculating the product of the average quality degradation when the feature item appears and the frequency of the feature item's appearance. The feature items with the highest contribution ranking are selected, and the specific controllable parameters or uncontrollable environmental factors corresponding to these feature items are identified. These factors are then aggregated into a set, which is the dominant driving factor set of the degradation mode features.

[0136] Based on a pre-defined multi-faceted reverse intervention scheme, when simulating the dynamic evolution path of the dominant driving factor set, the pre-defined scheme library contains multiple combinations of adjustment strategies for various adjustable parameters. For each adjustable factor in the dominant driving factor set, a series of adjustment strategies aimed at offsetting its degradation effects are selected from the pre-defined scheme library. These strategy combinations are applied to the system state starting from the initial moment of the vulnerable segment, and the evolution process of the system state over time after applying the strategies is deduced based on known state transition laws. The change sequence of link quality indicators over a future period is recorded, with each strategy combination corresponding to a change sequence. The set of all these change sequences constitutes the evolution path sequence of the dominant driving factor set.

[0137] When calculating the resilience gain value of the multi-path reverse intervention scheme based on the evolution path sequence, each evolution path sequence is analyzed. The average improvement of the link quality indicators in this sequence compared to the original vulnerable segment trajectory is calculated, along with the variance of the quality indicators' fluctuations. Dividing the average improvement by the variance yields a stability-adjusted gain score. This score is then compared to the resource overhead required to implement the strategy combination, estimated through parameter adjustments and energy consumption increments. Finally, a scalar value that comprehensively considers performance improvement, stability, and overhead is calculated; this value is the resilience gain value of the intervention scheme.

[0138] Based on the resilience gain value, when performing Pareto front screening on the multivariate inverse intervention schemes, each intervention scheme is plotted as a point on a two-dimensional plane with the resilience gain value as the vertical axis and resource cost as the horizontal axis. From all points, those points that are not dominated by other points in either the horizontal or vertical direction are selected; that is, for these points, there is no other point that simultaneously has a higher resilience gain value and a lower resource cost. The intervention schemes corresponding to these selected points constitute an optimal subset of schemes, which is the result of the Pareto front screening.

[0139] When performing multidimensional feature extraction on the screened intervention programs, for each intervention program in the Pareto front screening results, all parameter adjustment actions and their magnitudes in its strategy combination are extracted. Simultaneously, key performance improvement features, such as the delay time and duration of quality recovery, are extracted from the corresponding evolutionary path sequence. The parameter adjustment action features and performance improvement features of each program are concatenated to form a multidimensional feature vector. The multidimensional feature vectors of all programs are then aggregated to form a new feature vector set, which is the resilience enhancement feature set of the degradation mode features.

[0140] Based on the resilience enhancement feature set, when performing reverse derivation of the dynamic boundary of the communication state in the terminal, the parameter adjustment actions in all feature vectors of the resilience enhancement feature set are analyzed. For each adjustable communication state parameter, the adjustment direction and adjustment magnitude of that parameter in all feature vectors are identified. The consensus direction among all adjustment directions is taken as the recommended adjustment direction for that parameter, and the minimum and maximum magnitudes among all adjustment magnitudes are taken as the safe lower and upper bounds for the parameter adjustment. Simultaneously, the minimum link quality threshold value required to maintain the improvement effect is extracted from the performance improvement features. These adjustment directions, safety boundaries, and quality thresholds for each parameter are integrated to form a complete set of operational constraints and target requirements, which constitutes the constraint set of the terminal.

[0141] The parameters in the formulas all directly depend on the evolution path sequence and the preset system configuration parameters. The total time step of the evolution path sequence is directly taken from the total number of time points contained in the sequence. The variance of each quality indicator in the evolution path sequence is obtained by calculating the sum of the squares of the differences between all data points of that indicator and its average value, and then dividing by the number of data points. The average value of each quality indicator is obtained by summing the values ​​of all data points of that indicator and then dividing by the total number of data points. The zero-prevention constant is a pre-set, extremely small positive number used to avoid the denominator being zero in the division operation. The sensitivity weight coefficient of each quality indicator is a positive coefficient pre-assigned according to the importance of that indicator to the link quality. The dynamic correlation coefficient between any two quality indicators at any time step is obtained by calculating the Pearson correlation coefficient of the data sequences of these two indicators within a time window near that time. The ideal correlation coefficient between any two quality indicators is a set of target correlation coefficients pre-defined based on the theoretical optimal state or historical best performance. The total number of quality indicators is the number of different performance dimension parameters contained in the evolution path sequence.

[0142] The formula's significance lies in comprehensively quantifying the improvement in system performance resilience brought about by a reverse intervention scheme. The calculation process first calculates the ratio of the square of the variance to the mean of each quality indicator in the evolution path sequence, plus the remainder after removing the zero constant. This ratio is then multiplied by one to obtain a stability coefficient, which is obtained by exponentiation using the negative of the sensitivity weight coefficient of that indicator. The stability coefficients of all quality indicators are multiplied to obtain a comprehensive stability factor. Simultaneously, for each time step, the absolute value of the difference between the dynamic correlation coefficient and the preset ideal correlation coefficient between all pairs of quality indicators at that moment is calculated. The sum of the absolute differences of all indicator pairs is divided by the square of the total number of indicators to obtain a coordination difference degree. This coordination difference degree is then negatively multiplied using a natural exponential function to obtain a coordination coefficient. The comprehensive stability factor is multiplied by the coordination coefficient to obtain the local gain contribution at that time step. Finally, the local gain contributions of all time steps are summed and divided by the total time step length to obtain the average resilience gain value.

[0143] The formula shows that the resilience gain value monotonically increases as the stability of each quality indicator improves and the dynamic correlation between indicators approaches the ideal correlation. When the variance of a quality indicator is small relative to the square of its mean, its corresponding stability coefficient term approaches one, minimizing the weakening effect of the product of these terms. When the variance of a quality indicator increases significantly, its stability coefficient becomes less than one, thus reducing the overall product result. Furthermore, the larger the sensitivity weight coefficient of the indicator, the more significant the negative impact of the increased variance. Regarding coordination, if at a certain time step the dynamic correlation coefficients between all quality indicators are completely consistent with the preset ideal correlation coefficient, then the coordination difference is zero, and the corresponding coordination coefficient is the maximum value of one. If the dynamic correlation coefficient deviates from the ideal value, the coordination difference is positive, and the coordination coefficient will decay exponentially as the deviation increases. Therefore, an intervention scheme that maintains low fluctuations in the system's quality indicators and ensures that their dynamic relationships always approach the ideal model will yield the highest resilience gain value calculated using this formula.

[0144] The beneficial effects include the accurate identification of potentially vulnerable segments where both the rate and duration of quality degradation exceed thresholds through continuous monitoring and analysis of link quality prediction trajectories. Furthermore, multi-dimensional data analysis of the terminal environment links within these segments reveals the specific causes and characteristics of degradation. The most contributing key factors are extracted from the degradation patterns to form a set of dominant driving factors. Based on a pre-defined reverse intervention strategy library, the system evolution path under various strategy combinations is simulated. Pareto-optimal intervention schemes are selected by calculating resilience gain values ​​that comprehensively consider performance improvement, stability, and resource overhead. Subsequently, multi-dimensional features of parameter adjustment and performance improvement are extracted from these optimal schemes to form a resilience enhancement feature set. Based on this feature set, recommended adjustment directions, security boundaries, and necessary quality thresholds for each communication state parameter are derived. Finally, these are integrated into a complete set of constraints for proactive prevention and precise control, transforming reactive post-event response into proactive pre-event planning, significantly enhancing the system's forward-looking defense capabilities against potential degradation and improving resource allocation accuracy.

[0145] S5. Based on the set of constraints, perform cooperative strategy encoding on the terminal to obtain the cooperative instructions of the terminal;

[0146] In this embodiment of the invention, the step of encoding the cooperative strategy of the terminal based on the constraint set to obtain the cooperative instructions of the terminal includes:

[0147] The decision architecture of the terminal is obtained by performing multi-objective collaborative logic analysis on the set of constraints.

[0148] Based on the aforementioned decision-making architecture, the resource supply and demand relationship of the terminal is dynamically coupled and optimized to obtain a resource allocation scheme for the terminal.

[0149] Based on the resource allocation scheme, the migration path and control elements of the link status are aggregated together to obtain the collaborative parameter set of the terminal.

[0150] The set of collaborative parameters is encapsulated into collaborative instructions for the terminal.

[0151] When performing multi-objective collaborative logic analysis on the constraint set, different optimization objectives involved in the constraint set are identified, such as maximizing spectral efficiency, minimizing interference, and meeting the service quality threshold of a specific terminal. Potential conflicts and collaborative relationships between these objectives are analyzed; for example, increasing the power of one terminal may cause interference to neighboring terminals. Based on preset priority rules and the inherent technical correlations between objectives, a hierarchical decision-making framework is constructed. This framework explicitly defines how to weigh trade-offs and utilize collaborative effects when objectives conflict. This hierarchical decision-making framework constitutes the terminal's decision architecture.

[0152] Based on the aforementioned decision-making architecture, when dynamically optimizing the resource supply and demand relationship of the terminals, the demand for various types of wireless resources from all terminals is first summarized according to the current network status and compared with the total available resource supply of the network. Guided by the decision-making architecture and driven by the highest priority objective, the resource allocation scheme is iteratively adjusted. In each round of adjustment, it is checked whether the allocation scheme meets all constraints, and resources are reallocated for those constraints that are not met, while ensuring that the adjustment process conforms to the trade-off rules defined in the decision-making architecture. When the iterative adjustment yields a resource allocation scheme that simultaneously meets all constraints and conforms to the target priority ranking, this detailed allocation scheme is the resource allocation scheme for the terminals.

[0153] Based on the resource allocation scheme, when performing joint parameter aggregation on the link state migration path and control elements, the specific resource allocation results for each terminal in the resource allocation scheme are extracted, such as the assigned spectrum block, transmit power level, and beam number. These resource allocation parameters are combined with the channel states expected to be reached at various future moments predicted in the link state migration path. For each future moment, the resource parameters allocated to the terminal are combined and encoded with the expected channel state parameters for that moment to form a complete control parameter set for that moment. The control parameter sets for all future moments are arranged in chronological order to form a cross-time parameter sequence, which is the cooperative parameter set of the terminal.

[0154] When encapsulating the coordination parameter set into a coordination instruction for the terminal, each control parameter group in the coordination parameter set is converted into a specific information element specified by the protocol, according to a predefined communication protocol signaling format. Necessary signaling headers are added to these information elements, including the target terminal identifier, the timestamp of instruction effectiveness, the instruction sequence number, and a cyclic redundancy check (CRC) code for verifying data integrity. All information elements and signaling headers are assembled into a complete, fixed-length binary data block in the order specified by the protocol. This binary data block, conforming to the standard protocol format and containing complete control information, constitutes the terminal's coordination instruction.

[0155] The beneficial effect is that by performing multi-objective collaborative logic analysis on the constraint set, a hierarchical decision-making architecture is constructed that can clearly weigh conflicting objectives and utilize synergistic effects. Based on this architecture, dynamic coupling optimization iterations are implemented on the supply and demand relationship of terminal resources, generating a precise resource allocation scheme that satisfies all constraints and meets objective priorities. Based on this scheme, specific resource allocation parameters and predicted channel states of link state transition paths are jointly encoded and combined across time dimensions to form a collaborative parameter set covering complete control commands at all future moments. Finally, according to standard communication protocol formats, this parameter set is encapsulated into binary data blocks containing complete signaling headers and verification information, generating directly executable collaborative commands. This achieves efficient, reliable, and standardized conversion from multi-dimensional constraints to system-level executable control commands, ensuring the accurate implementation of optimization strategies and coordination and synchronization between terminals.

[0156] S6. Based on the cooperative instructions, perform global stabilization deployment on the link state to obtain the optimized link state of the terminal;

[0157] In this embodiment of the invention, the step of performing global stabilization deployment of the link state based on the cooperative instruction to obtain the optimized link state of the terminal includes:

[0158] The timing dependency of the collaborative instructions is parsed to obtain the timing execution plan of the terminal;

[0159] Based on the timing execution plan, the configuration parameters of the terminal are coordinated and deployed to obtain the synchronization link status of the terminal;

[0160] A steady-state assessment of the synchronization link status is performed to obtain an assessment report of the synchronization link status;

[0161] Based on the evaluation report, the link configuration of the terminal is adaptively corrected to obtain the optimized link status of the terminal.

[0162] When parsing the timing dependencies of the cooperative instructions, the binary data blocks of the cooperative instructions are first decoded to extract the individual control commands contained therein, along with their accompanying timestamps and prerequisite identifiers. The logical order of these commands is analyzed; for example, power adjustment can only be performed after channel measurement is completed, or data transmission can only begin after all terminals have completed beam alignment. Based on these dependencies, all commands are arranged into a directed acyclic graph (DAG), where nodes represent commands and edges represent dependencies. Then, based on the timestamp requirements and execution time estimates for each command, a specific execution start time and duration are assigned to each node in the graph, ensuring that dependencies are satisfied and the overall timeline is compact. Finally, a detailed time-sequence execution list is generated, which constitutes the timing execution plan for the terminal.

[0163] When coordinating the deployment of terminal configuration parameters based on the time-series execution plan, configuration commands are sent to each terminal one by one according to the time list in the time-series execution plan, driven by a precisely synchronized global clock. The sending time of each command strictly follows the start time in the plan, with a predetermined transmission and processing delay margin. After sending each batch of commands, confirmation feedback signals from each terminal are monitored to ensure that each terminal has correctly received and applied the configuration. When all commands in the plan have been sent and all necessary confirmations have been received, it is determined that all terminals have completed the parameter configuration according to the plan. At this time, all terminals in the entire network enter the operating state under the new parameters at the same time. This state in which all terminal parameters are consistent and synchronously updated is the synchronization link state of the terminals.

[0164] When performing a steady-state evaluation of the synchronization link status, during an observation period after configuration and deployment, link performance metrics data reported by all terminals are continuously collected, including but not limited to signal-to-noise ratio, bit error rate, throughput, and latency. The average value and fluctuation range of each metric during the observation period are calculated and compared with a preset steady-state target threshold. It is checked whether each metric remains within the target threshold range and whether the fluctuation amplitude converges to a stable range over time. Based on the comparison results and convergence status of all metrics, a structured document is generated. This document records the actual value, target value, compliance status, and overall stability conclusions for each metric. This structured document constitutes the evaluation report of the synchronization link status.

[0165] When adaptively correcting the terminal's link configuration based on the evaluation report, the indicators in the evaluation report that do not meet the steady-state target and their degree of deviation are analyzed. For each deviation, the corresponding configuration parameter adjustment strategy is found according to a predefined correction rule base. For example, if the signal-to-noise ratio is lower than the target, the transmit power is increased appropriately; if the bit error rate is too high, the modulation and coding scheme is adjusted. Based on the strategy, the terminal that needs adjustment and its parameter adjustment amount are calculated, and a set of correction commands are generated. These correction commands are used as new cooperative instructions for timing analysis and cooperative deployment again, and the steady-state evaluation is re-executed after deployment. This process is repeated until the evaluation report confirms that all indicators have reached and remained within the steady-state target range. At this point, the link state maintained by the terminal that meets all steady-state targets is the optimized link state of the terminal.

[0166] The beneficial effects are as follows: by analyzing the temporal dependencies of collaborative instructions, a temporal execution plan is constructed to ensure that the logical order of commands is synchronized with the execution time. Based on this, under the drive of a global clock, the configuration parameters of multiple terminals are precisely and collaboratively deployed, achieving a synchronized link state where all terminals in the network simultaneously enter the new parameter operating state. Furthermore, by continuously collecting the performance indicators of the synchronized link and comparing them with the steady-state target threshold, an evaluation report that comprehensively reflects the compliance and stability of each indicator is generated. Then, based on the deviations in the report, a parameter adjustment strategy is triggered according to preset correction rules, and correction instructions are generated. Through an iterative closed-loop process of redeployment and evaluation, until the link state of all terminals reaches and remains within the preset steady-state target range, a stable and reliable optimized link state is finally obtained. This achieves fully automated steady-state control from instruction issuance, synchronous execution, effect verification to closed-loop correction.

[0167] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0168] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for optimizing and adaptively adjusting the quality of 5G-A low-altitude communication links based on AI prediction, characterized in that, The method includes: S01. Perform cross-domain conflict resolution on the link status and multi-dimensional environmental data of the terminal in the low-altitude communication scenario to obtain conflict-free fused data of the terminal, including: In low-altitude communication scenarios, the link status of the terminal and multi-dimensional environmental data are dynamically aligned heterogeneously to obtain the spatiotemporal aligned data sequence of the terminal. Conflict identification is performed on the spatiotemporally aligned data sequence to obtain the set of conflicting elements in the spatiotemporally aligned data sequence; Remove the elements in the spatiotemporally aligned data sequence that correspond to the set of conflicting elements to obtain the conflict-free fused data of the terminal; S02. Perform high-dimensional situation extraction on the conflict-free fused data to obtain a deep state representation of the conflict-free fused data, including: Cross-domain causal influencing factors are decoupled from the ternary coupling relationship of terminal, environment, and link in the conflict-free fused data to obtain the set of influencing elements of the conflict-free fused data, including: Cross-domain phase synchronization analysis is performed on the ternary coupling relationship of terminal, environment, and link in the conflict-free fused data to obtain the phase oscillation sequence of the conflict-free fused data; Based on the phase oscillation sequence, the influence intensity value of the cross-domain factor in the ternary coupling relationship is calculated, wherein the formula for calculating the influence intensity value is: ； In the formula, cross-domain factors Cross-domain factors The influence intensity value, The total length of the trans-domain phase oscillation sequence. The preset delay time for causal analysis, For the current moment, Trans-domain factors in the phase oscillation sequence The instantaneous phase sequence, It is a sine trigonometric function. For the cross-domain factors The instantaneous phase value, Trans-domain factors in the phase oscillation sequence The instantaneous phase value, It is an exponential function. The sequence of first derivatives of the instantaneous phase sequence. Trans-domain factors in the phase oscillation sequence The sequence of second derivatives of the instantaneous phase, This is the absolute value operator; Based on the influence intensity value, the conflict-free fused data is subjected to sparsity filtering to obtain the causal influence element set of the conflict-free fused data; Based on the set of influencing elements, a structured deduction of the link state is performed to obtain a probability distribution map of the link state; The probability distribution map is subjected to topological manifold diffusion to obtain a stability description of the conflict-free fused data; Based on the stability description, the key mode reorganization is performed on the comprehensive communication situation of the low-altitude communication scenario to obtain the deep state representation of the conflict-free fused data. S03. Based on the deep state representation and the historical link quality of the terminal, perform intelligent graph deduction on the spatiotemporal evolution pattern of the link state to obtain the link quality prediction trajectory of the terminal, including: A topological analysis of the deep state representation and the historical link quality of the terminal is performed to obtain the spatiotemporal entanglement relationship of the terminal; Based on the spatiotemporal entanglement relationship, the transition conditions of the link state are extracted to obtain the constraint boundary set of the link state; Based on the constraint boundary set, the spatiotemporal topological relationship of the link state is synthesized to obtain the spatiotemporal evolution pattern map of the link state; Based on the spatiotemporal evolution pattern map, simulated path exploration is performed on the deep state representation to obtain a set of candidate state transition paths for the deep state representation; The candidate state transition path set is refined into a dominant path to obtain the link quality prediction trajectory of the terminal. S04. Based on the predicted link quality trajectory, a preventative constraint derivation is performed on the potential degradation modes of the terminal to obtain the constraint set of the terminal, including: Dynamic vulnerability insight is performed on the predicted link quality trajectory to obtain potential vulnerable segments of the predicted link quality trajectory; The degradation mode characteristics of the potentially vulnerable sections are obtained by performing degradation mode deconstruction on the potentially vulnerable sections. The degradation mode features are subjected to inverse optimization to construct conditions, resulting in a set of resilience enhancement features, including: Key factors are extracted from the degradation mode features to obtain the set of dominant driving factors for the degradation mode features; Based on a pre-defined multi-faceted reverse intervention scheme, the dynamic evolution path of the dominant driving factor set is simulated to obtain the evolution path sequence of the dominant driving factor set. Based on the evolutionary path sequence, the resilience gain value of the multi-faceted reverse intervention scheme is calculated, wherein the formula for calculating the resilience gain value is: ； In the formula, This is the toughness gain value. The total time step of the evolutionary path sequence. The first in the evolutionary path sequence The variance of each quality indicator The first in the evolutionary path sequence The average value of each quality indicator The preset zero-prevention constant is used. The first in the preset evolutionary path sequence Sensitivity weighting coefficients for each quality indicator In time step No. The quality indicator and the first The preset dynamic correlation coefficient between the quality indicators. For the first The quality indicator and the first The preset ideal correlation coefficient between the quality indicators The total number of quality indicators in the evolutionary path sequence. It is an exponential function; Based on the resilience gain value, the Pareto front screening of the multivariate reverse intervention schemes is performed, and multidimensional feature extraction is performed on the screened intervention schemes to obtain the resilience enhancement feature set of the degradation mode features. Based on the resilience enhancement feature set, the dynamic boundary of the communication state in the terminal is constrained and deduced in reverse to obtain the constraint condition set of the terminal. S05. Based on the set of constraints, perform cooperative strategy encoding on the terminal to obtain the cooperative instructions of the terminal; S06. Based on the cooperative instructions, the link state is globally stabilized to obtain the optimized link state of the terminal.

2. The AI-predictive-based 5G-A low-altitude communication link quality optimization and adaptive adjustment method as described in claim 1, characterized in that, The step of performing topological analysis on the deep state representation and the historical link quality of the terminal to obtain the spatiotemporal entanglement relationship of the terminal includes: Based on the historical link quality of the terminal, a multi-scale feature hierarchy is constructed for the deep state representation to obtain a multi-scale knowledge representation of the deep state representation. Based on the multi-scale knowledge representation, the co-occurrence patterns of cross-domain state variables in the deep state representation are correlated and structured to obtain the higher-order correlation tensor of the multi-scale knowledge representation. The state evolution patterns of the higher-order correlation tensor are summarized to obtain the set of transition patterns of the intrinsic states in the higher-order correlation tensor. Cross-scale fusion of the dynamic interaction dependencies of the transfer mode set yields the spatiotemporal entanglement relationship of the terminal.

3. The AI-predictive-based 5G-A low-altitude communication link quality optimization and adaptive adjustment method as described in claim 1, characterized in that, The step of encoding the cooperative strategy for the terminal based on the constraint set to obtain the cooperative instructions for the terminal includes: The decision architecture of the terminal is obtained by performing multi-objective collaborative logic analysis on the set of constraints. Based on the aforementioned decision-making architecture, the resource supply and demand relationship of the terminal is dynamically coupled and optimized to obtain a resource allocation scheme for the terminal. Based on the resource allocation scheme, the migration path and control elements of the link status are aggregated together to obtain the collaborative parameter set of the terminal. The set of collaborative parameters is encapsulated into collaborative instructions for the terminal.

4. The AI-predictive-based 5G-A low-altitude communication link quality optimization and adaptive adjustment method as described in claim 1, characterized in that, The step of globally stabilizing the link state based on the cooperative instructions to obtain the optimized link state of the terminal includes: The timing dependency of the collaborative instructions is parsed to obtain the timing execution plan of the terminal; Based on the timing execution plan, the configuration parameters of the terminal are coordinated and deployed to obtain the synchronization link status of the terminal; A steady-state assessment of the synchronization link status is performed to obtain an assessment report of the synchronization link status; Based on the evaluation report, the link configuration of the terminal is adaptively corrected to obtain the optimized link status of the terminal.

Images