Multi-dimensional communication service suitability evaluation method, system, equipment and medium

Abstract

The invention provides a multi-dimensional communication service suitability evaluation method, system and device and a medium, and the method comprises the steps: carrying out the network suitability evaluation of collected multi-dimensional communication service data through an suitability evaluation model, and obtaining a corresponding network suitability score; performing hierarchical division on the multi-dimensional communication service data based on preset communication protocol stack hierarchies to obtain network data corresponding to the multi-dimensional communication service data under different hierarchies; carrying out correlation analysis on the network data under different hierarchies by utilizing a graph neural network to obtain a cross-layer correlation score of the multi-dimensional communication service data; according to the network suitability score and the cross-layer relevance score, obtaining a comprehensive suitability score of the multi-dimensional communication service data; according to the method and the device, the dynamic relationship among the data of each layer of the communication protocol stack is analyzed through the graph neural network, and the cross-layer relevance is quantified, so that the suitability evaluation is not limited to the analysis of single-layer data, and can be combined with the cross-layer synergistic effect, thereby providing a more accurate suitability evaluation result.

Description

Technical Field

[0001] The present invention relates to the technical fields of communication network optimization and deep learning, and particularly relates to a multi-dimensional communication service adaptability evaluation method, system, device and medium. Background Art

[0002] At present, with the continuous development of 5G and future 6G communication technologies, the performance, topology and resources of communication networks have become increasingly complex. Modern communication networks not only involve traditional physical layer, transport layer and application layer, but also include emerging technologies such as relay networks and edge computing. And the network state often changes dynamically, affected by various factors such as network load, user location, bandwidth and environmental interference. Due to the complexity and variability of the network environment, traditional network adaptability evaluation methods can no longer meet the current communication service's requirements for performance evaluation and prediction.

[0003] Existing adaptability evaluation methods usually calculate based on static metrics or deterministic empirical rules. For example, some methods directly judge whether a certain service can adapt to the network environment by sampling the current network performance once; or evaluate service adaptability based on deterministic rules, such as the way that the bandwidth meets the threshold for adaptation. However, traditional communication service adaptability evaluation methods are difficult to quantify and analyze the cross-layer relationships between each layer of the communication protocol stack (such as the physical layer, transport layer, application layer), so that the synergistic effects between layers are not fully considered, thus affecting the accuracy of adaptability evaluation. Summary of the Invention

[0004] To solve the problem that the existing adaptability evaluation method cannot quantify cross-layer relationships, resulting in low accuracy of adaptability evaluation, the present invention proposes a multi-dimensional communication service adaptability evaluation method, including:

[0005] Based on the collected multi-dimensional communication service data, use a pre-constructed adaptability evaluation model to evaluate the network adaptability of the multi-dimensional communication service data, and obtain the corresponding network adaptability score;

[0006] Based on the preset communication protocol stack levels, divide the multi-dimensional communication service data into levels, and obtain the network data of the multi-dimensional communication service data under different levels;

[0007] Use a graph neural network to perform correlation analysis on the network data under different levels, and obtain the cross-layer correlation score of the multi-dimensional communication service data;

[0008] According to the network adaptability score and the cross-layer correlation score, obtain the comprehensive adaptability score of the multi-dimensional communication service data;

[0009] Among them, the adaptability evaluation model is constructed based on the dynamic weight matrix output by the deep adaptability evaluation network.

[0010] Optionally, based on the collected multi-dimensional communication service data, using the pre-constructed adaptability evaluation model to evaluate the network adaptability of the multi-dimensional communication service data, and obtaining the corresponding network adaptability score, including:

[0011] Arrange the collected multi-dimensional communication service data in time series to generate communication sequence data;

[0012] According to the communication sequence data, use the pre-constructed adaptability evaluation model to evaluate the adaptability of the communication sequence data, and obtain the network adaptability score of the multi-dimensional communication service data.

[0013] Optionally, the adaptability evaluation model includes the following construction process:

[0014] Use the historical communication sequence data as the input data of the training data;

[0015] Use the network adaptability score corresponding to the historical communication sequence data as the output data of the training data;

[0016] Based on the input data and output data of the training data, train the deep adaptability evaluation network to obtain the adaptability evaluation model.

[0017] Optionally, the training of the deep adaptability evaluation network based on the input data and output data of the training data to obtain the adaptability evaluation model includes:

[0018] Input the input data of the training data into the deep adaptability evaluation network, and output the dynamic weight matrix corresponding to the input data;

[0019] Generate an adaptability prediction score according to the input data and the dynamic weight matrix corresponding to the input data;

[0020] According to the adaptability prediction score and the output data, calculate the loss function value corresponding to the adaptability prediction score and the output data using the mean square error;

[0021] According to the loss function value, calculate the gradient data of the deep adaptability network using the batch gradient descent method, and according to the gradient data, use the Adam optimizer to optimize the parameters of the deep adaptability network until the deep adaptability evaluation network converges to obtain the adaptability evaluation model.

[0022] Optionally, the communication protocol stack levels at least include: physical layer, transport layer, and application layer.

[0023] Optionally, the correlation analysis of the network data at different levels using a graph neural network to obtain the cross-layer correlation score of the multi-dimensional communication service data includes:

[0024] Calculate the performance metric values of each level according to the network data at different levels and the preset performance metrics of each level;

[0025] Construct a network relationship graph with the performance metric values of each level as nodes corresponding to each level and the correlation relationships between the performance metric values of each level as edges;

[0026] According to the network relationship graph, perform correlation analysis using a graph neural network to obtain the cross-layer correlation score.

[0027] Optionally, the performing correlation analysis using a graph neural network according to the network relationship graph to obtain the cross-layer correlation score includes:

[0028] According to the network relationship graph, obtain the feature information of each node in the network relationship graph;

[0029] Use a predefined propagation mechanism to update the feature information of each node to obtain the feature update information of each node;

[0030] Aggregate the feature update information of each node to obtain the global embedding information of the network relationship graph;

[0031] According to the global embedding information, obtain the overall association information between the network data at different levels;

[0032] Generate a cross-layer correlation score according to the overall association information between the network data at different levels.

[0033] Optionally, the multi-dimensional communication service data includes one or more of the following: network performance data, service demand data, terminal characteristic data, and environment adaptability data;

[0034] The network performance data includes one or more of the following: bandwidth data, latency data, jitter data, and packet loss data;

[0035] The terminal characteristic data includes one or more of the following: computing power, storage capacity, and power consumption performance;

[0036] The environment adaptability data includes one or more of the following: channel interference data, network coverage data, and spectrum resource data.

[0037] Optionally, the expression corresponding to the comprehensive adaptability score is as follows:

[0038] S final= α·S t + β·S cross ;

[0039] where S final represents the comprehensive adaptability score; S t represents the network adaptability score; α represents the network adaptability weight; S cross represents the cross - layer correlation score; β represents the cross - layer correlation weight.

[0040] Based on the same inventive concept, the present invention also provides a multi - dimensional communication service adaptability evaluation system, including:

[0041] A network adaptability evaluation module, configured to perform network adaptability evaluation on the multi - dimensional communication service data based on the collected multi - dimensional communication service data by using a pre - constructed adaptability evaluation model, and obtain the corresponding network adaptability score;

[0042] A hierarchical division module, configured to perform hierarchical division on the multi - dimensional communication service data based on a preset communication protocol stack level, and obtain the network data of the multi - dimensional communication service data at different levels;

[0043] A cross - layer correlation evaluation module, configured to perform correlation analysis on the network data at different levels by using a graph neural network, and obtain the cross - layer correlation score of the multi - dimensional communication service data;

[0044] A comprehensive adaptability evaluation module, configured to obtain the comprehensive adaptability score of the multi - dimensional communication service data according to the network adaptability score and the cross - layer correlation score;

[0045] where the adaptability evaluation model is constructed based on a dynamic weight matrix output by a deep adaptability evaluation network.

[0046] Optionally, the network adaptability evaluation module includes:

[0047] A time - series arrangement sub - module, configured to perform time - series arrangement on the collected multi - dimensional communication service data to generate communication sequence data;

[0048] A dynamic adaptability evaluation sub - module, configured to perform adaptability evaluation on the communication sequence data by using a pre - constructed adaptability evaluation model according to the communication sequence data, and obtain the network adaptability score of the multi - dimensional communication service data.

[0049] Optionally, the evaluation system further includes: a model construction module, including:

[0050] An input setting sub - module, configured to use historical communication sequence data as input data for training data;

[0051] An output setting sub-module for using the network adaptability score corresponding to the historical communication sequence data as the output data of the training data;

[0052] A model training sub-module for training a depth adaptability evaluation network based on the input data and output data of the training data to obtain an adaptability evaluation model.

[0053] Optionally, the model training sub-module includes:

[0054] A data input unit for inputting the input data of the training data into the depth adaptability evaluation network and outputting a dynamic weight matrix corresponding to the input data;

[0055] A score prediction unit for generating an adaptability prediction score according to the input data and the dynamic weight matrix corresponding to the input data;

[0056] An error calculation unit for calculating a loss function value corresponding to the adaptability prediction score and the output data by using the mean square error according to the adaptability prediction score and the output data;

[0057] A parameter optimization unit for calculating gradient data of the depth adaptability network by using the batch gradient descent method according to the loss function value, and optimizing parameters of the depth adaptability network by using an Adam optimizer according to the gradient data until the depth adaptability evaluation network converges to obtain an adaptability evaluation model.

[0058] Optionally, the communication protocol stack levels at least include: a physical layer, a transport layer, and an application layer.

[0059] Optionally, the cross-layer correlation evaluation module includes:

[0060] A performance index calculation sub-module for calculating performance index values of each layer according to network data at different layers and preset performance indexes of each layer;

[0061] A relationship graph construction sub-module for constructing a network relationship graph with the performance index values of each layer as nodes corresponding to each layer and the association relationships between the performance index values of each layer as edges;

[0062] A correlation analysis sub-module for performing correlation analysis by using a graph neural network according to the network relationship graph to obtain a cross-layer correlation score.

[0063] Optionally, the correlation analysis sub-module includes:

[0064] A feature determination unit for obtaining feature information of each node in the network relationship graph according to the network relationship graph;

[0065] A feature update unit, configured to update the feature information of each node by using a predefined propagation mechanism to obtain the feature update information of each node;

[0066] An information aggregation unit, configured to aggregate the feature update information of each node to obtain the global embedding information of the network relationship graph;

[0067] An overall feature association unit, configured to obtain the overall association information between the network data at different levels according to the global embedding information;

[0068] A cross-layer association analysis unit, configured to generate a cross-layer association score according to the overall association information between the network data at different levels.

[0069] Optionally, the multi-dimensional communication service data includes one or more of the following: network performance data, service demand data, terminal characteristic data, and environment adaptability data;

[0070] The network performance data includes one or more of the following: bandwidth data, delay data, jitter data, and packet loss data;

[0071] The terminal characteristic data includes one or more of the following: computing power, storage capacity, and power consumption performance;

[0072] The environment adaptability data includes one or more of the following: channel interference data, network coverage data, and spectrum resource data.

[0073] Optionally, the expression corresponding to the comprehensive adaptability score is as follows:

[0074] S final = α·S t + β·S cross ;

[0075] Wherein, S final represents the comprehensive adaptability score; S t represents the network adaptability score; α represents the network adaptability weight; S cross represents the cross-layer association score; β represents the cross-layer association weight.

[0076] On the other hand, the present invention further provides an electronic device, including: at least one processor and a memory; the memory and the processor are connected by a bus;

[0077] The memory is configured to store one or more programs;

[0078] When the one or more programs are executed by the at least one processor, the above-mentioned multi-dimensional communication service adaptability evaluation method is implemented.

[0079] On the other hand, the present invention also provides a computer device-readable storage medium with an execution program stored thereon. When the execution program is executed, a multi-dimensional communication service adaptability evaluation method as described above is implemented.

[0080] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0081] The present invention provides a multi-dimensional communication service adaptability evaluation method, system, device and medium, including: based on the collected multi-dimensional communication service data, using a pre-constructed adaptability evaluation model to evaluate the network adaptability of the multi-dimensional communication service data, and obtaining the corresponding network adaptability score; based on a preset communication protocol stack hierarchy, dividing the multi-dimensional communication service data into levels to obtain the network data of the multi-dimensional communication service data at different levels; using a graph neural network to analyze the correlation of the network data at different levels to obtain the cross-layer correlation score of the multi-dimensional communication service data; according to the network adaptability score and the cross-layer correlation score, obtaining the comprehensive adaptability score of the multi-dimensional communication service data; wherein, the adaptability evaluation model is constructed based on the dynamic weight matrix output by the deep adaptability evaluation network; through the pre-constructed adaptability evaluation model of the present application, the network adaptability score of the multi-dimensional communication service data can be realized based on the dynamic weight matrix, so as to capture the dynamic changes of multi-dimensional factors. At the same time, through the hierarchical division based on the communication protocol stack, the network data at different levels is structurally processed, and the graph neural network is combined to analyze the correlation of the hierarchical data to quantify the collaborative relationship between different levels of the protocol stack, thereby generating a cross-layer correlation score. This cross-layer analysis further solves the problem that it is difficult for traditional methods to quantify the influence of cross-layer data interaction, making the adaptability evaluation more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 It is a schematic flow chart of a multi-dimensional communication service adaptability evaluation method provided by the present invention;

[0083] Figure 2 It is a schematic structural composition diagram of a multi-dimensional communication service adaptability evaluation system provided by the present invention;

[0084] Figure 3 It is a schematic structural diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0085] The present invention proposes a multi-dimensional communication service adaptability evaluation method, system, device and medium. The following further details the specific embodiments of the present invention with reference to the accompanying drawings.

[0086] Embodiment 1:

[0087] The present invention provides a method for evaluating the adaptability of multi-dimensional communication services. By using deep learning and dynamic simulation algorithms, combined with multi-dimensional communication service requirements and network environment changes, it realizes the dynamic analysis of the adaptability between communication services and network resources, predicts the performance of different types of communication service data in multiple scenarios, and generates optimized deployment suggestions. This method can be widely applied to scenarios such as network planning, service optimization, and fault warning. The specific process schematic diagram is as shown in Figure 1 shown, including:

[0088] Step 1: Based on the collected multi-dimensional communication service data, use the pre-constructed adaptability evaluation model to evaluate the network adaptability of the multi-dimensional communication service data, and obtain the corresponding network adaptability score;

[0089] Step 2: Based on the preset communication protocol stack levels, divide the multi-dimensional communication service data into levels to obtain the network data of the multi-dimensional communication service data under different levels;

[0090] Step 3: Use a graph neural network to perform correlation analysis on the network data under different levels to obtain the cross-layer correlation score of the multi-dimensional communication service data;

[0091] Step 4: According to the network adaptability score and the cross-layer correlation score, obtain the comprehensive adaptability score of the multi-dimensional communication service data;

[0092] Among them, the above-mentioned adaptability evaluation model is constructed based on the dynamic weight matrix output by the Deep Adaptation Evaluation Network (DAEN).

[0093] Generally, traditional methods for evaluating the adaptability of communication services mostly rely on static index calculations or empirical rule systems. For example, some methods judge whether a certain service is adapted to the current network environment by taking a single sample of network performance, or evaluate whether a service meets network conditions through deterministic rules. However, these methods have significant limitations in practical applications. First, they cannot effectively cope with the dynamically changing network environment. As the network topology, load, and user behavior continue to change, static evaluation methods cannot reflect the changes in network status in real time. Second, traditional methods often only focus on a single network dimension (such as bandwidth or latency) or the requirements of a specific service (such as the bandwidth requirement of a video stream), while ignoring other key factors such as the performance of terminal devices and environmental interference and other multi-dimensional factors, making it difficult to provide a comprehensive adaptability evaluation. Finally, even if traditional methods can identify the adaptability problems between a certain service and the network, they usually cannot provide precise optimization solutions.

[0094] To solve the above problems, the present invention collects multi-dimensional communication service data including network performance data, service requirement data, terminal characteristic data, and environmental adaptability data, and dynamically evaluates it using a pre-constructed adaptability evaluation model to generate a network adaptability score. For example, the above-mentioned network performance data may include information such as bandwidth, latency, jitter, and packet loss rate, the terminal characteristic data may include information such as computing power, storage capacity, and power consumption performance, and the environmental adaptability data may include information such as channel interference, network coverage, and spectrum resource utilization rate. In addition, in view of the hierarchical characteristics of the communication protocol stack, the present invention further hierarchically divides the multi-dimensional communication service data based on the communication protocol stack levels (such as the physical layer, transport layer, application layer), and combines the Graph Neural Network (GNN) to perform cross-layer correlation analysis on the data at different levels, thereby generating a cross-layer correlation score. By comprehensively combining the network adaptability score and the cross-layer correlation score, a comprehensive adaptability score is generated, providing comprehensive and dynamic guidance for the adaptability evaluation and resource optimization of communication services. However, the limitations of traditional methods are also manifested in the data organization and evaluation process. The static method of single sampling often fails to effectively utilize the time series characteristics in communication sequence data, resulting in the omission of important information in the dynamic changes of the network. To solve this problem, specifically:

[0095] In one implementation, the process of performing network adaptability evaluation on the multi-dimensional communication service data using the pre-constructed adaptability evaluation model based on the collected multi-dimensional communication service data in step 1 above to obtain the corresponding network adaptability score may include:

[0096] Arrange the collected multi-dimensional communication service data in time series to generate communication sequence data;

[0097] According to the communication sequence data, use the pre-constructed adaptability evaluation model to perform adaptability evaluation on the communication sequence data to obtain the network adaptability score of the multi-dimensional communication service data;

[0098] In this implementation, by arranging the collected multi-dimensional communication service data (which can be collected through mechanical energy in the communication network and terminal devices) in chronological order to generate communication sequence data, and using an adaptability evaluation model to evaluate the communication sequence data, the network adaptability score of the multi-dimensional communication service data can be obtained. The technical effect of this process is that it makes full use of the time series characteristics in the multi-dimensional communication service data, can capture the data change trend in the dynamic network environment, and provides an accurate evaluation of the communication network adaptability. Through the evaluation based on the communication sequence data, not only can the dynamic changes of network performance parameters (such as bandwidth, delay, jitter, and packet loss rate, which can be collected in real time from the base station, core network, and terminal devices) be reflected in real time, but also the service demand parameters (such as the demand models for high-definition video or vehicle networking), terminal capability parameters (such as computing power, storage capacity, and power consumption performance), and environmental adaptability parameters (such as channel interference and network coverage) can be combined to comprehensively evaluate the adaptability of the communication service. This implementation not only improves the dynamics and real-time performance of the adaptability evaluation through the generation and processing of time series data, but also provides an accurate basis for optimization and resource allocation in complex network environments. Through this process, the evaluation results are more reliable and comprehensive, thus significantly improving the adaptability performance of communication services in the dynamic network environment. In order to improve the speed and performance of model convergence, in this implementation, after collecting the multi-dimensional communication service data and before constructing the time series data, the data can also be normalized, and the specific process can include:

[0099] (1) Clean the data: Remove missing or abnormal data (for example, use interpolation or mean filling).

[0100] (2) Standardize the features: Normalize all parameters and unify them to the interval [0, 1];

[0101] (3) Construct the time series. After normalizing the data, arrange it in chronological order to generate the above-mentioned communication sequence data as the input for subsequent simulation. Specifically, the process of constructing the time series can include:

[0102] Establish a Markov chain model to generate the time series of the dynamic network environment;

[0103] Define the state set S = {s 1 , s 2 , …, s m}, such as states like "idle", "lightly congested", "severely congested", etc. (for example, use s 1 to represent idle, indicating low network load and smooth communication; s 2 to represent lightly congested, indicating medium network load and slightly delayed communication; ……; s mIndicates severe congestion, meaning the network load is too high, and there may be serious delays or packet losses in communication; m represents the total number of states). And through the transition probability matrix P = [p ab reflects the change probabilities between different states (for example, p ab represents the state s a to s b ). Based on the existing transition probability matrix, the future change trend of the network state can be predicted, which helps optimize the allocation of communication resources. And taking the time series as input data provides dynamic network environment information for the adaptability evaluation model, thereby improving the accuracy and timeliness of the evaluation.

[0104] The above evaluation process of network adaptability scoring based on communication sequence data depends on the accuracy and reliability of the adaptability evaluation model, and the construction of the adaptability evaluation model is the key to ensuring the evaluation effect. In order to enable the model to effectively cope with the dynamically changing network environment, capture the time series characteristics of multi-dimensional communication service data, and accurately evaluate the adaptability of communication services, the adaptability evaluation model needs to be trained on a large amount of historical data through deep learning technology to master the mapping relationship between input and output, and generate an adaptability model that can be dynamically adjusted. Therefore, the following will further describe the specific construction process of the adaptability evaluation model to clearly show how to learn from historical communication sequence data through a deep adaptability evaluation network, thereby constructing the model and achieving accurate network adaptability evaluation. Specifically:

[0105] In one implementation, the above adaptability evaluation model may include the following construction process:

[0106] Use historical communication sequence data as the input data of the training data;

[0107] Use the network adaptability scores corresponding to the historical communication sequence data as the output data of the training data;

[0108] Based on the input data and output data of the training data, train the deep adaptability evaluation network to obtain the adaptability evaluation model.

[0109] In this implementation, the model framework of the above deep adaptability evaluation network may include: an input layer, a hidden layer, and an output layer, which are:

[0110] Input layer: includes the service demand vector V b , the network performance vector N t , the terminal capability vector T c , and the environmental adaptability vector E;

[0111] Hidden layer: adopts a multi-layer fully connected network, and the activation function uses ReLU to extract non-linear features;

[0112] Output layer: Generate the dynamic weight matrix W DL ={w 1 , w 2 , …, w n}, where n represents the total number of data, and w n represents the weight of the nth data.

[0113] In this implementation, by constructing an adaptability evaluation model and using historical communication sequence data as input data and network adaptability scores (i.e., labels for supervised learning) as output data for training, a dynamic and adaptive adaptability evaluation model is achieved. During the model training process, multi-dimensional features such as service demand vectors, network performance vectors, terminal capability vectors, and environmental adaptability vectors are introduced through the input layer, enabling the model to fully capture multi-dimensional information in complex communication scenarios; the hidden layer uses a multi-layer fully connected network combined with the ReLU activation function to extract non-linear features, allowing the model to learn complex interaction relationships between multi-dimensional inputs; the dynamic weight matrix generated by the output layer can not only dynamically adjust the importance of different features but also flexibly adapt to the requirements of different communication scenarios according to real-time input data. Therefore, this implementation efficiently learns the temporal characteristics and multi-dimensional associations of multi-dimensional communication service data through deep learning techniques, generating an evaluation model that can dynamically adapt to changes in the network environment, not only improving the accuracy and comprehensiveness of adaptability evaluation but also providing a weight allocation mechanism for multi-dimensional data, making the model have stronger interpretability and optimization guidance capabilities, thus providing a scientific basis for adaptability evaluation and resource optimization in complex communication environments.

[0114] To implement the construction process of the above adaptability evaluation model, it is necessary to clarify how to use the deep adaptability evaluation network to train historical communication sequence data, so as to ensure that the model can dynamically generate adaptability scores and accurately reflect network adaptability. This process involves data label generation, learning the mapping relationship between data input and output, generating a dynamic weight matrix, optimizing the loss function, dynamically updating weights, and adjusting model parameters. Next, the implementation steps of training the deep adaptability evaluation network based on the input data and output data of the training data and finally obtaining the adaptability evaluation model will be described in detail. Specifically:

[0115] In this implementation, the process of training the deep adaptability evaluation network based on the input data and output data of the training data to obtain the adaptability evaluation model can include:

[0116] Input the input data of the training data into the deep adaptability evaluation network, and output the dynamic weight matrix corresponding to the input data;

[0117] Generate an adaptability prediction score according to the input data and the dynamic weight matrix corresponding to the input data;

[0118] According to the adaptability prediction score and the output data, the mean squared error is used to calculate the loss function value corresponding to the adaptability prediction score and the output data;

[0119] According to the loss function value, the batch gradient descent method is used to calculate the gradient data of the deep adaptability network, and according to the gradient data, the Adam optimizer is used to optimize the parameters of the deep adaptability network until the deep adaptability evaluation network converges, and an adaptability evaluation model is obtained.

[0120] In this implementation, the generation of data labels is the process of selecting training data. By using historical communication sequence data as the input of training data, the user experience score of the historical communication sequence data (for example, using S user denotes) is used as the supervised learning label;

[0121] Exemplarily, the expression of the above loss function can be as follows:

[0122]

[0123] where L represents the loss function value; m represents the total amount of training data; i = 1...m; represents the actual user experience score of the i-th data (i.e., the above network adaptability score); represents the adaptability prediction score of the i-th data;

[0124] In this implementation, by training the deep adaptability evaluation network, an adaptability evaluation model is accurately constructed, realizing the generation of a dynamic weight matrix and the efficient evaluation of the adaptability prediction score. By using the user experience score of historical communication sequence data as the supervised learning label and combining the mean squared error loss function to measure the difference between the predicted score and the actual score, the goal of model optimization is effectively clarified, and the gradient is calculated by the batch gradient descent method and the Adam optimizer is used to dynamically adjust the model parameters, enabling the model to converge quickly and reach the optimal state. Through this training process, the model can not only dynamically generate a weight matrix highly correlated with the input data, but also accurately quantify the contribution degree of the input features based on the weight matrix, thereby generating an adaptability prediction score closer to the actual business requirements. Finally, through this implementation, the prediction accuracy and dynamic adaptability of the model are improved, effectively coping with the complexity and dynamic changes of the network environment, and at the same time providing a scientific guiding basis for the optimization of communication services and resource scheduling, significantly enhancing the comprehensiveness and real-time nature of adaptability evaluation.

[0125] Through the above training process, the deep adaptability evaluation network can construct an adaptability evaluation model. This model realizes the adaptability prediction score of multi-dimensional communication service data by dynamically generating a dynamic weight matrix corresponding to the input data. To further improve the comprehensiveness and accuracy of the adaptability evaluation, based on the adaptability evaluation model, according to the hierarchical characteristics of the communication protocol stack, the multi-dimensional communication service data can be hierarchically divided, so as to more deeply explore the correlation between data at different levels and obtain a more comprehensive network adaptability evaluation result. Therefore, the process of hierarchically dividing the multi-dimensional communication service data based on the communication protocol stack level will be described next. Specifically:

[0126] In one implementation, the process of hierarchically dividing the multi-dimensional communication service data based on the preset communication protocol stack level to obtain the network data corresponding to the multi-dimensional communication service data at different levels in step 2 above may include:

[0127] The process of hierarchically dividing the multi-dimensional communication service data based on the preset communication protocol stack level first needs to clarify the hierarchical structure of the communication protocol stack, such as the physical layer, transport layer, network layer, and application layer, etc. By mapping the collected multi-dimensional communication service data features to different protocol levels, the orderly classification of data can be realized. This process classifies data according to the functional characteristics of each layer of the communication protocol stack and the attributes of data dimensions. For example, signal strength and channel interference are classified into the physical layer, latency and jitter are classified into the network layer, real-time requirements and user experience scores are classified into the application layer, thus forming a clearly hierarchical data structure. At the same time, during the data layering process, the influence and correlation of some cross-layer data are fully considered to ensure that each data feature can accurately correspond to a certain layer or a set of related layers of the protocol stack. Through this layering mapping process, the complex multi-dimensional communication service data can be organized and managed more clearly, laying a foundation for subsequent analysis steps and providing accurate input for cross-layer data correlation analysis, which helps to achieve a more accurate and comprehensive evaluation of the communication network adaptability.

[0128] Through the above process of hierarchical division of multi-dimensional communication service data based on the communication protocol stack levels, complex multi-dimensional communication service data can be effectively organized and managed, and a clear and accurate input basis is provided for subsequent cross-layer data correlation analysis. Since the data at each level of the protocol stack has a certain correlation in the actual communication network, for example, channel interference at the physical layer may affect the delay and packet loss rate at the transport layer, and the path selection at the network layer may also affect the user experience at the application layer. Therefore, evaluating only through the data at a single level often cannot comprehensively reflect the adaptability performance of the communication network. To more deeply explore the correlation and synergy between the data at each level of the protocol stack, the next step can be to perform correlation analysis on the network data at different levels by using a graph neural network, so as to generate a cross-layer correlation score and provide a more accurate and comprehensive reference for comprehensive adaptability evaluation. Specifically:

[0129] In one implementation, the process of using a graph neural network to perform correlation analysis on the network data at different levels in step 3 above to obtain the cross-layer correlation score of multi-dimensional communication service data may include:

[0130] Calculate the performance metric values of each level according to the network data at different levels and the preset performance metrics of each level;

[0131] Use the performance metric values of each level as the nodes corresponding to each level (such as the signal strength at the physical layer, the round-trip time at the transport layer, and the delay sensitivity at the application layer), and use the correlation relationships between the performance metric values of each level as the edges to construct a network relationship graph (such as the correlation relationship between signal strength and packet loss rate);

[0132] According to the network relationship graph, use a graph neural network to perform correlation analysis to obtain the cross-layer correlation score;

[0133] In this implementation, by using graph neural networks to perform correlation analysis on network data at different levels, accurate generation of cross-layer correlation scores is achieved, thereby significantly enhancing the comprehensiveness and depth of communication service adaptability evaluation. By calculating the performance metric values of each level based on preset performance metrics and using them as nodes, and at the same time constructing a network relationship graph according to the correlation relationships between these metric values (such as the correlation relationship between physical layer signal strength and transport layer packet loss rate), this implementation effectively presents the complex correlations between different levels of the communication protocol stack in a structured form. Subsequently, by using graph neural networks to perform correlation analysis on the network relationship graph, the synergistic effects and potential relationships between the data of each level of the protocol stack can be fully explored, and a quantitative cross-layer correlation score can be generated. This technical effect is reflected in many aspects: it can not only deeply reflect the interactive influence between data at different levels, but also capture the time-varying characteristics of cross-layer data in a dynamic network environment, providing a more comprehensive and accurate correlation basis for comprehensive adaptability evaluation. Finally, the solution improves the adaptability evaluation ability for complex communication networks through structured correlation quantification, providing more scientific and refined guidance for subsequent optimization and resource allocation.

[0134] Through the above process of using graph neural networks to perform correlation analysis on network data at different levels, the complex correlation relationships between the performance metrics of each level of the protocol stack can be effectively captured, and the structured characteristics of these correlation relationships can be reflected by constructing a network relationship graph. This method can not only quantify the degree of correlation between data at different levels, but also provides a structured data input basis for subsequent evaluation steps. However, in order to further explore the node feature information in the network relationship graph and reveal the overall correlation between data at different levels from a global perspective, it is necessary to utilize the characteristics of graph neural networks. Through the propagation and update of node feature information, global embedding information is generated, and then a cross-layer correlation score is generated. Specifically:

[0135] In one implementation, the process of using graph neural networks to perform correlation analysis based on the network relationship graph to obtain a cross-layer correlation score may include:

[0136] According to the network relationship graph, obtain the feature information of each node in the network relationship graph (for example, it can be represented by H (0) );

[0137] Use a predefined propagation mechanism to update the feature information of each node to obtain the feature update information of each node;

[0138] Aggregate the feature update information of each node to obtain the global embedding information of the network relationship graph;

[0139] According to the global embedding information, obtain the overall correlation information between network data at different levels;

[0140] Generate a cross-layer correlation score based on the overall correlation information between network data at different levels for comprehensive evaluation.

[0141] In this implementation, by using a graph neural network to update and aggregate the feature information of each node in the network relationship graph, a cross-layer correlation score is generated, significantly improving the accuracy and comprehensiveness of cross-layer data correlation analysis. By propagating and updating the feature information of each node in the network relationship graph, a predefined propagation mechanism can be adopted to perform weighted summation and non-linear transformation (such as using the ReLU activation function) based on the feature information of neighboring nodes in each layer, so as to capture the complex correlations and topological structures between nodes in the network relationship graph. Through layer-by-layer iteration, the updated node features can effectively fuse the context information of neighboring nodes, and then reflect higher-order cross-layer interaction relationships. After completing the update of node features, further aggregate the node feature information to generate the global embedding information of the network relationship graph, and quantify the overall correlation between data at different levels from an overall perspective. Finally, the cross-layer correlation score calculated based on the global embedding information can not only reflect the synergy between data at different levels of the protocol stack, but also dynamically adapt to changes in the network environment, improving the timeliness and accuracy of the evaluation results. Through this process, multi-level interaction relationships in a complex network environment can be efficiently captured, providing key indicators for comprehensive adaptability evaluation, and thus providing a more scientific and detailed guiding basis for the optimization of communication services and resource scheduling.

[0142] Exemplarily, the expression for updating the feature information of each node in the network relationship graph using the predefined propagation mechanism can be as follows:

[0143]

[0144] where, H i (k) represents the feature update information of node i in the k-th layer; N(i) represents the set of neighboring nodes of node i; d i represents the degree of node i; d j represents the degree of node j, and satisfies j ∈ N(i); W (k) represents the weight matrix of the graph neural network in the k-th layer (representing the learnable parameters of the model); σ represents the activation function (for example, ReLU is used as the activation function in the present invention); represents the feature information of node j in the (k - 1)-th layer; In this example, by defining a propagation mechanism to update the feature information of each node in the network relationship graph, in-depth mining of the local structure and global correlation of network data is realized, thereby effectively capturing the complex interaction relationships between different levels of the protocol stack. In this propagation mechanism, by combining the feature information of neighboring nodes, the degree of nodes, and the learnable weight matrix W(k) , it can dynamically adjust the weight distribution of each node feature to ensure that the influence of different nodes in the feature update process can be reasonably quantified. At the same time, by introducing a non-linear activation function (such as ReLU), it can further enhance the model's ability to express high-dimensional features and its ability to model non-linear relationships. This mechanism of multi-layer propagation and update enables the feature information of nodes to not only contain their own states, but also gradually integrate the feature information of neighbor nodes and their associations with other nodes, thereby generating more accurate and comprehensive feature representations.

[0145] Through the above process of using a graph neural network to perform correlation analysis on the network relationship graph and generate cross-layer correlation scores, the global embedding information of node features in the network relationship graph can be effectively extracted, the overall correlation between different levels of data can be quantified, and further an important correlation basis for adaptability evaluation is provided. However, in order to more comprehensively evaluate the adaptability of multi-dimensional communication service data, a single network adaptability score or cross-layer correlation score may not fully reflect the comprehensive adaptability performance of the communication network. Therefore, it is particularly important to generate a comprehensive adaptability score by combining the comprehensive evaluation results of the network adaptability score and the cross-layer correlation score. Specifically:

[0146] In one implementation, the expression corresponding to the comprehensive adaptability score in step 4 above can be as follows:

[0147] S final = α·S t + β·S cross ;

[0148] Wherein, S final represents the comprehensive adaptability score, with a range of [0, 1], 1 representing the best adaptability, and 0 representing complete non-adaptability; S t represents the network adaptability score calculated based on the dynamic weight matrix; α represents the network adaptability weight; S cross represents the cross-layer correlation score; β represents the cross-layer correlation weight; in this implementation, through the expression of the comprehensive adaptability score, the network adaptability score S t calculated based on the dynamic weight matrix is organically combined with the cross-layer correlation score S cross , endowing different scores with the adjustment ability of weights α and β, thereby realizing a comprehensive quantitative evaluation of the adaptability of communication services. The design of this expression enables the comprehensive adaptability score S finalThe range is limited between [0, 1], where 1 represents the best adaptability and 0 represents complete non - adaptability, thus providing an intuitive and interpretable adaptability evaluation result. When conducting a comprehensive adaptability assessment based on multi - dimensional communication service data, by combining the network adaptability score and the cross - layer correlation score, it can not only accurately reflect the adaptability degree between the current network environment and service requirements, but also capture the complex interaction relationships between different layers of the protocol stack, further improving the accuracy and comprehensiveness of the assessment. In addition, the introduction of weight parameters α and β endows the scheme with the ability to flexibly adjust the adaptability assessment in different scenarios. For example, in scenarios with high real - time requirements, the weight of S t can be increased, while in scenarios sensitive to cross - layer synergy, the weight of S cross can be increased. This flexibility effectively enhances the adaptability of the model, enabling it to adapt to diverse communication scenarios. Finally, through the combination of dynamic and cross - layer evaluation, the scientific quantification of the adaptability of communication services in complex network environments is achieved, providing a solid technical support for optimizing resource allocation and improving network service quality.

[0149] After obtaining the comprehensive adaptability score according to the above method, corresponding optimization schemes can be generated for the comprehensive adaptability score (for example, optimization can be carried out from the perspectives of network resource adjustment, terminal optimization, protocol parameter adjustment, and environment adaptation). For example, the optimization scheme from the perspective of network resource adjustment can be: dynamically allocate bandwidth to relieve congestion in hot spots; the optimization scheme from the perspective of terminal optimization can be: establish and upgrade terminal hardware to meet service requirements; the optimization scheme from the perspective of protocol parameter adjustment can be: optimize the congestion control parameters of the transport layer to improve the user experience; the optimization scheme from the perspective of environment adaptation improvement can be: deploy more base stations or adjust the spectrum resource allocation.

[0150] In summary, a multi-dimensional communication service adaptability evaluation method proposed by the present invention can effectively solve a number of technical problems of traditional network adaptability evaluation methods by combining a deep adaptability evaluation network and graph neural network technology. First, in view of the complexity and variability of the dynamic network environment, the present invention realizes the real-time dynamic evaluation of network adaptability by collecting and processing multi-dimensional communication service data and combining the generation and analysis of time series, thus overcoming the limitation that traditional static methods cannot reflect the dynamic changes of network states. Second, traditional methods often rely only on a single network dimension for adaptability evaluation, ignoring the influence of multi-dimensional factors such as terminal capabilities, service requirements, and environmental adaptability. The present invention significantly improves the comprehensiveness and accuracy of evaluation by constructing a dynamic weight matrix and comprehensively considering multi-dimensional features such as network performance data, terminal characteristic data, service requirement data, and environmental adaptability data. In addition, in view of the complex interaction relationships between different levels of the communication protocol stack, the present invention quantifies and captures the correlation characteristics between cross-layer data through correlation analysis based on graph neural networks, solving the problem that traditional methods cannot quantify cross-layer cooperation relationships. Finally, through the generated comprehensive adaptability score, the present invention can not only provide accurate adaptability evaluation results, but also provide a scientific basis for the optimization and resource allocation of communication networks, effectively solving the problem that the evaluation results of traditional methods lack optimization guidance. In summary, the present invention provides a comprehensive and efficient solution to problems such as dynamic network environment, single-dimensional evaluation limitations, insufficient cross-layer relationship quantification, and lack of optimization guidance, laying a solid technical foundation for communication service adaptability evaluation.

[0151] Embodiment 2:

[0152] Taking a specific embodiment (for example, in a scenario where within a certain urban area, a user is playing a high-definition video stream service, the network environment fluctuates randomly, and this service has high requirements for indicators such as bandwidth, latency, and jitter) to illustrate a multi-dimensional communication service adaptability evaluation method proposed by the present invention, which specifically includes:

[0153] Real-time data is collected through a wireless network transmission terminal and a base station to obtain parameters such as network performance parameters, service requirement parameters, terminal capability parameters, and environmental adaptability information. Assume that the real-time data collected is shown in Table 1.

[0154] Table 1 High-definition video service collected data information

[0155]

[0156] Perform normalized data preprocessing on the above parameters to construct a time series:

[0157] Step 1: Clean the data - Assume that the data has no missing or abnormal values and enter the standardization process.

[0158] Step 2: Standardize features - Normalize all parameters to the range [0, 1]. The calculation results are as follows:

[0159] Bandwidth: 20 Mbps / 100 Mbps = 0.2;

[0160] Latency: 50 ms / 100 ms = 0.5;

[0161] Jitter: 5 ms / 20 ms = 0.25;

[0162] Packet loss rate: 0.1% / 1% = 0.1;

[0163] Bandwidth requirement: 10 Mbps / 100 Mbps = 0.1;

[0164] Latency requirement: 50 ms / 100 ms = 0.5;

[0165] Packet loss rate tolerance: 0.1% / 1% = 0.1;

[0166] Computing power: 2.5 GHz / 3 GHz = 0.833;

[0167] Storage capacity: 64 GB / 128 GB = 0.5;

[0168] Power consumption: 2.5 W / 5 W = 0.5;

[0169] Channel interference: Low = 0.1;

[0170] Network coverage: 85% = 0.85;

[0171] Spectrum resource: 25 MHz / 100 MHz = 0.25.

[0172] Step 3: Construct time series - For the dynamic network environment data, assume that data at 5 time points are collected (once every 10 seconds), then the normalized time series is formed as follows:

[0173] Time window 1 (t = 0 seconds to 10 seconds):

[0174] Bandwidth: 20 Mbps → 0.2

[0175] Latency: 50 ms → 0.5

[0176] Jitter: 5 ms → 0.25

[0177] Packet loss rate: 0.1% → 0.1

[0178] Time window 2 (t = 10 seconds to 20 seconds):

[0179] Bandwidth: 18 Mbps → 0.18

[0180] Time delay: 52 ms → 0.52

[0181] Jitter: 6 ms → 0.3

[0182] Packet loss rate: 0.12% → 0.12

[0183] Time window 3 (t = 20 seconds to 30 seconds):

[0184] Bandwidth: 19 Mbps → 0.19

[0185] Time delay: 55 ms → 0.55

[0186] Jitter: 4 ms → 0.2

[0187] Packet loss rate: 0.09% → 0.09

[0188] Time window 4 (t = 30 seconds to 40 seconds):

[0189] Bandwidth: 21 Mbps → 0.21

[0190] Time delay: 48 ms → 0.48

[0191] Jitter: 5 ms → 0.25

[0192] Packet loss rate: 0.1% → 0.1

[0193] Time window 5 (t = 40 seconds to 50 seconds):

[0194] Bandwidth: 20 Mbps → 0.2

[0195] Time delay: 50 ms → 0.5

[0196] Jitter: 5 ms → 0.25

[0197] Packet loss rate: 0.1% → 0.1

[0198] Use the normalized data of these five time windows as input to generate time series data for subsequent model training and simulation.

[0199] By designing a deep learning model, dynamically adjust the weights of multi-dimensional metrics to generate an adaptive scoring dynamic weight matrix. It includes two parts: model architecture design and model training and optimization.

[0200] (1) Model architecture design, including three parts: input layer, hidden layer, and output layer:

[0201] Input layer: contains the following four vectors:

[0202] Service demand vector V b= [0.1, 0.5, 0.1], where the vector values are the service bandwidth requirement, latency requirement, and packet loss rate tolerance respectively;

[0203] Network performance vector N t = [0.2, 0.5, 0.25, 0.1], where the vector values are bandwidth, latency, jitter, and packet loss rate respectively;

[0204] Terminal capability vector T c = [0.833, 0.5, 0.5], where the vector values are computing power, storage capacity, and power consumption respectively;

[0205] Environment adaptability vector E = [0.1, 0.85, 0.25], where the vector values are channel interference, network coverage, and spectrum resources respectively;

[0206] Hidden layer: Assume a 3-layer fully connected network is adopted, with 5 nodes in each layer, and the ReLU activation function is used. After training, the model learns an appropriate dynamic weight matrix.

[0207] Output layer: Generate the dynamic weight matrix W DL = {w 1 , w 2 , …, w n}, where n represents the total number of data, and w n represents the weight of the nth data. Assume the following dynamic weights are obtained after the training process: W DL = [0.3, 0.2, 0.25, 0.1, 0.15].

[0208] (2) In model training and optimization, the data label uses the historical user experience score S user as the supervised learning label. The model can be trained with available historical data, and the model parameters are optimized through the following loss function:

[0209] Loss function design:

[0210]

[0211] where L represents the loss function value; m represents the total amount of training data; i = 1…m; represents the actual user experience score of the ith data; represents the adaptability prediction score of the ith data;

[0212] Training process: Use a large amount of labeled data for batch gradient descent training, and preferably use the Adam optimizer. Assume that the model has been trained with a large amount of historical data and a set of trained weights has been obtained. When new real-time data (such as network data collected every 10 seconds) is input, the model will generate an adaptive dynamic weight matrix according to the new input, adjust the weights of different dimensions, so that the evaluation results can accurately reflect the changes in network and business requirements.

[0213] Through the above training process, the deep adaptability evaluation network can construct an adaptability evaluation model. This model realizes the adaptability prediction score of multi-dimensional communication service data by dynamically generating a dynamic weight matrix corresponding to the input data. To further improve the comprehensiveness and accuracy of the adaptability evaluation, based on the adaptability evaluation model, according to the hierarchical characteristics of the communication protocol stack, the multi-dimensional communication service data can be hierarchically divided to more deeply explore the correlation between data at different levels. When analyzing the correlation, the graph neural network (GNN) can be used to analyze the correlation between layers of the communication protocol stack to make up for the deficiency that traditional models cannot quantify cross-layer relationships.

[0214] (1) Construct a multi-layer graph network model. The nodes can include:

[0215] Physical layer: Signal strength (RSSI) = 0.9;

[0216] Transport layer: Round Trip Time (RTT) = 50ms;

[0217] Application layer: Delay sensitivity = 0.4.

[0218] The edges include:

[0219] Physical layer → Transport layer (relationship between signal strength RSSI and RTT);

[0220] Transport layer → Application layer (relationship between RTT and delay sensitivity).

[0221] (2) Graph neural network (GNN) analysis process:

[0222] Assume the neighbor node relationship is:

[0223] N(Physical layer) = [Transport layer]

[0224] N(Physical layer) = [Transport layer, Application layer]

[0225] Use the graph neural network propagation mechanism to update the node features of each layer. Assume that after 5 propagation iterations, the model generates a cross-layer correlation score S cross = 0.75.

[0226] Based on the simulation algorithm, generate the time series of the dynamic network environment. Combine the results of deep learning weights and cross-layer correlation analysis, dynamically calculate the adaptability score, and generate optimization strategy suggestions, including three parts: dynamic network simulation, comprehensive adaptability score calculation, and optimization strategy generation. Specifically:

[0227] (I) Dynamic network simulation

[0228] Use the Markov chain model to simulate the network state transition. Assume that the current state is "lightly congested", and use the following transition probability matrix:

[0229]

[0230] Generate the future state sequence according to the current state of "lightly congested" to obtain an adaptability score S t = 0.65.

[0231] (II) Comprehensive adaptability score calculation

[0232] Combine the dynamic score and the cross-layer score to calculate the comprehensive adaptability score:

[0233] S final = α·S t + β·S cross ;

[0234] Among them, S t represents the adaptability score calculated based on the dynamic weight matrix; S cross represents the correlation score generated by cross-layer analysis; α and β are weight coefficients, which can be optimized through experiments. Specifically, α represents the network adaptability weight, and β represents the cross-layer correlation weight.

[0235] Assume that α = 0.6 and β = 0.4, then the comprehensive adaptability score S final can be expressed as follows:

[0236] S final = 0.6·0.65 + 0.4·0.75 = 0.69

[0237] (III) Optimization strategy generation

[0238] According to the final comprehensive adaptability score S final = 0.69, the following optimized strategies can be generated:

[0239] 1. Network resource adjustment: Increase bandwidth allocation, prioritize ensuring the bandwidth requirements of high-definition video streams, and reduce latency;

[0240] 2. Terminal optimization: It is recommended to upgrade the device hardware, especially to improve the computing power and storage capacity;

[0241] 3. Protocol parameter adjustment: Optimize the congestion control algorithm in the transport layer to reduce the packet loss rate and latency;

[0242] 4. Environment adaptation improvement: Deploy more base stations or increase spectrum resources to improve network coverage and channel quality.

[0243] The above specific embodiments can illustrate a multi-dimensional communication service adaptability evaluation method proposed by the present invention, which demonstrates the comprehensiveness, dynamics, and practicality of multi-dimensional communication service adaptability evaluation in a dynamic network environment. By real-time collecting network performance parameters, service demand parameters, terminal capability parameters, and environment adaptability information, and performing normalization processing and time series construction on them, the characteristics of the dynamic network environment can be captured, generating high-quality input data for subsequent evaluation. In the deep learning model, a deep adaptability evaluation network including dynamic weight matrix generation is designed, which can adjust the weights in real time according to the dynamic changes of different dimension parameters, and perform supervised learning in combination with historical user experience scores, so as to achieve accurate network adaptability scores. At the same time, aiming at the hierarchical characteristics of the communication protocol stack, the correlation between cross-layer data is analyzed through a graph neural network, effectively quantifying the complex interaction relationship between the physical layer, transport layer, and application layer, making up for the deficiency that traditional methods cannot capture cross-layer relationships. In addition, by combining dynamic weight scores and cross-layer correlation scores, a comprehensive adaptability score is calculated, providing a scientific quantitative index for network adaptability, and further generating optimization strategies. Using the Markov chain model for dynamic network simulation, combined with the comprehensive adaptability score, specific optimization strategies can be proposed in the simulation, such as adjusting network resource allocation, optimizing terminal performance, improving protocol parameters, and enhancing environment adaptability, so as to further ensure the smoothness and stability of high-demand services such as high-definition video stream playback in a complex network environment. The overall embodiments show that the present invention not only improves the real-time and accuracy of dynamic evaluation, but also can guide the actual network resource and service optimization through the comprehensive evaluation results, with extremely strong practical value and technical advantages, providing a comprehensive and scientific solution for communication service adaptability evaluation and optimization in a dynamic network environment.

[0244] Embodiment 3:

[0245] Based on the same inventive concept, the present invention also provides a multi-dimensional communication service adaptability evaluation system, the structural composition schematic diagram of which is as Figure 2 shown, including:

[0246] A network adaptability evaluation module, configured to perform network adaptability evaluation on multi-dimensional communication service data based on the collected multi-dimensional communication service data by using a pre-constructed adaptability evaluation model, and obtain corresponding network adaptability scores;

[0247] A hierarchical division module, which is used to divide multi-dimensional communication service data into different levels based on a preset communication protocol stack level, so as to obtain network data corresponding to different levels of the multi-dimensional communication service data;

[0248] A cross-layer correlation evaluation module, which is used to analyze the correlation of network data at different levels by using a graph neural network to obtain the cross-layer correlation score of the multi-dimensional communication service data;

[0249] A comprehensive adaptability evaluation module, which is used to obtain the comprehensive adaptability score of the multi-dimensional communication service data according to the network adaptability score and the cross-layer correlation score;

[0250] Among them, the adaptability evaluation model is constructed based on the dynamic weight matrix output by the deep adaptability evaluation network.

[0251] Exemplarily, the above-mentioned multi-dimensional communication service data may include one or more of the following: network performance data, service requirement data, terminal characteristic data, and environmental adaptability data;

[0252] The network performance data includes one or more of the following: bandwidth data, delay data, jitter data, and packet loss data;

[0253] The terminal characteristic data includes one or more of the following: computing power, storage capacity, and power consumption performance;

[0254] The environmental adaptability data includes one or more of the following: channel interference data, network coverage data, and spectrum resource data.

[0255] In one implementation, the above-mentioned network adaptability evaluation module may include:

[0256] A timing arrangement sub-module, which is used to arrange the collected multi-dimensional communication service data in time sequence to generate communication sequence data;

[0257] A dynamic adaptability evaluation sub-module, which is used to evaluate the adaptability of the communication sequence data according to the communication sequence data by using a pre-constructed adaptability evaluation model to obtain the network adaptability score of the multi-dimensional communication service data.

[0258] In one implementation, the above-mentioned evaluation system may further include: a model construction module, including:

[0259] An input setting sub-module, which is used to use historical communication sequence data as the input data of the training data;

[0260] An output setting sub-module, which is used to use the network adaptability score corresponding to the historical communication sequence data as the output data of the training data;

[0261] A model training sub-module, which is used to train the depth adaptability evaluation network based on the input data and output data of the training data, and obtain an adaptability evaluation model.

[0262] In this implementation manner, the above-mentioned model training sub-module may include:

[0263] A data input unit, which is used to input the input data of the training data into the depth adaptability evaluation network and output a dynamic weight matrix corresponding to the input data;

[0264] A scoring prediction unit, which is used to generate an adaptability prediction score according to the input data and the dynamic weight matrix corresponding to the input data;

[0265] An error calculation unit, which is used to calculate the loss function value corresponding to the adaptability prediction score and the output data by using the mean square error according to the adaptability prediction score and the output data;

[0266] A parameter optimization unit, which is used to calculate the gradient data of the depth adaptability network by using the batch gradient descent method according to the loss function value, and optimize the parameters of the depth adaptability network by using the Adam optimizer according to the gradient data until the depth adaptability evaluation network converges, so as to obtain an adaptability evaluation model.

[0267] Exemplarily, the above-mentioned communication protocol stack levels may at least include: a physical layer, a transport layer, and an application layer.

[0268] In one implementation manner, the above-mentioned cross-layer correlation evaluation module may include:

[0269] A performance index calculation sub-module, which is used to calculate the performance index values of each layer according to the network data at different layers and the preset performance indexes of each layer;

[0270] A relationship graph construction sub-module, which is used to construct a network relationship graph with the performance index values of each layer as the nodes corresponding to each layer and the association relationships between the performance index values of each layer as the edges;

[0271] A correlation analysis sub-module, which is used to perform correlation analysis by using a graph neural network according to the network relationship graph to obtain a cross-layer correlation score.

[0272] In one implementation manner, the above-mentioned correlation analysis sub-module may include:

[0273] A feature determination unit, which is used to obtain the feature information of each node in the network relationship graph according to the network relationship graph;

[0274] A feature update unit, which is used to update the feature information of each node by using a predefined propagation mechanism to obtain the feature update information of each node;

[0275] An information aggregation unit, configured to aggregate the feature update information of each node to obtain the global embedding information of the network relationship graph;

[0276] An overall feature association unit, configured to obtain the overall association information between network data at different levels according to the global embedding information;

[0277] A cross-layer association analysis unit, configured to generate a cross-layer association score according to the overall association information between network data at different levels.

[0278] Exemplarily, the expression corresponding to the above comprehensive adaptability score can be as follows:

[0279] S final = α·S t + β·S cross ;

[0280] Wherein, S final represents the comprehensive adaptability score; S t represents the network adaptability score; α represents the network adaptability weight; S cross represents the cross-layer association score; β represents the cross-layer association weight.

[0281] Embodiment 4:

[0282] As Figure 3 shown, the present invention further provides an electronic device, which may be a computer device, a single-chip microcomputer device, a smart mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, the processor, and the transceiver component are connected through a bus; the memory can be used to store an execution program, and the exemplary execution program may include instructions; the processor is used to execute the instructions stored in the memory. The memory can also be used to store data, and the data can be called and / or modified when the instructions are executed.

[0283] The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of a multi-dimensional communication service adaptability evaluation method in the above embodiments.

[0284] Embodiment 5:

[0285] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device-readable storage medium (Memory). The electronic device-readable storage medium is a memory device in the electronic device, used to store programs and data. It can be understood that the storage medium here can include both the built-in storage medium in the electronic device, and of course can also include the extended storage medium supported by the electronic device. The storage medium provides a storage space, and this storage space stores the operating system of the terminal. And, in this storage space, one or more instructions suitable for being loaded and executed by the processor are also stored. These instructions can be one or more execution programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory, or a non-volatile memory, such as at least one disk memory. By the processor loading and executing one or more instructions stored in the storage medium, the steps of a multi-dimensional communication service adaptability evaluation method in the above embodiments can be implemented.

[0286] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0287] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices produce means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0288] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0289] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0290] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit the scope of its protection. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that after reading the present invention, various changes, modifications, or equivalent replacements can still be made to the specific implementation manners of the application. However, these changes, modifications, or equivalent replacements are all within the scope of the protection of the claims pending for the application.

Claims

1. A multi-dimensional communication service adaptability evaluation method, characterized in that: include: Based on the collected multi-dimensional communication service data, a pre-built adaptability evaluation model is used to perform a network adaptability evaluation on the multi-dimensional communication service data to obtain a corresponding network adaptability score; Based on a preset communication protocol stack level, the multi-dimensional communication service data is divided into levels to obtain network data at different levels corresponding to the multi-dimensional communication service data; Using a graph neural network to perform correlation analysis on the network data at different levels to obtain a cross-layer correlation score for the multi-dimensional communication service data; Obtaining a comprehensive adaptability score of the multi-dimensional communication service data according to the network adaptability score and the cross-layer relevance score; Among them, the adaptability evaluation model is constructed based on the dynamic weight matrix output by the deep adaptability evaluation network.

2. The method according to claim 1, characterized in that The method of performing network adaptability evaluation on the multi-dimensional communication service data based on the collected multi-dimensional communication service data using a pre-built adaptability evaluation model to obtain a corresponding network adaptability score includes: Arrange the collected multi-dimensional communication service data in time sequence to generate communication sequence data; According to the communication sequence data, adaptability evaluation is performed on the communication sequence data using a pre-constructed adaptability evaluation model to obtain a network adaptability score for the multi-dimensional communication service data.

3. The method according to claim 1 or 2, characterized in that The adaptability evaluation model includes the following construction process: Using historical communication sequence data as input data for training data; Using the network adaptability score corresponding to the historical communication sequence data as output data of training data; Based on the input data and output data of the training data, the deep adaptability evaluation network is trained to obtain a adaptability evaluation model.

4. The method according to claim 3, characterized in that The input data and output data of the training data are used to train the deep adaptability evaluation network to obtain the adaptability evaluation model, including: Inputting the input data of the training data into a deep adaptability evaluation network, and outputting a dynamic weight matrix corresponding to the input data; generating a fitness prediction score according to the input data and a dynamic weight matrix corresponding to the input data; According to the adaptability prediction score and the output data, using mean square error to calculate the loss function value corresponding to the adaptability prediction score and the output data; According to the loss function value, the batch gradient descent method is used to calculate the gradient data of the deep adaptive network, and according to the gradient data, the Adam optimizer is used to optimize the parameters of the deep adaptive network until the deep adaptive evaluation network converges to obtain the adaptive evaluation model.

5. The method according to claim 1, characterized in that The communication protocol stack layers include at least: a physical layer, a transport layer and an application layer.

6. The method according to claim 1, characterized in that The using of the graph neural network to perform correlation analysis on the network data at different levels to obtain a cross-layer correlation score of the multi-dimensional communication service data includes: Calculate the performance index value of each layer according to the network data at the different layers and the preset performance index of each layer; The performance indicator values ​​of each level are used as nodes of the corresponding level, and the correlation between the performance indicator values ​​of each level are used as edges to construct a network relationship graph; According to the network relationship diagram, a graph neural network is used to perform correlation analysis to obtain a cross-layer correlation score.

7. The method according to claim 6, characterized in that The method of performing correlation analysis using a graph neural network according to the network relationship diagram to obtain a cross-layer correlation score includes: According to the network relationship graph, characteristic information of each node in the network relationship graph is obtained; Using a predefined propagation mechanism to update the feature information of each node, to obtain updated feature information of each node; Aggregating the feature update information of each node to obtain global embedding information of the network relationship graph; Obtaining overall association information between network data at different levels according to the global embedding information; A cross-layer correlation score is generated according to the overall correlation information between the network data at different levels.

8. The method according to claim 1, characterized in that The multi-dimensional communication service data includes one or more of the following: network performance data, service demand data, terminal characteristic data and environmental adaptability data; The network performance data includes one or more of the following: bandwidth data, delay data, jitter data and packet loss data; The terminal characteristic data includes one or more of the following: computing capability, storage capability and power consumption performance; The environmental adaptability data includes one or more of the following: channel interference data, network coverage data and spectrum resource data.

9. The method according to claim 1, characterized in that The expression corresponding to the comprehensive suitability score is as follows: S final =α·S t +β·S cross ; Among them, S final represents the comprehensive suitability score; S t represents the network adaptability score; α represents the network adaptability weight; S cross represents the cross-layer correlation score; β represents the cross-layer correlation weight.

10. A multi-dimensional communication service adaptability evaluation system, characterized in that: include: A network adaptability evaluation module is used to evaluate the network adaptability of the multi-dimensional communication service data collected by using a pre-built adaptability evaluation model to obtain a corresponding network adaptability score; A hierarchical division module, used to hierarchically divide the multi-dimensional communication service data based on a preset communication protocol stack hierarchy, and obtain network data at different hierarchies corresponding to the multi-dimensional communication service data; A cross-layer relevance evaluation module, used to perform relevance analysis on the network data at different levels using a graph neural network to obtain a cross-layer relevance score for the multi-dimensional communication service data; A comprehensive adaptability evaluation module, used to obtain a comprehensive adaptability score of the multi-dimensional communication service data according to the network adaptability score and the cross-layer relevance score; Among them, the adaptability evaluation model is constructed based on the dynamic weight matrix output by the deep adaptability evaluation network.

11. The system according to claim 10, characterized in that The network adaptability evaluation module comprises: The time sequence arrangement submodule is used to arrange the collected multi-dimensional communication service data in time sequence to generate communication sequence data; The dynamic adaptability evaluation submodule is used to perform adaptability evaluation on the communication sequence data according to the communication sequence data using a pre-built adaptability evaluation model to obtain a network adaptability score for the multi-dimensional communication service data.

12. The system according to claim 10 or 11, characterized in that The evaluation system further includes: a model building module, including: An input setting submodule, used to use historical communication sequence data as input data of training data; An output setting submodule, used to use the network adaptability score corresponding to the historical communication sequence data as output data of the training data; The model training submodule is used to train the deep adaptability evaluation network based on the input data and output data of the training data to obtain the adaptability evaluation model.

13. The system of claim 12, wherein: The model training submodule includes: A data input unit, used to input the input data of the training data into the deep adaptability evaluation network, and output a dynamic weight matrix corresponding to the input data; A score prediction unit, used to generate a suitability prediction score according to the input data and a dynamic weight matrix corresponding to the input data; an error calculation unit, configured to calculate, according to the adaptability prediction score and the output data, a loss function value corresponding to the adaptability prediction score and the output data by using a mean square error; A parameter optimization unit is used to calculate the gradient data of the deep adaptive network using the batch gradient descent method according to the loss function value, and optimize the parameters of the deep adaptive network using the Adam optimizer according to the gradient data until the deep adaptive evaluation network converges to obtain an adaptive evaluation model.

14. The system according to claim 10, characterized in that The communication protocol stack layers include at least: a physical layer, a transport layer and an application layer.

15. The system of claim 10, wherein: The cross-layer relevance evaluation module comprises: The performance indicator calculation submodule is used to calculate the performance indicator value of each layer according to the network data at the different layers and the preset performance indicators of each layer; A relationship graph construction submodule, used to construct a network relationship graph using the performance indicator values ​​of each level as nodes of the corresponding level and the correlation relationships between the performance indicator values ​​of each level as edges; The correlation analysis submodule is used to perform correlation analysis based on the network relationship diagram using a graph neural network to obtain a cross-layer correlation score.

16. The system of claim 15, wherein: The relevance analysis submodule includes: A feature determination unit, used to obtain feature information of each node in the network relationship graph according to the network relationship graph; A feature updating unit, used to update the feature information of each node using a predefined propagation mechanism to obtain feature update information of each node; An information aggregation unit, used to aggregate the feature update information of each node to obtain global embedding information of the network relationship graph; An overall feature association unit, used to obtain overall association information between network data at different levels according to the global embedding information; The cross-layer correlation analysis unit is used to generate a cross-layer correlation score according to the overall correlation information between the network data at different levels.

17. The system of claim 10, wherein: The multi-dimensional communication service data includes one or more of the following: network performance data, service demand data, terminal characteristic data and environmental adaptability data; The network performance data includes one or more of the following: bandwidth data, delay data, jitter data and packet loss data; The terminal characteristic data includes one or more of the following: computing capability, storage capability and power consumption performance; The environmental adaptability data includes one or more of the following: channel interference data, network coverage data and spectrum resource data.

18. The system of claim 10, wherein: The expression corresponding to the comprehensive suitability score is as follows: S final =α·S t +β·S cross ; Among them, S final represents the comprehensive adaptability score; St represents the network adaptability score; α represents the network adaptability weight; S cross represents the cross-layer correlation score; β represents the cross-layer correlation weight.

19. An electronic device, characterized in that: include: at least one processor and memory; The memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, a multi-dimensional communication service suitability evaluation method as claimed in any one of claims 1 to 9 is implemented.

20. A computing device readable storage medium, characterized in that: An execution program is stored thereon, and when the execution program is executed, a multi-dimensional communication service adaptability evaluation method as described in any one of claims 1 to 9 is implemented.

Images