A predictability judgment method for electromagnetic communication data

Abstract

This invention belongs to the field of wireless communication technology, and particularly relates to a method for predictability assessment of electromagnetic communication data. It includes the following steps: Step 1: Collect electromagnetic communication data from multiple receivers and transmitters over a period of time, and calculate the intrinsic predictability of the parameter time series at each point based on true entropy, sample entropy, and multi-scale entropy; Step 2: Select the time scale with the minimum entropy value, preprocess the electromagnetic communication data time series, and construct a spatiotemporal similarity graph to transform the electromagnetic communication data into graph data; Step 3: Locally construct a spatiotemporal graph convolutional network model with node-level attention, perform local prediction, and calculate the prediction accuracy; Step 4: Perform predictability assessment on the electromagnetic communication data. This invention proposes for the first time a comprehensive assessment framework that can both calculate intrinsic predictability and guide the establishment of prediction models, enabling a comprehensive evaluation of the predictability of electromagnetic communication data.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology, and in particular relates to a method for predicting the predictability of electromagnetic communication data in this field. Background Technology

[0002] Electromagnetic communication is one of the most widely used communication methods in modern society, encompassing various forms of electromagnetic waves such as radio waves, microwaves, and infrared rays. These waves can propagate in air, vacuum, and even certain solid and liquid media. In electromagnetic communication, raw data is first encoded into a format suitable for electromagnetic transmission. Then, through modulation techniques, this data is attached to a carrier signal of the electromagnetic wave for propagation. When the signal reaches the target location, the receiver captures these electromagnetic waves and demodulates them back to the original data signal for end-user use. Electromagnetic communication exhibits several significant advantages: firstly, it can cover a wide distance, enabling long-distance communication without a physical connection; secondly, it supports the transmission of various data formats, including voice, text, images, and video, meeting diverse communication needs; and thirdly, its data transmission speed is close to the speed of light, enabling near-real-time information sharing and greatly improving communication efficiency.

[0003] Electromagnetic communication data can essentially be considered a spatiotemporal sequence within a certain time period. One of its significant characteristics is temporal correlation, which shows the trend of data change over time. Spatiotemporal data analysis has become a research hotspot in the field of data mining. For electromagnetic communication data, it is of great significance to explore the potential relationships and make predictions. At present, a variety of methods have been applied to time series prediction, including: (1) Traditional statistical learning methods: These methods are mainly based on historical data and empirical rules, and analyze and predict the changes in future electromagnetic communication parameters by establishing mathematical models. Although traditional statistical methods have certain application value in the field of short-term prediction, their prediction accuracy and reliability are limited due to the complexity of electromagnetic communication. (2) Modern machine learning methods: Machine learning technology can utilize the spatiotemporal correlation of electromagnetic signals, especially the graph convolutional network method, which represents spatiotemporal data as a graph structure and learns spatiotemporal features through graph neural networks. This method can more effectively capture the correlation between spatiotemporal data, thereby improving the accuracy of prediction.

[0004] In conclusion, while traditional statistical learning methods have played a role in electromagnetic communication data prediction, modern machine learning methods, especially graph convolutional networks, demonstrate higher prediction accuracy and application potential due to their ability to deeply capture the complex relationships between spatiotemporal data.

[0005] With significant progress in electromagnetic communication data prediction algorithms, a crucial question arises: what is the theoretical upper limit of prediction accuracy for a given electromagnetic communication dataset? This question relates to the intrinsic predictability of the dataset, that is, the highest prediction accuracy that any model can theoretically achieve. Analyzing the intrinsic predictability of electromagnetic communication data is of great significance for providing model selection and practical application guidance for datasets in different scenarios. However, current analyses of predictability are mostly limited to studying spatiotemporal sequences from the perspective of intrinsic predictability calculation. Therefore, constructing a comprehensive judgment framework that can both calculate intrinsic predictability and guide the establishment of prediction models presents significant challenges and research value. Summary of the Invention

[0006] The main objective of this invention is to provide a method for predictability assessment of electromagnetic communication data, thereby solving the problems of data mining and prediction of electromagnetic communication data.

[0007] The present invention adopts the following technical solution:

[0008] A method for predictability assessment of electromagnetic communication data includes the following steps:

[0009] Step 1: Collect electromagnetic communication data from multiple receivers and transmitters over a period of time, and perform intrinsic predictability calculations on the parameter time series at each point based on true entropy, sample entropy, and multi-scale entropy;

[0010] Step 2: Based on the results of the inherent predictability calculation in Step 1, select the time scale with the minimum entropy value, preprocess the electromagnetic communication data time series at that scale, and construct a spatiotemporal similarity map to transform the electromagnetic communication data into graph data.

[0011] Step 3: Construct a spatiotemporal graph convolutional network model with node-level attention locally to train the data preprocessed in Step 2, perform local predictions, and calculate the prediction accuracy;

[0012] Step 4: Combine the inherent predictability calculation results from Step 1 with the model prediction accuracy from Step 3 to make a predictability judgment on the electromagnetic communication data.

[0013] Furthermore, step 1 specifically includes:

[0014] To determine the geographical locations of receivers and transmitters in an electromagnetic communication service scenario, multiple receivers are treated as multiple nodes, and graph data is constructed. We represent the electromagnetic communication network as graph G = (V, E), where V represents a set of N nodes within the electromagnetic communication network, and E represents a set of edges indicating connectivity between nodes. The adjacency matrix of graph G is used... It means that A i,jThe state of electromagnetic communication data at any time step t represents the connectivity between node i and node j, thus it can be regarded as a graph signal. Here, C represents the type of electromagnetic communication data parameters. Therefore, for each node n and a certain parameter c on it, there exists a time series x. For each node and each parameter, we perform an intrinsic predictability calculation based on true entropy, sample entropy, and multi-scale entropy.

[0015] Furthermore, step 2 specifically includes:

[0016] Based on the results of the inherent predictability calculation in step 1, the time scale with the minimum entropy value is selected, and the electromagnetic communication data time series is preprocessed at this scale. A spatiotemporal similarity graph is then constructed to transform the electromagnetic communication data into graph data.

[0017] First, compare the calculated results of the true entropy, sample entropy, and multi-scale entropy from the time series, and select the entropy value that is the lowest. If the entropy value belongs to the true entropy and sample entropy, then no scaling processing is performed on the time series; if the entropy value belongs to the multi-scale entropy, then the time scale corresponding to the lowest entropy value is selected, and the original time series is averaged on the scale.

[0018] After processing the original time series, a spatiotemporal similarity map is constructed using three aspects: state time series correlation, Wasserstein distance, and KL divergence.

[0019] Furthermore, step 3 specifically involves:

[0020] A spatiotemporal graph convolutional network model with node-level attention is built locally to train the data after preprocessing in step 2, perform local predictions, and calculate the prediction accuracy.

[0021] This method introduces a spatiotemporal graph convolutional network model with node-level attention. First, to capture the influence of each node on its neighbors at the current and nearby time steps and to expand the spatiotemporal receptive field, we construct a fully connected local spatiotemporal graph, where a given node is interconnected with all nodes at other time steps. However, this graph includes not only the edges of the original graph at each time step but also the edges connecting nodes in the graphs across different time steps. Therefore, the weights of these newly introduced edges must be determined.

[0022] Secondly, we implement a graph convolutional network with node-level attention to more effectively capture local spatiotemporal correlations. It consists of two main operations: a series of graph convolution operations and a series of (flipped) node-level attention operations. By executing these operations in parallel, we combine the attention mechanism with graph convolution, achieving mutual reinforcement. Through the attention mechanism, the receptive field of graph convolution can be extended to more relevant nodes, which may be far from the central node. On the other hand, graph convolution helps the attention mechanism capture complex graph topological information and effectively aggregates more relevant nodes from the neighborhood.

[0023] Finally, we used data from 12 historical time steps as input to combine with a spatiotemporal graph convolutional network model with node-level attention to predict the data for the next 12 time steps. We then evaluated the model's prediction accuracy by comparing its predictions with the actual values.

[0024] Furthermore, step 4 specifically involves:

[0025] By combining the intrinsic predictability calculation results from Step 1 and the model prediction accuracy from Step 3, the predictability of electromagnetic communication data is assessed. In Step 1, we revealed the regularity and disorder of a time series by calculating its intrinsic predictability value. This value not only reflects the highest prediction accuracy that any model can theoretically achieve on this specific dataset, but also provides valuable guidance for model selection and optimization for different scenarios. The model prediction accuracy shown in Step 3 provides researchers with an important reference point, indicating whether there is room for further optimization of the adopted method and whether the current model can be simplified. Therefore, by combining the results of Steps 1 and 3, we can comprehensively evaluate the predictability of electromagnetic communication data.

[0026] The beneficial effects of this invention are:

[0027] The method disclosed in this invention first collects electromagnetic communication data, treating the receiver as a node in a graph structure. It then introduces the calculation of inherent predictability into the field of electromagnetic communication data prediction, performing inherent predictability calculations based on true entropy, sample entropy, and multi-scale entropy on the time series of parameters at each point. This step is crucial for predictability assessment, and the results of the inherent predictability calculation reveal the intrinsic patterns in the electromagnetic communication data.

[0028] The method disclosed in this invention, based on inherent predictability calculations, selects a time scale with the minimum entropy value to preprocess the electromagnetic communication data time series at that scale, and proposes for the first time a method for constructing a spatiotemporal similarity graph to transform the electromagnetic communication data into graph data. Preprocessing the data significantly improves the prediction accuracy of the original data and fully represents the inherent patterns of the time series, laying the foundation for subsequent model training.

[0029] The method disclosed in this invention constructs a fully connected local spatiotemporal graph, thereby increasing the amount of information during prediction and enabling the model to learn richer feature representations. On this basis, a spatiotemporal graph convolutional network model with node-level attention is introduced for the first time, combining the attention mechanism with graph convolution from a completely new perspective, thereby achieving a mutually reinforcing effect.

[0030] The method disclosed in this invention combines the calculation results of inherent predictability with the prediction accuracy of the model to make an overall predictability judgment on electromagnetic communication data. This step is a summary of the preliminary work of this method, and for the first time in this field, a comprehensive judgment framework that can both calculate inherent predictability and guide the establishment of prediction models is proposed, thereby enabling a comprehensive evaluation of the predictability of electromagnetic communication data. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating the present invention;

[0032] Figure 2 This is a specific process for constructing a spatiotemporal similarity graph in one embodiment of the present invention;

[0033] Figure 3 This is a specific framework of a spatiotemporal graph convolutional network model with node-level attention involved in one embodiment of the present invention;

[0034] Figure 4 This is the predictability assessment result of the parameters involved in one embodiment of the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0036] Example 1: This embodiment discloses a method for predictability assessment of electromagnetic communication data, including the following steps: Figure 1 :

[0037] Step 1: Collect electromagnetic communication data from multiple receivers and transmitters over a period of time, and perform intrinsic predictability calculations on the parameter time series at each point based on true entropy, sample entropy, and multi-scale entropy;

[0038] Step 2: Based on the results of the inherent predictability calculation in Step 1, select the time scale with the minimum entropy value, preprocess the electromagnetic communication data time series at that scale, and construct a spatiotemporal similarity map to transform the electromagnetic communication data into graph data.

[0039] Step 3: Construct a spatiotemporal graph convolutional network model with node-level attention locally to train the data preprocessed in Step 2, perform local predictions, and calculate the prediction accuracy;

[0040] Step 4: Combine the inherent predictability calculation results from Step 1 with the model prediction accuracy from Step 3 to make a predictability judgment on the electromagnetic communication data.

[0041] Step 1 specifically involves:

[0042] To determine the geographical locations of receivers and transmitters in electromagnetic communication service scenarios, we treat multiple receivers as multiple nodes and construct graph data. We use VOACAP (a widely used ionospheric propagation prediction tool) to calculate electromagnetic communication data between transmitters and receivers. We place a signal transmitter in Shanghai and establish 20 signal receivers across China for transmission. From January to October 2023, we record three parameters from the receiver locations at 30-minute intervals: Maximum Available Frequency (MUF), Median Field Strength at Receiver (DBU), and Signal-to-Noise Ratio (SNR) at Receiver.

[0043] We represent an electromagnetic communication network as a graph G = (V, E), where V represents a set of N nodes within the network, and E represents a set of edges indicating connectivity between the nodes. The adjacency matrix of graph G is... It means that A i,j The state of electromagnetic communication data at any time step t represents the connectivity between node i and node j, thus it can be regarded as a graph signal. Where C represents the type of electromagnetic communication data parameter, and in this embodiment, C = 3 and N = 20. Therefore, for each node n and a certain parameter c on it, there exists a time series x. For each node and each parameter, we perform an intrinsic predictability calculation based on true entropy, sample entropy, and multi-scale entropy.

[0044] Entropy in time series analysis is a metric that measures the complexity and unpredictability of a time series. Based on the concept of information entropy, it quantifies the amount of information contained within a time series. In time series analysis, entropy helps us understand the degree of regularity in data, thus aiding in calculating the inherent predictability of the time series.

[0045] Given a time series x = (x1, ..., x2) t The formula for calculating the true entropy is:

[0046]

[0047] Here, Q represents a specific ordered subsequence that appears in x. p(Q) represents the proportion of the specific subsequence Q among all subsequences.

[0048] The calculation of sample entropy first requires defining relevant parameters m and r. For a time series x, the reconstructed m-dimensional vector x′={x(1),x(2),…,x(N-m+1)} is given, where x(i)=[x i ,x i+1 ,…,x i+m-1 For 1≤i≤N-m+1, define B m (r) is:

[0049]

[0050] in, The definition is: for i≠j, the proportion of x(j) with d[x(i),x(j)]≤r within x′. Here, d[x(i),x(j)] represents the distance between vectors x(i) and x(j), determined by the maximum difference between their corresponding elements.

[0051] Let k = m + 1, define A k (r):

[0052]

[0053] in, Definition and Consistent with the definition: For i ≠ j, the proportion of x(j) with d[x(i),x(j)] ≤ r within x′. The formula for calculating sample entropy is:

[0054]

[0055] The calculation of multiscale entropy requires the introduction of the time scale p. In short, selecting the time scale p means averaging the original time series every p steps to form a new series, and simultaneously calculating the sample entropy of this new series to obtain the multiscale entropy of the original series at time scale p. This allows us to obtain the set of multiscale entropies of the original series.

[0056]

[0057] in, This represents the sample entropy of the new sequence obtained after averaging the original sequence every p steps.

[0058] Intrinsic predictability IP is defined by the following formula:

[0059] S=-IPlog2(IP)-(1-IP)log2(1-IP)+(1-IP)log2(Nu-1)

[0060] Where S is the entropy of the time series (true entropy, sample entropy, or multi-scale entropy), different entropy values ​​correspond to different IP values, and N... u This represents the number of independent values ​​appearing in the time series. Based on the above formula, the intrinsic predictability (IP) is calculated using numerical methods (such as the squeeze theorem).

[0061] Step 2 specifically includes:

[0062] Based on the results of the inherent predictability calculation in step 1, the time scale with the minimum entropy value is selected. Preprocessing of the electromagnetic communication data time series at this scale is then performed, and a spatiotemporal similarity graph is constructed to transform the electromagnetic communication data into graph data. First, the calculated results of the true entropy, sample entropy, and multi-scale entropy from the time series are compared. The entropy with the minimum value is selected. If this entropy value belongs to the true entropy and sample entropy, a time scale with a value of 1 is selected, meaning no preprocessing is performed on the original time series. If this entropy value belongs to the multi-scale entropy, the time scale corresponding to the lowest entropy value is selected, and the original time series is averaged across the scales. Assuming the time scale corresponding to the lowest entropy value is s, an average value is calculated for the original electromagnetic signal time series every s steps to obtain a new electromagnetic time series, and the length of this new time series is 1 / s of the original series.

[0063] After processing the original time series, a spatiotemporal similarity map is constructed using three aspects: state-time series correlation, Wasserstein distance, and Kullback-Leibler divergence. Each receiving station possesses a state-time series. Where c∈[1,C], the connection strength between nodes can be quantified by measuring the correlation between these state time series. Given the state time series Y corresponding to node i and node j respectively i and Y j The correlation function between them can be defined as:

[0064]

[0065] In addition, normalized correlation values ​​can be calculated to standardize the measure of connectivity:

[0066]

[0067] For the c-th parameter, the state time series Represented as a set of vectors: Where D w T represents the total number of days in a time series record. d This represents the number of time steps recorded each day (assuming the dataset is recorded at half-hour intervals, T). d =48), Indicates the dth w The electromagnetic communication data recorded daily. The daily electromagnetic data capacity information is extracted using the following formula:

[0068]

[0069] Where ||·||² represents the Euclidean norm, thus transforming the vector sequence of receiving station i into a probability distribution. The Wasserstein distance between node i and node j is calculated as follows:

[0070]

[0071] in,

[0072]

[0073]

[0074]

[0075]

[0076] KL divergence is an asymmetric metric, therefore the KL distance between nodes is calculated as follows:

[0077]

[0078] in,

[0079]

[0080] Where P(X) and Q(X) represent two probability distributions. The spatial dependence between nodes can be more effectively evaluated through three dimensions, and these calculations are integrated to formulate a spatio-temporal similarity graph of electromagnetic communication data:

[0081]

[0082] Where,

[0083] A3 = Norm(A 3′ ), A 3′ [i, j] = d KL (i, j),

[0084] A2[i, j] = d WD (i, j)

[0085] Since the calculated value range of d KL (i, j) is not between 0 and 1, a normalization operation, i.e., Norm(·), needs to be used for matrix A 3′ . Figure 2 Shows the specific process of constructing a spatio-temporal similarity graph on the collected electromagnetic communication data set in this example.

[0086] The specific content of step 3 is as follows:

[0087] Locally construct a spatio-temporal graph convolutional network model with node-level attention, which is used to train the data preprocessed in step 2, perform local prediction, and calculate the prediction accuracy.

[0088] This method introduces a spatio-temporal graph convolutional network model with node-level attention. First, in order to capture the influence of each node on its neighbors at the current and nearby time steps and expand the spatio-temporal receptive field, we construct a fully connected local spatio-temporal graph, in which a given node is interconnected with all nodes at other time steps. However, this graph not only includes the edges of the original graph at each time step, but also includes the edges connecting the nodes of the graph between different time steps. Therefore, the weights of these newly introduced edges must be determined.

[0089] The state of electromagnetic communication parameters at any time step t can be represented as a graph signal matrix Correspondingly, the newly constructed graph has its graph signal matrix The weights of the new edges are defined by cosine similarity. Assume that i t′ and i t″ represent nodes at different time steps, where i, j ∈ [1, N], t′, t″ ∈ [1, 3], t′ < t″, and the weight of the edge connecting them is calculated as follows:

[0090]

[0091]

[0092] Secondly, we implement a graph convolutional network with node-level attention to more effectively capture local spatiotemporal correlations. It consists of two main operations: a series of graph convolution operations and a series of node-level attention / flip-mode attention operations. By executing these operations in parallel, we combine the attention mechanism with graph convolution, achieving mutual reinforcement. Through the attention mechanism, the receptive field of graph convolution can be extended to more relevant nodes, which may be far from the central node. On the other hand, graph convolution helps the attention mechanism capture complex graph topological information and effectively aggregate more relevant nodes from the neighborhood.

[0093] The graph convolution module operates as follows, and the mathematical operations can be formalized as follows:

[0094]

[0095] in, It is the input of the l-th graph convolutional layer. σ represents the activation function, W G and b G These are all trainable parameters for convolution operations.

[0096] Node-level attention operations are described below. The l-th node-level attention layer receives... As input, To compute the global attention to node i in the l-th attention layer, the representations of other nodes are stacked into a matrix. Attention calculation is as follows:

[0097]

[0098]

[0099]

[0100] Among them, W q W k W v It is the projection matrix. To enhance the model's expressiveness and reduce the risk of overfitting, the method is extended to incorporate multi-head attention:

[0101]

[0102] Where m represents the number of attention heads, Concat(·) is the concatenation operation, and Wm It is a projection matrix.

[0103] Another version of attention computation is the flip-pattern attention operation, which embeds each parameter independently as a variable label. To achieve this, the concatenation operation is performed by first aligning the representations in chronological order and then arranging them according to the node order. The l-th layer flips the node-level attention layer. As input, each Treated as a single tag, the remaining operations are consistent with node-level attention operations. The most time-consuming part of the different versions of the operation is the attention computation between tags; therefore, researchers should evaluate the computational time of different attention operations to select the appropriate version.

[0104] The aggregation operation is used to combine the outputs of each graph convolutional layer and the attention operation layer of the node-level attention / flip mode. It takes the element-wise maximum of the outputs of the graph convolutional layer and the attention operation layer of the node-level attention / flip mode, which can be mathematically expressed as:

[0105]

[0106]

[0107]

[0108]

[0109] To preserve the features derived from these two operations and eliminate a large amount of redundant information in the model, the Concat(·) operation and the Cropping(·) operation are used in sequence:

[0110]

[0111]

[0112] Continuing by deploying more than one set of operations, a spatiotemporal graph convolutional layer with node-level attention is defined, where the input of the l-th spatiotemporal graph convolutional layer is defined as... The following output can be generated by using a sliding window:

[0113]

[0114] Among them, M (l) This represents the output of the l-th spatiotemporal graph convolutional layer.

[0115] Finally, the model's prediction block transforms the output of the last spatiotemporal graph convolutional layer into the final output—the predicted values ​​for the last 12 time steps—through a fully connected layer. Then, we evaluate the model's prediction accuracy by comparing its predictions with the actual values.

[0116]

[0117]

[0118] Where Acc represents the final prediction accuracy, This represents the prediction model proposed in this method. This represents the actual value of the c-th parameter at the n-th node at the u-th time step. This represents the predicted value of the c-th parameter at the n-th node at the u-th time step.

[0119] Figure 3 This example demonstrates the specific architecture of a spatiotemporal graph convolutional network model with node-level attention, constructed to predict the evolution of electromagnetic communication data.

[0120] It should be noted that, in addition to the spatiotemporal graph convolutional network model, other neural networks can also be used for prediction.

[0121] Step 4 specifically refers to:

[0122] Combining the intrinsic predictability calculation results from Step 1 and the model prediction accuracy from Step 3, the predictability of electromagnetic communication data is assessed. In Step 1, we reveal the regularity and disorder of a time series by calculating its intrinsic predictability value. This value not only reflects the highest prediction accuracy that any model can theoretically achieve on this specific dataset, but also provides valuable guidance for model selection and optimization for different scenarios. The model prediction accuracy shown in Step 3 provides researchers with an important reference point, indicating whether there is room for further optimization of the adopted method and whether the current model can be simplified. We define the intrinsic predictability calculated in Step 1 as IP, and the model prediction accuracy in Step 3, i.e., the realized predictability of the data, as RP, where IP,RP∈(0,1).

[0123] We consider several scenarios. If IP < 0.3, we define the time series as unpredictable because in real-world business scenarios, if the predictability of a series is very low, it has virtually no practical application value.

[0124] If IP ≥ 0.3 and IP > RP, it can be concluded that the intrinsic predictability of this time series is greater than the prediction accuracy achievable by the proposed prediction model for this time series. Therefore, we can define this time series as a predictable time series, and the prediction model has room for further optimization. In practical applications, the model structure needs to be adjusted to obtain higher prediction accuracy.

[0125] If IP ≥ 0.3 and IP ≤ RP, it can be concluded that the intrinsic predictability of this time series is less than or equal to the prediction accuracy achievable by the proposed prediction model for that time series. Therefore, we can define this time series as predictable, and the proposed prediction model may be overly complex for this time series. In practical applications, we should evaluate the trade-off between computational resources and prediction accuracy, and typically choose a simpler, faster, and more efficient model to save time and resources.

[0126] In summary, by combining the results of steps 1 and 3, we are able to make a comprehensive assessment of the predictability of electromagnetic communication data.

[0127] like Figure 4 As shown, this example demonstrates a comprehensive predictability assessment on an electromagnetic communication dataset, calculating the dataset's inherent predictability and the model's prediction accuracy, including the maximum available frequency (MUF), receiver median field strength (DBU), and receiver signal-to-noise ratio (SNR). Figure 4 This demonstrates the predictability of different parameters on this dataset. Figure 4 Use tables as a presentation method for predictability assessment.

[0128] Furthermore, it should be noted that the specific embodiments described in this specification may be given different names, and the above description is merely illustrative of the structure of the present invention. Any minor or simple variations made based on the structure, features, and principles of the present invention are included within the scope of protection of the present invention. Those skilled in the art can make various modifications or additions to the described specific examples or adopt similar methods, as long as they do not deviate from the structure of the present invention or exceed the scope defined by the claims, all of which should fall within the scope of protection of the present invention.

Claims

1. A method for predictability assessment of electromagnetic communication data, characterized in that, Includes the following steps: Step 1: Collect electromagnetic communication data from multiple receivers and transmitters over a period of time, and perform intrinsic predictability calculations on the parameter time series at each point based on true entropy, sample entropy, and multi-scale entropy; Given time series The formula for calculating the true entropy is: in, represent A specific ordered time subsequence that has appeared in the text. Then it represents a specific subsequence The proportion of all subsequences; The calculation of sample entropy first requires defining relevant parameters. and For time series In other words, reconstruction dimensional vector ,in ,for ,definition for: in, The definition is: for In other words, of The quantity in The proportion within; among which, Representative vector and The distance is determined by the maximum difference between the corresponding elements; make ,definition : in, Definition and Consistent with the definition: for In other words, of The quantity in The proportion within; the formula for calculating sample entropy is: The calculation of multiscale entropy requires the introduction of a time scale parameter. In short, select a time scale This means that for each of the original time series The process involves calculating an average value to form a new sequence, and simultaneously calculating the sample entropy of this new sequence to obtain the time scale of the original sequence. The multi-scale entropy below can be used to obtain the set of multi-scale entropies of the original sequence: in, Represents each of the original sequences The sample entropy of the new sequence obtained after calculating an average value; Intrinsic predictability Solve using the following formula: in, The entropy of a time series is calculated; different entropy values ​​correspond to different... , This represents the number of independent values ​​appearing in the time series; based on this formula, the inherent predictability is calculated using numerical methods. ; Step 2: Based on the results of the inherent predictability calculation in Step 1, select the time scale with the minimum entropy value, preprocess the electromagnetic communication data time series at that scale, and construct a spatiotemporal similarity map to transform the electromagnetic communication data into graph data. Step 3: Construct a spatiotemporal graph convolutional network model with node-level attention locally to train the data preprocessed in Step 2, perform local predictions, and calculate the prediction accuracy; Step 4: Combine the inherent predictability calculation results from Step 1 with the model prediction accuracy from Step 3 to make a predictability judgment on the electromagnetic communication data. Specifically: Define the intrinsic predictability value calculated in step 1 as follows: The model prediction accuracy in step 3, i.e., the predictability of the data, is... ,in, ; like This time series is defined as unpredictable; like ,and For this time series, the intrinsic predictability value is greater than the prediction accuracy that the prediction model can achieve on this time series; this time series is defined as a predictable time series, and the prediction model has room for further optimization. like ,and For this time series, the intrinsic predictability value is less than or equal to the prediction accuracy that the prediction model can achieve on this time series; the time series is defined as a predictable time series, and for this time series, the proposed prediction model may be too complex. The trade-off between computational resources and prediction accuracy should be evaluated, and a simpler, faster and more efficient model should be selected to save time and resources.

2. The predictability determination method for electromagnetic communication data as described in claim 1, characterized in that, Step 1 specifically involves: Determine the geographical locations of receivers and transmitters in electromagnetic communication service scenarios, treat multiple receivers as multiple nodes, and construct graph data; represent the electromagnetic communication network as a graph. ,in Represents a group within an electromagnetic communication network A set of nodes, A graph represents a set of edges that indicate the connectivity between nodes. The adjacency matrix is ​​used express, Represents a node and nodes The connectivity of electromagnetic communication data at any time step The state below is considered a graph signal. ,in, It refers to the type of electromagnetic communication data parameters; for each node and a certain parameter on it In general, there exists a time series. For each node and each parameter, the inherent predictability is calculated based on the true entropy, sample entropy, and multi-scale entropy.

3. The predictability determination method for electromagnetic communication data as described in claim 1, characterized in that, Step 2 specifically involves: First, compare the calculated results of the true entropy, sample entropy, and multi-scale entropy from the time series, and select the entropy value that is the lowest. If the entropy value belongs to the true entropy and sample entropy, then no scaling processing is performed on the time series. If the entropy value belongs to multi-scale entropy, then select the time scale corresponding to the lowest entropy value and perform scale averaging on the original time series; After processing the original time series, a spatiotemporal similarity map is constructed using three aspects: state time series correlation, Wasserstein distance, and KL divergence.

4. The predictability determination method for electromagnetic communication data as described in claim 1, characterized in that, Step 3 specifically involves: A spatiotemporal graph convolutional network model with node-level attention is introduced. First, in order to capture the influence of each node on its neighbors at the current and nearby time steps and to expand the spatiotemporal receptive field, a fully connected local spatiotemporal graph is constructed, in which a given node is interconnected with all nodes at other time steps. This graph includes not only the edges of the original graph at each time step, but also the edges connecting the nodes of the graphs between different time steps. The weights of these newly introduced edges are then determined. Secondly, a graph convolutional network with node-level attention is implemented to more effectively capture local spatiotemporal correlations. It consists of two main operations: a series of graph convolution operations and a series of node-level attention operations. By executing these operations in parallel, the attention mechanism is combined with graph convolution to achieve mutual reinforcement. Through the attention mechanism, the receptive field of graph convolution is extended to more relevant nodes. On the other hand, graph convolution helps the attention mechanism capture complex graph topological information and helps to effectively aggregate more relevant nodes from the neighborhood. Finally, data from 12 historical time steps is used as input to combine a spatiotemporal graph convolutional network model with node-level attention to predict the data for the next 12 time steps; then, the prediction accuracy is evaluated by comparing the model's predictions with the actual values.

Images