Power transmission line icing prediction method based on dynamic space-time diagram and uncertainty fusion

Abstract

The application discloses a power transmission line icing prediction method based on dynamic space-time graph and uncertainty fusion, comprising: obtaining the tower node icing feature sequence and the power transmission line dynamic graph structure sequence of the historical time step, constructing the multi-time scale space-time graph representation, inputting the uncertainty perception space-time coding feature block composed of the dynamic graph structure coding path, the time dependence coding path and the uncertainty estimation path for coding, obtaining the structure coding feature, the time coding feature and the multi-source uncertainty representation, inputting the multi-source uncertainty representation into the weight generation network to dynamically calculate the fusion weight, adaptively weighting and fusing the two types of coding features, and finally outputting the future icing prediction mean and variance through the prediction layer. The application solves the problem that the traditional method cannot perceive multi-source risk and is inaccurate in prediction under complex working conditions through multi-source uncertainty joint modeling and adaptive space-time fusion, and has the advantages of high prediction credibility and strong robustness.

Description

Technical Field

[0001] This application relates to the field of transmission line icing prediction, and more specifically, to a method for predicting transmission line icing based on dynamic spatiotemporal diagrams and uncertainty fusion. Background Technology

[0002] Ice accumulation on transmission lines is one of the major natural disasters threatening the safe operation of the power grid. Ice and frost adhere to the surface of the lines, increasing the load and sag. In severe cases, this can lead to conductor galloping, wire breakage, insulator flashover, or tower collapse, causing widespread power outages.

[0003] The icing evolution process of transmission lines exhibits typical spatiotemporal dynamic graph data characteristics. Spatially, there are line topological connections between monitoring points on different towers. Adjacent or electrically close monitoring points experience similar micro-meteorological conditions and their icing evolution influences each other, forming a spatial correlation in the graph structure. Temporally, icing is continuously driven by multiple factors such as temperature, humidity, wind speed, wind direction, and precipitation, exhibiting characteristics of long-term accumulation and short-term abrupt changes. Structurally, the effective correlation between monitoring points dynamically changes with the spatial distribution of meteorological conditions, micro-topographical differences, and the icing stage.

[0004] Existing prediction methods typically employ graph neural networks to model spatial dependencies and combine them with recurrent neural networks or attention mechanisms to process time series data, enabling predictions of states such as icing thickness. However, in actual icing monitoring, sensors are susceptible to data loss or drift due to extreme low temperatures and power outages, leading to uncertainty in node states. The effective correlation strength between transmission line nodes is dynamically affected by line operating conditions, icing galloping, de-icing jumps, de-icing operations, micro-meteorological changes, and monitoring data quality, resulting in uncertainty in dynamic graph edge weights or candidate edge connectivity. Icing growth is influenced by multiple factors, including meteorology and topography, leading to distributional uncertainty in prediction results.

[0005] Furthermore, existing methods often treat graph structures as static, fixed inputs, or construct deterministic adjacency matrices based solely on historical averages, making it difficult to perceive the aforementioned differences in multi-source risks. This can easily lead to erroneous predictions under complex operating conditions such as sensor anomalies and sudden changes in line conditions.

[0006] Therefore, there is an urgent need for an icing prediction method that can jointly model the reliability of monitoring point data, dynamic topology reliability, and prediction distribution credibility, and adaptively fuse spatiotemporal features based on uncertainty. Summary of the Invention

[0007] This application provides a method for predicting icing on transmission lines based on dynamic spatiotemporal maps and uncertainty fusion, aiming to solve the problem that existing models are unable to simultaneously capture the spatiotemporal dynamic evolution of micro-meteorology of transmission lines and the multi-source uncertainty of icing growth, resulting in insufficient prediction accuracy.

[0008] The first aspect of this invention provides a method for predicting icing on transmission lines based on dynamic spatiotemporal graphs and uncertainty fusion, comprising: Step S1: Obtain the icing feature sequence of tower nodes and the dynamic diagram structure sequence of transmission lines at historical time steps; Step S2: Based on the icing feature sequence of the tower nodes and the dynamic diagram structure sequence of the transmission line, construct a spatiotemporal diagram representation with multiple time scales; Step S3: Encode the spatiotemporal graph representation into the prediction model composed of multiple uncertainty-aware spatiotemporal coding feature blocks to obtain dynamic graph structure coding features, time-dependent coding features, and multi-source uncertainty representations; wherein, each uncertainty-aware spatiotemporal coding feature block includes a dynamic graph structure coding path, a time-dependent coding path, and an uncertainty estimation path; the multi-source uncertainty representations include node state uncertainty, edge connection uncertainty, and prediction distribution uncertainty; Step S4: Input the multi-source uncertainty representation into the weight generation network, dynamically generate structural coding weights and temporal coding weights based on the node state uncertainty, edge connection uncertainty and prediction distribution uncertainty, and adaptively weight and fuse the dynamic graph structural coding features and the temporal dependency coding features based on the structural coding weights and temporal coding weights to obtain the final icing spatiotemporal coding features. Step S5: Input the final icing spatiotemporal coding features into the prediction layer to obtain the icing prediction results of the transmission line in the future time step and the uncertainty estimation results corresponding to the icing prediction results; The icing prediction results include the icing prediction mean output by the mean prediction branch, and the corresponding uncertainty estimation results include at least one of the prediction variance, prediction interval, risk score, and icing exceedance probability output by the variance prediction branch.

[0009] Preferably, step S2 specifically includes: The icing feature sequence of tower nodes and the dynamic diagram structure sequence of transmission lines are divided according to a preset time window to obtain the icing feature subsequence of tower nodes and the dynamic diagram structure subsequence of transmission lines. By matching the icing feature subsequence of tower nodes with the dynamic diagram structure subsequence of transmission lines at the same time scale, a spatiotemporal diagram representation with multiple time scales is obtained.

[0010] Preferably, the dynamic graph structure encoding path is used to capture the spatial dependencies and dynamic interactions between nodes; the temporal dependency encoding path is used to capture the patterns of node state changes over time; and the uncertainty estimation path is used to estimate the uncertainty information in the input, structure, and prediction results, and to feed the uncertainty information back to the feature fusion process.

[0011] Preferably, the encoding process of the dynamic graph structure encoding path is as follows: Based on the current time step of the transmission line dynamic graph structure subsequence, determine the candidate adjacency relationships between tower nodes; The confidence score of edge connections between tower nodes is calculated using the following expression:

[0012] in, In the first Time step node With nodes Confidence levels between them It is the Sigmoid activation function. The weight matrix is ​​a learnable matrix. In the first Time step node With nodes Node feature similarity between In the first Nodes within each time step With nodes The frequency of edge connections between them In the first Time step node With nodes Edge attribute information between them This indicates a feature concatenation operation. For learnable bias terms; Candidate adjacency relationships are weighted by the edge connection confidence between tower nodes to generate an uncertainty-aware dynamic graph adjacency matrix. The adjacency matrix of the uncertainty-aware line dynamic graph and the icing features of the tower nodes are input into the graph neural network to complete spatial message passing and obtain the dynamic graph structure encoding features.

[0013] Preferably, the encoding process of the time-dependent encoding path specifically includes: Perform time-location encoding on the icing feature sequence of tower nodes to obtain the icing feature sequence of tower nodes with temporal location information; Input the icing feature sequence of tower nodes with temporal location information into the time coding network, extract the long-term change trend features and short-term fluctuation features of each tower node within the historical time window, and output the time-dependent coding features. The time-coding network employs a multi-head time attention mechanism to capture the dependencies between any historical time steps.

[0014] Preferably, the estimation process of the uncertainty estimation path is as follows: The expression for estimating the uncertainty of the tower node state is:

[0015] in, For nodes The state is uncertain. For network estimation of node uncertainty, Represents a node In the window of history The characteristic variance within, Represents a node The proportion of missing historical data, Represents a node The hidden state, Indicates feature concatenation operation; The uncertainty of transmission line side connections is estimated by the following expression:

[0016] in, For nodes With nodes The edge connections between them are uncertain. For edge uncertainty estimation networks, For nodes With nodes The edge between them in the historical window The weighted variance For nodes With nodes Stability of edge connections between them This represents the adjacency matrix reconstruction error; Estimating the uncertainty of the intermediate prediction distribution of icing specifically includes: obtaining the intermediate prediction variance based on at least one of the dynamic graph structure coding features and the time-dependent coding features, or based on the intermediate spatiotemporal coding features output by the current uncertainty-aware spatiotemporal coding block, through the intermediate variance prediction branch; The intermediate prediction variance is expressed as:

[0017] in, Indicates the intermediate prediction variance. This is the Softplus activation function. For the intermediate variance prediction branch linear layer, It can be any one of the following: dynamic graph structure coding features, time-dependent coding features, and intermediate spatiotemporal coding features; The uncertainty of the tower node state, the uncertainty of the transmission line side connection, and the uncertainty of the intermediate predicted distribution of icing are input into the uncertainty fusion network. The multi-source uncertainty representation U is obtained through at least one of gating fusion, attention fusion, weighted summation or feature splicing.

[0018] Preferably, step S4 specifically includes: The multi-source uncertainty representation is input into the weight generation network, expressed as:

[0019] in, and These are the fusion weights corresponding to the dynamic graph structure encoding features and the time-dependent encoding features, respectively. For normalization function, It is a multilayer perceptron. Characterized by multi-source uncertainty; The dynamic graph structure encoding features and time-dependent encoding features are weighted and fused based on the fusion weight, and the expression is:

[0020] in, For the final spatiotemporal encoding features of icing, and These are dynamic graph structure encoding features and time-dependent encoding features, respectively.

[0021] Preferably, the weighted fusion of dynamic graph structure encoding features and time-dependent encoding features based on fusion weights further includes adaptive weight adjustment, specifically: If the uncertainty of the edge connection is greater than the first preset threshold, the value of the dynamic graph structure encoding weight is reduced; If the uncertainty of the node state or the uncertainty of the predicted distribution is greater than the second preset threshold, the value of the temporal coding weight is increased; The structural encoding weights decrease monotonically as the uncertainty of edge connections increases, while the temporal encoding weights increase monotonically as the uncertainty of node states or intermediate prediction distributions increases.

[0022] A second aspect of the present invention provides an electronic device, characterized in that it includes: a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements a transmission line icing prediction method based on dynamic spatiotemporal graphs and uncertainty fusion.

[0023] A third aspect of the present invention provides a computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and the computer program, when executed by a processor, is a transmission line icing prediction method based on dynamic spatiotemporal graphs and uncertainty fusion.

[0024] This application has at least the following advantages or beneficial effects: (1) This invention constructs a three-parallel coding architecture of dynamic graph structure coding, time-dependent coding and uncertainty estimation, and jointly models three types of uncertainty: node state, edge connection and prediction distribution. This solves the defect of traditional methods that cannot perceive multi-source risks and realizes the joint full probability output of the mean and variance of icing prediction.

[0025] (2) This invention solves the problem of prediction inaccuracy caused by traditional fixed fusion coefficients under complex conditions by generating a network with input weights based on multi-source uncertainty representations and dynamically calculating structural weights and time weights and implementing adaptive adjustment. It realizes the on-demand dynamic allocation of the contribution ratio of spatial and temporal features and has the beneficial effect of significantly improving prediction accuracy and stability in complex scenarios such as sensor anomalies and line galloping.

[0026] (3) This invention generates an uncertainty-aware adjacency matrix by calculating the edge connection confidence to suppress the contribution of unreliable edges in spatial message passing, and constructs a multi-timescale spatiotemporal graph representation to simultaneously capture ice-covering mutations, trends and periodic patterns. It solves the dual problems of traditional static graph structures being prone to error diffusion and incomplete information at a single time scale, and realizes robust coding of dynamic graph structures and comprehensive perception of the multi-periodic evolution of ice-covering. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 The flowchart of the transmission line icing prediction method based on dynamic spatiotemporal diagram and uncertainty fusion provided by the present invention.

[0029] Figure 2 The overall architecture diagram of the spatiotemporal dynamic map prediction model for transmission line icing based on uncertainty perception fusion provided in the embodiments of this application is shown.

[0030] Figure 3 This is a schematic diagram of the internal structure of the uncertainty-aware spatiotemporal coding and uncertainty adaptive fusion module provided in the embodiments of this application.

[0031] Figure 4 This is a schematic diagram of the structure of an electronic device provided by the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0033] Please refer to Figure 1 , Figure 1 This is a flowchart of the transmission line icing prediction method based on dynamic spatiotemporal graphs and uncertainty fusion proposed in this application. Figure 2 The diagram shown illustrates the overall architecture of the prediction model provided in this application embodiment. The prediction model consists of multiple stacked uncertainty-aware spatiotemporal coded feature blocks. The overall architecture includes: an input layer 210 for acquiring historical observation data of the dynamic spatiotemporal map of transmission lines; a multi-timescale construction layer 220 for constructing spatiotemporal map representations at short-term, medium-term, and long-term timescales; an uncertainty-aware spatiotemporal encoder 230, composed of N (N≥1) stacked uncertainty-aware spatiotemporal coded feature blocks; an uncertainty adaptive fusion layer 240 for dynamically calculating structural coding weights and temporal coding weights based on uncertainty representations, and performing weighted fusion of the two types of coding features; and a prediction layer 250, including a mean prediction branch and a variance prediction branch. The mean prediction branch is used to output the predicted mean of icing at future time steps based on the final icing spatiotemporal coded features. The variance prediction branch is used to output the corresponding prediction variance based on the final icing spatiotemporal coding features. The prediction interval, risk score, or probability of exceeding the icing threshold is determined based on the icing prediction mean and prediction variance; the output layer 260 is used to output the icing prediction results of the transmission line in the future time step and the uncertainty estimation results corresponding to the icing prediction results.

[0034] The specific implementation of the method of this application will be described in detail below with reference to the above-mentioned accompanying drawings, including the following steps: Step S1: Obtain the icing feature sequence of tower nodes and the dynamic diagram structure sequence of transmission lines at historical time steps.

[0035] The prediction of icing on transmission lines involves multiple tower nodes, which form a graph structure through the transmission lines. This graph structure dynamically adjusts with changes in line operation status, environmental conditions, and topology. Therefore, the historical observation data of the dynamic spatiotemporal graph includes icing feature sequences of tower nodes under multiple historical time markers and dynamic graph structure sequences of transmission lines.

[0036] For example, in the icing monitoring scenario, the icing characteristics of tower nodes include, but are not limited to: icing thickness, temperature, humidity, wind speed, wind direction, conductor tension, conductor tilt angle, insulator string sway angle, micro-meteorological data, and historical icing growth rate.

[0037] In one specific embodiment, the monitoring area includes M pole nodes, the first... The node feature matrix at each time step is represented as follows:

[0038] in, Indicates the first The pole node at the first Ice-related feature vectors at each time step For feature dimensions.

[0039] The dynamic diagram structure sequence of the transmission line reflects the connection relationship and interaction intensity between tower nodes at different time steps.

[0040] In one specific embodiment, in the first Each time step, the graph structure is represented as follows ,in For a set of nodes, Let be the set of edges. It is an adjacency matrix.

[0041] It should be noted that the elements in the adjacency matrix not only indicate whether there is a transmission line connection between nodes, but can also further represent the connection weight, which can be calculated based on factors such as line electrical distance, historical icing correlation, terrain connectivity, and micro-meteorological zoning.

[0042] In addition, time stamp information includes timestamps corresponding to each historical time step, date types (such as weekdays / holidays), season identifiers, and extreme weather conditions, which are used to help capture the periodic patterns and seasonal characteristics of icing evolution.

[0043] Step S2: Based on the icing feature sequence of the tower nodes and the dynamic diagram structure sequence of the transmission line, construct a spatiotemporal diagram representation with multiple time scales.

[0044] The icing evolution process simultaneously encompasses dynamic characteristics across multiple time scales. For example, on a short-term scale (within hours), sudden freezing rain or torrential rain may cause rapid icing thickening; on a medium-term scale (within days), it reflects the accumulation and melting trends of icing during a complete cold wave; and on a long-term scale (within weeks to months), it presents the periodic changes throughout the entire winter icing season. Therefore, a single time scale is insufficient to fully characterize these multi-layered evolutionary patterns.

[0045] Specifically, step S2 is as follows: The icing feature sequence of tower nodes and the dynamic diagram structure sequence of transmission lines are divided according to a preset time window to obtain the icing feature subsequence of tower nodes and the dynamic diagram structure subsequence of transmission lines. By matching the icing feature subsequence of tower nodes with the dynamic diagram structure subsequence of transmission lines at the same time scale, a spatiotemporal diagram representation with multiple time scales is obtained.

[0046] In some embodiments, spatiotemporal graph representations with multiple time scales can also be constructed using methods such as sliding windows, temporal pooling, temporal convolution, temporal sampling, or attention aggregation.

[0047] Step S3: Encode the spatiotemporal graph representation into the prediction model composed of multiple uncertainty-aware spatiotemporal coding feature blocks to obtain dynamic graph structure coding features, time-dependent coding features, and multi-source uncertainty representations; wherein, each uncertainty-aware spatiotemporal coding feature block includes a dynamic graph structure coding path, a time-dependent coding path, and an uncertainty estimation path; the multi-source uncertainty representations include node state uncertainty, edge connection uncertainty, and prediction distribution uncertainty.

[0048] like Figure 3 The diagram shown illustrates the integrated processing flow of uncertainty-aware spatiotemporal coding and adaptive fusion in this embodiment. The spatiotemporal graph representations at multiple time scales (corresponding to short-term, medium-term, and long-term time scales) are first input to an encoding layer composed of multiple stacked uncertainty-aware spatiotemporal coding feature blocks. Each coding feature block includes three parallel processing paths: Dynamic graph structure encoding path 310 is used to capture spatial dependencies and dynamic interaction relationships between nodes; Time-dependent encoding path 320 is used to capture the pattern of node state changes over time; Uncertainty estimation path 330 is used to estimate uncertainty information in the input, structure and prediction results, and feeds the uncertainty information back to the feature fusion process.

[0049] The encoding processes of the three paths described above are executed in parallel within the same encoded feature block. Subsequently, the uncertainty representation is input to the weight generation unit 340 to calculate the structural encoding weight and the temporal encoding weight. Based on this, the dynamic graph structural encoding features and temporal dependent encoding features are weighted and fused by the adaptive fusion unit 350, and finally the fused icing spatiotemporal encoding features are output.

[0050] It should be noted that in the icing prediction scenario, different tower nodes influence each other due to the spatial correlation of transmission line connections, topographic factors, and micro-meteorological conditions. For example, adjacent towers usually experience similar meteorological conditions, thus exhibiting spatial similarity in icing growth patterns; at the same time, the chain-like topology of transmission lines causes the icing load on the line to have a transmission effect along the conductor direction.

[0051] Specifically, the encoding process of the dynamic graph structure encoding path is as follows: Based on the current time step of the transmission line dynamic graph structure subsequence, determine the candidate adjacency relationships between tower nodes; The confidence score of edge connections between tower nodes is calculated using the following expression:

[0052] in, In the first Time step node With nodes Confidence levels between them It is the Sigmoid activation function. The weight matrix is ​​a learnable matrix. In the first Time step node With nodes Node feature similarity between In the first Nodes within each time step With nodes The frequency of edge connections between them In the first Time step node With nodes Edge attribute information between them This indicates a feature concatenation operation. For learnable bias terms; Candidate adjacency relationships are weighted by the edge connection confidence between tower nodes to generate an uncertainty-aware dynamic graph adjacency matrix. The adjacency matrix of the uncertainty-aware line dynamic graph and the icing features of the tower nodes are input into the graph neural network to complete spatial message passing and obtain the dynamic graph structure encoding features.

[0053] In this embodiment, the graph neural network adopts a graph convolutional network.

[0054] In some embodiments, the graph neural network may also use a graph attention network, a dynamic graph convolutional network, a spatiotemporal graph convolutional network, a message-passing neural network, or a Transformer-based graph structure coding network.

[0055] It should be noted that in icing prediction, key variables such as icing thickness, temperature, and wind speed all exhibit complex temporal evolution patterns, including rapid abrupt changes caused by cold waves as well as periodic changes caused by diurnal temperature variations and seasonal changes.

[0056] Specifically, the encoding process of the time-dependent encoding path is as follows: Perform time-location encoding on the icing feature sequence of tower nodes to obtain the icing feature sequence of tower nodes with temporal location information; Input the icing feature sequence of tower nodes with temporal location information into the time coding network, extract the long-term change trend features and short-term fluctuation features of each tower node within the historical time window, and output the time-dependent coding features. The time-coding network employs a multi-head time attention mechanism to capture the dependencies between any historical time steps.

[0057] It should be noted that transmission line icing prediction faces multiple uncertainties: First, micro-meteorological sensors and icing monitoring equipment may malfunction or experience data drift in harsh environments, leading to uncertainties in node status observations; second, the dynamic graph structure of transmission lines is affected by factors such as line icing galloping, de-icing operations, and line relocation, resulting in inherent uncertainties in edge connectivity; third, icing growth is influenced by a combination of complex factors, leading to inherent random uncertainties in future prediction results.

[0058] Specifically, the estimation process of the uncertainty estimation path is as follows: The expression for estimating the uncertainty of the tower node state is:

[0059] in, For nodes The state is uncertain. For network estimation of node uncertainty, Represents a node In the window of history The characteristic variance within, Represents a node The proportion of missing historical data, Represents a node The hidden state, Indicates feature concatenation operation; The uncertainty of transmission line side connections is estimated by the following expression:

[0060] in, For nodes With nodes The edge connections between them are uncertain. For edge uncertainty estimation networks, For nodes With nodes The edge between them in the historical window The weighted variance For nodes With nodes Stability of edge connections between them This represents the adjacency matrix reconstruction error; Estimating the uncertainty of the intermediate prediction distribution of icing specifically includes: obtaining the intermediate prediction variance based on at least one of the dynamic graph structure coding features and the time-dependent coding features, or based on the intermediate spatiotemporal coding features output by the current uncertainty-aware spatiotemporal coding block, through the intermediate variance prediction branch; The intermediate prediction variance is expressed as:

[0061] in, Indicates the intermediate prediction variance. This is the Softplus activation function. For the intermediate variance prediction branch linear layer, It can be any one of the following: dynamic graph structure coding features, time-dependent coding features, and intermediate spatiotemporal coding features; The uncertainty of the tower node state, the uncertainty of the transmission line side connection, and the uncertainty of the intermediate predicted distribution of icing are input into the uncertainty fusion network. The multi-source uncertainty representation U is obtained through at least one of gating fusion, attention fusion, weighted summation or feature splicing.

[0062] Compared to traditional fixed graph convolution or simple spatiotemporal feature stitching methods, step S3 explicitly models multi-source uncertainty, enabling the model to perceive input noise, structural instability and prediction risks, and dynamically adjust in subsequent fusion, significantly improving prediction robustness under complex working conditions such as sensor anomalies and line galloping.

[0063] Step S4: Input the multi-source uncertainty representation into the weight generation network, dynamically generate structural coding weights and temporal coding weights based on the node state uncertainty, edge connection uncertainty and prediction distribution uncertainty, and adaptively weight and fuse the dynamic graph structural coding features and the temporal dependent coding features based on the structural coding weights and temporal coding weights to obtain the final icing spatiotemporal coding features.

[0064] Specifically, step S4 is as follows: The multi-source uncertainty representation is input into the weight generation network, expressed as:

[0065] in, and These are the fusion weights corresponding to the dynamic graph structure encoding features and the time-dependent encoding features, respectively. For normalization function, It is a multilayer perceptron. Characterized by multi-source uncertainty; The dynamic graph structure encoding features and time-dependent encoding features are weighted and fused based on the fusion weight, and the expression is:

[0066] in, For the final spatiotemporal encoding features of icing, and These are dynamic graph structure encoding features and time-dependent encoding features, respectively.

[0067] In traditional icing prediction methods, spatial and temporal features are typically fused in a fixed proportion or through simple splicing, failing to dynamically adjust based on the actual quality of the input data. This application, however, quantifies the reliability of the graph structure and node observations through uncertainty estimation paths and feeds this information back to the fusion weight generation process, enabling the model to adaptively adjust the contribution ratio of spatial and temporal features under different scenarios.

[0068] Specifically, the weighted fusion of dynamic graph structure encoding features and time-dependent encoding features based on fusion weights also includes adaptive weight adjustment, specifically: If the uncertainty of the edge connection is greater than the first preset threshold, it indicates that the graph connection of the transmission line is unreliable (e.g., the line ice galloping causes frequent topology changes, or the line status data fluctuates violently under extreme weather conditions). The dependence on spatial structure information should be reduced, and the value of the dynamic graph structure encoding weight should be reduced. If the uncertainty of the node state or the uncertainty of the prediction distribution is greater than the second preset threshold, it indicates that the quality of the node observation data is poor or the prediction result has a large uncertainty. More attention should be paid to the stable trend in the historical event sequence, and the value of the time-series coding weight should be increased. The structural encoding weights decrease monotonically as the uncertainty of edge connections increases, while the temporal encoding weights increase monotonically as the uncertainty of node states or intermediate prediction distributions increases.

[0069] Compared with traditional models that use fixed fusion coefficients, step S4 can dynamically optimize the contribution ratio of spatial and temporal features based on the reliability of the current samples. This avoids over-reliance on unreliable information when there are drastic structural changes or severe data gaps, and effectively improves the prediction accuracy and stability in extreme scenarios.

[0070] Step S5: Input the final icing spatiotemporal coding features into the prediction layer to obtain the icing prediction results of the transmission line in the future time step and the uncertainty estimation results corresponding to the icing prediction results; The icing prediction results include the icing prediction mean output by the mean prediction branch, and the corresponding uncertainty estimation results include at least one of the prediction variance, prediction interval, risk score, and icing exceedance probability output by the variance prediction branch.

[0071] like Figure 3 As shown, Figure 2 A schematic diagram of the internal structure of the uncertainty-aware spatiotemporal encoder 230 and the uncertainty adaptive fusion layer 240. Spatiotemporal graph representation for each time scale. The dynamic graph structure encoding path 310 outputs dynamic graph structure encoding features. Time-dependent encoding path 320 outputs time-dependent encoded features The uncertainty estimation path 330 output includes a multi-source uncertainty representation of node state uncertainty, edge connection uncertainty, and intermediate prediction distribution uncertainty. The weight generation unit 340 generates weights based on the multi-source uncertainty characterization. Generate structure encoding weights and time-encoded weights The adaptive fusion unit 350 adjusts the weights accordingly. and Perform weighted summaries and output the final encoded features. .

[0072] Furthermore, in order to fully verify the effectiveness of the method of the present invention, prediction error, uncertainty estimation effect and ablation experiment were carried out with the baseline model on a real transmission line icing monitoring dataset. The model performance was evaluated from two dimensions: prediction accuracy and uncertainty estimation quality.

[0073] Baseline Model 1 employs a GCN+GRU architecture, utilizing a Graph Convolutional Network (GCN) to extract static spatial topological dependencies between tower nodes and a Gated Recurrent Unit (GRU) to model the evolution of icing features over time. Baseline Model 2 uses STGCN (Spatiotemporal Graph Convolutional Network), which achieves simultaneous joint extraction of spatial-temporal features by alternating graph convolution and temporal convolution operations in each layer. Baseline Model 3 uses Graph Transformer (a neural network model that applies the Transformer architecture to graph-structured data), employing a graph attention mechanism in conjunction with the Transformer architecture to jointly model spatiotemporal dependencies.

[0074] Table 1 shows the comparison of prediction errors between the method of this invention and the baseline model in the task of predicting icing on transmission lines: Table 1 Comparison of Prediction Errors

[0075] As shown in Table 1, the method of this invention outperforms all three types of baseline models in both mean absolute error and root mean square error. This result demonstrates that the present invention effectively avoids the propagation of errors under conditions such as sensor anomalies or line galloping by adaptively suppressing spatial message passing of unreliable edges through the dynamic graph structure encoding path; it captures long-range dependencies in the icing history sequence through a multi-head time attention mechanism in the time-dependent encoding path, achieving accurate extraction of icing evolution trends and short-term fluctuations; and it avoids prediction inaccuracies under complex conditions by using multi-source uncertainty joint modeling and adaptive fusion mechanisms.

[0076] Table 2 shows the comparison results between the method of this invention and the baseline model in terms of uncertainty estimation quality: Table 2 Comparison of Uncertainty Estimation Performance

[0077] As shown in Table 2, the method of this invention outperforms both baseline models in terms of negative log-likelihood and continuous rank probability fraction, indicating that the probability distribution predicted by the method of this invention matches the actual icing thickness distribution better and has superior overall prediction capability. Regarding the prediction interval coverage probability index, the method of this invention is closest to the set 90% confidence level, indicating that this invention effectively improves the calibration of the prediction interval by explicitly modeling multi-source uncertainties, avoiding the risk of underreporting caused by overconfidence in traditional methods. Regarding the average prediction interval width index, the method of this invention achieves the narrowest prediction interval while ensuring the highest coverage rate, indicating that this invention can provide a more accurate and valuable icing prediction range. Overall, the method of this application outperforms both baseline models in all indicators, verifying the effectiveness of the proposed multi-source uncertainty joint modeling scheme.

[0078] The comparative results of ablation experiments using the method of this invention are shown in Table 3: Table 3 Ablation Experiment Table

[0079] As can be seen from Table 3, the complete model of the present invention outperforms the variant model after any type of uncertainty is removed in terms of mean absolute error, root mean square error, and negative log-likelihood.

[0080] Among these challenges, removing node state uncertainty estimation makes it impossible to distinguish between trusted and untrusted nodes, and abnormal sensor data contaminates neighbor node predictions after participating in spatial message passing. Removing edge connection uncertainty estimation results in the continued use of the original topology even when the dynamic graph structure undergoes drastic changes, leading to distortion of spatial coding features. Removing prediction distribution uncertainty estimation lacks an assessment of the confidence level of the prediction results themselves, resulting in significant bias on out-of-distribution samples. Each of the three uncertainty estimation paths plays an irreplaceable modeling role, complementing and cooperating with each other. The absence of any one path leads to a considerable decline in prediction performance, validating the technical effectiveness of the multi-source uncertainty joint modeling and adaptive fusion scheme of this invention.

[0081] This disclosure also provides an electronic device, please refer to... Figure 4 , Figure 4 This is a schematic diagram of an electronic device illustrated in an embodiment of this disclosure. For example... Figure 4 As shown, the electronic device 100 includes a memory 110 and a processor 120. The memory 110 and the processor 120 are connected via a bus for communication. The memory 110 stores a computer program that can run on the processor 120 to implement the steps in the transmission line icing prediction method based on dynamic spatiotemporal graph and uncertainty fusion disclosed in this embodiment.

[0082] The disclosed embodiments also provide a computer-readable storage medium that, when executed by a processor of a computer device, enables the computer device to perform steps as described in the transmission line icing prediction method based on dynamic spatiotemporal graphs and uncertainty fusion of the present disclosure embodiments.

[0083] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, electronic devices, and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0084] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0085] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0086] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0087] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0088] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0089] The above provides a detailed description of the transmission line icing prediction method based on dynamic spatiotemporal graphs and uncertainty fusion provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for predicting icing on transmission lines based on dynamic spatiotemporal graphs and uncertainty fusion, characterized in that, include: Step S1: Obtain the icing feature sequence of tower nodes and the dynamic diagram structure sequence of transmission lines at historical time steps; Step S2: Based on the icing feature sequence of the tower nodes and the dynamic diagram structure sequence of the transmission line, construct a spatiotemporal diagram representation with multiple time scales; Step S3: Encode the spatiotemporal graph representation into the prediction model composed of multiple uncertainty-aware spatiotemporal coding feature blocks to obtain dynamic graph structure coding features, time-dependent coding features, and multi-source uncertainty representations; wherein, each uncertainty-aware spatiotemporal coding feature block includes a dynamic graph structure coding path, a time-dependent coding path, and an uncertainty estimation path; the multi-source uncertainty representations include node state uncertainty, edge connection uncertainty, and prediction distribution uncertainty; Step S4: Input the multi-source uncertainty representation into the weight generation network, dynamically generate structural coding weights and temporal coding weights based on the node state uncertainty, edge connection uncertainty and prediction distribution uncertainty, and adaptively weight and fuse the dynamic graph structural coding features and the temporal dependency coding features based on the structural coding weights and temporal coding weights to obtain the final icing spatiotemporal coding features. Step S5: Input the final icing spatiotemporal coding features into the prediction layer to obtain the icing prediction results of the transmission line in the future time step and the uncertainty estimation results corresponding to the icing prediction results; The icing prediction results include the icing prediction mean output by the mean prediction branch, and the corresponding uncertainty estimation results include at least one of the prediction variance, prediction interval, risk score, and icing exceedance probability output by the variance prediction branch.

2. The transmission line icing prediction method based on dynamic spatiotemporal graph and uncertainty fusion according to claim 1, characterized in that, Step S2, specifically: The icing feature sequence of tower nodes and the dynamic diagram structure sequence of transmission lines are divided according to a preset time window to obtain the icing feature subsequence of tower nodes and the dynamic diagram structure subsequence of transmission lines. By matching the icing feature subsequence of tower nodes with the dynamic diagram structure subsequence of transmission lines at the same time scale, a spatiotemporal diagram representation with multiple time scales is obtained.

3. The method for predicting icing of transmission lines based on dynamic spatiotemporal graphs and uncertainty fusion as described in claim 1, characterized in that, The dynamic graph structure encoding path is used to capture the spatial dependencies and dynamic interactions between nodes; the temporal dependency encoding path is used to capture the pattern of node state changes over time. The uncertainty estimation path is used to estimate the uncertainty information in the input, structure, and prediction results, and feeds the uncertainty information back to the feature fusion process.

4. The method for predicting icing of transmission lines based on dynamic spatiotemporal graphs and uncertainty fusion as described in claim 1, characterized in that, The encoding process of the dynamic graph structure encoding path is as follows: Based on the current time step of the transmission line dynamic graph structure subsequence, determine the candidate adjacency relationships between tower nodes; The confidence score of edge connections between tower nodes is calculated using the following expression: in, In the first Time step node With nodes Confidence levels between them It is the Sigmoid activation function. The weight matrix is ​​a learnable matrix. In the first Time step node With nodes Node feature similarity between In the first Nodes within each time step With nodes The frequency of edge connections between them In the first Time step node With nodes Edge attribute information between them This indicates a feature concatenation operation. For learnable bias terms; Candidate adjacency relationships are weighted by the edge connection confidence between tower nodes to generate an uncertainty-aware dynamic graph adjacency matrix. The adjacency matrix of the uncertainty-aware dynamic graph and the icing features of the tower nodes are input into the graph neural network to complete spatial message passing and obtain the dynamic graph structure encoding features.

5. The method for predicting icing of transmission lines based on dynamic spatiotemporal graphs and uncertainty fusion as described in claim 1, characterized in that, The encoding process of the time-dependent encoding path is as follows: Perform time-location encoding on the icing feature sequence of tower nodes to obtain the icing feature sequence of tower nodes with temporal location information; Input the icing feature sequence of tower nodes with temporal location information into the time coding network, extract the long-term change trend features and short-term fluctuation features of each tower node within the historical time window, and output the time-dependent coding features. The time-coding network employs a multi-head time attention mechanism to capture the dependencies between any historical time steps.

6. The method for predicting icing of transmission lines based on dynamic spatiotemporal graphs and uncertainty fusion as described in claim 1, characterized in that, The estimation process of the uncertainty estimation path is as follows: The expression for estimating the uncertainty of the tower node state is: in, For nodes The state is uncertain. For network estimation of node uncertainty, Represents a node In the window of history The characteristic variance within, Represents a node The proportion of missing historical data, Represents a node The hidden state, Indicates feature concatenation operation; The uncertainty of transmission line side connections is estimated by the following expression: in, For nodes With nodes The edge connections between them are uncertain. For edge uncertainty estimation networks, For nodes With nodes The edge between them in the historical window The weighted variance For nodes With nodes Stability of edge connections between them This represents the adjacency matrix reconstruction error; Estimating the uncertainty of the intermediate prediction distribution of icing specifically includes: obtaining the intermediate prediction variance based on at least one of the dynamic graph structure coding features and the time-dependent coding features, or based on the intermediate spatiotemporal coding features output by the current uncertainty-aware spatiotemporal coding block, through the intermediate variance prediction branch; The intermediate prediction variance is expressed as: in, Indicates the intermediate prediction variance. This is the Softplus activation function. For the intermediate variance prediction branch linear layer, It can be any one of the following: dynamic graph structure coding features, time-dependent coding features, and intermediate spatiotemporal coding features; The uncertainty of the tower node state, the uncertainty of the transmission line side connection, and the uncertainty of the intermediate predicted distribution of icing are input into the uncertainty fusion network. The multi-source uncertainty representation U is obtained through at least one of gating fusion, attention fusion, weighted summation or feature splicing.

7. The method for predicting icing of transmission lines based on dynamic spatiotemporal graphs and uncertainty fusion as described in claim 1, characterized in that, Step S4 is as follows: The multi-source uncertainty representation is input into the weight generation network, expressed as: in, and These are the fusion weights corresponding to the dynamic graph structure encoding features and the time-dependent encoding features, respectively. For normalization function, It is a multilayer perceptron. Characterized by multi-source uncertainty; The dynamic graph structure encoding features and time-dependent encoding features are weighted and fused based on the fusion weight, and the expression is: in, For the final spatiotemporal encoding features of icing, and These are dynamic graph structure encoding features and time-dependent encoding features, respectively.

8. The method for predicting icing of transmission lines based on dynamic spatiotemporal graphs and uncertainty fusion as described in claim 7, characterized in that, The weighted fusion of dynamic graph structure encoding features and time-dependent encoding features based on fusion weights also includes adaptive weight adjustment, specifically: If the uncertainty of the edge connection is greater than the first preset threshold, the value of the dynamic graph structure encoding weight is reduced; If the uncertainty of the node state or the uncertainty of the predicted distribution is greater than the second preset threshold, the value of the temporal coding weight is increased. The structural encoding weights decrease monotonically as the uncertainty of edge connections increases, while the temporal encoding weights increase monotonically as the uncertainty of node states or intermediate prediction distributions increases.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the transmission line icing prediction method based on dynamic spatiotemporal graph and uncertainty fusion as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the transmission line icing prediction method based on dynamic spatiotemporal graphs and uncertainty fusion as described in any one of claims 1 to 8.

Images