Power transmission line icing fault prediction method and system based on bayesian network

Abstract

The application discloses a power transmission line icing fault prediction method and system based on a Bayesian network, relates to the technical field of fault prediction, and comprises the following steps: acquiring multi-source monitoring data, constructing a unified time series data set; constructing a mechanism constraint rule set, constructing an initial Bayesian network structure, and generating a dynamic Bayesian network model after establishing the state transition dependency relationship between adjacent time slices; inputting the time series data set into the dynamic Bayesian network model, performing posterior probability updating on icing state nodes and hidden variable nodes, obtaining the time series probability distribution of each node, performing icing state recursive prediction of a future time period, establishing the prediction probability distribution of each node in the future time period, performing probability propagation of the dynamic Bayesian network model, and configuring icing fault risk early warning information. The application solves the technical problem that the existing technology cannot accurately predict the icing fault occurrence time and risk distribution, and achieves the technical effects of improving the accuracy of icing fault prediction and early warning.

Description

Technical Field

[0001] This invention relates to the field of fault prediction technology, specifically to a method and system for predicting icing faults in transmission lines based on Bayesian networks. Background Technology

[0002] Predicting icing faults on transmission lines typically relies on meteorological monitoring data, conductor operating status data, and historical icing observation data. However, these methods often fail to effectively integrate multi-source data, resulting in an inability to fully reflect the complex mechanisms of icing formation and the dynamic changes in the external environment. In practical applications, due to the lack of accurate modeling of influencing factors, the timing and risk distribution of icing faults are often impossible to predict accurately. This insufficient prediction accuracy prevents the early detection of potential risks, hindering timely and effective prevention and control measures and impacting the stable operation and safety of transmission lines. Summary of the Invention

[0003] This application provides a method and system for predicting icing faults in transmission lines based on Bayesian networks, which is used to address the technical problem that existing technologies cannot accurately predict the occurrence time and risk distribution of icing faults.

[0004] In view of the above problems, this application provides a method and system for predicting icing faults in transmission lines based on Bayesian networks.

[0005] The first aspect of this application provides a method for predicting icing faults in transmission lines based on Bayesian networks, the method comprising: Multi-source monitoring data along the target transmission line is acquired, including meteorological monitoring data, conductor operating status data, and historical icing monitoring data. Spatiotemporal alignment and feature standardization are performed on the multi-source monitoring data to construct a unified time-series dataset. A set of mechanistic constraint rules is constructed based on the icing formation mechanism, including constraints on temperature phase transition, supercooled water droplet existence, and the influence of wind speed on icing adhesion efficiency. An initial Bayesian network structure is constructed based on the set of mechanistic constraint rules, and a dynamic network is generated after establishing the state transition dependencies between adjacent time slices. A Bayesian network model is used. The time-series dataset is input into the dynamic Bayesian network model. Within the dynamic Bayesian network model, based on the current meteorological monitoring data and historical icing observation data, posterior probability updates are performed on the icing state nodes and latent variable nodes to obtain the time-series probability distribution of each node. Based on the time-series probability distribution and state transition dependencies, recursive prediction of the icing state for future periods is performed to establish the predicted probability distribution of each node for future periods. Probability propagation of the dynamic Bayesian network model is performed based on the predicted probability distribution, and icing fault risk warning information is configured using the probability propagation results.

[0006] A second aspect of this application provides a transmission line icing fault prediction system based on Bayesian networks, the system comprising: The data acquisition module acquires multi-source monitoring data along the target transmission line, including meteorological monitoring data, conductor operating status data, and historical icing monitoring data. It performs spatiotemporal alignment and feature standardization on the multi-source monitoring data to construct a unified time-series dataset. The constraint rule set construction module constructs a mechanistic constraint rule set based on the icing formation mechanism, including temperature phase transition constraints, supercooled water droplet existence constraints, and wind speed influence on icing adhesion efficiency constraints. The model generation module constructs an initial Bayesian network structure based on the mechanistic constraint rule set and establishes state transition dependencies between adjacent time slices. Then, a dynamic Bayesian network model is generated. The prediction module is used to input the time series dataset into the dynamic Bayesian network model. Within the dynamic Bayesian network model, based on the meteorological monitoring data at the current moment and the historical icing observation data, the posterior probability update is performed on the icing state nodes and latent variable nodes to obtain the time series probability distribution of each node. Based on the time series probability distribution and state transition dependencies, the icing state of future periods is recursively predicted to establish the predicted probability distribution of each node in the future period. The early warning information configuration module is used to perform probability propagation of the dynamic Bayesian network model according to the predicted probability distribution and to configure icing fault risk early warning information using the probability propagation results.

[0007] One or more technical solutions provided in this application have at least the following technical effects or advantages: This application acquires multi-source monitoring data along the target transmission line, including meteorological monitoring data, conductor operating status data, and historical icing monitoring data. It performs spatiotemporal alignment and feature standardization on the multi-source monitoring data to construct a unified time-series dataset. Based on the icing formation mechanism, it constructs a set of mechanism constraint rules, including temperature phase transition constraints, supercooled water droplet existence constraints, and wind speed influence on icing adhesion efficiency constraints. Based on the set of mechanism constraint rules, it constructs an initial Bayesian network structure and, after establishing the state transition dependencies between adjacent time slices, generates... A dynamic Bayesian network model is used. The time-series dataset is input into the dynamic Bayesian network model. Within the model, based on current meteorological monitoring data and historical icing observation data, posterior probability updates are performed on icing state nodes and latent variable nodes to obtain the time-series probability distribution of each node. Based on the time-series probability distribution and state transition dependencies, recursive predictions of icing states for future periods are performed to establish predicted probability distributions for each node in the future period. Probability propagation of the dynamic Bayesian network model is performed based on the predicted probability distributions, and icing fault risk warning information is configured using the probability propagation results. This invention solves the technical problem of inaccurate prediction of icing fault occurrence time and risk distribution in existing technologies. Through probability propagation and time-series data updates using a dynamic Bayesian network model, it achieves improved accuracy in icing fault prediction and provides early warning. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A schematic diagram of the transmission line icing fault prediction method based on Bayesian network provided in the embodiments of this application; Figure 2 A comparative diagram of verification data for the Bayesian network-based transmission line icing fault prediction method provided in the embodiments of this application; Figure 3 A schematic diagram of the structure of a transmission line icing fault prediction system based on Bayesian networks provided in this application embodiment.

[0010] Figure labeling: Data acquisition module 11, constraint rule set construction module 12, model generation module 13, prediction module 14, early warning information configuration module 15. Detailed Implementation

[0011] This application provides a method and system for predicting icing faults in transmission lines based on Bayesian networks. It addresses the technical problem of the inability to accurately predict the occurrence time and risk distribution of icing faults in existing technologies. By using probability propagation and time-series data updates through a dynamic Bayesian network model, it achieves the technical effect of improving the accuracy of icing fault prediction and providing early warning.

[0012] 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 a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0013] It should be noted that any variation of the terms "comprising" and "having" is intended to cover non-exclusive inclusion, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or modules that are not explicitly listed or that are inherent to such processes, methods, products, or devices.

[0014] Example 1, as Figure 1 As shown, this application provides a method for predicting icing faults in transmission lines based on Bayesian networks, the method comprising: Step S100: Obtain multi-source monitoring data along the target transmission line. The multi-source monitoring data includes meteorological monitoring data, conductor operation status data, and historical icing monitoring data. Perform spatiotemporal alignment and feature standardization processing on the multi-source monitoring data to construct a unified time-series dataset.

[0015] In this embodiment, multi-source monitoring data along the target transmission line is first acquired, including meteorological monitoring data, conductor operating status data, and historical icing monitoring data. The meteorological monitoring data is collected in real-time by meteorological stations installed along the transmission line, including meteorological information such as temperature, humidity, wind speed, and precipitation. The conductor operating status data is collected by the transmission line monitoring system, which monitors parameters such as current, voltage, and conductor temperature using sensors deployed along the line. The historical icing monitoring data is extracted from historical records of past icing events along the line.

[0016] Next, spatiotemporal alignment and feature standardization are performed on the multi-source monitoring data. First, based on the sampling time intervals of the multi-source monitoring data, time-scale unification processing is performed on meteorological monitoring data, conductor operation status data, and historical icing monitoring data. Through interpolation compensation methods, these data are adjusted to the same time interval, thus constructing a unified time series. Then, based on the spatial distribution of the target transmission line, spatial mapping is performed on the data from different monitoring nodes, transforming the data from each monitoring node into a unified spatial coordinate system. Through spatial mapping, scattered data points are integrated into a standardized spatial data representation. Finally, spatiotemporal alignment and feature standardization are performed by combining the unified time series and the segmented spatial data representation. Through this process, all data sources achieve consistency in both time and space, thus forming a complete and consistent time-series dataset.

[0017] Furthermore, the method provided in the application embodiments, which performs spatiotemporal alignment and feature standardization processing on the multi-source monitoring data, further includes: Based on the sampling time interval of multi-source monitoring data, meteorological monitoring data, conductor operation status data and historical icing monitoring data are processed with unified time scale, and a unified time series is constructed through interpolation compensation; based on the spatial distribution of the target transmission line, spatial mapping is performed on the data of different monitoring nodes, and segmented data representation is formed according to the spatial mapping results; spatiotemporal alignment and feature standardization are performed using the unified time series and segmented data representation.

[0018] In this embodiment, based on the sampling time interval of multi-source monitoring data, a time series resampling method is used to perform time-scale unification processing on meteorological monitoring data, conductor operation status data, and historical icing monitoring data. The original different acquisition frequencies of the three types of data are normalized to a preset unified time resolution. For missing values ​​in the resampled data sequence, interpolation compensation is performed using a linear interpolation method to fill the data gaps in the time dimension, so that the three types of data form a continuous and synchronous sequence on the time axis. Finally, a unified time series with completely consistent time dimension is constructed, realizing the homogenization and normalization of multi-source monitoring data in the time dimension.

[0019] Next, based on the spatial distribution of the target transmission line, the GIS spatial geographic information analysis method is used to perform spatial mapping on the multi-source monitoring data collected by different monitoring nodes along the transmission line. The geographic coordinate information of each monitoring node is mapped to the actual spatial location of the transmission line. The data collected by each monitoring node is matched one-to-one with the corresponding spatial points of the line. The monitoring data corresponding to the completed spatial locations are classified and integrated according to the actual layout sections of the transmission line. Based on the spatial mapping results, a segmented data expression that fits the actual layout of the transmission line is formed, thus completing the structured reconstruction of the multi-source monitoring data in the spatial dimension.

[0020] Finally, using the constructed unified time series and the established segmented data representation, spatiotemporal alignment and feature standardization are performed on the multi-source monitoring data. First, the monitoring data of each time node in the unified time series is matched with the corresponding transmission line spatial monitoring nodes and deployment sections in the segmented data representation, so that the temporal and spatial characteristics of the monitoring data form a complete correspondence, eliminating the heterogeneity of the data in the spatiotemporal dimension. Then, the spatiotemporally aligned multi-source monitoring data is subjected to Z-score standardization to unify the dimensions and normalized numerical ranges of different types of data. Finally, the standardized multi-source monitoring data is integrated and sorted to construct a unified time series dataset.

[0021] Step S200: Construct a mechanism constraint rule set based on the icing formation mechanism. The mechanism constraint rule set includes temperature phase change constraints, supercooled water droplet existence condition constraints, and wind speed influence on icing adhesion efficiency constraints.

[0022] In this embodiment, the pre-stored ice formation mechanism is first retrieved. This mechanism fully covers the core physical processes and the laws governing the action of each link in the ice formation process, including water vapor phase change, supercooled water droplet impact and adhesion, and wind speed affecting ice accumulation.

[0023] When constructing the mechanistic constraint rule set based on the icing formation mechanism, starting from the physical mechanism of icing formation, temperature, supercooled water droplets, and wind speed are identified as the core factors affecting icing formation on transmission lines. Based on the role and physical characteristics of each factor in the icing formation process, corresponding constraint rules are formulated for each. These rules are then integrated to form a complete mechanistic constraint rule set. Among them, the temperature phase change constraint is formulated around the influence of ambient temperature on the phase change of water vapor. It clarifies that when the ambient temperature drops to 0℃, water vapor around the transmission line will undergo a phase change and condense into ice on the conductor surface. Specific constraint parameters are quantified, requiring the ambient temperature to remain below 0℃ for at least 4 hours to achieve effective icing formation. Furthermore, when the ambient temperature rises above 0℃, the ice on the conductor surface will begin to melt. The influence of the temperature change rate on icing formation is also incorporated, setting a cooling rate of less than 1℃ per hour to effectively promote icing formation. The constraints on the existence of supercooled water droplets are based on the physical properties of supercooled water droplets as the core material basis for icing formation. The key environmental and physical parameters for supercooled water droplets to participate in icing formation are clearly defined and quantified. It is stipulated that the particle size of supercooled water droplets participating in icing formation must be between 10 micrometers and 50 micrometers, the ambient humidity along the transmission line must reach above 80%, and the supercooled water droplets must be in the atmospheric altitude range of 50 meters to 1000 meters. Through the above parameter constraints, it is ensured that supercooled water droplets under specific meteorological conditions can effectively collide with the conductor and participate in the icing formation process. The constraint on the impact of wind speed on icing adhesion efficiency is based on the correlation between wind speed and the probability of supercooled water droplets hitting the conductor. By quantifying the probability of supercooled water droplets hitting the conductor in different wind speed ranges, its specific impact on icing adhesion efficiency is clarified. When the wind speed is less than 1 m / s, the probability of supercooled water droplets hitting the conductor is less than 0.1%, at which point icing adhesion on the transmission line hardly occurs. When the wind speed is in the range of 2 to 5 m / s, the probability of supercooled water droplets hitting the conductor is 10% to 15%, which is the optimal wind speed range for icing adhesion, with strong icing adhesion and the highest adhesion efficiency. When the wind speed exceeds 8 m / s, the probability of supercooled water droplets hitting the conductor drops to less than 2%, the icing adhesion effect is significantly suppressed, and the adhesion efficiency is greatly reduced.

[0024] Based on the detailed quantitative constraint parameters and specific rules for temperature, supercooled water droplets, and wind speed, the constraints on temperature phase change, the existence conditions of supercooled water droplets, and the influence of wind speed on icing adhesion efficiency are integrated and compiled to finally construct a complete set of mechanism constraint rules.

[0025] Step S300: Construct an initial Bayesian network structure based on the set of mechanistic constraint rules, and generate a dynamic Bayesian network model after establishing the state transition dependency relationship between adjacent time slices.

[0026] In this embodiment, when constructing the initial Bayesian network structure based on the mechanistic constraint rule set, firstly, icing state nodes, conductor operation state nodes, and latent variable nodes are generated according to the constraints of temperature phase transition, supercooled water droplet existence, and wind speed influence on icing adhesion efficiency. A conditional probability table is then established for each node, constructed statistically based on historical icing monitoring data and conductor operation state data. Next, a local causal topology is established according to the causal relationships in the mechanistic constraint rule set, thus forming the initial Bayesian network structure. This initial Bayesian network structure is then replicated along the time axis into multiple time slice nodes. Based on historical icing evolution data and the probability distribution of the local causal topology, state transition dependencies between adjacent time slices are established, forming the framework of a dynamic Bayesian network. In each time slice, conditional dependencies are established between unobservable latent variable nodes and their corresponding icing state nodes and conductor operation state nodes, and the prior probabilities of the latent variable nodes are initialized based on historical data. Finally, the structural integrity and causal consistency of the dynamic Bayesian network skeleton are verified, ultimately generating the dynamic Bayesian network model.

[0027] Furthermore, in the method provided in the application embodiments, the initial Bayesian network structure is constructed based on the set of mechanistic constraint rules, and a dynamic Bayesian network model is generated after establishing the state transition dependencies between adjacent time slices, further including: Based on the temperature phase transition constraints, supercooled water droplet existence constraints, and wind speed influence on icing adhesion efficiency constraints in the mechanistic constraint rule set, icing state nodes, conductor operation state nodes, and latent variable nodes are generated, and a conditional probability table is established for each node. This table is constructed based on historical icing monitoring data and conductor operation state data statistics. A local causal topology structure is established for each node according to the causal relationships defined in the mechanistic constraint rule set, forming an initial Bayesian network structure. This initial Bayesian network structure is replicated along the time axis into multiple time slice nodes. Based on historical icing evolution data and the probability distribution of the local causal topology, state transition dependencies between adjacent time slices are established to form a dynamic Bayesian network skeleton. In each time slice, conditional dependencies are established between unobservable latent variable nodes and their corresponding icing state nodes and conductor operation state nodes, and the prior probabilities of the latent variable nodes are initialized based on historical icing monitoring data. The structural integrity and causal consistency of the dynamic Bayesian network skeleton are verified, generating a dynamic Bayesian network model.

[0028] In this embodiment, based on the temperature phase transition constraints, supercooled water droplet existence constraints, and wind speed influence on icing adhesion efficiency constraints in the mechanism constraint rule set, characteristic indicators of corresponding icing development patterns, actual conductor operating conditions, and intangible influencing factors during the icing process are extracted. Icing state nodes, conductor operating state nodes, and latent variable nodes are sequentially divided. Historical icing monitoring data and conductor operating state data are selected as the calculation basis. First, missing value removal, outlier value filtering, and continuous value discretization are performed on all raw data. Then, the processed data are grouped and statistically analyzed according to the condition range defined by the mechanism constraints. The frequency of each node under different parent node state combinations is recorded. The conditional probability of each item is calculated by dividing the frequency of a single state by the total sample size of the corresponding group. All calculated probability values ​​are arranged according to node category and state type. A corresponding data table is compiled for each node. Finally, the generation of all network nodes and the construction of the corresponding conditional probability table for each node are completed.

[0029] Next, the generated icing state nodes, conductor running state nodes, and latent variable nodes will be arranged according to the physical causal relationships clearly defined in the mechanism constraint rules. Directed line segments will be used to directly represent the influence transmission path between nodes, and local causal topologies corresponding to different constraint conditions will be built one by one. Then, all the scattered local causal topologies will be spliced ​​and integrated to unify the hierarchical relationship and association direction of the nodes, eliminate logical intersections and conflicts between structures, and finally form the initial Bayesian network structure.

[0030] Subsequently, the initial Bayesian network structure is replicated along the time axis at fixed time steps, so that each replicated network unit corresponds to an independent moment, forming multiple time slice nodes distributed in different time dimensions. Historical icing evolution time series data are retrieved, and combined with the probability distribution characteristics of local causal topology, the frequency of state transitions of the same node between adjacent time slices is counted. The state transition probability is obtained by dividing the transition frequency by the total sample size of the current state. Based on the calculated transition probability, a quantitative correlation relationship between adjacent time slice nodes is established, and finally a dynamic Bayesian network skeleton containing time series change characteristics is formed.

[0031] Subsequently, within each independent time slice, unobservable latent variable nodes are directionally connected to the corresponding icing state nodes and conductor operation state nodes in the same time slice. This clarifies the probabilistic influence path of latent variable nodes on the two types of observation nodes, thereby establishing the conditional dependency relationship between latent variable nodes and observation nodes. At the same time, based on historical icing monitoring data, the frequency of occurrence of each discrete state of the latent variable nodes is statistically analyzed. The corresponding probability value is calculated by dividing the frequency of state occurrence by the total sample size and directly assigned, thus completing the initialization setting of the prior probability of the latent variable nodes.

[0032] Finally, a structural integrity check was performed on the dynamic Bayesian network skeleton. All nodes, conditional dependencies, and state transition dependencies were checked one by one to confirm that there were no missing nodes, no missing connections, and no missing probability parameters. Then, a causal consistency check was performed. The association direction of all nodes in the network, the probability transmission logic, and the temperature phase change constraints, supercooled water droplet existence constraints, and wind speed influence on icing adhesion efficiency constraints in the mechanism constraint rule set were compared one by one to confirm that there were no logical conflicts or causal contradictions. After both checks passed, the network structure and probability parameters were finally solidified to generate the dynamic Bayesian network model.

[0033] Step S400: Input the time series dataset into the dynamic Bayesian network model. Within the dynamic Bayesian network model, based on the meteorological monitoring data at the current moment and the historical icing observation data, perform posterior probability updates on the icing state nodes and latent variable nodes to obtain the time series probability distribution of each node. Based on the time series probability distribution and state transition dependencies, perform recursive prediction of the icing state for future periods to establish the predicted probability distribution of each node for future periods.

[0034] In this embodiment, when inputting the time-series dataset into the dynamic Bayesian network model, the meteorological monitoring data and historical icing observation data at the current moment in the time-series dataset are first assigned corresponding values ​​according to the data interfaces of each observation node in the dynamic Bayesian network model. The meteorological monitoring data includes measured values ​​of ambient temperature, supercooled water droplet size distribution, ambient humidity, atmospheric altitude, and wind speed; the historical icing observation data includes measured values ​​of icing thickness, conductor temperature, current, and sag. The measured values ​​are then divided into corresponding discrete intervals according to the pre-defined numerical range division rules for each node, and input as observation evidence into the observation nodes of the corresponding time slices in the dynamic Bayesian network model.

[0035] Within the dynamic Bayesian network model, posterior probability updates are performed on icing state nodes and latent variable nodes based on the current meteorological monitoring data and historical icing observation data. Specifically, firstly, using the discrete intervals of the observed data for ambient temperature, supercooled water droplets, and wind speed within the current time slice as known conditions, the conditional probability values ​​for each discrete interval of the icing state node under the given set of parent node discrete interval combinations are retrieved from the conditional probability table of the icing state node. The conditional probability values ​​for each discrete interval of the icing state node are multiplied by the probability of their corresponding parent node combination appearing in the historical data, resulting in a set of unnormalized joint probability values. These joint probability values ​​are then summed, and the posterior probability value for each discrete interval is obtained by dividing the joint probability value of each discrete interval by this sum. Similarly, when calculating the posterior probability of latent variable nodes, the conditional probability values ​​for each discrete interval of the latent variable node under the current observation evidence are retrieved from the latent variable node conditional probability table. Each discrete interval's conditional probability value is multiplied by the initial probability of the latent variable node without observation evidence, and the resulting unnormalized joint probability values ​​are summed. Then, the joint probability value for each discrete interval is divided by this sum to obtain the posterior probability value for each discrete interval of the latent variable node. After completing the above calculations, the temporal probability distributions of the icing state nodes and latent variable nodes in the current time slice are obtained.

[0036] Subsequently, based on the posterior probability distributions of the icing state nodes obtained from the current time slice and the posterior probability distributions of the latent variable nodes, combined with the state transition dependencies between adjacent time slices pre-established in the dynamic Bayesian network model, a recursive prediction of the icing state for future time periods is performed. Specifically, the probability values ​​for transitioning from each discrete interval of the previous time step to each discrete interval of the current time step, recorded in the transition probability table of the icing state nodes, and the probability values ​​for transitioning from each discrete interval of the previous time step to each discrete interval of the current time step, recorded in the transition probability table of the latent variable nodes, are retrieved. The posterior probability values ​​of each discrete interval of the icing state nodes in the current time slice are combined with the posterior probability values ​​of each discrete interval of the latent variable nodes to form the joint probability value of the combination of discrete intervals at the current time step. For the icing state nodes in the next time slice, the predicted probability value of a certain discrete interval is obtained by multiplying the joint probability value of all discrete interval combinations at the current time step by the transition probability value of the discrete interval of the icing state of the previous time step corresponding to that combination to the target icing state discrete interval, and then summing all the product results. For the latent variable nodes in the next time slice, the same method is used: the joint probability value of all discrete interval combinations at the current time is multiplied by the transition probability value of the latent variable discrete interval corresponding to that combination from the previous time-series latent variable discrete interval to the target latent variable discrete interval, and all product results are summed. After the calculation is completed, the predicted probability values ​​of each discrete interval of the icing state node and latent variable node in the next time slice are obtained.

[0037] Finally, the calculated predicted probability values ​​for the next time slice are normalized. The predicted probability values ​​for each discrete interval of the icing state node are summed to obtain a total. Then, the predicted probability value for each discrete interval is divided by this total, so that the sum of the probability values ​​for each discrete interval is 1. This is taken as the predicted probability distribution result for this time slice. The same normalization operation is performed on the latent variable nodes. The normalized predicted probability distribution of the icing state node and the predicted probability distribution of the latent variable node are recombined into a joint probability distribution, which serves as the starting distribution for the next round of recursion. The above multiplication and summation calculation and normalization process of the transition probability values ​​are repeated to recursively calculate the predicted probability distribution of multiple subsequent time slices. During the recursion process, the predicted probability distribution of the latent variable node and the predicted probability distribution of the icing state node are jointly updated, and finally, the predicted probability distribution of each node in multiple time slices in the future period is established.

[0038] Furthermore, in the method provided in the application embodiments, inputting the time-series dataset into the dynamic Bayesian network model further includes: Meteorological monitoring data and historical icing observation data at the current moment are input as observation evidence into the corresponding observation nodes in the dynamic Bayesian network model. Combining the conditional probability table and prior probabilities of each observation node, Bayesian update calculations are performed on the icing state nodes and latent variable nodes to obtain the posterior probability distribution of each node at the current moment. Based on the posterior probability distribution, the joint probability distribution of the icing state nodes and latent variable nodes is extracted, and this joint probability distribution is used as the prior input for the next time slice. Utilizing the state transition dependencies between adjacent time slices, recursive calculations are performed on the prior input to generate the predicted probability distribution in the next time slice. The predicted probability distribution is normalized, and recursive iteration continues. During the recursive iteration process, the probability distributions of the latent variable nodes and the probability distributions of the icing state nodes are jointly updated to establish the predicted probability distribution of each node for future periods.

[0039] In this embodiment of the application, when the meteorological monitoring data at the current moment and the historical icing observation data are input as observation evidence into the corresponding observation node in the dynamic Bayesian network model, the meteorological monitoring data such as ambient temperature, supercooled water droplet size distribution, ambient humidity, atmospheric height, and wind speed, as well as the historical icing observation data such as icing thickness, conductor temperature, current, and sag, are first converted into corresponding discrete interval values ​​according to the discrete interval division rules preset by each observation node, and these values ​​are assigned to the corresponding observation node in the current time slice in the dynamic Bayesian network model.

[0040] Next, when performing Bayesian update calculations on the icing-state nodes and latent variable nodes using the conditional probability tables and prior probabilities of each observation node, the conditional probability values ​​for each discrete interval of the icing-state nodes under the current observation node value combination are retrieved from the conditional probability table of the icing-state nodes, and the conditional probability values ​​for each discrete interval of the latent variable nodes under the current observation node value combination are retrieved from the conditional probability table of the latent variable nodes. These conditional probability values ​​are multiplied by their corresponding prior probabilities, i.e., the initial probability distributions of the icing-state nodes and latent variable nodes before the introduction of current observation evidence, to obtain unnormalized joint probability values. All unnormalized joint probability values ​​are summed to obtain a total, and then the joint probability value of each discrete interval is divided by this total to obtain the posterior probability values ​​for each discrete interval of the icing-state nodes and the latent variable nodes, thus obtaining the posterior probability distribution of each node at the current time.

[0041] When extracting the joint probability distribution of icing state nodes and latent variable nodes based on the posterior probability distribution, the posterior probability values ​​of each discrete interval of the icing state node are combined with the posterior probability values ​​of each discrete interval of the latent variable node. That is, the probability value of each discrete interval of the icing state node is multiplied by the probability value of each discrete interval of the latent variable node to form the joint probability value under all possible state combinations, which constitutes the joint probability distribution of the icing state node and the latent variable node. This joint probability distribution is then used as the prior input for the next time slice.

[0042] When performing recursive calculations on the prior input using the state transition dependencies between adjacent time slices, the probability values ​​for transitioning from each discrete interval of the previous time step to each discrete interval of the current time step, recorded in the transition probability table of the icing state node, and the probability values ​​for transitioning from each discrete interval of the previous time step to each discrete interval of the current time step, recorded in the transition probability table of the latent variable node, are retrieved. The probability values ​​of each state combination in the joint probability distribution, which serves as the prior input, are multiplied by the transition probability values ​​for transitioning from the icing state discrete interval corresponding to that combination to the target icing state discrete interval, and by the transition probability values ​​for transitioning from the latent variable discrete interval corresponding to that combination to the target latent variable discrete interval. All product results are then summed according to the target state combination to obtain the unnormalized predicted probability values ​​of each state combination of the icing state node and the latent variable node in the next time slice, generating the predicted probability distribution in the next time slice.

[0043] When normalizing the predicted probability distribution, the predicted probability values ​​of each discrete interval of the icing state node are summed to obtain a total. Then, the predicted probability value of each discrete interval is divided by this total, making the sum of the probability values ​​of each discrete interval equal to 1. The same normalization operation is performed on the latent variable nodes. The normalized predicted probability distribution of the icing state node and the predicted probability distribution of the latent variable node are recombined into a joint probability distribution, which serves as the prior input for the next round of recursion, and the recursive iteration continues. During the recursive iteration, the probability distribution of the latent variable node and the probability distribution of the icing state node are jointly updated. That is, in each iteration, the two are used to perform state transition calculations and normalization in the form of a joint probability distribution, maintaining the consistency of their correlation in the recursive process. The predicted probability distributions of multiple subsequent time slices are calculated recursively, and finally, the predicted probability distributions of each node in multiple time slices in the future period are established.

[0044] Step S500: Perform probability propagation of the dynamic Bayesian network model according to the predicted probability distribution, and configure icing fault risk warning information using the probability propagation results.

[0045] In this embodiment, firstly, probability propagation is performed in the dynamic Bayesian network model along a predetermined causal dependency based on the predicted probability distribution, mapping the icing state and latent variable state to fault risk nodes to obtain the fault risk probability distribution of the target transmission line in the corresponding time period; then, based on the fault risk probability distribution and the predicted probability distribution, the joint evolution relationship between the icing state and the fault risk is extracted to generate the icing fault development path; finally, the fault risk probability distribution and the icing fault development path are used as the probability propagation results to configure icing fault risk early warning information.

[0046] Furthermore, in the method provided in the application embodiments, the probability propagation of the dynamic Bayesian network model is performed according to the predicted probability distribution, and the icing fault risk warning information is configured using the probability propagation results, which further includes: Based on the predicted probability distribution, probability propagation is performed along the established causal dependencies in the dynamic Bayesian network model to map the icing state and latent variable state to fault risk nodes, thereby obtaining the fault risk probability distribution of the target transmission line within the corresponding time period. Based on the fault risk probability distribution and the predicted probability distribution, the joint evolution relationship between the icing state and the fault risk is extracted to generate the icing fault development path. The fault risk probability distribution and the icing fault development path are used as the probability propagation results to configure icing fault risk early warning information.

[0047] In this embodiment, when performing probability propagation along a predetermined causal dependency in a dynamic Bayesian network model based on the predicted probability distribution, the conditional probability table of the fault risk node is first retrieved from the dynamic Bayesian network model. This conditional probability table records the conditional probability values ​​corresponding to each discrete interval of the fault risk node under different combinations of discrete intervals of the icing state node and discrete intervals of the latent variable node. Using the predicted probability distributions of the current time slice and future time slices as input, the predicted probability values ​​of each discrete interval of the icing state node are combined with the predicted probability values ​​of each discrete interval of the latent variable node to form a joint probability value under all possible state combinations. Then, each set of joint probability values ​​is multiplied by the conditional probability value of each discrete interval of the fault risk node under that combination condition, and all product results are accumulated according to the discrete interval of the fault risk node to obtain the probability value of each discrete interval of the fault risk node. This completes the probability propagation process of mapping the icing state and latent variable state to the fault risk node, thereby obtaining the fault risk probability distribution of the target transmission line in each time slice within the corresponding time period.

[0048] Next, when extracting the joint evolution relationship between icing state and fault risk based on the fault risk probability distribution and the predicted probability distribution, the probability values ​​of each discrete interval of the fault risk node and the predicted probability values ​​of each discrete interval of the icing state node are combined to form a joint probability distribution of icing state and fault risk under all possible state combinations. This joint probability distribution describes the probability of icing state and fault risk occurring simultaneously in different time slices. The state combination with the largest joint probability value is extracted slice by slice along the time axis, and the dominant state combination of icing state and fault risk in each time slice is recorded. The dominant state combinations of adjacent time slices are connected sequentially according to time to form a sequence reflecting the coordinated change trend of icing state and fault risk over time, which is the joint evolution relationship between icing state and fault risk. The state combination sequence in the above joint evolution relationship is organized in chronological order, with the time axis as the horizontal axis and the state combination of icing state and fault risk as the vertical axis, to construct the icing fault development path.

[0049] Finally, the fault risk probability distribution and icing fault development path are used as the probability propagation results to configure icing fault risk warning information. In this process, based on the fault risk probability distribution, the fault risk of each node of the target transmission line is classified according to a preset risk probability threshold range to obtain corresponding risk level identifiers. Then, the risk level identifiers are associated with the time evolution information in the icing fault development path to determine the occurrence time interval corresponding to each risk level. Finally, icing fault risk warning information is configured according to the risk level identifiers and occurrence time intervals.

[0050] Furthermore, in the method provided in the application embodiments, the failure risk probability distribution and icing failure development path are used as probability propagation results to configure icing failure risk early warning information, which further includes: Based on the fault risk probability distribution, the fault risk of each node of the target transmission line is classified according to the preset risk probability threshold range to obtain the corresponding risk level identifier; the risk level identifier is associated with the time evolution information in the icing fault development path to determine the occurrence time interval corresponding to each risk level; and icing fault risk warning information is configured according to the risk level identifier and the occurrence time interval.

[0051] In this embodiment, when classifying the fault risk of each node of the target transmission line according to a preset risk probability threshold range based on the fault risk probability distribution, the probability values ​​of each discrete interval of the fault risk node in each time slice, i.e., the fault risk probability distribution, are first obtained from the dynamic Bayesian network model. The preset risk probability threshold range is divided into a range of 0 to 0.3 corresponding to low risk level, a range of 0.3 to 0.7 corresponding to medium risk level, and a range of 0.7 to 1.0 corresponding to high risk level. For each time slice, the discrete interval with the largest probability value of the fault risk node is extracted, and the probability value corresponding to this discrete interval is compared with the above three preset intervals to determine which interval the probability value falls into, thereby marking the time slice as the corresponding risk level identifier.

[0052] Next, when associating risk level identifiers with the temporal evolution information in the icing fault development path, the first step is to retrieve the state combinations of icing status and fault risk recorded for each time slice in the icing fault development path. The risk level identifiers for each time slice are then matched chronologically with the corresponding state combinations in the icing fault development path, ensuring that each time slice simultaneously possesses both a risk level identifier and information on the state combinations of icing status and fault risk. Changes in risk level identifiers are examined along the time axis, recording the critical time points where these identifiers change, including the moments when they change from low to medium risk, from medium to high risk, and from high to medium or low risk. Based on these critical time points, the continuous time period covered by each risk level identifier is determined, which is the occurrence time interval corresponding to each risk level.

[0053] Finally, when configuring icing fault risk warning information based on risk level identifiers and occurrence time intervals, each risk level identifier is combined with its corresponding occurrence time interval and arranged into warning information entries in chronological order. Each entry includes a risk level identifier and the expected start and end times for that risk level. The arranged warning information entries are then output as icing fault risk warning information.

[0054] Furthermore, in the method provided in the application embodiments, configuring fault risk warning information based on the risk level identifier and the occurrence time interval also includes: Based on the temporal evolution of icing state in the icing fault development path, the icing state change trend is extracted, and the extracted result is matched with the fault risk probability distribution in the corresponding time period to obtain risk change trend information; the fault risk warning information configured according to the risk change trend information is dynamically updated.

[0055] In this embodiment, when extracting the trend of icing state changes based on the temporal evolution of icing state in the icing fault development path, the discrete interval sequence of icing state nodes recorded in each time slice is first retrieved. This sequence arranges the dominant discrete interval of icing state in each time slice in chronological order. Starting from the second time slice, the discrete interval of icing state in the current time slice is compared with the discrete interval of icing state in the previous time slice to determine the numerical relationship between the two. If the value of the current discrete interval is greater than that of the previous discrete interval, it is marked as an upward trend; if it is less, it is marked as a downward trend; if they are equal, it is marked as a flat trend. The change marks of each time slice are arranged in chronological order to form a sequence of icing state change trends.

[0056] Next, the extracted results are matched with the fault risk probability distribution within the corresponding time period. In this process, the probability values ​​of each discrete interval of the fault risk node recorded in the fault risk probability distribution for each time slice are first retrieved. For each time slice, the discrete interval with the highest probability value of the fault risk node is extracted as the dominant risk level for that time slice. The change markers in the icing state change trend sequence are aligned with the dominant risk level of the corresponding time slice, ensuring that each time slice contains both the icing state change markers and the dominant risk level. For each consecutive time interval, the correspondence between the icing state change markers and the dominant risk level within that interval is statistically analyzed. If the icing state shows a continuous upward trend and the dominant risk level increases synchronously, the risk change trend corresponding to that time interval is marked as an upward risk trend; if the icing state shows a continuous downward trend and the dominant risk level decreases synchronously, it is marked as a downward risk trend; if the icing state change trend and the dominant risk level change have no obvious correlation, it is marked as a risk fluctuation trend. The above marking results are organized by time interval to obtain the risk change trend information.

[0057] Finally, when dynamically updating the configured fault risk warning information based on risk change trend information, the first step is to retrieve the already configured fault risk warning information, which includes risk level identifiers and their corresponding occurrence time intervals. The time intervals of risk increase trends marked in the risk change trend information are overlaid and compared with the occurrence time intervals in the fault risk warning information. For warning information within the risk increase trend time interval, a message indicating that the risk level may occur earlier is added; for warning information within the risk decrease trend time interval, a message indicating that the risk level may be delayed or reduced is added; for warning information within the risk fluctuation trend time interval, a message indicating that continuous monitoring of risk changes is required is added. The updated warning information replaces the corresponding entries in the original warning information, forming the dynamically updated icing fault risk warning information.

[0058] Furthermore, the method provided in the application embodiments, which uses probability propagation results to configure icing fault risk warning information, also includes: Configure a visual early warning signal for the target transmission line based on the icing fault risk early warning information; and use the visual early warning signal to perform visual early warning dispatch management.

[0059] In this embodiment, when configuring the visual warning signal for the target transmission line based on the icing fault risk warning information, the time interval corresponding to the high-risk level included in the icing fault risk warning information is configured as a red flashing graphic, and the red flashing graphic is superimposed on the corresponding section of the transmission line electronic map; the time interval corresponding to the medium-risk level included in the icing fault risk warning information is configured as a yellow solid-light graphic, and the yellow solid-light graphic is superimposed on the corresponding section of the transmission line electronic map; the time interval corresponding to the low-risk level included in the icing fault risk warning information is configured as a green solid-light graphic, and the green solid-light graphic is superimposed on the corresponding section of the transmission line electronic map, thus completing the configuration of the visual warning signal.

[0060] Next, when using the visual early warning signal to perform visual early warning dispatch management, the risk level and its corresponding time interval are determined according to the current system time. The corresponding graphic style is retrieved from the configured visual early warning signal and displayed on the corresponding section of the transmission line electronic map to complete the visual early warning dispatch management.

[0061] Furthermore, the method provided in the application embodiments also includes: The icing fault risk warning information is recorded, a warning dispatch database is established, and the warning dispatch database is encrypted and stored.

[0062] In this embodiment, when recording icing fault risk warning information, the risk level identifier, occurrence time interval, and line segment number of the target transmission line are first extracted from the icing fault risk warning information. This information is then structured according to a preset field format to form a single warning record data entry. When writing the single warning record data to a database for storing warning records, a connection is made to a pre-deployed database, and a warning record data table is created in this database. This data table includes a risk level identifier field, a start time field, an end time field, a line segment number field, and a record timestamp field. The structured warning record data is then inserted into this data table field by field, completing the establishment of the warning record database and the writing of the data.

[0063] Next, when performing encrypted storage on the early warning record database, a symmetric encryption algorithm is used to encrypt the storage file of the early warning record database. During the encryption process, the encryption key is stored in a separate key storage location. After encryption, the database file is stored in ciphertext form on the server disk. The database file is only decrypted and read by calling the decryption key through the key storage location when querying early warning records, thus realizing the encrypted storage of the early warning record database.

[0064] To further verify the application effect of the Bayesian network-based transmission line icing fault prediction method in fault risk early warning, meteorological monitoring data, conductor operation status data, and historical icing monitoring data collected over 45 consecutive days during the high-incidence period of icing along three sections of the target transmission line were selected as verification samples, forming a total of 2160 time-series samples after spatiotemporal alignment and feature standardization. Some samples were used for the statistical construction of conditional probability tables and state transition dependencies, while the remaining samples served as verification data. The verification object was whether the target transmission line would experience medium-risk or higher icing fault risks in the future. Comparative analyses were conducted using empirical threshold judgment methods, static Bayesian network judgment methods, and the dynamic Bayesian network model described in the embodiments of this application.

[0065] The verification results show that the empirical threshold judgment method is consistent with the actual records at 78.2%, with an error warning rate of 18.6% and an average early warning time of 1.8 hours; the static Bayesian network judgment method is consistent with the actual records at 84.6%, with an error warning rate of 13.4% and an average early warning time of 2.7 hours; and the dynamic Bayesian network model described in this application embodiment is consistent with the actual records at 92.1%, with an error warning rate of 7.8% and an average early warning time of 4.6 hours.

[0066] like Figure 2 As shown, Figure 2The comparison results of different judgment methods under the verification data are shown. The bars represent the degree of consistency with actual records and the proportion of false warnings, respectively, while the line represents the average advance warning time. Figure 2 As can be seen, the dynamic Bayesian network model described in this application embodiment has the highest consistency with actual records, reaching 92.1%, which is significantly better than the empirical threshold judgment method and the static Bayesian network judgment method; it has the lowest false alarm rate, at only 7.8%, lower than the other two control methods; and it has the longest average early warning time, reaching 4.6 hours, indicating that this application embodiment can output effective early warning information earlier. The above results show that, through multi-source monitoring data fusion, mechanistic constraint rule set construction, dynamic Bayesian network probability propagation, and fault risk early warning information configuration, this application embodiment not only improves the accuracy of icing fault risk identification but also reduces the probability of false alarms and enhances the early warning lead time, thereby verifying the effectiveness and reliability of the method in practical applications.

[0067] In summary, the embodiments of this application have at least the following technical effects: This application acquires multi-source monitoring data along the target transmission line, including meteorological monitoring data, conductor operating status data, and historical icing monitoring data. It performs spatiotemporal alignment and feature standardization on the multi-source monitoring data to construct a unified time-series dataset. Based on the icing formation mechanism, it constructs a set of mechanism constraint rules, including temperature phase transition constraints, supercooled water droplet existence constraints, and wind speed influence on icing adhesion efficiency constraints. Based on the set of mechanism constraint rules, it constructs an initial Bayesian network structure and, after establishing the state transition dependencies between adjacent time slices, generates... A dynamic Bayesian network model is used. The time-series dataset is input into the dynamic Bayesian network model. Within the model, based on current meteorological monitoring data and historical icing observation data, posterior probability updates are performed on icing state nodes and latent variable nodes to obtain the time-series probability distribution of each node. Based on the time-series probability distribution and state transition dependencies, recursive predictions of icing states for future periods are performed to establish predicted probability distributions for each node in the future period. Probability propagation of the dynamic Bayesian network model is performed based on the predicted probability distributions, and icing fault risk warning information is configured using the probability propagation results. This invention solves the technical problem of inaccurate prediction of icing fault occurrence time and risk distribution in existing technologies. Through probability propagation and time-series data updates using a dynamic Bayesian network model, it achieves improved accuracy in icing fault prediction and provides early warning.

[0068] Example 2 is based on the same inventive concept as the transmission line icing fault prediction method based on Bayesian networks in the previous examples, such as... Figure 3As shown, this application provides a transmission line icing fault prediction system based on Bayesian networks. The system and method embodiments in this application are based on the same inventive concept. The system includes: Data acquisition module 11 is used to acquire multi-source monitoring data along the target transmission line, including meteorological monitoring data, conductor operating status data, and historical icing monitoring data. The multi-source monitoring data undergoes spatiotemporal alignment and feature standardization to construct a unified time-series dataset. Constraint rule set construction module 12 is used to construct a mechanism constraint rule set based on the icing formation mechanism. This mechanism constraint rule set includes constraints on temperature phase transition, supercooled water droplet existence, and the influence of wind speed on icing adhesion efficiency. Model generation module 13 is used to construct an initial Bayesian network structure based on the mechanism constraint rule set and establish state transition dependencies between adjacent time slices. After that, a dynamic Bayesian network model is generated; the prediction module 14 is used to input the time series dataset into the dynamic Bayesian network model, and within the dynamic Bayesian network model, based on the meteorological monitoring data at the current time and the historical icing observation data, to perform posterior probability updates on the icing state nodes and latent variable nodes, to obtain the time series probability distribution of each node, and to perform recursive prediction of the icing state in future periods based on the time series probability distribution and state transition dependencies, and to establish the predicted probability distribution of each node in future periods; the early warning information configuration module 15 is used to perform probability propagation of the dynamic Bayesian network model according to the predicted probability distribution, and to configure icing fault risk early warning information using the probability propagation results.

[0069] Furthermore, the system is also used to implement the following functions: Based on the predicted probability distribution, probability propagation is performed along the established causal dependencies in the dynamic Bayesian network model to map the icing state and latent variable state to fault risk nodes, thereby obtaining the fault risk probability distribution of the target transmission line within the corresponding time period. Based on the fault risk probability distribution and the predicted probability distribution, the joint evolution relationship between the icing state and the fault risk is extracted to generate the icing fault development path. The fault risk probability distribution and the icing fault development path are used as the probability propagation results to configure icing fault risk early warning information.

[0070] Furthermore, the system is also used to implement the following functions: Based on the fault risk probability distribution, the fault risk of each node of the target transmission line is classified according to the preset risk probability threshold range to obtain the corresponding risk level identifier; the risk level identifier is associated with the time evolution information in the icing fault development path to determine the occurrence time interval corresponding to each risk level; and icing fault risk warning information is configured according to the risk level identifier and the occurrence time interval.

[0071] Furthermore, the system is also used to implement the following functions: Based on the temporal evolution of icing state in the icing fault development path, the icing state change trend is extracted, and the extracted result is matched with the fault risk probability distribution in the corresponding time period to obtain risk change trend information; the fault risk warning information configured according to the risk change trend information is dynamically updated.

[0072] Furthermore, the system is also used to implement the following functions: Based on the temperature phase transition constraints, supercooled water droplet existence constraints, and wind speed influence on icing adhesion efficiency constraints in the mechanistic constraint rule set, icing state nodes, conductor operation state nodes, and latent variable nodes are generated, and a conditional probability table is established for each node. This table is constructed based on historical icing monitoring data and conductor operation state data statistics. A local causal topology structure is established for each node according to the causal relationships defined in the mechanistic constraint rule set, forming an initial Bayesian network structure. This initial Bayesian network structure is replicated along the time axis into multiple time slice nodes. Based on historical icing evolution data and the probability distribution of the local causal topology, state transition dependencies between adjacent time slices are established to form a dynamic Bayesian network skeleton. In each time slice, conditional dependencies are established between unobservable latent variable nodes and their corresponding icing state nodes and conductor operation state nodes, and the prior probabilities of the latent variable nodes are initialized based on historical icing monitoring data. The structural integrity and causal consistency of the dynamic Bayesian network skeleton are verified, generating a dynamic Bayesian network model.

[0073] Furthermore, the system is also used to implement the following functions: Meteorological monitoring data and historical icing observation data at the current moment are input as observation evidence into the corresponding observation nodes in the dynamic Bayesian network model. Combining the conditional probability table and prior probabilities of each observation node, Bayesian update calculations are performed on the icing state nodes and latent variable nodes to obtain the posterior probability distribution of each node at the current moment. Based on the posterior probability distribution, the joint probability distribution of the icing state nodes and latent variable nodes is extracted, and this joint probability distribution is used as the prior input for the next time slice. Utilizing the state transition dependencies between adjacent time slices, recursive calculations are performed on the prior input to generate the predicted probability distribution in the next time slice. The predicted probability distribution is normalized, and recursive iteration continues. During the recursive iteration process, the probability distributions of the latent variable nodes and the probability distributions of the icing state nodes are jointly updated to establish the predicted probability distribution of each node for future periods.

[0074] Furthermore, the system is also used to implement the following functions: Configure a visual early warning signal for the target transmission line based on the icing fault risk early warning information; and use the visual early warning signal to perform visual early warning dispatch management.

[0075] Furthermore, the system is also used to implement the following functions: The icing fault risk warning information is recorded, a warning dispatch database is established, and the warning dispatch database is encrypted and stored.

[0076] Furthermore, the system is also used to implement the following functions: Based on the sampling time interval of multi-source monitoring data, meteorological monitoring data, conductor operation status data and historical icing monitoring data are processed with unified time scale, and a unified time series is constructed through interpolation compensation; based on the spatial distribution of the target transmission line, spatial mapping is performed on the data of different monitoring nodes, and segmented data representation is formed according to the spatial mapping results; spatiotemporal alignment and feature standardization are performed using the unified time series and segmented data representation.

[0077] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0078] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for predicting icing faults in transmission lines based on Bayesian networks, characterized in that, The method includes: Acquire multi-source monitoring data along the target transmission line, including meteorological monitoring data, conductor operation status data, and historical icing monitoring data. Perform spatiotemporal alignment and feature standardization processing on the multi-source monitoring data to construct a unified time-series dataset. Based on the ice formation mechanism, a set of mechanism constraint rules is constructed, which includes temperature phase change constraints, supercooled water droplet existence conditions constraints, and wind speed influence on ice adhesion efficiency constraints. An initial Bayesian network structure is constructed based on the aforementioned mechanism constraint rule set, and a dynamic Bayesian network model is generated after establishing the state transition dependency relationship between adjacent time slices. The time series dataset is input into the dynamic Bayesian network model. Within the dynamic Bayesian network model, based on the meteorological monitoring data at the current moment and the historical icing observation data, the posterior probability update is performed on the icing state nodes and latent variable nodes to obtain the time series probability distribution of each node. Based on the time series probability distribution and state transition dependency, the icing state of the future period is recursively predicted to establish the predicted probability distribution of each node in the future period. Based on the predicted probability distribution, perform probability propagation of the dynamic Bayesian network model, and use the probability propagation results to configure icing fault risk warning information.

2. The method for predicting icing faults in transmission lines based on Bayesian networks as described in claim 1, characterized in that, Based on the predicted probability distribution, a probability propagation of a dynamic Bayesian network model is performed, and the probability propagation results are used to configure icing fault risk warning information, including: Based on the predicted probability distribution, probability propagation is performed in the dynamic Bayesian network model along the established causal dependencies to map the icing state and latent variable state to the fault risk node, thereby obtaining the fault risk probability distribution of the target transmission line in the corresponding time period. Based on the aforementioned fault risk probability distribution and predicted probability distribution, the joint evolution relationship between icing state and fault risk is extracted to generate an icing fault development path. The probability distribution of the fault risk and the development path of the icing fault are used as the results of probability propagation to configure icing fault risk early warning information.

3. The method for predicting icing faults in transmission lines based on Bayesian networks as described in claim 2, characterized in that, Using the aforementioned fault risk probability distribution and icing fault development path as the probability propagation results, icing fault risk early warning information is configured, including: Based on the fault risk probability distribution, the fault risk of each node of the target transmission line is classified according to the preset risk probability threshold range to obtain the corresponding risk level identifier. The risk level identifier is associated with the time evolution information in the icing fault development path to determine the occurrence time interval corresponding to each risk level; Configure icing fault risk warning information according to the risk level identifier and the time interval of occurrence.

4. The method for predicting icing faults in transmission lines based on Bayesian networks as described in claim 3, characterized in that, Based on the aforementioned risk level identifier and occurrence time interval, the configuration of fault risk warning information also includes: Based on the temporal evolution of icing state in the icing fault development path, the icing state change trend is extracted, and the extracted results are matched with the fault risk probability distribution in the corresponding time period to obtain risk change trend information. The fault risk warning information configured based on the aforementioned risk change trend information is dynamically updated.

5. The method for predicting icing faults in transmission lines based on Bayesian networks as described in claim 1, characterized in that, An initial Bayesian network structure is constructed based on the aforementioned mechanistic constraint rule set. After establishing the state transition dependencies between adjacent time slices, a dynamic Bayesian network model is generated, including: Based on the temperature phase transition constraints, supercooled water droplet existence constraints, and wind speed influence on icing adhesion efficiency constraints in the mechanism constraint rule set, icing state nodes, conductor operation state nodes, and latent variable nodes are generated, and a conditional probability table is established for each node. The conditional probability table is constructed based on historical icing monitoring data and conductor operation state data statistics. Establish a local causal topology structure for each node according to the causal relationships defined in the mechanism constraint rules set, and form the initial Bayesian network structure. The initial Bayesian network structure is replicated along the time axis into multiple time slice nodes. Based on the historical icing evolution data and the probability distribution of local causal topology, state transition dependencies between adjacent time slices are established to form a dynamic Bayesian network skeleton. In each time slice, conditional dependencies are established between the unobservable latent variable nodes and the corresponding icing state nodes and conductor operation state nodes, and the prior probabilities of the latent variable nodes are initialized based on historical icing monitoring data. The structural integrity and causal consistency of the dynamic Bayesian network skeleton are verified, and a dynamic Bayesian network model is generated.

6. The method for predicting icing faults in transmission lines based on Bayesian networks as described in claim 1, characterized in that, Inputting the time-series dataset into the dynamic Bayesian network model further includes: The current meteorological monitoring data and historical icing observation data are used as observation evidence and input into the corresponding observation nodes in the dynamic Bayesian network model. Combined with the conditional probability table and prior probability of each observation node, Bayesian update calculation is performed on the icing state nodes and latent variable nodes to obtain the posterior probability distribution of each node at the current time. The joint probability distribution of the icing state node and the latent variable node is extracted based on the posterior probability distribution, and the joint probability distribution is used as the prior input for the next time slice. By utilizing the state transition dependencies between adjacent time slices, recursive calculations are performed on the prior input to generate the predicted probability distribution in the next time slice; The predicted probability distribution is normalized and recursively iterated. During the recursive iteration, the probability distribution of the latent variable nodes and the probability distribution of the icing state nodes are jointly updated to establish the predicted probability distribution of each node in the future time period.

7. The method for predicting icing faults in transmission lines based on Bayesian networks as described in claim 1, characterized in that, Configuring icing fault risk early warning information using probability propagation results also includes: Configure a visual early warning signal for the target transmission line based on the icing fault risk early warning information; The visualized early warning signal is used to perform visualized early warning issuance management.

8. The method for predicting icing faults in transmission lines based on Bayesian networks as described in claim 1, characterized in that, The icing fault risk warning information is recorded, a warning dispatch database is established, and the warning dispatch database is encrypted and stored.

9. The method for predicting icing faults in transmission lines based on Bayesian networks as described in claim 1, characterized in that, Perform spatiotemporal alignment and feature normalization processing on the multi-source monitoring data, including: Based on the sampling time interval of multi-source monitoring data, meteorological monitoring data, conductor operation status data and historical icing monitoring data are processed in a unified time scale, and a unified time series is constructed through interpolation compensation. Based on the spatial distribution of the target transmission line, spatial mapping is performed on the data of different monitoring nodes, and segmented data representation is formed based on the spatial mapping results; Spatiotemporal alignment and feature standardization are performed using unified time series and segmented data representation.

10. A transmission line icing fault prediction system based on Bayesian networks, characterized in that, The system is used to execute the transmission line icing fault prediction method based on Bayesian networks as described in any one of claims 1-9, and the system includes: The data acquisition module is used to acquire multi-source monitoring data along the target transmission line. The multi-source monitoring data includes meteorological monitoring data, conductor operation status data, and historical icing monitoring data. The module performs spatiotemporal alignment and feature standardization processing on the multi-source monitoring data to construct a unified time-series dataset. The constraint rule set construction module is used to construct a mechanism constraint rule set based on the ice formation mechanism. The mechanism constraint rule set includes temperature phase change constraints, supercooled water droplet existence condition constraints, and wind speed influence on ice adhesion efficiency constraints. The model generation module is used to construct an initial Bayesian network structure based on the set of mechanistic constraint rules, and generate a dynamic Bayesian network model after establishing the state transition dependency relationship between adjacent time slices. The prediction module is used to input the time series dataset into the dynamic Bayesian network model. Within the dynamic Bayesian network model, based on the meteorological monitoring data at the current moment and the historical icing observation data, the posterior probability update is performed on the icing state nodes and latent variable nodes to obtain the time series probability distribution of each node. Based on the time series probability distribution and state transition dependency, the icing state of the future period is recursively predicted to establish the predicted probability distribution of each node in the future period. The early warning information configuration module is used to perform probability propagation of the dynamic Bayesian network model based on the predicted probability distribution, and to configure icing fault risk early warning information using the probability propagation results.

Images