Meteorological big data-based insulated conductor icing risk prediction method and system

Abstract

The invention discloses an insulated conductor icing risk prediction method and system based on meteorological big data, and relates to the technical field of power system icing risk prediction and meteorological big data intelligent analysis, and the method comprises the steps: collecting multi-dimensional sensing data, and constructing a sliding time window; constructing an adversarial neural network and fusing LSTM and convolution to extract sequence features; and evaluating a risk value based on dynamic output risk probability calculation and weight aggregation. Through collaborative modeling of the GAN and the CNN / LSTM, extreme sample enhancement, noise suppression and time sequence dependence extraction are realized, and reliable risk output is maintained under the non-stationary climate, so that the effects of reducing missing report and false report and improving the extreme weather recognition rate are achieved. Through node weighted aggregation and dual-condition triggering, system-level risk quantification and executable control are achieved, and the method is used for being connected with an electric heating deicing and dispatching system, so that the effects of advanced intervention, energy-saving operation and control within a risk threshold value are achieved.

Description

Technical Field

[0001] This invention relates to the field of power system icing risk prediction and meteorological big data intelligent analysis technology, specifically a method and system for predicting icing risk of insulated conductors based on meteorological big data. Background Technology

[0002] Ice accumulation on insulated conductors is one of the core factors leading to power transmission line faults. Ice accumulation can trigger a series of unstable phenomena such as conductor galloping, ice flashover, and ice-shedding jumping, and in severe cases, it can even lead to conductor breakage and tower collapse, posing a significant threat to the safe and stable operation of the power system. Especially under extreme weather conditions, a break in an insulated conductor at any point can easily paralyze a local or even large-scale power grid. Therefore, conducting early warning systems for insulated conductor ice accumulation, deploying de-icing devices in advance, and scientifically predicting the development trend of ice accumulation are of great significance for ensuring the safe operation of the power grid and the efficiency of power dispatch. With the application of the Internet of Things, sensors, and meteorological big data, the power system is gradually introducing multi-source data fusion and intelligent processing methods; however, existing prediction models are still in their initial stages, and their overall accuracy and real-time performance need further improvement.

[0003] Existing methods for predicting icing risk primarily rely on physical quantity monitoring and empirical model extrapolation. For example, icing development is predicted by measuring physical indicators such as conductor tension, sag, and inclination, combined with single-point meteorological parameters. However, these methods have short prediction timelines, typically covering only a few hours into the future. Furthermore, data collection and analysis are highly dependent on manual intervention, making them susceptible to subjective biases and resulting in insufficient prediction stability. Simultaneously, physical computational models are computationally intensive, making real-time online prediction difficult. Static risk assessment methods also fail to adapt to the shifting trends of icing patterns under climate change, leading to insufficient long-term prediction accuracy. More critically, existing methods lack the ability to dynamically acquire data from multi-dimensional sensors and extract features through sliding time windows. They also fail to incorporate adversarial networks into the prediction structure to improve robustness and generalization, and cannot achieve dynamic weighted aggregation of node risks, making it difficult to output a global risk value at the system level. Therefore, a comprehensive icing risk prediction solution that combines meteorological big data, multi-source sensors, and deep learning models is currently lacking, necessitating the development of an innovative method to fill this technological gap. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that existing methods for predicting icing on insulated conductors rely on a single monitoring data source, which cannot reflect complex meteorological differences. The physical models have a large computational load, making it difficult to meet real-time prediction requirements. The machine learning models have poor noise resistance and lack a dynamic update mechanism. The problem is how to achieve dynamic prediction and global risk assessment of conductor icing risk based on multi-dimensional sensor data and intelligent algorithms.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a method for predicting the risk of icing on insulated conductors based on meteorological big data, comprising collecting multidimensional sensor data and constructing a sliding time window; constructing an adversarial neural network and fusing LSTM and convolution to extract sequence features; and evaluating the risk value based on dynamic output risk probability calculation and weight aggregation.

[0007] As a preferred embodiment of the method for predicting the risk of icing on insulated conductors based on meteorological big data as described in this invention, the following steps are included: collecting multi-dimensional sensor data, including temperature, humidity, wind speed, altitude, and liquid water content, through IoT environmental sensing nodes distributed along the insulated conductor, to form a time-series dataset; uniformly preprocessing the collected data, including unit conversion, missing value imputation, and normalization, to form a standardized meteorological input vector; and inputting the data at each time point into a sliding time window of fixed length L, denoted as:

[0008] W k ={x kS ,x kS+1 ,…,x kS+L-1}

[0009] Where S is the sliding step size, the window slides to replace the oldest S data points and add the newest S data points each time, forming a dynamic update mechanism, x kS+L-1 This represents the last data point in the current window, with a starting point of kS and an ending point of kS+L-1, where k is the index number of the sliding time window.

[0010] As a preferred embodiment of the method for predicting the risk of icing on insulated conductors based on meteorological big data as described in this invention, the construction of the sliding time window includes, from each sliding time window W k Key meteorological factors were extracted and indicators for risk characterization were constructed, including the ratio of low temperature duration T. low :

[0011]

[0012] Where I(·) is an indicator function used to determine whether the temperature is less than or equal to 0℃, and T t The temperature value at time t; the joint probability H under high humidity conditions. high Represented as:

[0013] H high =P(humidity ≥ 90% ∩ T ≤ 0℃)

[0014] Where P(·) is a probability function, the joint probability of high humidity conditions is used to determine whether high humidity co-occurs under low temperature conditions, as a prerequisite for freezing; the liquid water content is estimated as LWC:

[0015] LWC=α·R·v -1

[0016] Where α is the calibration coefficient, R is the precipitation, and v is the wind speed.

[0017] As a preferred embodiment of the method for predicting the risk of icing on insulated conductors based on meteorological big data as described in this invention, the construction of the adversarial neural network includes: employing a generative adversarial network (GAN) framework, containing generator and discriminator structures, and inputting historical meteorological time window samples W. k or latent variables The generator structure uses an LSTM or Transformer network to capture temporal dependencies and outputs the next potential time window:

[0018]

[0019] in, To synthesize the data for the (k+1)th time window output by the adversarial network generator, θ G Given the set of all parameters of the generator network, compare the generator's output samples with historical real samples. The input is fed into the discriminator, whose network structure includes a CNN to extract local features and outputs:

[0020] [D real ,R k ] = D(W k ;θ D )

[0021] Among them, D real ∈[0,1] represents the true probability, R k θ is the icing risk score calculated by the discriminator. D This is the parameter set of the discriminator.

[0022] As a preferred embodiment of the method for predicting the risk of icing on insulated conductors based on meteorological big data as described in this invention, the method of fusing LSTM and convolutional extraction of sequence features includes inputting meteorological data processed by a sliding time window into a neural network prediction model. The network structure includes a convolutional extraction layer (CNN) for extracting local meteorological change features.

[0023] H CNN=ReLU(Conv1D(X) t W conv ))

[0024] Among them, H CNN X represents the output features of the CNN layer. t Given the input sequence, d is the convolution kernel, k' is the kernel size, d is the input dimension, and m is the number of output channels. Dimensionality; Long Short-Term Memory (LSTM) network, used to model long-range dependencies in time series:

[0025] h t ,c t =LSTM(H CNN W lstm )

[0026] Among them, h t For the hidden state output of the LSTM network at time step t, c t W represents the memory cell state of the LSTM network at time step t. lstm The set of weight parameters for the LSTM network; the risk prediction output layer outputs the hidden state to the fully connected layer:

[0027] R t =σ(W f ·h t +b f )

[0028] Among them, the normalized risk value R is output. t ∈[0,1].

[0029] As a preferred embodiment of the method for predicting the risk of icing on insulated conductors based on meteorological big data as described in this invention, the dynamic output risk probability calculation includes calculating the risk value R for each time period predicted and output by the neural network model. t Convert to cell node risk value R i And set weights f according to the degree of influence of each node i in the power grid structure on the system. i Construct an aggregated risk model:

[0030]

[0031] Among them, S ice risk The total icing risk value for the entire power grid system is given by n, where n is the number of insulated conductors or tower nodes in the current system. The system dynamically distributes risk scores to different power generation control systems or de-icing equipment modules. When the total system risk value S... ice risk When the set threshold is exceeded, an alert will be automatically triggered and the next strategy calculation will proceed.

[0032] The risk value is recalculated every time a new data window is updated.

[0033] As a preferred embodiment of the method for predicting the risk of icing on insulated conductors based on meteorological big data as described in this invention, the weighted aggregation assessment of risk values ​​includes constructing a power supply adjustment module that interfaces with the power dispatching system and receiving S... ice risk The risk value is used as input; when the system determines that the current risk value is below the threshold, the current power supply method and the operating status of the electric heating de-icing equipment are maintained; when the system determines that the current risk value is above the threshold, and the future time window T... low >0.6 and H high When the value is greater than 0.4, the system will prioritize the start-up of the electric heating module in the power generation system, shorten the heating cycle and increase the power output.

[0034] Another objective of this invention is to provide a risk prediction system for icing of insulated conductors based on meteorological big data. As a preferred embodiment of the risk prediction system for icing of insulated conductors based on meteorological big data described in this invention, it includes: a data processing module, a feature extraction module, and a risk prediction module; the data processing module is used to collect multi-dimensional sensor data and construct a sliding time window; the feature extraction module is used to construct an adversarial neural network and fuse LSTM and convolution to extract sequence features; the risk prediction module is used to calculate the risk probability based on dynamic output and evaluate the risk value through weight aggregation.

[0035] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a method for predicting the risk of icing on insulated conductors based on meteorological big data.

[0036] A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of a method for predicting the risk of icing on insulated wires based on meteorological big data.

[0037] The beneficial effects of this invention are as follows: The insulated conductor icing risk prediction method based on meteorological big data provided by this invention achieves real-time quantitative characterization of micrometeorology and icing precursors through the construction of a sliding time window and three types of physically consistent features. This supports subsequent intelligent prediction and threshold control, thereby achieving data stabilization and interpretable input. Through collaborative modeling of GAN and CNN / LSTM, extreme sample enhancement, noise suppression, and temporal dependency extraction are achieved to maintain reliable risk output under non-stationary climate conditions, thereby reducing false alarms and improving the recognition rate of extreme weather. Through node weighted aggregation and dual-condition triggering, system-level risk quantification and executable control are achieved, which can be used to interface with electrothermal de-icing and scheduling systems, thereby achieving early intervention, energy-saving operation, and control within risk thresholds. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the 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.

[0039] Figure 1 The above is an overall flowchart of a method for predicting the risk of icing on insulated conductors based on meteorological big data, provided in Embodiment 1 of the present invention. Detailed Implementation

[0040] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0041] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for predicting the risk of icing on insulated conductors based on meteorological big data is provided, comprising:

[0042] S1: Collect multidimensional sensor data and construct a sliding time window.

[0043] Furthermore, the collection of multi-dimensional sensor data includes collecting temperature, humidity, wind speed, altitude, and liquid water content through IoT environmental sensing nodes distributed along the insulated conductor, forming a time-series dataset; the collected data undergoes unified preprocessing, including unit conversion, missing value imputation, and normalization, to form a standardized meteorological input vector; after data processing, the data at each time point is input into a sliding time window of fixed length L, denoted as:

[0044] W k ={x kS ,x kS+1 ,…,x kS+L-1}

[0045] Where S is the sliding step size, the window slides to replace the oldest S data points and add the newest S data points each time, forming a dynamic update mechanism, x kS+L-1 This represents the last data point in the current window, with a starting point of kS and an ending point of kS+L-1, where k is the index number of the sliding time window.

[0046] It should be noted that constructing a sliding time window includes, from each sliding time window W kKey meteorological factors were extracted and indicators for risk characterization were constructed, including the ratio of low temperature duration T. low :

[0047]

[0048] Where I(·) is an indicator function used to determine whether the temperature is less than or equal to 0℃, and T t The temperature value at time t; the joint probability H under high humidity conditions. high Represented as:

[0049] H high =P(humidity ≥ 90% ∩ T ≤ 0℃)

[0050] Where P(·) is a probability function, the joint probability of high humidity conditions is used to determine whether high humidity co-occurs under low temperature conditions, as a prerequisite for freezing; the liquid water content is estimated as LWC:

[0051] LWC=α·R·v -1

[0052] Where α is the calibration coefficient, R is the precipitation, and v is the wind speed.

[0053] It should also be noted that by distributing IoT sensors along insulated wires, environmental data such as temperature, humidity, wind speed, altitude, and liquid water content are acquired in real time. These raw data undergo unified preprocessing operations, including unit standardization, missing data imputation, and overall normalization. Subsequently, this processed data is organized chronologically into a fixed-length time window and continuously updated according to a set step size, thus forming a dynamic time-series input. Each time window not only includes the latest collected data but also retains a certain historical data segment. This structure helps to comprehensively reflect the persistence and changing trends of meteorological conditions over a short period. Through this mechanism, a time-series, standardized risk characteristic input is achieved from scattered meteorological observation points, providing stable, interpretable, and continuously updated data support for subsequent algorithm analysis. The ultimate benefit is the ability to reflect the dynamic characteristics of complex meteorological environments in real time, reducing prediction bias caused by errors at a single time point or a single sensor, and establishing a reliable data foundation for risk prediction.

[0054] S2: Construct an adversarial neural network and fuse LSTM and convolution to extract sequence features.

[0055] Furthermore, constructing an adversarial neural network involves employing a generative adversarial network (GAN) framework, including generator and discriminator structures, with historical meteorological time window samples W as input. k or latent variables The generator structure uses an LSTM or Transformer network to capture temporal dependencies and outputs the next potential time window:

[0056]

[0057] in, To synthesize the data for the (k+1)th time window output by the adversarial network generator, θ G Given the set of all parameters of the generator network, compare the generator's output samples with historical real samples. The input is fed into the discriminator, whose network structure includes a CNN to extract local features and outputs:

[0058] [D real ,R k ] = D(W k ;θ D )

[0059] Among them, D real ∈[0,1] represents the true probability, R k θ is the icing risk score calculated by the discriminator. D This is the parameter set of the discriminator.

[0060] It should be noted that the fusion of LSTM and convolutional feature extraction involves inputting meteorological data processed by a sliding time window into a neural network prediction model. The network structure includes a convolutional extraction layer (CNN) used to extract local meteorological change features.

[0061] H CNN =ReLU(Conv1D(X) t W conv ))

[0062] Among them, H CNN X represents the output features of the CNN layer. t Given the input sequence, d is the convolution kernel, k' is the kernel size, d is the input dimension, and m is the number of output channels. Dimensionality; Long Short-Term Memory (LSTM) network, used to model long-range dependencies in time series:

[0063] h t ,c t =LSTM(H CNN W lstm )

[0064] Among them, h t For the hidden state output of the LSTM network at time step t, c t W represents the memory cell state of the LSTM network at time step t. lstmThe set of weight parameters for the LSTM network; the risk prediction output layer outputs the hidden state to the fully connected layer:

[0065] R t =σ(W f ·h t +b f )

[0066] Among them, the normalized risk value R is output. t ∈[0,1].

[0067] It should also be noted that after acquiring dynamic time window data, an adversarial neural network architecture is introduced to enhance the model's robustness and data coverage. The generator module simulates and supplements sample data under extreme weather conditions to compensate for scarce scenarios that are difficult to collect in actual observations; the discriminator module, while distinguishing between generated and real data, directly outputs a score for icing risk, achieving integrated data authenticity verification and risk quantification. Simultaneously, the model structure combines convolutional neural networks and long short-term memory networks: convolutional networks effectively extract short-term local meteorological change features, such as sudden temperature drops or sharp increases in humidity; long short-term memory networks are responsible for capturing long-term temporal dependencies, such as sustained low temperatures or multi-day accumulated humidity conditions. The combination of these two allows the model to consider both short-term anomalies and long-term trends, maintaining predictive stability under different meteorological environments. Through this mechanism, the model ultimately achieves the effect of ensuring prediction reliability and generalization ability even under data scarcity and noise interference, significantly reducing the incidence of missed and false alarms.

[0068] S3: Risk value is evaluated based on dynamic output risk probability calculation and weight aggregation.

[0069] Furthermore, the dynamic output risk probability calculation includes calculating the risk value R for each time period predicted by the neural network model. t Convert to cell node risk value R i And set weights f according to the degree of influence of each node i in the power grid structure on the system. i Construct an aggregated risk model:

[0070]

[0071] Among them, S ice risk The total icing risk value for the entire power grid system is given by n, where n is the number of insulated conductors or tower nodes in the current system. The system dynamically distributes risk scores to different power generation control systems or de-icing equipment modules. When the total system risk value S... ice risk When the set threshold is exceeded, an early warning is automatically triggered and the next step of strategy calculation is initiated; the risk value is recalculated every time a new data window is updated.

[0072] It should be noted that the weighted aggregation assessment of risk values ​​includes constructing a power supply adjustment module that interfaces with the power dispatching system and receiving S... ice risk The risk value is used as input; when the system determines that the current risk value is below the threshold, the current power supply method and the operating status of the electric heating de-icing equipment are maintained; when the system determines that the current risk value is above the threshold, and the future time window T... low >0.6 and H high When the value is greater than 0.4, the system will prioritize the start-up of the electric heating module in the power generation system, shorten the heating cycle and increase the power output.

[0073] It should also be noted that after completing data feature extraction and risk scoring output, the predicted hourly risk values ​​are transformed into risk indicators at the line or node level, and different weights are assigned to each node based on its importance in the power grid, its power supply range, and its potential impact. A comprehensive risk value for the entire system is formed through weighted aggregation. This risk value is not static but dynamically adjusted as the time window updates, ensuring the continuity and real-time nature of the prediction results. Furthermore, this risk value is input into the power supply adjustment module connected to the power dispatch system. When the risk level is below a threshold, the system maintains its current operating state; when the risk level exceeds the threshold, and the meteorological conditions within the future time window predict a combination of sustained low temperature and high humidity, the system will trigger countermeasures in advance, such as shortening the working cycle of electrothermal de-icing and increasing de-icing power, to proactively reduce the risk of icing. Through this mechanism, closed-loop management from risk prediction to actual control is ultimately achieved, not only improving the safety and reliability of the power grid under extreme weather conditions but also effectively avoiding excessive energy consumption, achieving a balance between safety and economy.

[0074] Example 2, an embodiment of the present invention, provides a system for predicting the risk of icing on insulated conductors based on meteorological big data, including a data processing module, a feature extraction module, and a risk prediction module.

[0075] The data processing module is used to collect multidimensional sensor data and construct a sliding time window; the feature extraction module is used to construct an adversarial neural network and fuse LSTM and convolution to extract sequence features; the risk prediction module is used to calculate the risk probability based on dynamic output and evaluate the risk value by weight aggregation.

Claims

1. A method for predicting the risk of icing on insulated conductors based on meteorological big data, characterized in that, include: Collect multidimensional sensor data and construct a sliding time window; Construct an adversarial neural network and fuse LSTM and convolution to extract sequence features; Risk values ​​are assessed based on dynamic output risk probability calculation and weighted aggregation.

2. The method for predicting the risk of icing on insulated conductors based on meteorological big data as described in claim 1, characterized in that: The collected multidimensional sensor data includes, By distributing IoT environmental sensing nodes along the insulated conductor, temperature, humidity, wind speed, altitude, and liquid water content are collected to form a time series dataset. The collected data undergoes unified preprocessing, including unit conversion, missing value imputation, and normalization, to form a standardized meteorological input vector. After data processing, the data at each time point are input into a sliding time window of fixed length L, denoted as: W k ={x kS ,x kS+1 ,…,x kS+L-1 } Where S is the sliding step size, the window slides to replace the oldest S data points and add the newest S data points each time, forming a dynamic update mechanism, x kS+L-1 This represents the last data point in the current window, with a starting point of kS and an ending point of kS+L-1, where k is the index number of the sliding time window.

3. The method for predicting the risk of icing on insulated conductors based on meteorological big data as described in claim 2, characterized in that: The construction of the sliding time window includes From each sliding time window W k Key meteorological factors were extracted and indicators for risk characterization were constructed, including the ratio of low temperature duration T. low : Where I(·) is an indicator function used to determine whether the temperature is less than or equal to 0℃, and T t The temperature value at time t; The joint probability H under high humidity conditions high Represented as: H high =P(humidity ≥ 90% ∩ T ≤ 0℃) Wherein, P(·) is a probability function, and the joint probability of high humidity conditions is used to determine whether high humidity coexists under low temperature conditions, as a prerequisite for icing to occur; Liquid water content is estimated and expressed as LWC: LWC=α·R·v -1 Where α is the calibration coefficient, R is the precipitation, and v is the wind speed.

4. The method for predicting the risk of icing on insulated conductors based on meteorological big data as described in claim 3, characterized in that: The construction of the adversarial neural network includes, A Generative Adversarial Network (GAN) framework is adopted, including generator and discriminator structures, with historical meteorological time window samples W as input. k or latent variables The generator structure uses an LSTM or Transformer network to capture temporal dependencies and outputs the next potential time window: in, To synthesize the data for the (k+1)th time window output by the adversarial network generator, θ G Given the set of all parameters of the generator network, compare the generator's output samples with historical real samples. The input is fed into the discriminator, whose network structure includes a CNN to extract local features and outputs: [D real ,R k ]=D(W k ;θ D ) Among them, D real ∈[0,1] represents the true probability, R k θ is the icing risk score calculated by the discriminator. D This is the parameter set of the discriminator.

5. The method for predicting the risk of icing on insulated conductors based on meteorological big data as described in claim 4, characterized in that: The fusion of LSTM and convolution to extract sequence features includes... The meteorological data processed by the sliding time window is input into the neural network prediction model. The network structure includes a convolutional extraction layer (CNN) to extract local meteorological change features. H CNN =ReLU(Conv1D(X t ;W conv )) Among them, H CNN X represents the output features of the CNN layer. t Given the input sequence, d is the convolution kernel, k' is the kernel size, d is the input dimension, and m is the number of output channels. Indicates dimension; Long Short-Term Memory (LSTM) networks are used to model long-range dependencies in time series data. h t ,c t =LSTM(H CNN ;W lstm ) Among them, h t For the hidden state output of the LSTM network at time step t, c t W represents the memory cell state of the LSTM network at time step t. lstm This is the set of weight parameters for the LSTM network. The risk prediction output layer outputs the hidden state to the fully connected layer: R t =σ(W f ·h t +b f ) Among them, the normalized risk value R is output. t ∈[0,1].

6. The method for predicting the risk of icing on insulated conductors based on meteorological big data as described in claim 5, characterized in that: The dynamic output risk probability calculation includes, The risk value R predicted by the neural network model for each time period. t Convert to cell node risk value R i And set weights f according to the degree of influence of each node i in the power grid structure on the system. i Construct an aggregated risk model: Among them, S icerisk The total icing risk of the entire power grid system is currently denoted by n, where n is the number of insulated conductor or tower nodes in the current system. The system dynamically distributes risk scores to different power generation control systems or de-icing equipment modules. When the total system risk value S... icerisk When the set threshold is exceeded, an alert will be automatically triggered and the next strategy calculation will proceed. The risk value is recalculated every time a new data window is updated.

7. The method for predicting the risk of icing on insulated conductors based on meteorological big data as described in claim 6, characterized in that: The weighted aggregation assessment risk value includes, Construct a power supply adjustment module that interfaces with the power dispatching system, receiving S icerisk Risk value as input; When the system determines that the current risk value is below the threshold, it maintains the current power supply method and the operating status of the electric heating de-icing equipment. When the current risk value is determined to be higher than the threshold, and the future time window T low >0.6 and H high When the value is greater than 0.4, the system will prioritize the start-up of the electric heating module in the power generation system, shorten the heating cycle and increase the power output.

8. A system for predicting the risk of icing on insulated conductors based on meteorological big data, employing the method for predicting the risk of icing on insulated conductors based on meteorological big data as described in any one of claims 1 to 7, characterized in that: It includes a data processing module, a feature extraction module, and a risk prediction module; The data processing module is used to collect multidimensional sensor data and construct a sliding time window; The feature extraction module is used to construct an adversarial neural network and fuse LSTM and convolution to extract sequence features; The risk prediction module is used to calculate risk probabilities based on dynamic output and to evaluate risk values ​​by weight aggregation.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for predicting the risk of icing of insulated conductors based on meteorological big data as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for predicting the risk of icing of insulated conductors based on meteorological big data as described in any one of claims 1 to 7.

Images