Transmission conductor icicle corona discharge identification and prediction method, system and device based on deep learning, and storage medium

Abstract

The invention discloses a transmission conductor icicle corona discharge identification and prediction method, system and device based on deep learning, and a storage medium, and relates to the technical field of power system safety monitoring, and the method comprises the steps: collecting multi-modal data, extracting and fusing each modal feature, and constructing a multi-dimensional feature vector; training a hybrid deep learning model comprising a corona discharge identification model and a time sequence prediction model; performing online incremental updating on the hybrid deep learning model by using new data continuously collected in actual operation; generating an identification probability and a prediction trend based on the updated mixed deep learning model, calculating a risk probability based on a Bayesian theory, and automatically generating a corresponding risk level and early warning information; the accuracy and timeliness of identification and prediction of icicle corona discharge of the transmission conductor are improved, manual inspection can be effectively replaced, safe and stable operation of a power grid is guaranteed, and electric energy loss and flashover accident risks are reduced.

Description

Technical Field

[0001] This invention relates to the field of power system safety monitoring technology, and in particular to a method, system, device and storage medium for identifying and predicting icicle corona discharge on transmission lines based on deep learning. Background Technology

[0002] Icing on transmission lines is a significant factor affecting the safe and stable operation of power systems. When icicles form on the surface of the conductors, the electric field strength at the tips of the icicles increases sharply, easily leading to corona discharge. Corona discharge not only causes energy loss but can also trigger flashover accidents, seriously threatening the safety of the power grid.

[0003] Existing technologies such as infrared thermal imaging detection, ultraviolet imaging detection, and acoustic detection suffer from problems such as single-mode detection being prone to false alarms or missed alarms, lacking effective prediction and early warning capabilities, poor environmental adaptability, low level of intelligence, and inability to adapt to different lines and meteorological conditions. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention provides a method, system, device and storage medium for identifying and predicting icicle corona discharge on power transmission lines based on deep learning.

[0005] Therefore, the technical problem solved by this invention is: how to achieve high-precision identification and early warning by integrating multiple detection methods and time-series prediction technology, thereby solving the problems of poor accuracy and insufficient prediction ability of existing technologies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for identifying and predicting icicle corona discharge on power transmission lines based on deep learning, comprising: Multimodal data, including acoustic, thermal imaging, meteorological, and electrical signals, are collected, and features of each modality are extracted and fused to construct a multidimensional feature vector. Based on multidimensional feature vectors, a hybrid deep learning model including a corona discharge recognition model and a time-series prediction model is trained. The hybrid deep learning model is updated online incrementally using new data continuously collected during actual operation; Based on the updated hybrid deep learning model, the system generates identification probabilities and predictive trends, calculates risk probabilities based on Bayesian theory, and automatically generates corresponding risk levels and early warning information.

[0007] As a preferred scheme for a deep learning-based method for identifying and predicting icicle corona discharge on power transmission lines, wherein: The acquisition includes multimodal data from acoustic, thermal imaging, meteorological, and electrical signals. Features from each modality are extracted and fused to construct a multidimensional feature vector, including: Mel frequency cepstral coefficient features are extracted from acoustic signals: the acoustic signals are filtered, framed, power spectrum is calculated, logarithmic energy is calculated by filtering, and discrete cosine transform is performed to obtain feature coefficients; features are extracted from thermal imaging images using a convolutional neural network feature extraction network.

[0008] As a preferred scheme for a deep learning-based method for identifying and predicting icicle corona discharge on power transmission lines, wherein: The acquisition includes multimodal data from acoustic, thermal imaging, meteorological, and electrical signals. Extracting and fusing features from each modality to construct a multidimensional feature vector also includes: The extracted modal features are fused using an attention mechanism. The corresponding attention weights are calculated based on the feature vectors of each modality. The feature vectors of each modality are then weighted and summed based on the attention weights to obtain the fused feature vector, which is the constructed multidimensional feature vector.

[0009] As a preferred scheme for a deep learning-based method for identifying and predicting icicle corona discharge on power transmission lines, wherein: The method of using newly collected data continuously during actual operation to perform online incremental updates to the hybrid deep learning model includes: An adaptive learning algorithm based on online gradient descent is used to update the hybrid deep learning model according to a strategy that dynamically adjusts the learning rate over time. This strategy causes the learning rate to gradually decay as the number of time steps increases.

[0010] The beneficial effects of this preferred technical solution are as follows: the adaptive learning algorithm of online gradient descent combined with the strategy of dynamic decay of the learning rate over time enables the model to adjust the parameters in a timely manner according to the changes in the data as it continuously receives new data. At the same time, it avoids the problems of model instability caused by an excessively large learning rate or slow convergence speed caused by an excessively small learning rate, ensuring that the model maintains good adaptability and accuracy in long-term operation.

[0011] As a preferred scheme for a deep learning-based method for identifying and predicting icicle corona discharge on power transmission lines, wherein: The updated hybrid deep learning model generates identification probabilities and predicted trends, calculates risk probabilities based on Bayesian theory, and automatically generates corresponding risk levels and early warning information, including: The updated hybrid deep learning model is used to generate the recognition probability of corona discharge and the predicted trend of time series.

[0012] The beneficial effects of this preferred technical solution are as follows: by generating the identification probability and time series prediction trend of corona discharge through the updated hybrid deep learning model, specific data support can be provided for subsequent risk assessment, making the risk assessment more scientific and accurate, and helping to discover potential corona discharge risks in a timely manner.

[0013] As a preferred scheme for a deep learning-based method for identifying and predicting icicle corona discharge on power transmission lines, wherein: The process of generating identification probabilities and predicted trends based on the updated hybrid deep learning model, calculating risk probabilities based on Bayesian theory, and automatically generating corresponding risk levels and early warning information also includes: Based on Bayesian theory, the probability of risk occurrence is calculated by combining observed feature vectors and risk events. Specifically, based on the probability of observed features appearing when a known risk event occurs, the probability of the risk event itself occurring, and the probability of observed features appearing, the probability of risk occurrence is calculated using Bayesian formula when a specific feature is observed.

[0014] The beneficial effects of this preferred technical solution are as follows: Bayesian theory can comprehensively consider the probabilistic relationships of various factors and make a more reasonable estimate of the probability of risk occurrence using known information. Compared with a single judgment method, it can more accurately assess the risk of ice corona discharge on transmission lines, providing a reliable basis for subsequent risk level classification and early warning.

[0015] As a preferred scheme for a deep learning-based method for identifying and predicting icicle corona discharge on power transmission lines, wherein: The process of generating identification probabilities and predicted trends based on the updated hybrid deep learning model, calculating risk probabilities based on Bayesian theory, and automatically generating corresponding risk levels and early warning information also includes: Based on the calculated risk probability, the risks are divided into different levels according to the preset risk level classification standards; and corresponding early warning information is automatically generated based on the classified risk levels.

[0016] Secondly, the present invention provides a deep learning-based system for identifying and predicting icicle corona discharge in power transmission lines, comprising: The multimodal data fusion feature construction module is used to collect multimodal data including acoustic, thermal imaging, meteorological and electrical signals, extract features of each modality and fuse them to construct a multidimensional feature vector; The hybrid deep learning model training module is used to train a hybrid deep learning model that includes a corona discharge recognition model and a time-series prediction model based on multi-dimensional feature vectors. The online incremental update module is used to perform online incremental updates on the hybrid deep learning model using new data continuously collected during actual operation. The risk assessment and early warning generation module is used to generate identification probabilities and predicted trends based on the updated hybrid deep learning model, calculate risk probabilities based on Bayesian theory, and automatically generate corresponding risk levels and early warning information.

[0017] Thirdly, the present invention provides a computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the deep learning-based method for identifying and predicting icicle corona discharge on power transmission lines.

[0018] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of a deep learning-based method for identifying and predicting icicle corona discharge in power transmission lines.

[0019] The beneficial effects of this invention: The intelligent identification and prediction method for corona discharge caused by icing on transmission lines provided by this invention has significant practical application value. In scenarios where corona discharge caused by icing on transmission lines threatens power grid safety, this invention solves the problem of false alarms or missed alarms in single-modal detection by acquiring multi-modal data and integrating information from acoustic, thermal imaging, meteorological, and electrical sources. Feature extraction and fusion construct multi-dimensional feature vectors, combined with deep learning model training and online adaptive learning, improve the accuracy of identification and prediction. The LSTM-Transformer hybrid architecture can capture long-term time dependencies and complex correlations, achieving high-precision time-series prediction. Risk assessment and early warning generate risk levels and warning information based on the prediction results, allowing the dispatch center to take measures in advance; it achieves 24-hour continuous monitoring and real-time intelligent analysis, effectively replacing inefficient and unsafe manual inspections, ensuring the safe and stable operation of the power grid. Attached Figure Description

[0020] 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.

[0021] Figure 1 This is an overall flowchart of a deep learning-based method for identifying and predicting icicle corona discharge on power transmission lines, provided by the present invention.

[0022] Figure 2This is a visualization result of the identification and prediction of icicle corona discharge on power transmission lines in a simulation example of a deep learning-based method for identifying and predicting icicle corona discharge on power transmission lines provided by this invention. Detailed Implementation

[0023] 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.

[0024] Example 1, referring to Figure 1 This is the first embodiment of the present invention, which provides a method for identifying and predicting icicle corona discharge on power transmission lines based on deep learning, including: S1: Collect multimodal data including acoustic, thermal imaging, meteorological and electrical signals, extract features of each modality and fuse them to construct a multidimensional feature vector; S2: Based on multi-dimensional feature vectors, train a hybrid deep learning model that includes a corona discharge recognition model and a time-series prediction model; S3: Utilize new data continuously collected during actual operation to perform online incremental updates on the hybrid deep learning model; S4: Based on the updated hybrid deep learning model, it generates recognition probabilities and predicts trends, calculates risk probabilities based on Bayesian theory, and automatically generates corresponding risk levels and early warning information.

[0025] It should be noted that through steps S1-S4, effective acquisition of multimodal data, feature fusion, and training and updating of deep learning models are achieved. By making full use of the multimodal fusion neural network and LSTM-Transformer hybrid architecture, combined with adaptive learning algorithms and Bayesian theory, corona discharge phenomena can be accurately identified, their development trends can be effectively predicted, risk probabilities can be calculated and risk levels can be classified, and corresponding early warning information can be automatically generated.

[0026] Example 2, refer to Figure 1 As one embodiment of the present invention, based on the previous embodiment, a deep learning-based method for identifying and predicting icicle corona discharge on power transmission lines is provided, comprising: In this embodiment, step S1 above involves acquiring multimodal data including acoustic, thermal imaging, meteorological, and electrical signals, extracting features from each modality, fusing them, and constructing a multidimensional feature vector, including: Multimodal data, including acoustic, thermal imaging, meteorological, and electrical signals, are collected through a multimodal fusion neural network. The multimodal fusion neural network includes an acoustic feature encoder, an image feature encoder, a numerical feature processor, and an attention fusion layer. Each encoder extracts the deep features of its corresponding modality, and feature fusion is performed through an attention mechanism.

[0027] Specifically, the collected raw data undergoes preprocessing: For acoustic signals, pre-emphasis filtering is performed, which is represented as: Among them, X filtered (t) represents the filtered acoustic signal; h (k) represents the impulse response of the digital filter; X (t) represents the original acoustic signal; N Indicates the filter order.

[0028] For thermal imaging images, enhancement processing is performed, as shown below: in, I enhanced ( x,y () represents the enhanced image pixel value. I original ( x,y ) represents the original image pixel value, α represents the contrast adjustment coefficient, and β represents the brightness adjustment coefficient.

[0029] Next, feature extraction is performed. For acoustic signals, MFCC (Mel frequency cepstral coefficients) features are extracted: first, the original acoustic signal is pre-emphasized and filtered; then, Hamming windowing is used for frame segmentation; next, the power spectrum is calculated using Fast Fourier Transform; after filtering by the Mel filter bank, the logarithmic energy is calculated; finally, Discrete Cosine Transform is performed to obtain the MFCC feature coefficients. The formula is as follows: in, MFCC (m) represents the m-th MFCC, the coefficient S(k) represents the output of the k-th Mel filter bank, K represents the number of Mel filter banks, and m represents the cepstral coefficient index.

[0030] For thermal imaging images, features are extracted using a CNN feature extraction network, as follows: in, F cnn This represents the feature map extracted by the CNN. W conv Indicates the convolution kernel weights, I inputIndicates the input image. b conv represents the bias term, and * represents the convolution operation.

[0031] Finally, an attention mechanism is used for multimodal feature fusion, represented as follows: in, α i This represents the attention weight for the i-th modality. F i This represents the eigenvector of the i-th mode. W a and W f For attention network parameters, b f Indicates the bias term. M Indicates the number of modes. F fused This represents the fused feature vector, i.e., the constructed multidimensional feature vector.

[0032] In another possible implementation, multimodal data acquisition can be carried out by distributing sensor networks at different key nodes of the transmission line. For acoustic signals, multiple high-sensitivity microphones are installed at different heights and locations on the towers to comprehensively capture acoustic information from different directions and distances; for thermal imaging data, thermal imaging cameras are set up at key connection points of the conductors and areas prone to icing to ensure real-time monitoring of areas where corona discharge may occur; meteorological sensors are distributed along the transmission line in different terrain and climate zones to obtain more accurate meteorological data; electrical signal acquisition equipment is installed at key locations in substations and transmission lines to accurately collect electrical parameters such as voltage and current.

[0033] In another possible implementation, multimodal data acquisition can be achieved by combining mobile acquisition devices with fixed sensors. In addition to installing sensors at fixed locations, drones can be used to periodically inspect power transmission lines. These drones are equipped with acoustic, thermal imaging, and electrical signal acquisition devices to collect data during flight. Simultaneously, vehicle-mounted mobile monitoring systems can collect relevant data while traveling on roads surrounding the power transmission lines, thus supplementing data gaps from fixed sensors in certain areas or time periods.

[0034] In another possible implementation, wavelet transform can be used to extract modal features from the acoustic signal. Wavelet decomposition of the acoustic signal yields wavelet coefficients at different scales and frequencies, reflecting the time-frequency characteristics of the signal. By selecting an appropriate wavelet basis and decomposition level, feature information related to corona discharge can be extracted. For thermal imaging images, a region-growing-based feature extraction method can be used. Based on the grayscale values ​​and spatial adjacency of pixels in the image, pixels with similar features are merged into regions, and then features such as area, perimeter, and average grayscale value of each region are extracted.

[0035] In another possible implementation, modal features can also be extracted using the Linear Predictive Cepstral Coefficient (LPCC) feature extraction method for acoustic signals. A linear predictive model of the acoustic signal is established through linear predictive analysis, and then cepstral coefficients are calculated from the predicted coefficients of the model to obtain the feature representation of the acoustic signal. For thermal imaging images, a deep learning-based autoencoder can be used for feature extraction. The thermal imaging image is input into the autoencoder, which compresses the image into a low-dimensional feature vector, and the decoder then reconstructs the image from the feature vector. During training, the error between the reconstructed image and the original image is minimized, thereby learning the essential features of the thermal imaging image.

[0036] In this embodiment, step S2 above, which trains a hybrid deep learning model including a corona discharge recognition model and a time-series prediction model based on multi-dimensional feature vectors, includes: The training of the hybrid deep learning model consists of two parts: using a multimodal fusion neural network for corona discharge identification; and using an LSTM-Transformer hybrid architecture for time series prediction.

[0037] The LSTM-Transformer hybrid architecture includes: LSTM Temporal Modeling: LSTM layers are used to capture long-term temporal dependencies. The cell state update formula is as follows: in, f t , i t , o t These represent the forget gate, input gate, and output gate, respectively. C t , h t These represent the cell state and the hidden state, respectively. W f , W i , W C, W o This is the weight matrix. b f , b i , b C , b o For bias vectors, σ It is the sigmoid activation function.

[0038] Transformer self-attention mechanism: The Transformer encoder learns complex relationships in time series data through a multi-head attention mechanism. The multi-head self-attention calculation is as follows: in, Q , K , V These represent the query, key, and value matrices, respectively. d k Represents the dimension of the key vector. head i = Attention ( QW i Q , KW i K , VW i V ), W i Q , W i K , W i V , W O This represents the weight matrix.

[0039] In another possible implementation, the hybrid deep learning model can also be trained using transfer learning. First, the corona discharge identification model and the time-series prediction model are pre-trained on large-scale, similar power system data, which can come from transmission lines in other regions or related power experiment data. Then, the pre-trained model parameters are transferred to the current transmission line icicle corona discharge identification and prediction task, and fine-tuned on the multi-dimensional feature vector data of this task. This reduces training time and improves the initial performance of the model.

[0040] In another possible implementation, training the hybrid deep learning model can also employ a multi-stage training strategy. First, the corona discharge recognition model and the time-series prediction model are trained independently, optimizing their respective structures and parameters. Then, the two models are jointly trained. By designing an appropriate loss function, the two models can mutually reinforce each other during collaborative work, improving overall recognition and prediction performance. Different learning rates and training epochs can be used during the joint training process to achieve the optimal training effect.

[0041] In this embodiment, step S3 above, which utilizes newly collected data during actual operation to perform online incremental updates to the hybrid deep learning model, includes: After the model is deployed in real-world applications, new data will be continuously collected. At this point, an adaptive learning algorithm using online gradient descent is used to update the hybrid deep learning model, improving its adaptability and prediction accuracy. The learning rate adjustment formula for the adaptive learning algorithm is: in, η 0 is the initial learning rate. α t is the decay coefficient, and t is the time step.

[0042] In another possible implementation, online incremental updates can also employ a batch-based update approach. After a certain amount of new data is continuously collected, this new data is grouped into a batch, and then the hybrid deep learning model is updated using a batch gradient descent algorithm. During the update process, the learning rate and other hyperparameters can be dynamically adjusted based on the characteristics of the new data and the current state of the model to ensure that the model can quickly adapt to changes in the new data.

[0043] In another possible implementation, online incremental updates can also employ a model fusion-based update strategy. When new data arrives, a new sub-model is trained using the new data, and then fused with the original hybrid deep learning model. The fusion method can be a weighted average, where the weights of the new sub-model and the original model are determined based on their performance on the validation data. This allows for the rapid incorporation of features from the new data while preserving useful information from the original model, improving the model's adaptability and accuracy.

[0044] In this embodiment, step S4 above, based on the updated hybrid deep learning model, generates identification probabilities and predicted trends, calculates risk probabilities based on Bayesian theory, and automatically generates corresponding risk levels and early warning information, including: The updated hybrid deep learning model is used to generate the recognition probability of corona discharge and the predicted trend of the time series.

[0045] The probability of risk is calculated based on Bayesian theory, using the following formula: Where X is the observed feature vector and Risk is the risk event.

[0046] Based on the calculated risk probability, risk levels are classified according to the following criteria: It will automatically generate early warning information corresponding to the risk level.

[0047] Example 3: The above is an illustrative scheme of a deep learning-based method for identifying and predicting ice corona discharge on power transmission lines. It should be noted that the technical solution of a deep learning-based system for identifying and predicting ice corona discharge on power transmission lines belongs to the same concept as the above-described deep learning-based method for identifying and predicting ice corona discharge on power transmission lines. Details not described in detail in the deep learning-based system for identifying and predicting ice corona discharge on power transmission lines in this embodiment can be found in the description of the deep learning-based method for identifying and predicting ice corona discharge on power transmission lines described above.

[0048] This embodiment also provides a deep learning-based system for identifying and predicting icicle corona discharge in power transmission lines, including: The multimodal data fusion feature construction module is used to collect multimodal data including acoustic, thermal imaging, meteorological and electrical signals, extract features of each modality and fuse them to construct a multidimensional feature vector; The hybrid deep learning model training module is used to train a hybrid deep learning model that includes a corona discharge recognition model and a time-series prediction model based on multi-dimensional feature vectors. The online incremental update module is used to perform online incremental updates on the hybrid deep learning model using new data continuously collected during actual operation. The risk assessment and early warning generation module is used to generate identification probabilities and predicted trends based on the updated hybrid deep learning model, calculate risk probabilities based on Bayesian theory, and automatically generate corresponding risk levels and early warning information.

[0049] This embodiment also provides an electronic device applicable to a method for identifying and predicting icicle corona discharge on power transmission lines based on deep learning, including: The system includes a memory and a processor. The memory stores computer-executable instructions, and the processor executes these instructions to implement a deep learning-based method for identifying and predicting icicle corona discharge on power transmission lines, as proposed in the above embodiments.

[0050] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a deep learning-based method for identifying and predicting icicle corona discharge in power transmission lines, as proposed in the above embodiment.

[0051] The storage medium proposed in this embodiment belongs to the same inventive concept as the deep learning-based method for identifying and predicting icicle corona discharge in power transmission lines proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0052] Example 4, refer to Figure 2 As an embodiment of the present invention, a method for identifying and predicting icicle corona discharge on power transmission lines based on deep learning is provided. To verify the beneficial effects of the present invention, a simulation experiment is conducted for scientific demonstration.

[0053] On a 500kV transmission line in a mountainous area, icing frequently occurs in winter due to the high altitude and low temperatures. Traditional manual inspection methods are inefficient and pose safety hazards. A multimodal monitoring system based on this method was deployed and applied. Deploy a sensor network: Set up a monitoring point every 10 kilometers, including ultrasonic sensors, infrared thermal imagers, weather stations, etc. Data acquisition: 24-hour continuous monitoring of acoustic signals, thermal imaging, meteorological parameters, and electrical parameters; Intelligent analysis: Real-time identification of corona discharge phenomena and prediction of development trends through deep learning models; Early warning system: Automatically sends early warning information to the dispatch center based on the risk level.

[0054] Case results as follows Figure 2 As shown.

[0055] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying and predicting icicle corona discharge on power transmission lines based on deep learning, characterized in that, include: Multimodal data, including acoustic, thermal imaging, meteorological, and electrical signals, are collected, and features of each modality are extracted and fused to construct a multidimensional feature vector. Based on multidimensional feature vectors, a hybrid deep learning model including a corona discharge recognition model and a time-series prediction model is trained. The hybrid deep learning model is updated online incrementally using new data continuously collected during actual operation; Based on the updated hybrid deep learning model, the system generates identification probabilities and predictive trends, calculates risk probabilities based on Bayesian theory, and automatically generates corresponding risk levels and early warning information.

2. The method for identifying and predicting icicle corona discharge on power transmission lines based on deep learning as described in claim 1, characterized in that, The acquisition includes multimodal data from acoustic, thermal imaging, meteorological, and electrical signals. Features from each modality are extracted and fused to construct a multidimensional feature vector, including: Mel frequency cepstral coefficient features are extracted from acoustic signals: the acoustic signals are filtered, framed, power spectrum is calculated, logarithmic energy is calculated by filtering, and discrete cosine transform is performed to obtain feature coefficients; features are extracted from thermal imaging images using a convolutional neural network feature extraction network.

3. The method for identifying and predicting icicle corona discharge on power transmission lines based on deep learning as described in claim 2, characterized in that, The acquisition includes multimodal data from acoustic, thermal imaging, meteorological, and electrical signals. Extracting and fusing features from each modality to construct a multidimensional feature vector also includes: The extracted modal features are fused using an attention mechanism. The corresponding attention weights are calculated based on the feature vectors of each modality. The feature vectors of each modality are then weighted and summed based on the attention weights to obtain the fused feature vector, which is the constructed multidimensional feature vector.

4. The method for identifying and predicting icicle corona discharge on power transmission lines based on deep learning as described in claim 3, characterized in that, The method of using newly collected data continuously during actual operation to perform online incremental updates to the hybrid deep learning model includes: An adaptive learning algorithm based on online gradient descent is used to update the hybrid deep learning model according to a strategy that dynamically adjusts the learning rate over time. This strategy causes the learning rate to gradually decay as the number of time steps increases.

5. The method for identifying and predicting icicle corona discharge on power transmission lines based on deep learning as described in claim 4, characterized in that, The updated hybrid deep learning model generates identification probabilities and predicted trends, calculates risk probabilities based on Bayesian theory, and automatically generates corresponding risk levels and early warning information, including: The updated hybrid deep learning model is used to generate the recognition probability of corona discharge and the predicted trend of time series.

6. The method for identifying and predicting icicle corona discharge on power transmission lines based on deep learning as described in claim 5, characterized in that, The process of generating identification probabilities and predicted trends based on the updated hybrid deep learning model, calculating risk probabilities based on Bayesian theory, and automatically generating corresponding risk levels and early warning information also includes: Based on Bayesian theory, the probability of risk occurrence is calculated by combining observed feature vectors and risk events. Specifically, based on the probability of observed features appearing when a known risk event occurs, the probability of the risk event itself occurring, and the probability of observed features appearing, the probability of risk occurrence is calculated using Bayesian formula when a specific feature is observed.

7. The method for identifying and predicting icicle corona discharge on power transmission lines based on deep learning as described in claim 6, characterized in that, The process of generating identification probabilities and predicted trends based on the updated hybrid deep learning model, calculating risk probabilities based on Bayesian theory, and automatically generating corresponding risk levels and early warning information also includes: Based on the calculated risk probability, the risks are divided into different levels according to the preset risk level classification standards; and corresponding early warning information is automatically generated based on the classified risk levels.

8. A deep learning-based system for identifying and predicting icicle corona discharge on power transmission lines, using the method described in any one of claims 1 to 7, characterized in that, include: The multimodal data fusion feature construction module is used to collect multimodal data including acoustic, thermal imaging, meteorological and electrical signals, extract features of each modality and fuse them to construct a multidimensional feature vector; The hybrid deep learning model training module is used to train a hybrid deep learning model that includes a corona discharge recognition model and a time-series prediction model based on multi-dimensional feature vectors. The online incremental update module is used to perform online incremental updates on the hybrid deep learning model using new data continuously collected during actual operation. The risk assessment and early warning generation module is used to generate identification probabilities and predicted trends based on the updated hybrid deep learning model, calculate risk probabilities based on Bayesian theory, and automatically generate corresponding risk levels and early warning information.

9. An electronic device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores computer-executable instructions that, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images