Intelligent identification and online monitoring method for lightning stroke characteristics of intelligent power transmission line

Abstract

The invention relates to the field of power systems, and discloses an intelligent power transmission line lightning stroke characteristic intelligent identification online monitoring method comprising the following steps: collecting multi-source data, preprocessing, modeling lightning stroke probability and spatial distribution by using a chaos dynamics model, extracting and classifying characteristics based on deep learning, and optimizing a prediction strategy in combination with reinforcement learning. Real-time early warning and automatic response are achieved; the invention also provides an intelligent transmission line lightning stroke characteristic intelligent identification online monitoring system. The system comprises a data acquisition module, a data processing module, a chaotic dynamics modeling module, a deep learning identification module, a reinforcement learning optimization module and an early warning and response module. According to the method, the real-time collection and preprocessing technology of multi-source data is adopted, data cleaning and standardization are carried out in combination with multi-dimensional information such as meteorological, electrical and historical lightning stroke data, the accuracy and consistency of the data are ensured, and the reliability of a prediction result is effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of power systems, and particularly to an online monitoring method for intelligent identification of lightning strike characteristics of intelligent transmission lines. Background Art

[0002] Lightning strike is one of the natural disasters. Especially for power systems, lightning strike events will have a great impact on transmission lines, which may lead to power grid failures, equipment damage, or even large-scale power outages. Traditional lightning strike prediction methods mostly rely on meteorological data and historical lightning strike data, but these methods often have problems such as single data, limited processing capabilities, and insufficient prediction accuracy.

[0003] In the prior art, most lightning strike prediction methods rely on predictions based on meteorological models. These methods usually use meteorological data (such as temperature, humidity, wind speed, etc.) and lightning strike historical records, combined with some statistical or simple regression analysis models to predict the occurrence of lightning strikes. The advantage of this method is that it is relatively simple and has a small amount of calculation, but its disadvantages are also obvious. Especially under complex meteorological conditions, it often fails to accurately capture the dynamic characteristics of lightning strike occurrences. Therefore, traditional meteorological prediction models often fail to effectively improve the accuracy of lightning strike predictions. Especially in the case of complex and variable lightning clouds, false alarms or missed alarms often occur; moreover, traditional lightning strike prediction models based on statistical analysis mostly find patterns through statistical processing of historical lightning strike data. Such methods often rely on a large amount of historical data and predict the occurrence of lightning strike events through pattern recognition. However, the method of simply relying on historical data is easily affected by extreme data or accidental events, resulting in inaccurate prediction results. At the same time, the statistical model lacks a deep understanding of time changes and spatial distributions and cannot fully consider the non-linear and complex spatio-temporal characteristics of lightning strike occurrences; furthermore, there are also some lightning strike prediction technologies based on machine learning, but these methods usually adopt traditional machine learning models, such as support vector machines (SVM), decision trees, etc. Although they can improve the prediction accuracy to a certain extent, these methods still have some limitations. For example, traditional machine learning models need to manually select features and are difficult to handle the correlation problems of high-dimensional data. In addition, when facing a large amount of complex lightning strike data, these methods often have low processing efficiency and fail to effectively fuse multi-source data for prediction and analysis. For this reason, those skilled in the art have proposed an online monitoring method for intelligent identification of lightning strike characteristics of intelligent transmission lines to solve the above problems. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides an online monitoring method for intelligent identification of lightning strike characteristics of intelligent transmission lines, which solves the problems of insufficient accuracy and lag in response existing in traditional lightning strike prediction methods.

[0005] To achieve the above object, the present invention is realized through the following technical solutions: an intelligent on-line monitoring method for identifying lightning strike characteristics of an intelligent transmission line, comprising the following steps: Collect multi-source data from different sensors, including meteorological data, electrical data, and historical lightning strike data; Preprocess the multi-source data, including denoising, normalization, and standardization; Model the occurrence of lightning strikes based on a chaotic dynamics model, where the chaotic dynamics model uses Logistic mapping and Henon mapping to describe the probability change and spatial distribution characteristics of lightning strikes; Extract and identify the characteristics of lightning strikes based on a deep learning model, where the deep learning model includes a convolutional neural network CNN and a long short-term memory network LSTM for extracting spatial and temporal characteristics and classifying them; Combine reinforcement learning to optimize the prediction strategy of the model and adjust the model parameters through feedback; During the real-time monitoring process, when the lightning strike risk reaches a predetermined threshold, give an early warning and execute an automatic response.

[0006] Preferably, the collected multi-source data includes: Meteorological data, including temperature, humidity, wind speed, air pressure, and lightning cloud maps; Electrical data, including current, voltage, load, and power consumption; Historical lightning strike data, including the time, location, intensity, and duration of lightning strikes.

[0007] Preferably, the preprocessing steps include: Perform time series processing on the electrical data in the multi-source data to ensure the time synchronization of the data; Denoise the meteorological data and lightning strike data to remove outliers and errors; Perform normalization and standardization processing on all data to ensure that different data sources have a consistent scale and distribution form.

[0008] Preferably, the steps of the chaotic dynamics model include: Use the Logistic mapping model to describe the probability of lightning strike occurrence, and simulate the sensitivity of lightning strike events under different meteorological conditions through control parameters; Use the Henon mapping model to describe the spatial distribution characteristics of lightning strike events, considering the influence of time and space factors on the occurrence of lightning strike events.

[0009] Preferably, the steps of the deep learning model include: Use the convolutional neural network CNN to extract features from the lightning cloud map and the transmission line image, and automatically learn the spatial features in the image; Use the long short-term memory network (LSTM) to perform time series analysis on meteorological data and electrical data to capture time-dependent relationships; combine the CNN and LSTM models, and identify and classify lightning strike characteristics of the extracted features through a fully connected layer.

[0010] Preferably, the steps of the reinforcement learning include: Train the prediction model through the deep Q network (DQN) to optimize the prediction strategy; Update the model parameters according to the real-time feedback lightning strike data to improve the prediction accuracy; Reinforcement learning automatically adjusts the model prediction strategy through a feedback mechanism to improve the prediction ability of the system and reduce the false alarm rate and missed alarm rate.

[0011] Preferably, the steps of the lightning strike risk warning include: When the lightning strike risk predicted by the model exceeds a predetermined threshold, the system issues a warning; The warning information is notified to the power grid operation and maintenance personnel through text messages, emails or a real-time monitoring platform; In the high-risk stage, the system can automatically execute switching to standby lines and adjusting load automation response measures to reduce the impact of lightning strikes.

[0012] Preferably, the steps of the feedback adjustment include: Dynamically adjust the model parameters based on the deviation between the actual lightning strike data and the model prediction results; Calculate the model prediction error through a feedback mechanism, and update the weights through an optimization algorithm to improve the stability and accuracy of the model.

[0013] Preferably, the training process of the model includes: Use historical lightning strike data and real-time monitoring data to train the deep learning model; The model adjusts its internal parameters through continuous training and verification, and detects its accuracy through cross-validation; Reinforcement learning improves the prediction ability of the model through multiple rounds of training and optimization, and reduces the false alarm rate and missed alarm rate.

[0014] An intelligent online monitoring system for identifying lightning strike characteristics of intelligent transmission lines is also provided, including: A data acquisition module for collecting meteorological data, electrical data, historical lightning strike data and transmission line image data; A data processing module for preprocessing the collected data, including denoising, normalization and standardization; A chaotic dynamics modeling module for constructing a nonlinear dynamics model of lightning strike occurrence, including Logistic mapping and Henon mapping; The deep learning identification module, which includes a Convolutional Neural Network (CNN) and a Long Short-Term Memory Network (LSTM), is used to extract and identify lightning strike features in data; The reinforcement learning optimization module is used to optimize the strategy of the prediction model and adjust parameters according to feedback; The warning and response module issues a warning and automatically executes response measures when the lightning strike risk exceeds a predetermined threshold.

[0015] The present invention provides an online monitoring method for intelligent identification of lightning strike features in intelligent transmission lines. It has the following beneficial effects: 1. The present invention adopts the real-time collection and preprocessing technology of multi-source data, combines multi-dimensional information such as meteorological, electrical, and historical lightning strike data for data cleaning and standardization, ensures the accuracy and consistency of the data, and achieves the technical effect of providing high-quality input for the subsequent prediction model. Compared with the traditional lightning strike prediction methods that rely on a single data source or have low data quality, the present invention effectively improves the reliability of the prediction results through multi-source data fusion and precise preprocessing.

[0016] 2. The present invention introduces the chaotic dynamics models Logistic mapping and Henon mapping to simulate the occurrence probability and spatial distribution characteristics of lightning strikes, capture the non-linear characteristics of lightning strike events, and achieves the technical effect of more accurately describing the spatio-temporal changes of lightning strikes. Different from the traditional methods that only rely on simple statistical analysis, the chaotic dynamics models can more comprehensively reflect the complex evolution of lightning strike risks and solve the problem that the existing methods cannot fully consider the spatio-temporal correlation.

[0017] 3. The present invention combines the deep learning models CNN and LSTM for feature extraction and identification, can simultaneously extract the spatial features of lightning cloud images and the temporal features of meteorological and electrical data, and achieves the technical effect of improving the prediction accuracy of lightning strike risks in complex environments. Compared with the traditional technical solutions based on linear regression or rule models, the deep learning method of the present invention can automatically learn and identify complex patterns from large-scale data, greatly improving the accuracy and flexibility of the prediction.

[0018] 4. The present invention adopts reinforcement learning to optimize the prediction strategy and adaptively adjusts it in combination with a feedback mechanism, enabling the model to optimize the decision-making process according to real-time data changes, and achieving the technical effects of dynamic optimization and real-time response. Different from the static optimization models in the prior art, the present invention can continuously optimize the prediction results in different environments through the adaptive ability of reinforcement learning, solving the problem of lag in traditional methods when dealing with complex and changing power grid environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a schematic flow chart of the method of the present invention; Figure 2Schematic diagram of the system architecture of the present invention. Detailed implementation manners

[0020] Next, in combination with the accompanying drawings of the present invention specification, the technical solutions in the embodiments of the present invention will be clearly and completely described. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

[0021] Please refer to the attached Figure 1 , the embodiment of the present invention provides an intelligent identification and on-line monitoring method for lightning strike characteristics of an intelligent transmission line, including the following steps: S1. Collect multi-source data from different sensors, including meteorological data, electrical data, and historical lightning strike data; Specifically, in the implementation of the present invention, data collection is the first and crucial operation. Through an efficient data collection process, sufficient input data can be provided for subsequent lightning strike risk prediction, model training, and real-time monitoring. The diversity and accuracy of the data directly affect the prediction ability of the model and the stability of the system. Therefore, the data collection work must be precise and efficient, covering various conditions and environmental factors of lightning strikes.

[0022] In this embodiment, data collection includes, but is not limited to, multi-source data from different sensors. These data are crucial for building the model and subsequent early warning work. The collected data mainly includes meteorological data, electrical data, and historical lightning strike data. Each data type provides different dimensional perspectives for comprehensively analyzing the possibility of lightning strikes.

[0023] Meteorological data is one of the bases of the entire data collection system. Meteorological data is obtained in real time through meteorological stations or other monitoring devices and includes multiple environmental factors such as temperature, humidity, wind speed, and air pressure. Meteorological conditions, especially the formation of thunderclouds and meteorological changes, directly affect the occurrence of lightning strike events. Therefore, meteorological data is crucial for subsequent prediction of lightning strikes. In this embodiment, the collection frequency of meteorological data is once per minute, which can provide sufficient information for lightning strike prediction within a short period.

[0024] Electrical data reflects the state of the transmission line through parameters such as current, voltage, load, and power consumption. When a lightning strike occurs, the electrical characteristics of the transmission line usually change significantly, such as current and voltage fluctuations. Therefore, the real-time monitoring of these electrical parameters can not only help analyze the consequences of lightning strikes but also provide reference information for prediction. Specifically, the electrical data in this embodiment is collected once per second to reflect the operation of the power grid in real time.

[0025] Historical lightning strike data includes detailed records of all lightning strike events over a period of time. These records usually include information such as the time, location, intensity, and duration of the lightning strike. By analyzing historical lightning strike data, the potential patterns of lightning strike events can be identified. Specifically, historical lightning strike data can help understand the temporal and spatial distribution characteristics of lightning strikes and train prediction models, thereby improving the accuracy of future lightning strike predictions.

[0026] In some embodiments, image data such as lightning cloud maps may also be collected as a supplementary data source, which provides strong support for power grid operation and maintenance personnel in monitoring before lightning strikes. Lightning cloud maps provide visual evidence of lightning activity, helping operation and maintenance personnel to confirm whether there is lightning activity before lightning strikes. The frequency of image data collection can be adjusted according to actual conditions. Normally, the frequency of lightning cloud map data collection is once every 5 minutes.

[0027] All collected data is transmitted to the central data storage system via wireless networks to ensure that the data can be summarized and processed in real time and efficiently. In order to ensure the real-time and accuracy of the data, the collection equipment must have a certain anti-interference capability and be able to operate stably in various environments.

[0028] In order to further improve the validity and accuracy of the data, the present invention can also use a distributed sensor network for data collection. The sensors are arranged in power facilities in different areas. This distributed collection method can better cover the lightning strike risks of various power grids and different areas, and provide more comprehensive monitoring data.

[0029] Specifically, all collected data will be initially verified and filtered before being transmitted to the central processing platform. The verification process ensures the validity of the data and removes outliers and erroneous data, thereby avoiding the impact of data errors on subsequent processing processes. During data transmission and storage, encryption technology is used to ensure data security and prevent data loss and leakage.

[0030] In a specific embodiment, all data will enter the data preprocessing module after preliminary verification to prepare for subsequent analysis and modeling. The accuracy and integrity of the data are crucial, so the preprocessing module will strictly perform operations such as data cleaning and format conversion to ensure that the data ultimately sent to subsequent steps is of high quality.

[0031] Generally speaking, the data collection and verification process will be accelerated in weather conditions with a higher risk of lightning strikes, especially during periods of high thunderstorm activity. This is because during these periods, the real-time changes in meteorological, electrical parameters and historical lightning strike data are particularly dramatic. Timely data collection and processing are crucial for accurately predicting lightning strike events.

[0032] S2, preprocessing of multi-source data, including denoising, normalization and standardization; Specifically, in the implementation process of the present invention, the data preprocessing link in step S2 is crucial. In the aforementioned step S1, the collection of multi-source data has been completed, including meteorological data, electrical data, and historical lightning strike data. These data will serve as the basis for subsequent analysis and modeling. Therefore, they must be fully cleaned and processed to ensure data quality and accuracy. Data preprocessing can not only remove noise and outliers but also standardize the data form, providing reliable data input for subsequent chaotic dynamics modeling and deep learning model training.

[0033] In this embodiment, the data preprocessing includes the following key steps: denoising, normalization, and standardization. These processing steps can effectively eliminate unnecessary interference in the data and make the data meet the requirements of subsequent models. Specifically, these processing steps play an important role in improving the accuracy of lightning strike risk prediction and the stability of the model.

[0034] Generally, during the data collection process, due to factors such as environmental interference and sensor errors, certain noise may be introduced. If this noise is not removed, it may affect subsequent analysis and modeling, resulting in inaccurate prediction results.

[0035] As an option, in this embodiment, mean filtering and Gaussian filtering methods are used to denoise the collected meteorological data and electrical data. Mean filtering replaces the original data value by calculating the average value of the neighborhood around the data point, thereby reducing the impact of individual abnormal data on the overall data. Gaussian filtering uses a Gaussian distribution to perform weighted averaging on the data to more smoothly remove noise.

[0036] Specifically, the Gaussian filtering formula is: Where: y(t) is the filtered output data; x(i) is the original data; G(t - i) is the Gaussian weighting function, defined as: Where: σ is the standard deviation, which controls the smoothness of the filter; t and i are time steps, where t is the current moment and i is the past moment, and t - i represents the time difference between the moments; G(t - i) is the value of the Gaussian weighting function.

[0037] Normalization processing: Data from different sources (such as meteorological data, electrical data, and historical lightning strike data) often have different dimensions or ranges. For example, temperature data may be between 0 and 40 degrees Celsius, while current data may fluctuate between milliamperes and amperes. To avoid the impact of different data dimensions on subsequent model training, it is necessary to perform normalization processing on the data.

[0038] As an option, in this embodiment, two methods, maximum normalization and Z-Score standardization, are used to normalize the data. The maximum normalization method divides each data point by the maximum value in the dataset, so that all data are compressed into the interval [0, 1]. The Z-Score standardization method adjusts the data to a distribution with a mean of 0 and a standard deviation of 1 by calculating the mean and standard deviation of the data.

[0039] Specifically, the formula for maximum normalization is: where: x normalized is the normalized data; x is the original data; max(x) is the maximum value in the dataset.

[0040] The formula for Z-Score standardization is: where: x ′ is the standardized data; μ is the mean of the data; σ is the standard deviation of the data.

[0041] Through these normalization steps, all input data can be processed with the same standard, avoiding the model training bias caused by different dimensions.

[0042] Standardization process: Specifically, in some cases, simply normalizing may not be sufficient to make the data suitable for subsequent machine learning models, especially deep learning models. Therefore, the standardization step is introduced to further optimize the performance of the data.

[0043] As an option, this embodiment adopts Min-Max standardization as the standardization method. This method scales the data according to a certain ratio to ensure that the data is still within a reasonable numerical range after processing. The formula for Min-Max standardization is as follows: where: x ′ is the standardized data; x is the original data; min(x) and max(x) are the minimum and maximum values in the dataset respectively.

[0044] The advantage of this method is that it can adjust all data to a unified scale, not only avoiding the dimension difference between data, but also helping to improve the learning efficiency of the model.

[0045] Generally, after data collection and preprocessing, the data should be verified and checked to ensure the stability and consistency of data quality. In this embodiment, data verification includes checking the integrity, consistency, and accuracy of the data.

[0046] In a possible implementation, a data verification algorithm is used to check the processed data to ensure that there is no lost or incorrect data. The specific methods may include: verifying whether each item of data is within a reasonable range, checking whether the data types are consistent, and ensuring that there are no null values or outliers.

[0047] Specifically, the verification algorithm can be based on two methods: threshold detection and regression detection. Threshold detection checks whether the data is abnormal by setting a reasonable numerical range; regression detection judges outliers according to the laws of historical data. Through this method, it can be ensured that the data is complete and accurate before entering the subsequent steps.

[0048] S3. Model the occurrence of lightning strikes based on the chaotic dynamics model. The chaotic dynamics model uses the Logistic map and the Henon map to describe the probability change and spatial distribution characteristics of lightning strikes; Specifically, in the aforementioned steps S1 and S2, the collection and preprocessing of data lay the foundation for subsequent modeling and prediction. Specifically, step S1 ensures the accuracy and real-time nature of multi-source data, while step S2 optimizes the data through methods such as denoising, normalization, and standardization to ensure that it can adapt to subsequent modeling steps. Next, step S3 involves modeling the occurrence of lightning strikes based on the chaotic dynamics model, which is a crucial step in the entire method, enabling the system to understand and predict the temporal changes and spatial distribution of lightning strike events.

[0049] In this embodiment, we use the Logistic map and the Henon map in the chaotic dynamics model to describe the temporal changes and spatial distribution characteristics of lightning strikes. These models can provide precise dynamic changes during the data analysis process, capture the potential laws of lightning strike occurrences, and thus provide strong support for subsequent prediction and identification.

[0050] The chaotic dynamics model is an important tool for describing nonlinear systems. It can reveal the underlying regularities behind seemingly random behaviors in complex systems through some simple dynamic equations. The occurrence of lightning strike events is affected by many factors, such as meteorological conditions, electrical loads, etc. There are complex interactions among these factors. Therefore, we choose the chaotic model to capture this complexity.

[0051] Generally, the chaotic dynamics model can simulate the evolution process of the system under different inputs, reveal the relationship between the probability of lightning strike events and factors such as meteorological conditions and electrical loads. By constructing a lightning strike occurrence model based on chaos theory, the occurrence and evolution process of lightning strikes can be effectively predicted.

[0052] As an option, in this embodiment, the Logistic map is used to describe the probability change of lightning strikes. The Logistic map is a classical non-linear dynamics model used to simulate the growth process in a system with limited resources. By controlling parameters, this model can reflect the complex dynamic behavior of lightning strike probability with respect to environmental factors.

[0053] Specifically, the Logistic map formula is: Where: represents the probability of lightning strikes at the n-th moment, with a value range of [0, 1]; r log is the control parameter that determines the chaos degree of the system and reflects the sensitivity of lightning strike probability to meteorological conditions; is the lightning strike probability at the next time step.

[0054] Specifically, the Logistic map can show the dynamic change of lightning strike probability. When the control parameter r takes a relatively large value, the system will enter a chaotic state, resulting in violent fluctuations in the lightning strike probability; while when r takes a relatively small value, the system tends to be stable, and the change of lightning strike probability is relatively gentle. By analyzing the time series data of lightning strike events, an accurate dynamic model can be provided for subsequent lightning strike risk prediction.

[0055] In addition to the Logistic map, the Henon map is also applied to describe the spatial distribution characteristics of lightning strike events. Lightning strike events not only show fluctuations in time but also have specific distribution patterns in space. As a two-dimensional non-linear dynamics model, the Henon map can reveal the distribution of lightning strike events at different geographical locations and provide a powerful tool for the identification of spatial patterns.

[0056] Specifically, the Henon map describes the spatio-temporal distribution of lightning strike events through the following formula: Where: represents the state variable at the n-th step, describing the state of the system; y n represents another state variable at the n-th step, used to describe different environmental conditions related to ; a henon and b henon are the parameters of the map, controlling the evolution process of the system in time and space; is the state variable of the system at the (n + 1)-th step, representing the predicted state at the next time step.

[0057] In this model, x n may represent the meteorological conditions (such as temperature or humidity) in a certain area, while y nIt may represent the electrical load status of the area. Through the Henon mapping, it is possible to capture the hot spots and distribution patterns of lightning strikes in the spatial dimension, which helps to identify high-risk lightning areas.

[0058] Specifically, in this embodiment, the Logistic mapping and the Henon mapping not only independently model the temporal changes and spatial distributions, but can also be used in combination to jointly construct a complete lightning strike prediction model. Through the multi-level modeling of the temporal changes and spatial distributions of lightning strike events, the system can accurately predict the occurrence of lightning strikes.

[0059] By analyzing meteorological data, electrical data, and historical lightning strike data, and combining the output results of the Logistic mapping and the Henon mapping, it is possible to provide effective input data for the subsequent training of the deep learning model. The output of the chaotic dynamics model and the preprocessed data are used together as the training dataset and input into the subsequent deep learning model for feature extraction and classification.

[0060] S4. Extract and identify the features of lightning strikes based on the deep learning model. The deep learning model includes a convolutional neural network (CNN) and a long short-term memory network (LSTM), which are used to extract spatial and temporal features and perform classification. Specifically, in the aforementioned steps S1 and S2, we have successfully completed the collection and preprocessing of data to ensure the accuracy and applicability of the data. In step S3, the chaotic dynamics model effectively models the temporal and spatial features of lightning strike events, laying a foundation for subsequent feature extraction and analysis. Next, step S4 involves using the deep learning model to extract and identify the features of lightning strikes, which is a key link in the present invention and can automatically identify the potential features of lightning strikes and make accurate classifications in complex meteorological and power data.

[0061] In this embodiment, we adopt two deep learning methods, the convolutional neural network (CNN) and the long short-term memory network (LSTM). These two methods are respectively used to extract spatial features and temporal features. The CNN model is particularly suitable for extracting spatial features from image data, while the LSTM model can effectively capture the time-varying features of lightning strike risks through its processing ability for time series data.

[0062] Convolutional neural network (CNN): The lightning cloud map and the image data of the transmission line are one of the important bases for judging the occurrence of lightning strikes. The morphology, distribution of the lightning clouds in the image data, and the status of the power grid facilities often contain the precursor information of the occurrence of lightning strikes. Therefore, in this embodiment, using CNN to extract features from the image data can effectively identify the spatial patterns of lightning strikes.

[0063] As an option, in this embodiment, the CNN network extracts local and global features in the image through multiple convolutional and pooling operations. The convolutional operation processes the input image through a convolutional kernel, converting the important features in the image into a feature map. The pooling operation is used for dimensionality reduction and improving the invariance of features, reducing the computational amount.

[0064] Specifically, the convolutional layer of the CNN performs a convolutional operation on the image, and the filter extracts the features of the image, such as information about the shape, size, intensity, etc. of the thundercloud. After multiple convolutional and pooling operations, the obtained feature map contains spatial pattern information about the thundercloud. Through the fully connected layer, the CNN can use the extracted features for classification to further judge the risk of lightning strike events.

[0065] The typical structure of the CNN includes multiple convolutional layers and pooling layers, and finally outputs the prediction result of the lightning strike risk through a fully connected layer. For example, the output of the convolutional layer can be expressed as: Y = σ sigmoid (WX + b); where: X is the matrix of the input image; W is the convolutional kernel (or filter); b is the bias term; σ sigmoid is the activation function, such as ReLU; Y is the feature map after the convolutional operation.

[0066] Through this method, the CNN can automatically extract the potential features of lightning strikes in the image data and classify them, improving the accuracy of prediction.

[0067] Long short-term memory network (LSTM): As an option, the LSTM is used to process time series data, such as meteorological data and electrical data. The occurrence of lightning strike events is usually affected by meteorological conditions and electrical fluctuations, and these effects are often dynamically changing. Therefore, the LSTM model is particularly suitable for capturing these time-varying features.

[0068] Specifically, through its three internal gating structures (forget gate, input gate, and output gate), the LSTM can effectively remember the information of long-term dependencies and transfer useful information between different time steps. The structure of the LSTM network enables it to handle the long-term and short-term dependencies in the time series data and identify the key time points of lightning strike occurrences.

[0069] The update rule of the LSTM model is as follows: f t = σ sigmoid (W f [h t-1 , x t + b f ); i t = σ sigmoid (W i[h t-1 ,x t +b i ); o t =σ sigmoid (W o [h t-1 ,x t +b o ); h t =o t ·tanh(C t ); Where: h t-1 is the hidden state at the previous time step; x t is the input at the current time step; f t is the forget gate, controlling the forgetting of information; i t is the input gate, determining the storage of the current input information; is the candidate cell state, determining how much new information to store; C t is the cell state, remembering long-term information; o t is the output gate, controlling the current output; h t is the hidden state at the current time step; W f , W i , W C , W o are the weight matrices of each gate control; b f , b i , b C , b o are the bias terms of each gate control.

[0070] The LSTM model can utilize historical meteorological data, electrical data, and past lightning strike data, combine time dependencies, automatically learn the prediction rules of lightning strikes, and train these time series data through the LSTM network. The system can identify the time trend and potential patterns of lightning strikes.

[0071] Model Training and Optimization: Specifically, after feature extraction, the output features of the CNN and LSTM will be used as the input for lightning strike prediction and enter the subsequent classification and regression models. During the training process, the parameters of the CNN and LSTM networks are updated through the backpropagation algorithm to optimize the prediction performance of the model. When training the model, historical data with labels is used for supervised learning, and the weights and bias terms are gradually adjusted to minimize the prediction error.

[0072] As an option, during the training process, the Dropout technique can also be adopted to prevent overfitting, thereby improving the generalization ability of the model. The cross-validation method is used during the training process to evaluate the effect of the model and ensure its accuracy in practical applications.

[0073] S5. Optimize the prediction strategy of the model in combination with reinforcement learning, and adjust the model parameters through feedback; Specifically, in the aforementioned step S4, the lightning strike characteristics have been extracted and identified through a convolutional neural network (CNN) and a long short-term memory network (LSTM), generating spatial and temporal characteristics of lightning strike events. These characteristics not only provide data support for the risk prediction of lightning strike events but also lay a foundation for subsequent optimization and adjustment of the prediction strategy. On this basis, step S5 involves optimizing the prediction strategy of the model through reinforcement learning to improve the accuracy and real-time response ability of lightning strike prediction.

[0074] In this embodiment, we introduce a deep Q-network (DQN) to optimize the prediction strategy of the model. Reinforcement learning is a feedback-based learning technique. The model can gradually adjust the prediction strategy by interacting with the environment, reducing false alarms and missed alarms. In this way, the system can dynamically adjust the prediction results according to real-time data and optimize the early warning mechanism for lightning strike risks.

[0075] Introduction of reinforcement learning: Generally, in traditional machine learning models, model training is often carried out on a static training set, while reinforcement learning can automatically adjust model parameters according to real-time feedback through interaction with the environment. For lightning strike prediction, the environment includes information such as real-time changing meteorological conditions, electrical parameters, and historical lightning strike events, and the feedback comes from the comparison between the prediction results and the actual occurrence of lightning strike events.

[0076] As an option, this embodiment adopts a deep Q-network (DQN) as the implementation method of reinforcement learning. DQN combines the powerful expressive ability of deep learning with the adaptability of reinforcement learning, and can automatically optimize its decision-making strategy in the lightning strike prediction model to improve the accuracy of prediction results.

[0077] Specifically, DQN trains the model through a cyclic interaction mechanism of state, action, and reward. In DQN, the state represents the current input data of the model, such as meteorological data, electrical data, etc.; the action represents the prediction strategy selected by the model; and the reward is the difference between the model prediction result and the actual situation. By maximizing the reward, DQN can continuously optimize the prediction strategy of the model.

[0078] Optimization process of the Q-learning algorithm: Specifically, DQN is based on the Q-learning algorithm and estimates the value of each state-action pair Q(s t , a t ) ← Q(s t , a t ) + α(r t + γ max a Q(s t+1 , a) - Q(s t , a t )); Where: Q(s t , a t ) is the current value of taking action a t under state s t ; α is the learning rate, controlling the step size of each update; r t is the immediate reward of the current action, reflecting the difference between the prediction result of the model and the actual situation; γ is the discount factor, indicating the weight of future rewards; Q(s t+1 , a) is the maximum expected reward corresponding to the best action selected by the model under the next state s t+1 .

[0079] In a possible implementation, DQN continuously updates the Q-value table through repeated training and optimization to improve the adaptability of the model to lightning strike prediction. For example, under meteorological conditions with a high risk of lightning strikes, the model will gradually adjust the prediction strategy to reduce false negatives and false positives.

[0080] Model parameter adjustment: Specifically, during the training process, the reinforcement learning model makes self-adjustments based on feedback information. For example, if the prediction result at a certain time point is inaccurate, the system will adjust the model's strategy through the reward mechanism to enhance the accuracy of predictions under similar conditions. In this embodiment, the training process may also include techniques such as experience replay and target network to improve training efficiency and stability.

[0081] Experience replay refers to storing the state-action pairs encountered during the training process in an experience pool and randomly sampling them for learning during subsequent training. This process helps to break the correlation between data and improve the training efficiency.

[0082] The target network refers to introducing a target Q-network and periodically copying the parameters of the current Q-network to the target network to reduce the fluctuations during Q-value updates and improve the stability of training.

[0083] Optimization strategy: As an option, during the reinforcement learning training process, the system not only focuses on the prediction accuracy of lightning strikes but also considers how to respond in a timely manner when lightning strikes occur. Through reinforcement learning, the model can adjust the prediction strategy according to real-time data and ensure that early warnings can be issued in a timely manner. Through continuous optimization, DQN can enable the lightning strike prediction system to still make efficient and accurate decisions in the face of complex environments.

[0084] Through the optimization process of reinforcement learning, the model can gradually master the laws of lightning strikes, avoid the overfitting problem in traditional prediction methods. As the training progresses, the model can identify more potential lightning strike risk factors and make reasonable early warning responses when encountering similar environments.

[0085] S6. During the real-time monitoring process, when the lightning strike risk reaches a predetermined threshold, issue an early warning and execute an automated response.

[0086] Specifically, in the aforementioned steps S1 to S5, the collection and preprocessing of data, the modeling of the lightning strike occurrence model, the feature extraction of the deep learning model, and the optimization of the prediction strategy through reinforcement learning have been successfully completed. These steps lay a solid foundation for the real-time early warning and automated response functions of the present invention. In step S6, the system uses the outputs of all the aforementioned modules, combines the real-time monitoring data, completes the real-time prediction of lightning strike risks, and automatically activates the early warning mechanism when the lightning strike risk reaches a predetermined threshold, and further executes an automated response to ensure the safe operation of the power system.

[0087] In this embodiment, based on the lightning strike prediction results of the model output and combined with the real-time environmental monitoring data, the system can quickly respond to the occurrence of lightning strike events and send real-time early warnings to the grid operation and maintenance personnel. At the same time, when the predicted risk exceeds the set threshold, the system automatically executes a series of preventive measures, such as grid load adjustment, standby line switching, etc., to minimize the damage caused by lightning strikes.

[0088] Real-time early warning mechanism: Generally, the occurrence of lightning strikes is a random and difficult-to-predict event. Therefore, the system must be able to respond in advance when a lightning strike is about to occur. This mechanism requires the system to have high real-time performance and accuracy to ensure that grid operation personnel can obtain early warning information in a timely manner and take effective countermeasures.

[0089] As an option, in this embodiment, the system adopts a threshold-based early warning strategy. Specifically, when the lightning strike risk reaches a predetermined threshold, the system will automatically trigger the early warning mechanism to notify the grid operation and maintenance personnel. This threshold is obtained based on historical lightning strike data and real-time meteorological data analysis and can be dynamically adjusted according to actual needs. The early warning information includes key information such as the predicted time, location, and intensity of the lightning strike.

[0090] Specifically, the early warning mechanism can send notifications to the operation and maintenance personnel in various ways, including: SMS notification: Send a lightning strike risk early warning to the operation and maintenance personnel via SMS to ensure that timely warnings can be obtained even when the power system is offline.

[0091] Email Notification: Describe the predicted lightning strike risk and warning information in detail via email, enabling operation and maintenance personnel to make a quick response based on this information.

[0092] Real-time Platform Alarm: Through the real-time monitoring platform integrated into the power grid management system, provide an intuitive graphical interface to display lightning strike prediction information and send real-time alarms to operation and maintenance personnel through the platform.

[0093] These warning notifications can inform power grid operation and maintenance personnel in the shortest time, enabling them to make a quick response based on the real-time lightning strike risk and avoid or mitigate the damage caused by lightning strikes.

[0094] Automated Response Mechanism: In a possible implementation, once the system predicts that the lightning strike risk reaches the set threshold and issues a warning, the automated response mechanism is immediately activated to automatically perform a series of power grid adjustment operations to mitigate the impact of lightning strikes on the power system. These automated response measures can be executed quickly and accurately without manual intervention.

[0095] Specifically, the automated response includes the following aspects: Load Dispatching and Switching: In the case of a high lightning strike risk, the system will automatically perform load dispatching based on the real-time data of the load. By switching the power grid load from high-risk areas to low-risk areas, the impact of lightning strikes on the power system can be effectively reduced.

[0096] The decision-making basis for load dispatching is not only based on lightning strike prediction data but also comprehensively considers the current load situation of the power grid, the dynamic changes in power consumption, and other relevant electrical parameters.

[0097] Switching of Backup Lines: When the lightning strike risk is high, the system will automatically select a backup line to ensure that the power supply will not be interrupted when a lightning strike occurs. The system intelligently selects the most suitable backup line for switching based on the topology structure and load conditions of the power grid to ensure the continuity and stability of power transmission.

[0098] Automatic Operation of Power Grid Switching Equipment: During high-risk lightning strike periods, the system may automatically disconnect the connection of certain electrical equipment through switching equipment to avoid the impact of lightning strikes on power facilities. The operation of the switching equipment is judged by the system in real time based on lightning strike prediction information and the current operating state of the power grid to ensure the normal operation of the power grid.

[0099] Feedback Mechanism and Optimization: As an option, after the system executes early warning and automatic response, the feedback mechanism starts to work, further optimizing the early warning system and response strategy. By comparing the actual lightning strike events with the prediction results, the system can identify the errors in the prediction and make appropriate adjustments to the model. For example, if the system predicts a high lightning strike risk but no actual lightning strike occurs, the system will adjust the parameters in the prediction model through feedback data to reduce future false alarms.

[0100] Specifically, the system continuously collects feedback data and combines it with reinforcement learning algorithms to optimize the early warning threshold and response strategy. For example, in some regions or certain weather conditions, the occurrence pattern of lightning strikes may be relatively special. The feedback mechanism can help the system identify these patterns and make adjustments to make the early warning and response more accurate.

[0101] The intelligent on-line monitoring system for identifying lightning strike characteristics of intelligent transmission lines described below can be correspondingly referred to the intelligent on-line monitoring method for identifying lightning strike characteristics of intelligent transmission lines described above.

[0102] Please refer to the appendix Figure 2 , the intelligent on-line monitoring system for identifying lightning strike characteristics of intelligent transmission lines, includes: A data acquisition module, which is used to collect meteorological data, electrical data, historical lightning strike data and transmission line image data; A data processing module, which is used to preprocess the collected data, including denoising, normalization and standardization; A chaotic dynamics modeling module, which is used to build a non-linear dynamics model of lightning strike occurrence, including Logistic mapping and Henon mapping; A deep learning identification module, which includes a convolutional neural network CNN and a long short-term memory network LSTM, and is used to extract and identify lightning strike characteristics in the data; A reinforcement learning optimization module, which is used to optimize the strategy of the prediction model and adjust the parameters according to the feedback; An early warning and response module, which issues an early warning and automatically executes response measures when the lightning strike risk exceeds a predetermined threshold.

[0103] Specifically, the data acquisition module: This module is responsible for collecting the following data: Meteorological data: Real-time acquisition of information such as temperature, humidity, wind speed, and air pressure, providing meteorological conditions for lightning strike prediction; Electrical data: including data such as current, voltage, and load, monitoring the operation status of the power grid; Historical lightning strike data: including information such as lightning strike time, location, and intensity, supporting model training and lightning strike risk prediction; Transmission line image data: obtaining the status of power facilities through images, assisting in lightning strike analysis. All data will be uploaded to the central processing system in real time to ensure timeliness.

[0104] Data Processing Module: This module preprocesses the collected data, specifically including: Denoising: Removing data noise using methods such as Gaussian filtering; Normalization: Making different data types have a unified scale through maximum normalization and Z-Score standardization; Standardization: Ensuring consistent data standards, which is beneficial for subsequent analysis. These processing steps provide clean and consistent data for subsequent modeling and analysis.

[0105] Chaotic Dynamics Modeling Module: This module describes the temporal and spatial characteristics of lightning strikes based on chaotic dynamics models: Logistic mapping: Simulating the non-linear dynamics of lightning strike probability changing with meteorological and electrical environments; Henon mapping: Describing the distribution characteristics of lightning strike events in space to help identify high-risk areas. These models provide spatio-temporal characteristic data of lightning strikes for deep learning.

[0106] Deep Learning Identification Module: This module uses deep learning methods to extract lightning strike characteristics, mainly including: Convolutional Neural Network (CNN): Extracting key spatial features in lightning cloud images and transmission line images; Long Short-Term Memory Network (LSTM): Processing time series data to capture the time-varying patterns of lightning strikes. The combination of these two methods enables the system to process spatial and temporal features simultaneously, improving the prediction accuracy.

[0107] Reinforcement Learning Optimization Module: This module optimizes the prediction strategy through reinforcement learning: Strategy optimization: Adjusting the prediction model through Deep Q-Network (DQN) to reduce false alarms and missed alarms; Feedback learning: Adjusting the model according to the difference between actual data and prediction results to improve prediction accuracy. Reinforcement learning enables the system to adaptively adjust and improve the prediction effect.

[0108] Warning and Response Module: This module automatically initiates warning and response when the lightning strike risk exceeds the threshold: Lightning strike risk warning: Notifying grid operation and maintenance personnel in real time through text messages, emails, and the platform; Automatic response: Automatically adjusting the load, switching to standby lines, or starting protection devices to ensure grid safety.

[0109] The system of this embodiment can be used to execute the method embodiment above. Its principle and technical effects are similar and will not be elaborated here.

[0110] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principle and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent online monitoring method for intelligent identification of lightning strike characteristics of intelligent transmission lines, characterized in that: The following steps are involved: Collect multi-source data from different sensors, including meteorological data, electrical data, and historical lightning strike data; Preprocessing the multi-source data, including denoising, normalization and standardization; The occurrence of lightning strike is modeled based on a chaotic dynamics model, wherein the chaotic dynamics model uses Logistic mapping and Henon mapping to describe the probability change and spatial distribution characteristics of lightning strike; Extracting and identifying the characteristics of lightning strikes based on a deep learning model, wherein the deep learning model includes a convolutional neural network (CNN) and a long short-term memory (LSTM) network for extracting spatial and temporal features and performing classification; Combine reinforcement learning to optimize the model's prediction strategy and adjust model parameters through feedback; During real-time monitoring, when the lightning strike risk reaches a predetermined threshold, an early warning is issued and an automated response is performed.

2. The method for intelligent online monitoring of lightning strike characteristics of intelligent transmission lines according to claim 1 is characterized in that: The collected multi-source data include: Meteorological data, including temperature, humidity, wind speed, air pressure, and thunderstorm cloud maps; Electrical data, including current, voltage, load, and power consumption; Historical lightning strike data, including the time, location, intensity and duration of lightning strikes.

3. The method for intelligent online monitoring of lightning strike characteristics of intelligent transmission lines according to claim 1 is characterized in that: The pre-processing step comprises: Perform time series processing on electrical data from multiple sources to ensure the time synchronization of data; De-noise meteorological data and lightning strike data to remove outliers and errors; All data are normalized and standardized to ensure that different data sources have consistent scale and distribution.

4. The method for intelligent online monitoring of lightning strike characteristics of intelligent transmission lines according to claim 1 is characterized in that: The steps of the chaotic dynamics model include: The Logistic mapping model is used to describe the probability of lightning strikes, and the sensitivity of lightning strikes under different meteorological conditions is simulated by controlling parameters; The Henon mapping model is used to describe the spatial distribution characteristics of lightning events, considering the influence of time and space factors on the occurrence of lightning events.

5. The method for intelligent online monitoring of lightning strike characteristics of intelligent transmission lines according to claim 1 is characterized in that: The steps of the deep learning model include: Use convolutional neural network (CNN) to extract features from lightning cloud images and power transmission line images, and automatically learn spatial features in images; Use long short-term memory networks (LSTMs) to perform time series analysis on meteorological and electrical data to capture temporal dependencies. Combining the CNN and LSTM models, the extracted features are used to identify and classify lightning strike features through the fully connected layer.

6. The method for intelligent identification and online monitoring of lightning strike characteristics of intelligent power transmission lines according to claim 1 is characterized in that: The steps of reinforcement learning include: The prediction model is trained through the deep Q network DQN to optimize the prediction strategy; Update model parameters based on real-time feedback of lightning strike data to improve prediction accuracy; Reinforcement learning automatically adjusts the model prediction strategy through a feedback mechanism to improve the system's prediction ability and reduce false positives and false negatives.

7. The method for intelligent identification and online monitoring of lightning strike characteristics of intelligent power transmission lines according to claim 1 is characterized in that: The lightning strike risk warning step comprises: When the lightning strike risk predicted by the model exceeds a predetermined threshold, the system issues an early warning; Warning information is notified to power grid operation and maintenance personnel via SMS, email or real-time monitoring platform; During high-risk phases, the system can automatically execute automated response measures such as switching backup lines and adjusting loads to mitigate the impact of lightning strikes.

8. The method for intelligent identification and online monitoring of lightning strike characteristics of intelligent power transmission lines according to claim 1 is characterized in that: The feedback adjustment step comprises: Dynamically adjust model parameters based on the deviation between actual lightning strike data and model prediction results; The model prediction error is calculated through a feedback mechanism, and the weights are updated through an optimization algorithm to improve the stability and accuracy of the model.

9. The method for intelligent online monitoring of lightning strike characteristics of intelligent transmission lines according to claim 1, characterized in that: The training process of the model includes: Use historical lightning strike data and real-time monitoring data to train deep learning models; The model adjusts internal parameters through continuous training and verification, and tests its accuracy through cross-validation; Reinforcement learning improves the model's predictive ability and reduces false positive and false negative rates through multiple rounds of training and optimization.

10. An intelligent transmission line lightning strike feature intelligent identification online monitoring system, applied to the intelligent transmission line lightning strike feature intelligent identification online monitoring method according to any one of claims 1 to 9, characterized in that: include: Data acquisition module, used to collect meteorological data, electrical data, historical lightning strike data and transmission line image data; Data processing module, used to pre-process the collected data, including denoising, normalization and standardization; Chaotic dynamics modeling module, used to build nonlinear dynamics models of lightning strikes, including Logistic mapping and Henon mapping; Deep learning recognition module, including convolutional neural network (CNN) and long short-term memory (LSTM) network, is used to extract and identify lightning strike features in the data; A reinforcement learning optimization module is used to optimize the prediction model’s strategy and adjust parameters based on feedback; The early warning and response module issues an early warning and automatically executes response measures when the lightning strike risk exceeds a predetermined threshold.

Images