Intelligent Detection System and Method for Transmission Lines of Distribution Network Based on AI Algorithm

Abstract

The present invention discloses an intelligent detection system and method for power distribution network transmission lines based on AI algorithms, including a data acquisition module, a data preprocessing module, a multi-modal data fusion module, an AI fault diagnosis module, a control execution module, and a communication module. By integrating various sensor data and adopting multi-modal data weighted fusion technology, it realizes comprehensive monitoring of the operating status of transmission lines. Based on the improved dynamic multi-modal adaptive optimization model DMAOM, it analyzes historical and current data in real time and has the ability of self-learning. A flexible adaptive control strategy is introduced to dynamically adjust the control scheme according to real-time data. In case of anomalies, the system can quickly execute current limiting, load transfer, or circuit breaking operations. Through the feedback mechanism, the system can continuously optimize the monitoring and response strategies after each fault handling. The present invention significantly improves the safety, stability, and operating efficiency of power distribution network transmission lines, and has important practical application value and broad market prospects.

Description

Technical Field

[0001] The present invention relates to the technical field of monitoring and control of power distribution network transmission lines, and particularly to an intelligent detection system and method for power distribution network transmission lines based on AI algorithms, which are specifically applied to the state monitoring, fault prediction and optimized control of load scheduling of transmission lines. Background Art

[0002] In modern power transmission systems, with the expansion of the power grid scale and the growth of power demand, the reliability and safety management of transmission lines have become crucial. Traditional transmission line monitoring systems mainly rely on a single type of sensor (such as temperature and humidity sensors, current sensors, etc.) to collect data and monitor the lines based on fixed control strategies. These systems can complete basic state monitoring and control operations and have certain practicality under stable operating conditions.

[0003] However, with the increasingly complex operating environment of transmission lines, many deficiencies have emerged in existing monitoring systems. First, the single-modal sensor acquisition method cannot comprehensively reflect the operating state of transmission lines. Especially in cases where environmental conditions change frequently or load fluctuations are large, data acquisition is lagging and inaccurate. Second, the fixed control strategy lacks the ability of adaptive adjustment, resulting in the system being unable to respond in a timely manner when a fault occurs. In addition, existing systems fail to effectively integrate multi-modal data (such as temperature and humidity, traveling waves, current, etc.) and cannot perform in-depth analysis and fault prediction, resulting in limited early warning capabilities of the system and prone to major faults.

[0004] In order to improve the operating efficiency and safety of transmission lines, there is an urgent need to be able to integrate dynamic data from multiple sensors in real time, perform fault prediction through intelligent algorithms, and automatically adjust control strategies according to real-time situations, thereby enhancing the system's ability to respond to complex operating states, significantly reducing the occurrence of line faults, and improving the stability and reliability of the power grid. Summary of the Invention

[0005] In view of the problems of the existing power distribution network transmission line monitoring system, such as single data acquisition, lack of adaptive adjustment and insufficient fault prediction ability, the present invention proposes an intelligent detection system and method for power distribution network transmission lines based on AI algorithms. The system adopts multi-modal data fusion technology, collects data through sensors such as temperature and humidity, current and fault detection, and combines AI algorithms such as convolutional neural networks and long short-term memory networks to dynamically adjust control strategies and achieve real-time fault prediction. The system has self-learning ability and can continuously optimize control strategies according to the historical operating data and real-time monitoring data of transmission lines, significantly improving the safety and stability of transmission lines.

[0006] To achieve the above object, the present invention is realized through the following technical solutions:

[0007] An intelligent detection system for power transmission lines in a distribution network based on AI algorithms, the system comprising:

[0008] A data acquisition module for real-time collecting operation status data and environmental parameter data of the power transmission line through a plurality of sensors, the sensors including a temperature and humidity sensor, a traveling wave distance measuring sensor, a current sensor, and a fault detection sensor, and the data acquisition module performing data transmission through a distributed network of a wireless local area network WLAN or a cellular network;

[0009] A data preprocessing module for preprocessing the temperature, humidity, traveling wave, current, and fault detection data sent by the data acquisition module, the preprocessing including using a data denoising algorithm based on wavelet transform, abnormal data filtering, and preliminary fault feature extraction;

[0010] A multi-modal data fusion module for receiving the cleaned data and preliminary fault features transmitted by the data preprocessing module, comprehensively analyzing the temperature, humidity, traveling wave, and current data from different sensors through a multi-modal data fusion algorithm, and combining and extracting the preliminary fault features to generate a state evaluation result and fault prediction information of the power transmission line;

[0011] An AI fault diagnosis module for receiving the state evaluation result and fault prediction information generated by the multi-modal data fusion module, deeply diagnosing the operation state of the power transmission line based on an improved dynamic multi-modal adaptive optimization model DMAOM, and combining historical fault data and real-time analysis results to generate a detailed fault diagnosis report;

[0012] A control execution module, including a circuit breaker, a switch, and dispatching equipment, for receiving the fault diagnosis report generated by the AI fault diagnosis module and performing control operations of fault isolation, load transfer, or line protection according to the information in the report;

[0013] A communication module for transmitting data and control instructions between the data acquisition module, the data preprocessing module, the multi-modal data fusion module, the AI fault diagnosis module, and the control execution module, and realizing real-time interaction of each module of the system through wireless communication or wired communication.

[0014] Preferably, the temperature and humidity sensor and the traveling wave distance measuring sensor in the data acquisition module can adaptively adjust the sampling frequency by detecting real-time data of external temperature and humidity changes or line load status, and the data acquisition module performs data transmission with the data preprocessing module through a low-power Bluetooth BLE module.

[0015] Preferably, the data preprocessing module uses a Convolutional Neural Network (CNN) model to extract fault features from the preprocessed temperature, humidity, traveling wave, and current data. The CNN model is pre-trained with historical fault data and adaptively updated based on new fault data through dynamic adjustment based on online learning during system operation.

[0016] Preferably, the multi-modal data fusion module uses a combination of a Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), self-attention mechanism, and Graph Neural Network (GNN) along with a reinforcement learning algorithm to perform weighted fusion analysis on temperature, humidity, traveling wave, and current data from different sensors. The weighting coefficients are based on the Bayesian model in multi-modal data fusion and are dynamically adjusted through an online learning model according to the environmental temperature, humidity, and load status of the transmission line.

[0017] Preferably, the implementation mechanism of the multi-modal data weighted fusion module includes:

[0018] S1: Abnormal detection and noise rejection mechanism. The intelligent abnormal detection mechanism based on Support Vector Machine (SVM) is used to reject the noise data in the sensors. The dynamic interpolation algorithm based on time series is used to automatically align the multi-modal data, and adaptive normalization processing is performed to ensure data consistency.

[0019] S2: Local feature and temporal dynamic relationship extraction mechanism. Combining a Convolutional Neural Network (CNN) and a Long Short-Term Memory Network (LSTM), the local features and temporal dynamic relationships of each modal data are extracted to optimize the correlation between data.

[0020] S3: Weight setting and adaptive adjustment mechanism. Through the unsupervised learning algorithm based on K-means clustering, the initial weights of each modal data are set based on historical data, and the weights of each modal data are dynamically adjusted through the self-attention mechanism.

[0021] S4: Multi-level fusion structure. The multi-level fusion structure of a Deep Neural Network (DNN) is used to extract and fuse the relationships between different modal data. Combining the topological information of the transmission line, the fused data is weighted by a Graph Neural Network (GNN).

[0022] S5: Abnormal data identification and fault tolerance mechanism. The intelligent abnormal detection mechanism is used to identify and reject abnormal situations in the sensor data, and the redundancy and fault tolerance mechanism of multi-modal data is used to ensure the reliability and robustness of data processing.

[0023] S6: Computational optimization mechanism. The amount of computation is reduced through model pruning and quantization, and combined with distributed computing, some computational tasks are offloaded to distributed nodes.

[0024] S7: Time series modeling and uncertainty quantification mechanism, which combines the long short-term memory network (LSTM) to enhance the time series modeling of the system, and quantifies and analyzes the uncertainty in fault prediction through Bayesian deep learning technology.

[0025] Preferably, the AI fault diagnosis module dynamically adjusts the system response mode by learning the effects of different fault handling solutions in real time through the improved dynamic multi-modal adaptive optimization model (DMAOM), and generates a detailed fault diagnosis report including the fault type, occurrence time, and expected impact range by combining historical fault data and the fault prediction information generated by the multi-modal data fusion module, and provides targeted repair suggestions.

[0026] Preferably, the implementation steps of the improved dynamic multi-modal adaptive optimization model (DMAOM) are as follows:

[0027] Step 1: Integrate historical fault data and real-time multi-modal data including temperature, current, voltage, and humidity sensor data, and use the dynamic data fusion algorithm based on the Bayesian model to perform weighted integration of real-time data and historical data. The calculation formula is as follows:

[0028] X fuse = αX real +(1 - α)X hist

[0029] where, X fuse is real-time data; X real is the sensor data of real-time multi-modal data; X hist is historical data; α is the fusion weight coefficient;

[0030] Apply the self-attention mechanism to extract important global features in fault diagnosis:

[0031]

[0032] where, Q, K, and V are feature matrices; d k is the dimension of the key matrix; softmax is the normalization function, which calculates the similarity between the key and the query and converts it into a probability distribution;

[0033] Step 2: Through self-supervised learning technology, combined with unlabeled data and contrastive learning methods, dynamically learn the fault patterns of the distribution network, and optimize the diagnosis model through the contrastive learning loss function at each time step:

[0034]

[0035] where, L i is the contrastive loss value of the i-th sample; x i is the encoded vector representation of the current sample; and xi The relevant positive samples, related to x i the relevant negative samples; sim(,) is the similarity between two samples; τ is the temperature parameter;

[0036] Step 3: The fault handling strategy is dynamically optimized through reinforcement learning. After each fault handling, the system optimizes the strategy according to the current state s t , the executed action a t for policy optimization:

[0037]

[0038] where π new is the new policy; s t is the state at time step t; Q(s t , a t ) is the Q-value of the current state and action, representing the expected return after executing action a t under state s t ; α is the entropy coefficient, controlling the balance between the reward and the policy entropy; logπ(s t |a t ) is the entropy of the policy, representing the randomness of choosing action a t under state s t ;

[0039] Step 4: The system receives real-time feedback on the processed effects through a feedback loop mechanism, continuously optimizing the fault diagnosis and handling strategies. The feedback information f t is introduced into the loss function:

[0040] L new = L(f t ) + βL previous

[0041] where L new is the new loss value for optimizing the model; L(f t ) is the current loss calculated based on the feedback information; L previous is the loss of the previous iteration; β is the feedback weight coefficient, controlling the balance between the new feedback and the historical loss;

[0042] Step 5: Apply the multi-modal fault handling optimization mechanism based on the diffusion model to gradually optimize the fault prediction results through the inverse diffusion process:

[0043] p θ (x t-1 ∣x t ) = N(x t-1 ; μ θ (x t , t), Σ θ (xt , t))

[0044] Among them, p θ (x t-1 ∣x t ) is the probability distribution of generating the data at time step t - 1 under the condition of the noisy data x t ; N(x t-1 ; μ θ (x t , t), Σ θ (x t , t)) is a normal distribution, which is used to model the generated data distribution in the reverse diffusion process; μ θ (x t , t) is the mean parameterized by the neural network, representing the mean estimate of the generation; Σ θ (x t , t) is the variance parameterized by the neural network, representing the uncertainty in the generation process; x t is the noisy data at time step t.

[0045] Preferably, the control execution module includes:

[0046] Current limiting operation unit: According to the preset current threshold and voltage threshold, when the current or voltage exceeds the preset range, perform current limiting operation, and the current limiting threshold is dynamically adjusted according to the load status and environmental parameters;

[0047] Load transfer and circuit breaker operation unit: When the current limiting operation fails to reduce the current or voltage within the safe range within the preset time and exceeds the specified number of times, perform load transfer or circuit breaker operation;

[0048] Self - inspection unit: In the case of multiple failures of current limiting or load transfer operations, the system automatically enters the self - inspection mode, and the internal fault detection module monitors the hardware and software status in the control execution system in real time;

[0049] Overload protection unit: When the system load exceeds the preset limit, automatically cut off some circuits to prevent overload damage;

[0050] Remote monitoring interface: Connect to the remote monitoring terminal through the Internet, allowing operators to view the system status in real time and perform remote operations.

[0051] Preferably, a method for intelligent detection of a power distribution network transmission line based on an AI algorithm, characterized in that the method includes the following steps:

[0052] Step 1: Real-time collect the operation status data and environmental parameter data of the transmission line through multiple sensors. The sensors include temperature and humidity sensors, traveling wave distance measuring sensors, current sensors, and fault detection sensors. The data is transmitted to the data preprocessing module through a wireless local area network (WLAN) or a cellular network.

[0053] Step 2: In the data preprocessing module, process the data. The preprocessing includes data denoising based on wavelet transform, abnormal data filtering, and preliminary fault feature extraction.

[0054] Step 3: Input the cleaned data into the multimodal data fusion module. Combine convolutional neural network (CNN), long short-term memory network (LSTM), and graph neural network (GNN) algorithms to perform multimodal data fusion analysis, and generate the status evaluation result and fault prediction information of the transmission line.

[0055] Step 4: Apply the improved dynamic multimodal adaptive optimization model (DMAOM). Combine historical fault data and real-time multimodal data, use the Bayesian model to perform dynamic multimodal data weighted fusion on the data, extract global features, and dynamically optimize the fault diagnosis model through self-supervised learning and reinforcement learning.

[0056] Step 5: AI fault diagnosis. Based on the fault prediction information and historical data generated by the DMAOM model, use deep learning and reinforcement learning algorithms to generate a fault diagnosis report. The report includes the fault type, occurrence time, expected impact range, and provides repair suggestions.

[0057] Step 6: Control execution. According to the fault diagnosis report, perform current limiting, load transfer, or circuit breaker operations. After multiple current limiting or load transfer operations fail, automatically enter the self-check mode to detect the hardware and software status through real-time monitoring.

[0058] Preferably, the dynamic multimodal adaptive optimization model (DMAOM) is implemented through the following steps:

[0059] Step S1: Input various sensor data of the transmission line, including temperature and humidity sensors, traveling wave distance measuring sensors, current sensors, and fault detection sensors, into the DMAOM model for multimodal data fusion. The significant improvements of the DMAOM model include:

[0060] Step S1.1: Use the Bayesian model to dynamically adjust the fusion weights of multimodal data according to the real-time monitored environmental parameters, including temperature, humidity, and load status. The initial weights are set through the historical data set and updated based on the new real-time data.

[0061] Step S1.2: Extract global features from multi-modal data through self-supervised learning techniques, identify the correlations between different data modalities in the transmission line, and extract important features related to fault prediction;

[0062] Step S1.3: Based on the data correlations in the multi-modal data fusion process, detect abnormal data in real time and correct the anomalies through an adaptive correction mechanism;

[0063] Step S2: Adaptively optimize the fault prediction model through reinforcement learning. The significant improvements in the model optimization include:

[0064] Step S2.1: After each fault prediction, the system adjusts the model parameters according to the feedback information, which includes the accuracy of fault occurrence, response time, and system control effect;

[0065] Step S2.2: Combine the long short-term memory network LSTM for multi-step time series modeling to further optimize the correlation analysis between historical data and real-time data;

[0066] Step S3: Based on the outputs of Step S1 and Step S2, generate fault prediction information and a detailed fault diagnosis report, which includes the fault type, occurrence time, affected range, and optimized repair suggestions.

[0067] Term Explanation:

[0068] Multi-modal data fusion: Multi-modal data fusion refers to integrating data from different sensors (such as temperature, humidity, current, traveling wave), and synthesizing information from multiple modalities to obtain a more comprehensive and accurate assessment of the transmission line status.

[0069] Implementation method: The system performs weighted fusion on the data from different sensors through deep learning algorithms such as convolutional neural network CNN and long short-term memory network LSTM. The specific implementation methods include:

[0070] First, extract features from multi-modal data such as temperature, humidity, and current;

[0071] Use CNN to extract local features and LSTM to capture the temporal dynamic relationships of the data;

[0072] Dynamically adjust the weights of different data sources through a Bayesian model to ensure that important data modalities can be preferentially processed in different environments.

[0073] Actual application scenario: When the load status of the line fluctuates, the weight of the current data may increase, while the weights of the temperature and humidity data are adjusted according to environmental changes. This dynamic weighted fusion improves the accuracy of data analysis.

[0074] DMAOM is an adaptive optimization model that can dynamically adjust the system response mode according to different fault handling effects. It realizes real-time optimization and fault diagnosis by integrating multimodal data, combining historical fault data and real-time data.

[0075] Implementation method: DMAOM is implemented through the following steps:

[0076] Integration of historical and real-time data: Based on the Bayesian model, real-time data (such as temperature, current) and historical fault data are weighted and integrated to generate a more accurate assessment of the current state.

[0077] Self-supervised learning: By learning from unlabeled data, fault patterns are dynamically identified. At each time step, the model optimizes the accuracy of fault detection through contrastive learning.

[0078] Feedback optimization: DMAOM can optimize the model parameters based on the feedback mechanism after each fault handling, making the next fault detection more accurate.

[0079] Actual application scenario: When a fault occurs in a transmission line, DMAOM can quickly judge the nature of the fault by adaptive adjustment, combined with historical data, and continuously optimize the fault handling plan according to the system feedback to avoid similar faults from occurring again.

[0080] Wavelet transform is a commonly used signal processing technology that can decompose complex signals into multiple frequency components, facilitating noise removal and retaining important signal features.

[0081] Implementation method: In the present invention, wavelet transform is used to process the noise in sensor data such as current and traveling waves. The specific steps of the algorithm include:

[0082] First, the collected raw data is wavelet decomposed, and the signal is divided into high-frequency and low-frequency components;

[0083] The high-frequency noise components are removed, and the low-frequency signal is retained;

[0084] The clean signal is restored through wavelet reconstruction, thereby reducing the impact of noise on system fault detection.

[0085] Actual application scenario: When the signal collected by the current sensor is mixed with high-frequency noise due to electromagnetic interference, wavelet transform can effectively remove these noises, enabling the system to analyze the current data more accurately and avoid false alarms or missed fault reports.

[0086] Convolutional neural network is a deep learning algorithm commonly used in image processing and pattern recognition. It can extract local features through convolutional operations and is suitable for processing data with spatial or temporal structures.

[0087] Implementation method: In the present invention, CNN is used to extract features from multi-modal sensor data:

[0088] First, local features are extracted from the input data through convolutional layers;

[0089] Pooling layers are used to reduce the data dimension and retain key features;

[0090] Finally, the extracted features are integrated through fully connected layers, and high-dimensional feature vectors are output for subsequent fault detection.

[0091] Actual application scenario: When current data and traveling wave data are both abnormal, CNN can automatically identify the abnormal patterns in these data, extract fault-related features, and help the system judge the line status faster.

[0092] Reinforcement learning is a machine learning method. Through interaction with the environment, the agent learns the optimal decision-making strategy through trial and error. The agent executes actions according to the state and adjusts the strategy according to the obtained rewards.

[0093] Implementation method: In the present invention, reinforcement learning is used to optimize the control strategy of transmission lines. The specific implementation includes:

[0094] State: The current state of the transmission line, such as load, current, temperature, etc.

[0095] Action: Actions that the system can take, including current limiting, load transfer, circuit breaking, etc.

[0096] Reward: The system adjusts the strategy according to the feedback after executing the action (such as whether the fault handling is successful). For example, the system obtains a positive reward after successfully transferring the load, and a negative reward if the fault cannot be handled in time.

[0097] Actual application scenario: When a certain line is overloaded, the reinforcement learning algorithm can adjust the current limiting or load transfer strategy according to the system historical data, and continuously optimize the control decision through the feedback mechanism.

[0098] Self-supervised learning is a learning method that does not require labeled data. The system generates supervision signals through its own structure and learns using the internal structure of the data or the contrast method.

[0099] Implementation method: In the present invention, self-supervised learning is used to identify fault patterns in transmission lines:

[0100] By inputting unlabeled sensor data, the system spontaneously generates pseudo-labels for constructing training samples;

[0101] Through contrastive learning, the model gradually identifies potential fault patterns, thereby improving the accuracy of fault prediction.

[0102] Actual application scenario: When the system is in the early operation stage and there is insufficient historical fault data, self-supervised learning can utilize unlabeled sensor data to automatically generate fault patterns for optimizing the system's fault identification.

[0103] Support Vector Machine (SVM) is a machine learning algorithm commonly used in classification tasks. In anomaly detection, SVM is used to identify anomaly points in data to ensure that the system can promptly detect abnormal working states.

[0104] Implementation method: In the present invention, SVM is used to detect anomalies in sensor data: First, normal data is trained to construct a high-dimensional classification boundary; when new data is input, the system determines whether it is outside the boundary, and if it exceeds the boundary, it is determined as abnormal data.

[0105] Actual application scenario: When the current sensor collects abnormal current fluctuations, SVM can quickly identify and determine whether these fluctuations are abnormal, helping the system make timely adjustments.

[0106] Time series interpolation algorithm is used to fill in missing values or inconsistent time intervals in data acquisition to ensure that multi-modal data can be accurately aligned for subsequent analysis.

[0107] Implementation method: In the present invention, the time series interpolation algorithm is used to solve the problem of inconsistent sampling rates of different sensors:

[0108] When data is missing at certain time points, the system automatically uses linear interpolation or spline interpolation algorithms to generate estimated values;

[0109] The interpolated data is then aligned with the data of other sensors to ensure data synchronization.

[0110] Actual application scenario: When the current sensor fails to collect data at a certain time point, the system generates an estimated current value at that time point through the interpolation algorithm, thus avoiding the impact of data loss on fault detection.

[0111] Self-attention mechanism is a deep learning technique that calculates the dependency relationships between different data points, assigns different weights to each data point, and is used to capture global features.

[0112] Implementation method: In the present invention, the self-attention mechanism is used to analyze important features in multi-modal data:

[0113] The system constructs query matrix, key matrix, and value matrix, and calculates the similarity between them;

[0114] The system converts the similarity into weights through the softmax function, and determines which data is more important according to the weight distribution.

[0115] Actual application scenario: When the system detects multimodal data, the self-attention mechanism can identify the data that is most important for current fault diagnosis, such as current or temperature changes at certain critical moments.

[0116] Graph neural network is a deep learning algorithm used to process graph-structured data. In the present invention, GNN is used to combine the topological structure of the transmission line and analyze the relationships between sensors and data flows.

[0117] Implementation method: In GNN, sensors are represented as nodes of the graph, and the data transmission or physical connection between sensors is represented as edges of the graph. By propagating and aggregating the features of nodes, GNN can learn the global information of the entire network and make predictions.

[0118] Actual application scenario: When multiple sensors in the transmission line detect anomalies simultaneously, GNN can analyze the associations between sensors based on the topological structure of the line, thereby accurately determining the fault location and type.

[0119] Model pruning and quantization are optimization techniques for deep learning models, aiming to reduce the computational amount and storage space, making the model more efficient, especially suitable for embedded systems or edge computing devices.

[0120] Implementation method: Model pruning: By removing unimportant neurons or connections in the network, the complexity of the model is reduced;

[0121] Quantization: Convert the floating-point operations in the model into fixed-point operations to further reduce the computational overhead and memory occupancy.

[0122] Actual application scenario: In the present invention, pruning and quantization techniques are applied to edge computing devices of distributed nodes, reducing power consumption and computational latency, ensuring that the system can process multimodal data in real time and respond quickly to faults.

[0123] Compared with the prior art, the beneficial effects of the present invention are as follows: By integrating real-time data from multiple sensors and adopting a multi-modal data weighted fusion module, the present invention can comprehensively reflect the operating state of the transmission line. Through the weighted mechanism based on the Bayesian model, the system can dynamically adjust the fusion weights of different modal data, significantly improving the accuracy of fault detection, and thus more effectively monitoring and analyzing the health status of the line in a complex environment. Based on the improved dynamic multi-modal adaptive optimization model DMAOM, the fault diagnosis module can analyze historical data and current data in real time and has good self-learning ability. Through the deep learning model, the system can not only identify and predict potential faults, but also continuously optimize the model parameters during the processing, take preventive measures in advance, reduce the risk of fault occurrence, and ensure the stability of the power grid. The present invention has flexible adaptive control ability and can dynamically adjust the control strategy according to real-time monitoring data. This ability enables the system to respond in a timely manner and optimize the operation of the power grid when the load and external environment change, significantly improving the safety and efficiency of the power grid. In the event of an anomaly, the system can quickly formulate and execute response strategies, such as current limiting, load transfer, or circuit breaker operation. Through an efficient dynamic response mechanism, the reliability of the power grid in emergencies is ensured, and the risk of large-scale power outages caused by faults is reduced. The feedback mechanism introduced in the present invention enables the system to continuously optimize the monitoring and response strategies based on the actual situation after each fault handling. This mechanism, combined with the dynamic learning ability of DMAOM, improves the intelligence level of the system, enabling it to cope with more unknown challenges in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0124] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings. Among them:

[0125] Figure 1 is a schematic diagram of the overall structure of the system according to an embodiment of the present invention;

[0126] Figure 2 is a detailed processing flow of multi-modal data fusion in an embodiment of the present invention;

[0127] Figure 3 is a working flowchart of the control execution module in an embodiment of the present invention;

[0128] Figure 4 is a flowchart of the method according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0129] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the described embodiments of the present invention fall within the scope of protection of the present invention.

[0130] Embodiment 1

[0131] As Figure 1 shown, this is an embodiment of the present invention, which provides an intelligent detection system for power transmission lines in a distribution network based on an AI algorithm, including:

[0132] A data acquisition module, configured to collect real-time operation status data and environmental parameter data of the power transmission line through multiple sensors. The sensors include a temperature and humidity sensor, a traveling wave distance measurement sensor, a current sensor, and a fault detection sensor. The data acquisition module performs data transmission through a distributed network of a wireless local area network (WLAN) or a cellular network;

[0133] Specifically, the temperature and humidity sensor and the traveling wave distance measurement sensor in the data acquisition module can detect real-time data of external temperature and humidity changes or line load status, and adopt an adaptive sampling frequency. The data acquisition module performs data transmission with the data preprocessing module through a low-power Bluetooth (BLE) module.

[0134] This module is responsible for collecting real-time operation status data and environmental parameter data of the power transmission line through multiple sensors. It is the basic data source of the entire system and directly affects the accuracy of subsequent processing and decision-making.

[0135] Types of sensors:

[0136] The data acquisition module integrates multiple sensors to comprehensively monitor the operation status of the power transmission line. Including:

[0137] Temperature and humidity sensor: Used to collect temperature and humidity data of the environment around the power transmission line. Changes in temperature and humidity may directly affect the load capacity and fault incidence of the line.

[0138] Traveling wave distance measurement sensor: Used to detect traveling wave phenomena in the power transmission line, so as to accurately measure the electrical fluctuation conditions in the line. This information is very important for fault location.

[0139] Current sensor: Used to monitor the current situation of the power transmission line in real time to detect the load status of the line and identify whether there is a risk of overload or short circuit.

[0140] Fault detection sensors: used to monitor and identify potential fault points in transmission lines, and provide early fault warnings in conjunction with other sensor data.

[0141] Data transmission method:

[0142] The data acquisition module transmits data through a wireless local area network (WLAN) or a cellular network. The use of a distributed network architecture ensures that the data collected at different sensor locations is transmitted to the central processing unit in a timely manner, ensuring the real-time and reliability of the data.

[0143] Application scenarios: In actual applications, temperature and humidity sensors and current sensors can be deployed at key nodes of power transmission lines to monitor changes in the environment and load in real time. When the line load exceeds a certain threshold, the current sensor will trigger an alarm and transmit the data to the preprocessing module for further analysis.

[0144] Furthermore, a data preprocessing module preprocesses the temperature, humidity, traveling wave, current and fault detection data sent by the data acquisition module, wherein the preprocessing includes adopting a data denoising algorithm based on wavelet transform, abnormal data filtering and preliminary fault feature extraction;

[0145] Specifically, the data preprocessing module uses a convolutional neural network (CNN) model to extract fault features from the preprocessed temperature, humidity, traveling wave and current data. The CNN model is pre-trained with historical fault data and adaptively updated according to new fault data through dynamic adjustment based on online learning when the system is running.

[0146] The data preprocessing module cleans, denoises and extracts preliminary features from the raw data from the data acquisition module. This step ensures that the data input to the subsequent modules is accurate and reliable, which helps to improve the overall stability of the system and the accuracy of fault detection.

[0147] Preprocessing algorithm:

[0148] Wavelet transform denoising algorithm: It is mainly used to remove high-frequency noise in the sensor acquisition process. Wavelet transform can effectively separate the noise components in the data, thereby retaining key signal features. For example, in the current sensor signal, there may be high-frequency interference. Wavelet transform can remove this interference and retain the real current fluctuation information.

[0149] Abnormal data filtering: Based on the set rules or anomaly detection algorithms, invalid or abnormal values ​​in the data are eliminated. For example, if a sensor suddenly has extreme data, it may be a sensor failure or an environmental emergency. This module can filter out this data.

[0150] Initial Fault Feature Extraction: By initially processing temperature, humidity, traveling wave, and current data, some possible fault features are extracted. These features can be used to preliminarily determine whether there are potential fault hazards in the transmission line and provide a basis for subsequent data fusion.

[0151] Application Scenario: When the current sensor detects a rapid increase in current, the system will automatically call the wavelet transform denoising algorithm to ensure that the acquired current waveform data is clear and reliable. Then, the preprocessing module will determine whether there are abnormal situations according to the preset threshold and extract the corresponding fault features.

[0152] Furthermore, as Figure 2 shown, the multimodal data fusion module is used to receive the cleaned data and initial fault features transmitted by the data preprocessing module, comprehensively analyze the temperature, humidity, traveling wave, and current data from different sensors through the multimodal data fusion algorithm, and combine with the extracted initial fault features to generate the state evaluation result and fault prediction information of the transmission line;

[0153] Specifically, the multimodal data fusion module uses a combination of convolutional neural network CNN, long short-term memory network LSTM, self-attention mechanism, and graph neural network GNN with a reinforcement learning algorithm to perform multimodal data weighted fusion analysis on the temperature, humidity, traveling wave, and current data from different sensors. The weighting coefficients are based on the Bayesian model in multimodal data fusion and are dynamically adjusted through an online learning model according to the ambient temperature, humidity, and load status of the transmission line.

[0154] Specifically, the implementation mechanism of the multimodal data weighted fusion module includes:

[0155] S1: Abnormal Detection and Noise Rejection Mechanism, which eliminates the noise data in the sensors through an intelligent abnormal detection mechanism based on support vector machine SVM, uses a dynamic interpolation algorithm based on time series to achieve automatic alignment of multimodal data, and ensures data consistency through adaptive normalization processing;

[0156] S2: Local Feature and Temporal Dynamic Relationship Extraction Mechanism, which combines convolutional neural network CNN and long short-term memory network LSTM to extract the local features and temporal dynamic relationships of each modal data and optimize the correlation between data;

[0157] S3: Weight Setting and Adaptive Adjustment Mechanism, which sets the initial weights of each modal data based on historical data through an unsupervised learning algorithm based on K-means clustering, and dynamically adjusts the weights of each modal data through the self-attention mechanism;

[0158] S4: Multi-level fusion structure. The multi-level fusion structure of the deep neural network (DNN) is used to extract and fuse the relationships between different modal data. Combining the topological information of the transmission line, the graph neural network (GNN) is used to perform weighted processing on the fused data.

[0159] S5: Abnormal data identification and fault tolerance mechanism. The intelligent anomaly detection mechanism is used to identify and eliminate abnormal conditions in the sensor data, and the redundant fault tolerance mechanism of multi-modal data is used to ensure the reliability and robustness of data processing.

[0160] S6: Computational optimization mechanism. Model pruning and quantization are used to reduce the computational amount, and combined with distributed computing, some computational tasks are sunk to distributed nodes.

[0161] S7: Temporal modeling and uncertainty quantification mechanism. The long short-term memory network (LSTM) is combined to enhance the time series modeling of the system, and Bayesian deep learning technology is used to quantitatively analyze the uncertainty in fault prediction.

[0162] The multi-modal data fusion module receives data from the data preprocessing module. By fusing data from different sensors, it generates a comprehensive evaluation result of the transmission line status and fault prediction information.

[0163] Multi-modal data fusion algorithm:

[0164] The system combines data of multiple modalities such as temperature, humidity, traveling wave, and current, and adopts multi-modal weighted fusion technology. This technology dynamically adjusts the fusion method based on the weights of different data sources to ensure the effectiveness and accuracy of data fusion. For example, when the temperature or humidity is abnormal, the system will appropriately increase the weights of temperature and humidity data to enhance the fault prediction for this factor.

[0165] During the fusion process, the algorithm extracts the correlation features between different sensors. By analyzing the interaction between these features, a more accurate evaluation result of the transmission line status can be obtained.

[0166] Status evaluation and fault prediction: The results generated by fusing data include the current operating status evaluation of the transmission line and potential fault prediction information. For example, when the temperature and current data show abnormal changes, the system may predict that there is a risk of overload or short circuit caused by temperature in the line.

[0167] Furthermore, the AI fault diagnosis module receives the status evaluation result and fault prediction information generated by the multi-modal data fusion module, and conducts in-depth fault diagnosis on the operating status of the transmission line based on the improved dynamic multi-modal adaptive optimization model (DMAOM). Combining historical fault data and real-time analysis results, it generates a detailed fault diagnosis report.

[0168] Specifically, the AI fault diagnosis module dynamically adjusts the system response mode by learning the effects of different fault handling solutions in real time through the improved dynamic multi-modal adaptive optimization model DMAOM, combines the historical fault data and the fault prediction information generated by the multi-modal data fusion module to generate a detailed fault diagnosis report including the fault type, occurrence time, and expected impact range, and provides targeted repair suggestions.

[0169] Among them, the implementation steps of the improved dynamic multi-modal adaptive optimization model DMAOM are as follows:

[0170] Step 1: Integrate the historical fault data and real-time multi-modal data including temperature, current, voltage, and humidity sensor data, and use the dynamic data fusion algorithm based on the Bayesian model to perform weighted integration on the real-time data and historical data. The calculation formula is as follows:

[0171] X fuse =αX real +(1-α)X hist

[0172] Among them, X fuse is the real-time data; X rea l is the sensor data of the real-time multi-modal data; X hist is the historical data; α is the fusion weight coefficient;

[0173] Apply the self-attention mechanism to extract important global features in fault diagnosis:

[0174]

[0175] Among them, Q, K, and V are feature matrices; d k is the dimension of the key matrix; softmax is the normalization function, which calculates the similarity between the key and the query and converts it into a probability distribution;

[0176] Step 2: Through self-supervised learning technology, combined with unlabeled data and contrast learning methods, dynamically learn the fault patterns of the distribution network. At each time step, optimize the diagnosis model through the contrast learning loss function:

[0177]

[0178] Among them, L i is the contrast loss value of the i-th sample; x i is the encoded vector representation of the current sample; The positive sample related to x i , The negative sample related to x i ; sim(,) is the similarity between two samples; τ is the temperature parameter;

[0179] Step 3: The fault handling strategy is dynamically optimized through reinforcement learning. After each fault handling, the system optimizes the strategy based on the current state s t , the action a t performed:

[0180]

[0181] where π new is the new policy; s t is the state at time step t; Q(s t , a t ) is the Q-value of the current state and action, representing the expected return after executing action a t in state s t ; α is the entropy coefficient, controlling the balance between the reward and the policy entropy; logπ(s t |a t ) is the entropy of the policy, representing the randomness of choosing action a t in state s t ;

[0182] Step 4: The system receives real-time feedback on the processed effects through a feedback loop mechanism and continuously optimizes the fault diagnosis and handling strategy. The feedback information f t is introduced into the loss function:

[0183] L new = L(f t ) + βL previous

[0184] where L new is the new loss value for optimizing the model; L(f t ) is the current loss calculated based on the feedback information; L previous is the loss of the previous iteration; β is the feedback weight coefficient, controlling the balance between the new feedback and the historical loss;

[0185] Step 5: Apply the multi-modal fault handling optimization mechanism based on the diffusion model to gradually optimize the fault prediction results through the inverse diffusion process:

[0186] p θ (x t-1 |x t ) = N(x t-1 ; μ θ (x t , t), Σ θ (x t , t))

[0187] where p θ (x t-1 |x t ) is the noisy data x at time step tt The probability distribution of generating data at time step t-1 under certain conditions; N(x t-1 ; μ θ (x t ,t), Σ θ (x t ,t)) is a normal distribution, used to model the generated data distribution in the inverse diffusion process; μ θ (x t ,t) is the mean parameterized by the neural network, representing the mean estimate of the generation; Σ θ (x t ,t) is the variance parameterized by the neural network, representing the uncertainty in the generation process; x t is the noisy data at time step t.

[0188] The AI fault diagnosis module receives the status evaluation and fault prediction information provided by the multi-modal data fusion module, and generates a detailed fault diagnosis report through further AI analysis.

[0189] Dynamic multi-modal adaptive optimization model (DMAOM):

[0190] This module is based on the improved DMAOM model, combines historical fault data and real-time data, and further deeply diagnoses the faults of the transmission line. The DMAOM model can generate a detailed fault report by comparing historical data and analyzing the abnormal performance of the current line.

[0191] Self-learning ability: The AI fault diagnosis module has self-learning ability, and can continuously optimize the fault identification model according to the fault history records of the transmission line to improve the accuracy of diagnosis.

[0192] Fault diagnosis report: The report content includes information such as the type of fault, occurrence time, and expected influence range. For example, if the current of a certain section of the line shows abnormal fluctuations, the system can judge based on historical fault data that it may be due to insulator damage caused by overload, and provide recommended repair methods.

[0193] Furthermore, as Figure 3 shown, the control execution module, including circuit breakers, switches, and dispatching equipment, receives the fault diagnosis report generated by the AI fault diagnosis module, and performs control operations such as fault isolation, load transfer, or line protection according to the information in the report;

[0194] Specifically, the control execution module includes:

[0195] Current limiting operation unit: According to the preset current threshold and voltage threshold, when the current or voltage exceeds the preset range, perform current limiting operation, and the current limiting threshold is dynamically adjusted according to the load status and environmental parameters;

[0196] Load transfer and circuit breaker operation unit: When the current limiting operation fails to reduce the current or voltage within the safe range within the preset time and exceeds the specified number of times, perform load transfer or circuit breaker operation;

[0197] Self-check unit: In the case of multiple failures of current limiting or load transfer operations, the system automatically enters the self-check mode, and the internal fault detection module monitors the hardware and software status in the control execution system in real time;

[0198] Overload protection unit: When the system load exceeds the preset limit, automatically cut off some circuits to prevent overload damage;

[0199] Remote monitoring interface: Connect to the remote monitoring terminal through the Internet, allowing operators to view the system status in real time and perform remote operations.

[0200] Overview:

[0201] The control execution module receives the fault diagnosis report generated by the AI fault diagnosis module and performs corresponding control operations to prevent further expansion of the fault.

[0202] Control operations:

[0203] Fault isolation: When the system detects a fault in a certain circuit, the control execution module can quickly cut off the faulty circuit through a circuit breaker or switching equipment to prevent the accident from expanding.

[0204] Load transfer: If a line overload is detected, the system can automatically transfer the load to other lines to reduce the load on the faulty line.

[0205] Line protection: In some cases, the system will perform protection measures, such as adjusting the line current, to prevent equipment damage caused by current overload.

[0206] Application scenario: When a certain line detects an overload and the temperature rises, the system will immediately start the current limiting operation. If the current limiting is ineffective, the control execution module will quickly cut off the line and transfer the load to the standby line.

[0207] Furthermore, the communication module is responsible for transmitting data and control instructions between various modules of the system. It ensures that data acquisition, preprocessing, fusion, fault diagnosis, and control operations can interact in real time.

[0208] Communication method: Supports wireless communication and wired communication. For example, data can be transmitted in real time through a wireless local area network WLAN or a cellular network. The communication module ensures that information can be transmitted efficiently and with low latency between different parts of the system, ensuring real-time performance.

[0209] Real-time interaction: The real-time interaction ability between system modules ensures that when a fault occurs, each module can quickly respond and work together, reducing the accident handling time and improving the fault response efficiency.

[0210] Embodiment 2

[0211] As Figure 4 shown, this is another embodiment of the present invention. This embodiment provides an intelligent detection method for transmission lines of a distribution network based on an AI algorithm. The method includes the following steps:

[0212] Step 1: Real-time collect the operation status data and environmental parameter data of the transmission line through multiple sensors. The sensors include temperature and humidity sensors, traveling wave distance measuring sensors, current sensors, and fault detection sensors. The data is transmitted to the data preprocessing module through a wireless local area network (WLAN) or a cellular network.

[0213] Step 2: In the data preprocessing module, process the data. The preprocessing includes data denoising based on wavelet transform, abnormal data filtering, and preliminary fault feature extraction.

[0214] Step 3: Input the cleaned data into the multi-modal data fusion module, and combine convolutional neural network (CNN), long short-term memory network (LSTM), and graph neural network (GNN) algorithms to perform multi-modal data fusion analysis, and generate the state evaluation result and fault prediction information of the transmission line.

[0215] Step 4: Apply the improved dynamic multi-modal adaptive optimization model (DMAOM), combine historical fault data and real-time multi-modal data, use the Bayesian model to perform dynamic multi-modal data weighted fusion on the data, extract global features, and dynamically optimize the fault diagnosis model through self-supervised learning and reinforcement learning.

[0216] Step 5: AI fault diagnosis. Based on the fault prediction information and historical data generated by the DMAOM model, use deep learning and reinforcement learning algorithms to generate a fault diagnosis report. The report includes the fault type, occurrence time, expected impact range, and provides repair suggestions.

[0217] Step 6: Control execution. According to the fault diagnosis report, perform current limiting, load transfer, or circuit breaker operations. After multiple current limiting or load transfer operations fail, automatically enter the self-check mode to detect the hardware and software status through real-time monitoring.

[0218] Furthermore, the dynamic multi-modal adaptive optimization model (DMAOM) is implemented through the following steps:

[0219] Step S1: Input various sensor data of the transmission line, including temperature and humidity sensors, traveling wave ranging sensors, current sensors, and fault detection sensors, into the DMAOM model for multimodal data fusion. The significant improvements of the DMAOM model include:

[0220] Step S1.1: Use the Bayesian model to dynamically adjust the fusion weights of multimodal data according to the real-time monitored environmental parameters, including temperature, humidity, and load status. The initial weights are set through the historical data set and updated based on the new real-time data.

[0221] Step S1.2: Extract global features from multimodal data through self-supervised learning technology, identify the correlations between different data modalities in the transmission line, and extract important features related to fault prediction.

[0222] Step S1.3: Based on the data correlations in the multimodal data fusion process, detect abnormal data in real time and correct the anomalies through an adaptive correction mechanism.

[0223] Step S2: Adaptive optimization of the fault prediction model through reinforcement learning. The significant improvements in the model optimization include:

[0224] Step S2.1: After each fault prediction, the system adjusts the model parameters according to the feedback information, which includes the accuracy of fault occurrence, response time, and system control effect.

[0225] Step S2.2: Combine the long short-term memory network LSTM for multi-step time series modeling to further optimize the correlation analysis between historical data and real-time data.

[0226] Step S3: Based on the outputs of Step S1 and Step S2, generate fault prediction information and a detailed fault diagnosis report, which includes the fault type, occurrence time, affected range, and optimized repair suggestions.

[0227] Example 3

[0228] Application of Multimodal Data Fusion and AI Fault Diagnosis in Transmission Line Fault Prediction

[0229] In this embodiment, the system collects the operation status data and environmental parameter data of the line in real time through multiple sensors deployed on the transmission line, including temperature and humidity sensors, traveling wave ranging sensors, and current sensors, etc. These data are transmitted to the data preprocessing module of the system through the wireless local area network WLAN or cellular network.

[0230] The data preprocessing module first denoises the raw data collected by the sensor and filters out abnormal data. The wavelet transform algorithm is used to denoise the current and traveling wave data to eliminate the interference of high-frequency noise on data analysis. At the same time, the abnormal data detected by the system will be eliminated through a rule-based filtering algorithm to ensure that the data sent to the downstream module is valid data.

[0231] The preprocessed data is transmitted to the multimodal data fusion module. This module performs weighted fusion of multimodal data such as temperature and humidity, traveling waves, and current through convolutional neural networks (CNN), long short-term memory networks (LSTM), and graph neural networks (GNN). The weighting coefficient is dynamically adjusted by the Bayesian model, and is adaptively optimized based on current environmental parameters such as temperature, humidity, and load status.

[0232] During the data fusion process, the system correlates and analyzes the current changes and traveling wave data at different times with the temperature and humidity data to extract the real-time status characteristics of the transmission line. At the same time, the system combines LSTM to model the temporal dynamic relationship of the data to capture the fluctuation characteristics of the line in a short time, especially to predict the abnormal state of the line under sudden load increase or drastic environmental change.

[0233] The fused data enters the AI ​​fault diagnosis module, and the system uses the improved dynamic multimodal adaptive optimization model DMAOM to perform in-depth diagnosis of the operating status of the transmission line. The DMAOM model not only considers the current real-time data, but also combines historical fault data, extracts fault modes from it using a deep learning model, and predicts possible fault types and occurrence times.

[0234] The fault diagnosis module further optimizes the accuracy and response speed of the fault prediction model through self-supervised learning algorithms. While predicting faults, the system also generates a diagnostic report containing the fault type, impact range, fault probability, and possible fault recovery solutions. This report can help power companies adjust their dispatch strategies in a timely manner to avoid potential grid failures.

[0235] Based on the report generated by the fault diagnosis module, the control execution module will issue corresponding control instructions through the dispatch center. If the diagnosis report clearly indicates that a certain line may fail due to overload or abnormal temperature, the system will first perform current limiting operations. When the current limiting operation fails to alleviate the abnormal state, the system will automatically perform load transfer operations or circuit breaking operations to ensure stable operation of the power grid.

[0236] The system is equipped with a feedback mechanism to further optimize the fault handling strategy by continuously monitoring the effects of current limiting, circuit breaking or load transfer operations. If the current operation does not achieve the expected effect, the system will adjust the model parameters again based on the newly acquired data, continuously improving the accuracy and response speed of fault handling through the feedback loop.

[0237] The advantage of this embodiment is that it can quickly and accurately predict potential faults in transmission lines through multi-modal data fusion technology, combining real-time monitoring and historical fault data, and take timely fault handling measures. This AI-based fault diagnosis system not only greatly improves the fault prediction ability of transmission lines, but also can dynamically adjust control strategies through the optimized model to ensure the stability and security of the power grid.

[0238] The difference from the prior art is that the present invention utilizes a multi-modal data fusion algorithm and a dynamic adaptive optimization model DMAOM to effectively integrate various sensor data, and continuously improves the diagnosis and prediction ability of the system through self-supervised learning and reinforcement learning, significantly enhancing the early warning and handling ability of transmission line faults in complex environments.

[0239] In summary, the present invention significantly improves the monitoring, prediction, and control capabilities of distribution network transmission lines by integrating multi-modal data fusion technology and an AI-based intelligent fault diagnosis system. Through the multi-modal data weighted fusion module, the system can dynamically integrate data from different sensors, making full use of the information of various data modalities, thereby improving the accuracy of fault detection and the overall performance of the system. Based on the improved dynamic multi-modal adaptive optimization model DMAOM, the present invention can analyze historical and current data in real time and has the ability of self-learning and adaptive adjustment. This enables the system to identify potential faults, take preventive measures in a timely manner, and reduce the risk of fault occurrence. The system can automatically adjust control strategies according to real-time monitoring data to ensure that the power grid still operates efficiently and safely under load changes or external environmental fluctuations. In case of abnormalities, the present invention can quickly formulate and execute response strategies such as current limiting, load transfer, or circuit breaking, reducing the risk of large-scale power outages caused by faults and ensuring the stability and reliability of the power grid. By introducing a feedback mechanism and dynamic learning ability, the system can continuously optimize monitoring and response strategies after each fault handling, improving its intelligence level and adapting to the future changing power grid environment. The present invention not only provides strong technical support for the safe operation of the distribution network, but also has important practical application value in improving the efficiency and reliability of the power system, with broad market prospects and promotion potential.

[0240] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0241] Any process or method description depicted in the flowchart or described otherwise herein can be understood to represent a module, segment, or portion of code including one or more executable instructions for implementing a specific logical function or process. And the scope of the preferred embodiments of the present application includes additional implementations, where functions may be executed in a substantially simultaneous manner or in a reverse order according to the functions involved, rather than in the order shown or discussed.

[0242] As described above, the above are only specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of various changes or substitutions, and these should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An intelligent control and measurement system for power distribution network transmission lines based on AI algorithm, characterized in that: The system comprises: A data acquisition module is used to collect the operation status data and environmental parameter data of the power transmission line in real time through multiple sensors, the sensors include temperature and humidity sensors, traveling wave ranging sensors, current sensors and fault detection sensors, and the data acquisition module transmits data through a distributed network of a wireless local area network WLAN or a cellular network; A data preprocessing module, which preprocesses the temperature, humidity, traveling wave, current and fault detection data sent by the data acquisition module, wherein the preprocessing includes adopting a data denoising algorithm based on wavelet transform, abnormal data filtering and preliminary fault feature extraction; A multimodal data fusion module is used to receive the cleaned data and preliminary fault characteristics transmitted by the data preprocessing module, conduct a comprehensive analysis of the temperature, humidity, traveling wave and current data from different sensors through a multimodal data fusion algorithm, and extract the preliminary fault characteristics to generate a state assessment result and fault prediction information of the transmission line; The AI ​​fault diagnosis module receives the state evaluation results and fault prediction information generated by the multimodal data fusion module, performs in-depth fault diagnosis on the operation state of the transmission line based on the improved dynamic multimodal adaptive optimization model DMAOM, and generates a detailed fault diagnosis report in combination with historical fault data and real-time analysis results; A control execution module, including a circuit breaker, a switch, and a dispatching device, receives a fault diagnosis report generated by the AI ​​fault diagnosis module, and performs control operations of fault isolation, load transfer, or line protection according to information in the report; A communication module, used to transmit data and control instructions between the data acquisition module, the data preprocessing module, the multimodal data fusion module, the AI ​​fault diagnosis module and the control execution module, and realize real-time interaction among the modules of the system through wireless communication or wired communication; The dynamic multi-modal adaptive optimization model DMAOM is implemented by the following steps: Step S1: inputting various sensor data of the transmission line including temperature and humidity sensor, traveling wave ranging sensor, current sensor and fault detection sensor into the DMAOM model to perform multimodal data fusion; Step S2: Adaptively optimize the fault prediction model through reinforcement learning; Step S3: Based on the outputs of step S1 and step S2, fault prediction information and a detailed fault diagnosis report are generated, wherein the report includes the fault type, occurrence time, impact scope, and optimized repair suggestions.

2. According to claim 1, the intelligent control and measurement system for power distribution network transmission lines based on AI algorithm is characterized in that: The temperature and humidity sensor and the traveling wave ranging sensor in the data acquisition module can detect real-time data of external temperature and humidity changes or line load status and adaptively adjust the sampling frequency. The data acquisition module transmits data with the data preprocessing module through a low-power Bluetooth BLE module.

3. According to claim 1, the intelligent control and measurement system for power distribution network transmission lines based on AI algorithm is characterized in that: The data preprocessing module uses a convolutional neural network (CNN) model to extract fault features from the preprocessed temperature, humidity, traveling wave and current data. The CNN model is pre-trained with historical fault data and adaptively updated according to new fault data through dynamic adjustment based on online learning when the system is running.

4. The intelligent control and measurement system for power distribution network transmission lines based on AI algorithm according to claim 1 is characterized in that: The multimodal data fusion module uses convolutional neural network CNN, long short-term memory network LSTM, self-attention mechanism and graph neural network GNN combined with reinforcement learning algorithm to perform multimodal data weighted fusion analysis on temperature, humidity, traveling wave and current data from different sensors. The weighting coefficient is based on the Bayesian model in multimodal data fusion and is dynamically adjusted through an online learning model according to the ambient temperature, humidity and load status of the transmission line.

5. The intelligent control and measurement system for power distribution network transmission lines based on AI algorithm according to claim 4 is characterized in that: Implementation mechanism of the multimodal data weighted fusion module include: S1: Anomaly detection and noise removal mechanism, which removes noise data from sensors through an intelligent anomaly detection mechanism based on support vector machine (SVM), uses a dynamic interpolation algorithm based on time series to achieve automatic alignment of multimodal data, and ensures data consistency through adaptive normalization processing; S2: Local feature and temporal dynamic relationship extraction mechanism, combining convolutional neural network (CNN) and long short-term memory (LSTM) network to extract local features and temporal dynamic relationship of each modal data; S3: Weight setting and adaptive adjustment mechanism, through the unsupervised learning algorithm based on K-means clustering, the initial weight of each modal data is set based on historical data, and the weight of each modal data is dynamically adjusted through the self-attention mechanism; S4: Multi-level fusion structure, using the multi-level fusion structure of deep neural network DNN to extract and fuse the relationship between different modal data, combined with the topological information of the transmission line, and weighted processing of the fused data through graph neural network GNN; S5: Abnormal data identification and fault tolerance mechanism, which identifies and eliminates abnormal situations in sensor data through intelligent anomaly detection mechanism, and ensures the reliability and robustness of data processing through redundant fault tolerance mechanism of multimodal data; S6: Computation optimization mechanism, which reduces the amount of computation through model pruning and quantization and combines distributed computing to shift some computing tasks to distributed nodes; S7: Time series modeling and uncertainty quantification mechanism, combining long short-term memory network LSTM to enhance the time series modeling of the system, and quantifying the uncertainty in fault prediction through Bayesian deep learning technology.

6. The intelligent control and measurement system for power distribution network transmission lines based on AI algorithm according to claim 1 is characterized in that: The AI ​​fault diagnosis module learns the effects of different fault handling solutions in real time through the improved dynamic multimodal adaptive optimization model DMAOM, dynamically adjusts the system response mode, combines historical fault data and the fault prediction information generated by the multimodal data fusion module to generate a detailed fault diagnosis report including fault type, occurrence time and expected impact range, and provides targeted repair suggestions.

7. The intelligent control and measurement system for power distribution network transmission lines based on AI algorithm according to claim 6 is characterized in that: The improved dynamic multi-modal adaptive optimization model DMAOM is implemented in the following steps: Step 1: Integrate historical fault data and real-time multimodal data including temperature, current, voltage, and humidity sensor data, and use the dynamic data fusion algorithm based on the Bayesian model to perform weighted integration of real-time data and historical data. The calculation formula is as follows: ； in, For real-time data; Sensor data for real-time multimodal data; For historical data; is the fusion weight coefficient; Apply self-attention mechanism to extract important global features in fault diagnosis: ； Among them, Q, K, and V are feature matrices; is the dimension of the key matrix; softmax is a normalization function that calculates the similarity between the key and the query and converts it into a probability distribution; Step 2: Through self-supervised learning technology, combined with unlabeled data and contrastive learning methods, the fault mode of the distribution network is dynamically learned. At each time step, the diagnostic model is optimized through the contrastive learning loss function: ； in, is the contrast loss value of the i-th sample; is the encoding vector representation of the current sample; and The relevant positive samples, and Related negative samples; sim(,) is the similarity between two samples; is the temperature parameter; Step 3: The fault handling strategy is dynamically optimized through reinforcement learning. After each fault handling, the system , Actions performed To optimize the strategy: ； in, for new strategies; is the state at time step t; is the Q value of the current state and action, indicating that Next action Expected return after is the entropy coefficient, which controls the balance between reward and policy entropy; is the entropy of the strategy, indicating that in the state Next select action The randomness of Step 4: The system receives real-time feedback on the effects of processing through a feedback loop mechanism, continuously optimizes fault diagnosis and processing strategies, and provides feedback information. is introduced into the loss function: ； in, is the new loss value used to optimize the model; is the current loss calculated based on the feedback information; is the loss of the previous iteration; is the feedback weight coefficient, which controls the balance between new feedback and historical loss; Step 5: Apply the multi-modal fault handling optimization mechanism based on the diffusion model to gradually optimize the fault prediction results through the reverse diffusion process: ； in, is the noisy data at time step t The probability distribution of the data at time step t−1 generated under the condition; is a normal distribution, which is used to model the distribution of generated data in the inverse diffusion process; is the mean of the neural network parameterization, which represents the generated mean estimate; is the variance of the neural network parameterization, which represents the uncertainty in the generation process; is the noisy data at time step t.

8. The intelligent control and measurement system for power distribution network transmission lines based on AI algorithm according to claim 1 is characterized in that: The control execution module includes: Current limiting operation unit: according to the preset current threshold and voltage threshold, when the current or voltage exceeds the preset range, the current limiting operation is performed, and the current limiting threshold is dynamically adjusted according to the load state and environmental parameters; Load transfer and circuit breaking operation unit: when the current limiting operation fails to reduce the current or voltage to a safe range within the preset time and exceeds the specified number of times, the load transfer or circuit breaking operation is performed; Self-check unit: In case of multiple current limiting or load transfer operation failures, the system automatically enters self-check mode, and monitors the hardware and software status in the control execution system in real time through the internal fault detection module; Overload protection unit: When the system load exceeds the preset limit, it automatically cuts off part of the line to prevent overload damage; Remote monitoring interface: connected to the remote monitoring terminal via the Internet, allowing operators to view system status in real time and perform remote operations.

9. A control and measurement method for a power distribution network transmission line intelligent control and measurement system based on an AI algorithm according to any one of claims 1 to 8, characterized in that: The method comprises the following steps: Step 1: collect the operation status data and environmental parameter data of the power transmission line in real time through multiple sensors, the sensors include temperature and humidity sensors, traveling wave ranging sensors, current sensors and fault detection sensors, and transmit the data to the data preprocessing module through the wireless local area network WLAN or cellular network; Step 2: In a data preprocessing module, the data is processed, and the preprocessing includes data denoising based on wavelet transform, abnormal data filtering, and preliminary fault feature extraction; Step 3: Input the cleaned data into the multimodal data fusion module, combine the convolutional neural network CNN, long short-term memory network LSTM and graph neural network GNN algorithms to perform multimodal data fusion analysis and generate the state assessment results and fault prediction information of the transmission line; Step 4: Apply the improved dynamic multimodal adaptive optimization model DMAOM, combine historical fault data and real-time multimodal data, use the Bayesian model to perform dynamic multimodal data weighted fusion on the data, extract global features, and dynamically optimize the fault diagnosis model through self-supervised learning and reinforcement learning; Step 5: AI fault diagnosis, based on the fault prediction information and historical data generated by the DMAOM model, uses deep learning and reinforcement learning algorithms to generate a fault diagnosis report, which includes the fault type, occurrence time, expected impact range, and provides repair suggestions; Step 6: Control execution. According to the fault diagnosis report, perform current limiting, load transfer or circuit breaking operations. After multiple current limiting or load transfer operations fail, it automatically enters the self-check mode to detect the hardware and software status through real-time monitoring.

10. The control and measurement method according to claim 9, characterized in that: The dynamic multi-modal adaptive optimization model DMAOM is implemented by the following steps: Step S1: Inputting various sensor data of the transmission line including temperature and humidity sensor, traveling wave ranging sensor, current sensor and fault detection sensor into the DMAOM model to perform multimodal data fusion. The significant improvements of the DMAOM model include: Step S1.1: Use the Bayesian model to dynamically adjust the fusion weights of multimodal data based on real-time monitored environmental parameters including temperature, humidity and load status. The initial weights are set through historical data sets and updated based on new real-time data. Step S1.2: Perform global feature extraction on multimodal data through self-supervised learning technology, identify the correlation between different data modes in the transmission line, and extract important features related to fault prediction; Step S1.3: Based on the data correlation in the multimodal data fusion process, abnormal data is detected in real time, and the abnormality is corrected through an adaptive correction mechanism; Step S2: Adaptively optimize the fault prediction model through reinforcement learning. The significant improvements of the model optimization include: Step S2.1: After each fault prediction, the system adjusts the model parameters according to the feedback information, which includes the accuracy of the fault occurrence, the response time and the system control effect; Step S2.2: Combine the long short-term memory network LSTM to perform multi-step time series modeling to further optimize the correlation analysis between historical data and real-time data; Step S3: Based on the outputs of step S1 and step S2, fault prediction information and a detailed fault diagnosis report are generated, wherein the report includes the fault type, occurrence time, impact scope and optimized repair suggestions.

Images