It is of great significance to locate faults in distribution systems quickly. Traditional methods for locating fault sections in distribution networks are mostly based on rules or physical models. However, with the increasing complexity of the topological structure of distribution networks and the increasing number of distributed power sources, traditional methods are difficult to meet the current needs of locating fault sections in distribution networks. Based on this, this paper proposes a fault section location scheme for distribution networks based on wavelet energy entropy and artificial neural networks. First, the effective values of voltage and current data are extracted to train the model, and correlation analysis is used to optimize the features. An improved IEEE33 node data set containing distributed power sources was built, and different types of fault data at different section locations were simulated. The proposed method was used for training and verification, which demonstrated the rationality and applicability of the proposed method. It could reach 97.38%.

High-voltage overhead transmission lines are susceptible to abnormal conditions such as icing, wind deviation, and galloping due to extreme weather conditions, which has a serious impact on the safe and stable operation of the power system. Therefore, it is necessary to establish an online monitoring system to identify and warn the abnormal conditions that may occur on overhead transmission lines in real time. In this paper, a scheme of transmission line state identification based on electromagnetic sensing data and machine learning method is proposed. Firstly, the catenary model of transmission line is modeled and analyzed, and the mathematical expression of the model in abnormal state is defined. Then, based on the law of spatial magnetic field and electric field distribution of overhead transmission lines, the installation strategy of sensors is discussed, and the feature quantities that can be used for machine learning are constructed. Finally, the electromagnetic field data is calculated and generated based on the specific sensor deployment scheme, and the task instance of transmission line state identification is completed based on the idea of machine learning.

Current measurement is a core link in power systems, industrial automation, and electronic equipment monitoring. High-precision measurement is essential for system safety, energy efficiency optimization, and equipment protection. The magnetic field distribution of finite-length conductors deviates from the circularly symmetric model of ideal infinite-length conductors due to their geometric shape and end effects, resulting in a distortion in the proportional relationship between magnetic field intensity and current. Traditional methods face significant challenges in scenarios with finite-length conductors. This paper addresses the array current measurement error problem caused by finite-length conductors. By establishing an accurate magnetic field calculation model, this paper systematically studies the influence of wire length, array position, and current loop structure on measurement accuracy. The numerical discretization method is used to handle the magnetic field calculation problem of rectangular cross-section busbars. The current measurement error under different working conditions is analyzed by parametric simulation, focusing on the core role of the length-to-radius ratio $(L/R)$ • The rectangular loop can significantly suppress the error. When the parallel wire spacing $\text{Lm}/\mathrm{R} > 10$ and the number of sensors $\mathrm{N}\geq 7$, the error of the complete loop is reduced by more than 85% compared with that of a single wire.

It is of great significance to accurately identify the fault type and quickly locate the fault in the distribution system. The access of distributed power sources leads to a decrease in the accuracy of the traditional fault identification method based on single-end or double-end measurement equipment. With the development of sensor technology, the cost of sensors has gradually decreased, the measurement accuracy has gradually increased, and wide-area distributed sensing has become possible. Distributed sensing can collect more fault information from the system and improve the accuracy of fault identification. However, the increase of distributed sensing will also bring disadvantages, that is, the amount of data has greatly increased, and how to effectively use the increased data needs to be studied. The development of artificial intelligence technology has also made the performance of machine learning algorithms in complex data feature extraction and analysis increasingly enhanced. Based on this, this paper proposes a fault type identification and section location scheme based on distributed sensing and KNN algorithm. First, the effective value of voltage and current data is extracted for type identification, and the sequence component is added to optimize the features. On the basis of fault type identification, considering the advantages of distributed sensing, the fault features are artificially constructed to achieve fault section location. The overall accuracy is as high as 99.5%. An improved IEEE33 node data set containing distributed power sources was built, and different types of fault data in different section locations were simulated. The proposed method was used for training and verification, which demonstrated the rationality and applicability of the proposed method.

This paper addresses the challenge of fault type identification in distribution networks, where small-current grounding systems often lead to weak fault current characteristics and limit the performance of traditional methods. To overcome issues of low feature discriminability, poor noise immunity, and weak generalization, a fault identification scheme based on root mean square (RMS) values, symmetrical component analysis, and the Adaboost ensemble learning algorithm is proposed. An improved IEEE 33-bus distribution network model with distributed generation is developed in PSCAD/EMTDC to generate diverse fault data and construct a fault fingerprint database. Feature matrices derived from multiple sensor nodes are trained and optimized using Adaboost combined with KNN, achieving a classification accuracy of 99.12%. The results verify the robustness, effectiveness, and practical applicability of the proposed method under complex operating conditions.

The integration of numerous distributed energy resources and power electronic devices introduces a wide spectrum of frequency disturbances, which significantly challenge the stability of modern power systems. Therefore, there is an urgent need to enhance the current monitoring level of modern power systems. This paper proposes a novel busbar current inversion scheme based on an elliptical magnetic sensor array. By establishing a simulation model, the effect of structural parameters of the elliptical array on its current measurement accuracy was analysed. The anti-interference capability of the elliptical array in complex environments such as busbar displacement and crosstalk was studied, and principles for designing array parameters under different current sensor standards were established. Experiments conducted on the proposed current sensing scheme demonstrated that the designed current array has a range of 0-150 A, with a current measurement error below 0.1% without external interference and not exceeding 1% during busbar displacement. Under conditions of crosstalk, the measurement accuracy achieved was class 0.5. The sensor array possesses high measurement accuracy, robust anti-interference capability, low power consumption, compact size and a noncontact nature. It exhibits significant potential for extensive application in novel business scenarios within the power system.

To address the challenges of weak electrical characteristics, complex fault types, and transient diversity caused by grid-connected distributed power sources in low-current grounded distribution networks, a method for locating distribution network fault sections based on distributed sensing and random forests was proposed. An improved IEEE33 node model was constructed on PECAD/EMTDC. Distributed sensing nodes and multi-type and multi-location faults were configured. Effective value and sequence component features were extracted to construct a section-level fault fingerprint library. A random forest multi-classifier was used to output the probabilities of each section and determine the section where the fault is located. Noise was added to verify the robustness of the method. Simulation results demonstrate that the method exhibits good robustness and interpretability, with a fault section location accuracy of up to 99.4%, enabling highly reliable section location under complex operating conditions.

Distribution network is generally based on the small current grounding system. When a short-circuit grounding fault occurs, the fault characteristics of the low-current grounding system are not prominent. Considering the access of distributed power supply, the characteristics of current flow in the fault system are quite different from before. The traditional fault section location algorithm based on zero sequence current direction is no longer applicable. Wavelet energy entropy is a signal analysis method based on wavelet transform, which measures the signal entropy by calculating the energy distribution of the signal on different frequency scales, and is suitable for analyzing the fault signal with complex time-frequency characteristics. With the development of artificial intelligence technology, machine learning and larger scale deep learning have become increasingly popular, and the ability to extract data features in massive and complex data sets has been increasing. Therefore, a fault location algorithm based on wavelet energy entropy and Alexnet convolutional neural network is proposed in this paper. In this paper, an improved IEEE33 node fault model with distributed power supply is built with PSCAD/EMTDC, and the fault data of different sections are simulated. The wavelet energy entropy of the fault signal is calculated, and the feature matrix is constructed by combining the wavelet energy entropy of different sensor nodes. Alexnet convolutional neural network is used to train the eigenmatrix, and the parameters of the model are trained, optimized and verified, and the rationality and applicability of the proposed method are demonstrated.