The use of artificial neural network for low latency of fault detection and localisation in transmission line
One of the most critical concerns in power system reliability is the timely and accurate detection of transmission line faults. Therefore, accurate detection and localisation of these faults are necessary to avert system collapse. This paper focuses on using Artificial Neural Networks in fault detection and localisation to attain accuracy, precision and speed of execution. A 330 kV, 500 km three-phase transmission line was modeled to extract faulty current and voltage data from the line. The Artificial Neural Network technique was used to train this data, and an accuracy of 100% was attained for fault detection and about 99.5% for fault localisation at different distances with 0.0017 μs of detection and an average error of 0% -- 0.5%. This model performs better than Support Vector Machine and Principal Component Analysis with a higher fault detection time. This proposed model serves as the basis for transmission line fault protection and management systems.
1. Introduction
The electrical power system is composed of interconnected parts such as power generation, transmission, and distribution. The transmission line is an essential component of the power system since it transports energy from the producing plant to the end customers. These components are linked via transmission lines, which are prone to malfunction and can only be managed remotely through complex processes . The transmission line's difficulties are exacerbated by aging gear, lightning, human contact, and extreme weather conditions. However, power quality is the most critical component in an electrical network. When a transmission line fails, the electricity quality drops, which directly impacts power output .
Fault detection and localisation are critical for safeguarding a transmission line's network. As a result, proper precautions must be taken to provide optimum protection and avoid system failure. The fault must be identified to prevent the transmission line from damage, and the fault location must be precise for speedy line isolation. However, fault detection input may considerably aid problem localisation for faster fault clearing and power restoration . Identifying the location of a transmission line failure in a power system is crucial for rapid response and power supply dependability.
Transmission line identification and localisation have mostly been performed using classic machine learning and deep learning approaches. As a switching mechanism for fault protection, the typical way employs a distance protection relay over current and voltage relays. Mobile robots are used to identify line problems and monitor transmission lines. Furthermore, the fuzzy logic technique Neuro-fuzzy method wavelet and fuzzy approach wavelet and fuzzy approach While the Artificial Neural Network (ANN) is a component of the machine learning technique , the Support Vector Machine (SVM) , and decision tree (DT) are also used. When time and frequency data are needed, the WT is helpful but vulnerable to noise and harmonics. Therefore, it has certain limits. The method of producing a reference wavelet is time-consuming and involves a high sampling rate. Furthermore, the number of decompositions is generated via experiments and is mainly used for defect detection . Back-propagation neural networks were employed as an alternate defect detection and localisation approach in Ref. . This may be used to create a long transmission line distance relay protection mechanism. However, the model's fault classification accuracy is weak.
Many hybrid approaches, such as the S-transform and ANN, have been used to improve performance in fault localisation and detection . This approach was used to discover transmission line faults despite the fact that it did not take into consideration a multi-class data set of fault data . Furthermore, the ANN and SVM were used in the defect detection process, and since they demand a large quantity of data for training, they are difficult to maintain and take a long time.
WT is also used in fault detection. However, distinguishing between various fault states is difficult. Even though the bulk of these strategies is very new, there are a number of challenges. The high processing cost and the unsuitability of the Hilbert-Huang Transform (HHT) for high-frequency signals, as seen in Ref. .
Principal Component Analysis is a rapid and easy machine learning approach that reduces re-projection error and is noise-resistant (PCA). However, if the number of dimensions exceeds the number of data points, the convergence matrix will always be enormous, making it impossible to get . The PCA is also used to map data from high dimensional space to low dimensional subspace to decrease the data's dimensionality and better understand its variance. According to studies, by training their algorithms on smaller datasets, the bulk of these methodologies provide fairly accurate results. Furthermore, for training, they use data from either a single phase to ground fault or a double phase to line fault . Single Phase, Double Phase, and Three Phase to Ground Fault datasets will be used concurrently to identify and localize the transmission line fault, with a focus on machine learning and deep learning approaches.
DWT and DT have poor performance for high-performance faults and limited temporal resolution capabilities when considering fault location. Using data mining and wavelets , Decision Tree (DT) and K-Nearest Neighbors (KNN) are used, although they do not quantify fault location. The S-transform approach was the only one investigated in Ref. . For fault identification and detection, morphology in mathematics and the Recursive Least-square (RLS) method are employed to identify fault characteristics based on mathematical morphology. This approach is difficult to maintain since it is employed in incredibly intricate situations that are not home in nature. Another disadvantage of the previously discussed approaches was their inability to focus more on fault localisation. Localisation of faults facilitates rapid diagnosis and power restoration during an outage by pinpointing the exact location of the problem.
Transmission line fault analysis normally requires three main activities for a successful fault management system: fault sensing or detection, categorizing the problem into various categories and identifying the spot to disclose the zone where the fault occurred . The extraction of fault features is being considered. This may be accomplished by first modeling the network in MATLAB/SIMULINK to extract fault instances from the transmission line. The next step is to identify and localize the flaws using data provided by the simulated model and an ANN-trained classifier .
Because of the delay in fault detection and the increased role of communication and computers in transmission systems , this study aims to offer a novel ANN-based approach for rapid, reliable, and accurate fault identification and localisation in transmission lines. Also, to detect many fault circumstances, such as defective voltage and current, simultaneously minimize fault detection time delay. The suggested algorithm's performance was assessed by simulating several errors and training them with the ANN model, and the results were encouraging. In addition, the suggested model will be used to develop transmission line fault management and protection in power systems.
One of the technique's significant disadvantages is the model's inability to train on non-numerical data. Therefore, interpreting the findings is always challenging as matching results with real-life circumstances and issue statements.
2. The Artificial Neural Network technique
Artificial Neural Networks have traditionally been used with great success in various sectors of fault analysis. This is one of the most extensively utilized artificial intelligence technologies, which is critical in constructing a strong power system failure management model. An ANN model typically has three major layers: input, hidden, and output. The input layer receives data or signals from the model, which can be fault current or voltage sent into the model. The hidden layer extracts patterns associated with the analyzed process or system. The output is in charge of creating and displaying the final network output. These layers handle most internal network processing, which involves the results processed from other layers. ANN has several advantages that allow it to be widely used in developing fault management models that are exceptionally effective .
Conclusion
This paper has investigated the use of artificial neural networks as a technique for fault detection and localisation in the transmission line. A 330 kV, 500 km, 50 Hz three-phase transmission line was modeled using Matlab/Simulink to generate the RMS value of faulty voltage and current signals from the defective line. About 12 different fault scenarios were considered, and 33,336 data samples of faulty current and voltage were taken from various locations in the transmission line to detect faults using ANN and to use the module for fault localisation. The different faults were discussed, especially single, double, and three-phase. The data were trained for accuracy, speed and precision, with an accuracy of 100\% for fault detection and 99\% fault localisation at separate locations. The time for fault detection is important in fault protection and this paper has focused on the speed of execution for prompt detection of fault. This technique produces excellent results compared to other conventional methods like the SVM and DWT.
However, the ANN technique has some drawbacks like large volumes of data are required for optimal performance and difficulty in determining the proper network structure for best performance of the model. This paper has some limitations which include also, Noise was not taken into account during simulation though it was mentioned.
In practice, The results generated from this can also be recommended as a basis for the design of effective fault management and protection of power systems.
0 Comments