Fault Detection in Solar Power System with Internet of Things using Multi resolution Sinusoidal Neural Network - Snow Geese Optimization Approach
Pramod Kumar, K.E. Purushothaman, R V Arunmozhi, Jayachitra S, D. Leela Rani, S. Kaliappan · 2024
Fault detection in power systems, including Photovoltaic (PV) systems, using Internet of Things (IoT) involves deploying sensors to monitor key parameters and analyzing the data identify anomalies. The accuracy of these systems can be compromised by issues such as noisy or low-quality sensor data, which may lead to false positives or missed faults. Additionally, the complexity of managing large volumes of data from numerous IoT devices can strain the system, further affecting detection accuracy. To overcome these drawbacks, this paper proposes an efficient approach for fault uncovering in solar power system grounded on IoT. The data are collected from solar power generation dataset. Subsequently, the data are nourished to preprocessing. In preprocessing segment removes the absent values in the data utilizing Reverse Lognormal Kalman Filter (RLKF). The consequence after the preprocessing data is transported to the Multiresolution Sinusoidal Neural Network (MSNN). The line-line faults, ground faults, line-ground faults and partial shading are successfully predicted and classified by using MSNN. The Snow Geese Optimization (SGO) is rummage-sale to optimize the weight structure of MSNN. The projected MSNN-SGO is utilized within the MATLAB platform. Performance metrics counting precision, accuracy, Root Mean Squared Error (RMSE), F1-score, and Mean Squared Error (MSE) were examined in order to determine the proposed method. The proposed DAGCN-COA technique yields 14.89%, 16.89%, and 18.23% higher precision, 16.65%, 18.85%, and 17.89% higher accuracy, 22.36%, 15.42% and 28.27% lower RMSE when analysed the existing methods. The proposed MSNN-SGO technique is associated with the existing methods for instance spatial-temporal recurrent graph neural network (STRGNN), convolutional neural network (CNN), and long short-term memory network (LSTMN), respectively.