| Work Detail |
Scientists have developed a machine learning model - which uses K-Means techniques and long-term memory - which aims to overcome the “fault detection and classification” in the operation and maintenance of large-scale photovoltaic solar parks. Scientists from Malaysia and Thailand have developed a novel machine learning model to predict the maintenance needs of large-scale solar photovoltaic plants. According to a recently published scientific paper , the model uses two machine learning techniques – K-Means and long short-term memory (LSTM) – and has a root mean square error (RMSE) of 0.7766. The objective of the tool is to overcome “the detection and classification of faults” that traditional operating systems usually have, according to the document. “Conventional operation and maintenance (O&M) systems for solar photovoltaic systems do not use machine learning for fault detection and classification,” the document states. “This poses problems for plant operators, especially those managing large-scale PV plants, who often rely on manual methods to analyze large amounts of electrical data and inspect numerous string panels. Consequently, the O&M cost is high.” K-Means is a data segmentation algorithm that groups similar data points into clusters. The researchers use the algorithm to match the electrical current of string modules with environmental factors such as global irradiance and module temperature. Next, a center or middle point is created for the cluster to represent typical behavior. Next, the LSTM technique comes into action, trained using historical data. The goal of the technique is to detect anomalies in the expected electrical current of the string modules, which would alert operators of maintenance needs. “LSTMs can handle variable-length sequential data using a gating mechanism to decide which information is important to keep and which to discard at each time step, and thus make predictions based on past trends and patterns of the input sequence,” explains the article. “LSTMs can do this using a special type of memory cell that can store information for long periods of time, as well as gates that control the flow of information in and out of the cell.” The data for training, as well as the accuracy analysis of the method, is based on information from a large-scale solar photovoltaic plant located in central Malaysia. A turnkey and a subinverter were used as test cases, which monitored 420 string modules and a total of 8,400 photovoltaic modules. Compared to the collected data, the model has a root mean square error (RMSE) of 0.7766. The relative error is then compared to the figure established by the reference model based on artificial neural networks (ANN). “LSTMs and ANNs are often compared because they both belong to neural networks and are commonly used in various tasks in natural language processing, computer vision, and speech recognition,” the article states. They found that LSTM is more accurate, with a mean LSTM relative error of 4.316% and ANN relative error of 4.363%. The algorithm is described in the study “ Anomaly detection using K-Means and long-short term memory for predictive maintenance of large-scale solar (LSS) photovoltaic plant ” predictive maintenance of large-scale photovoltaic plants [LSS]), recently published in Energy Reports . The research group consisted of scientists from the University of Technology Malaysia and Chiang Mai Rajabhat University. |
