Search

To realize the distributed generation and to make the partnership between the dispatchable units and variable renewable resources work efficiently, accurate and flexible monitoring needs to be implemented. Due to digital transformation in the energy industry, a large amount of data is and will be captured every day, but the inability to process them in real time challenges the conventional monitoring and maintenance practices. Access to automated and reliable data-filtering tools seems to be crucial for the monitoring of many distributed generation units, avoiding false warnings and improving the reliability. This study aims to evaluate a machine-learning-based methodology for autodetecting outliers from real data, exploring an interdisciplinary solution to replace the conventional manual approach that was very time-consuming and error-prone. The raw data used in this study was collected from experiments on a 100-kW micro gas turbine test rig in Norway. The proposed method uses Density-Based Spatial Clustering of Applications with Noise (DBSCAN) to detect and filter out the outliers. The filtered datasets are used to develop artificial neural networks (ANNs) as a baseline to predict the normal performance of the system for monitoring applications. Results show that the filtering method presented is reliable and fast, minimizing time and resources for data processing. It was also shown that the proposed method has the potential to enhance the performance of the predictive models and ANN-based monitoring.