Machine Learning Implementation at Mölndal Energi AB Evaluating the Possibilities of Implementing Machine Learning within the Production System at Mölndal Energi AB

Abstract

This thesis aimed to evaluate the possibilities of implementing machine learning within the production system at Mölndal Energi. The outcome of the thesis included an overview of potential applications and recommendations for implementation, along with two proof of concepts demonstrating simplified versions of potential applications. The aim was to support Mölndal Energi in improving certain areas of their production system and give the company a foundation for further development and increased competitiveness. The literature review revealed predictive maintenance, demand forecasting, scheduling and energy storage as relevant applications. It was concluded that predictive maintenance would be the most relevant to implement due to the high potential of improvement. Mölndal Energi already use, both directly and indirectly through Göteborg Energi, external machine learning applications for both demand forecasting and scheduling. There are several other areas of interest that could be investigated further in future work, such as the usage of digital twins to simulate and debug systems, as well as inspection and fault detection using drones and augmented reality. Two proof of concepts were developed using the programming language Python in a Jupyter Notebook environment. In the first proof of concept, a machine learning model was developed to predict the maximum load of a boiler based on operational data. It used an LGBM algorithm for quantile regression and the results were evaluated by comparing the predicted output with actual output for a number of random samples, which showed reasonable results. Yet, further testing would be required before deployment. A second proof of concept was developed to identify deviations within fuel data. Three regressor algorithms were compared (LGBM, XGB and RF) and results were evaluated using tables and plots of the deviations. Once again results seemed reasonable, however further testing would be required before operational usage. In future work, both proof of concepts could be improved by comparing more algorithms and fine-tuning the parameters further. The anomaly detection model could also be applied to similar areas by changing the dataset.