Machine learning based hosting capacity determination methodology for low voltage distribution networks

For the past few years, the addition of renewable energy sources has been on the rise, but the unregulated addition of these sources can cause severe harm to the grid. Therefore, it is necessary to have a predefined limit for a grid, beyond which no further addition of renewables should be allowed without reinforcement. That limit is called the hosting capacity (HC), which is addressed in the literature by search-based methods with heavy computational burdens. This manuscript first presents a framework to find the HC for a grid. The same framework is then applied to 503 different simulated but realistic LV distribution feeders in Finland to find their HCs and the most effective network parameters in defining a network specific HC. Next, different machine learning models, that is, Decision Tree (DT), Random Forest (RF), Linear Regression (LR), K nearest neighbours (KNN), Logistic Regression, and Support Vector Machine (SVM) are implemented on the generated data. For the classification case, the accuracy values for logistic regression, KNN, and SVM were 0.89, 0.84, and 0.81, respectively. The findings demonstrate that the developed machine learning based technique will enable distribution network operators in finding the HC without applying any deterministic or probabilistic approaches.