DSGC AI

Decentral Smart Grid Control (DSGC) is a strategy used with electrical grids to classify whether an electrical grid is stable or unstable. AI is often used as a tool to reduce costs in this strategy. In this project, I optimized the hyperparameters for logistic regression, random forests, xgboost, and neural network models with this goal in mind.

Sklearn was used to create the logistic regression and random forest models, and the GridSearchCV module was used to optimize hyperparameters for each model except neural network. Xgboost was another model used, and to optimize neural networks hand tuning with a manual grid search was used with Keras to build the neural network. Matplotlib was also used for the data analysis earlier in the notebook, and Seaborn to visualize the confusion matrices.

I experienced how to approach normal machine learning problems, given a dataset and the task to predict a certain feature. I learned how to use data visualization to see trends between data, as well as plotting the distributions of any certain feature. I also experimented with automating the hyperparameter optimization for keras neural networks; however, the runtime was too long so I ended up hand tuning it. More models could be tested with this dataset, I only tested logistic regression as a baseline, as well as random forest, xgboost, and neural network since they were mentioned in the paper associated with this dataset.