A Machine Learning Approach for Real-time Battery Optimal Operation Mode Prediction and Control

This paper introduces a machine learning approach for real-time battery optimal operation mode prediction in residential PV applications. First, from the historical data, the optimal battery operation mode for each operation interval is derived. Then, a best performing algorithm for the prediction of the optimal modes is obtained. Performances are tested with different number of features in the training test and different training lengths. Then, the features will be used to predict future operation mode in real-time operations. A comparison on bill savings is made with the model-predictive control approach using the residential load and PV data from the Pecan Street project website under the Hawaiian electricity rate. Simulation results show a 9 points improvement in performance.