Manan Khattar
Spring 2020 Research Findings
Creating a Working Python Interface for the Dynamic System
- Simulation, Workstation, Person classes
- Encoding the linear relationships between states and exogenous inputs
- Generating sample data for exogenous inputs and creating an API to interface with them
- Assisting Lucas in creating a framework for the parameter fit update
right
Creating a Mock Response for Faster RL Testing
- Ideated and mocked several possible responses to points given in terms of energy usage.
- Incorporated behavioral components - gradient in energy usage, baseline energy vs intended energy (weighted average)
right
Working with RNNs for Planning Model Design
- Testing RNN (LSTM) architectures for time series prediction of energy, given timestamp and points.
- Current RMSE: ~40
- Hyperparameters to think about:
- Number of timesteps to be used in training
- Whether or not to include past energy values in training set
- Number of layers (bilayered LSTM?)
fit, right
uuid: 202005040920 date: May 4, 2020 tags: #raise