slzhang-git / DRL

A Deep Reinforcement Learning Method for Electricity Optimization in Smart Homes

Home-Energy-Recommendation-System--HERS

Paper: Home Energy Recommendation System(HERS): A Deep Reinforcement Learning Method for Electricity Optimization in Smart Homes

1-Dependencies: Numpy Matplotlib Pandas Math Random Tensorflow Keras Pytorch

2-Data Generation: Throughout the code we have used padding=50; # non-informative addded characters to a string to the ends of it for formatting purposes.

a)Activity Labels:

Activity labels are taken from the Activity Recognition with Ambient Sensing (ARAS) dataset. link: https://www.cmpe.boun.edu.tr/aras/

We extracted the 15-minute interval activity labels from the daily txt files provided in the dataset. 

This is implemented in house_B_activity_data.ipynb

b) Device Operation Data:

We generated our device operation data from the activity labels as explained in the main paper. This is implemented in smart_home_data.ipynb

c) Results:

We implemented our proposed deep RL based Home Energy Recommendation System(HERS) in HERS.ipynb

We applied the popular Advantage Actor Critic(A2c) algorithm.

Hyperparameters are: GAMMA = 0.99

For the 4 devices; input states and output actions are taken as explained in the main paper. The network architectures are: (Hidden layers are shown as (number of neurons, activation function))

1. AC Actor Model structure: (learning_rate=3e-6), 1 (input) -(12, linear)-(12, relu)-(2, softmax) Critic Model structure: (learning_rate=3e-6), 1 (input) -(12, linear)-(12, relu)-(1, linear)

2. DW Actor Model structure: (learning_rate=3e-6), 18 (input) -(48, linear)-(120, linear)-(48, relu)-(18, softmax) Critic Model structure: (learning_rate=3e-6), 18 (input) -(48, linear)-(120, linear)-(48, relu)-(18, linear)

3. WD Actor Model structure: (learning_rate=3e-6), 18 (input) -(48, linear)-(120, linear)-(48, relu)-(18, softmax) Critic Model structure: (learning_rate=3e-6), 18 (input) -(48, linear)-(120, linear)-(48, relu)-(18, linear)

4. EV Actor Model structure: (learning_rate=3e-6), 3 (input) -(48, linear)-(120, linear)-(48, relu)-(5, softmax) Critic Model structure: (learning_rate=3e-6), 3 (input) -(48, linear)-(120, linear)-(48, relu)-(5, linear)