Using Machine Learning/Deep Learning to identify home appliances consuming excessive power

Overview

This Machine Learning model is helpful in aiding home automation and home appliaances when there is limited supply of power. Typical home appliances are considered and power consumption of these devices are noted down based on properties like type of device, time (hourly basis), power consumption, number of people and time they have used the appliances in an hour.

Dependencies

1. Python 3.x
2. Numpy, Scipy, Pandas, Matplotlib
3. TensorFlow 1.x+, Keras 2.x+
4. NVIDIA CUDA 9.x, cuDNN 7.x (Only if using Tensorflow-GPU)
5. Sci-kit learn

Important Files and Folders

home
│   README.md   
│
└───data
│      sample - contains sample dataset
│      trial_x - all types of different datasets
│
└───regression
│      *.ipynb - Regression ML/DL algorithms
│   
└───model
│      regression models saved
│
└───docs
│      documentation and reports
│
└───main
       visualization - data analysis
       data generation - generate data and cleaning
       experiment - results and custom algorithms

Usage

1. main/data generation/generate_home_data.py - generates home_data.csv

   • Generates dataset with all columns
   • This dataset will be used to build machine learning model

2. main/data generation/generate_home_data_test.py - generates home_data_test.csv

   • Generates dataset with all columns
   • This dataset will be used to predict power values

3. main/visualization/total_power_consumption.ipynb - creates date_time_group.csv

   • This is the final output which will be used to visualise and compare machine learning algorithms

4. main/visualization/total_power_consumption_final.ipynb - creates final_output_group.csv

   • This is the final output which will be used to visualise and compare machine learning algorithms

5. regression/<*>.ipynb

   • Machine learning algorithms used on home_data.csv to build models in /model/

6. predict/<*>.ipynb

   • Machine learning algorithms used on home_data_test.csv to predict and generate new datasets
   • home_data_predict_x are dataset generated with prediction values

7. main/experiment/trail_x/trail_x_x.py

   • Custom algorithm to give messages to users - generates final_output_x.csv in data/trail_x

8. main/experiment/trail_x/results_x_x.py

   • Visualising all algorithms and performances
   • Graphs plotted to identify power consumption before machine learning and after using machine learning and custom algorithm

Results

Available in main/experiment/trail_x/results_x_x where 'x' can be either sequence or a machine learning algorithm

Credits

Vinod Agrawal

CTO, Faststream Technologies

License

Copyright(c) 2018, Faststream Technologies

Authors:

• Sudhanva Narayana
• Shreyas S