Data Alchemist

Domestic power consumption depends largely on use of household appliances. Appliances having moving mechanical parts (like motors in washing machines and compressors in refrigerators and air conditioners) and which involve heating (like heaters, electric ironing machines, water geysers, etc.) consume more electricity than electronics (like Televisions, LED lights, etc.). Energy ratings and age of these appliances, residential wiring age and quality, home insulation, natural light within the house further influence energy consumption considerably. It is generally observed that climate changes (humidity and temperature) have a strong correlation with domestic electricity consumption. However, it is also subject to unprecedented and unexpected situations like COVID 19. Due to lack of substantial appliances related data across all households, we considered weather data as the major contributor to consumption. While daily temperature (max, min, avg) might be fed to the machine-learning model to get a consumption figure, the challenge requires to predict energy use for the entire succeeding year, for which daily temperatures are neither available, nor can they be calculated, derived or predicted. Previous day’s consumption also helps in predicting it for current day. We assumed that while actual temperature and consumption might vary in the succeeding year, the variation (seasonality and trend) in maximum, minimum and average temperature, along with the daily change in consumption would remain similar and follow the same pattern. To even out the sudden spikes in consumption for a particular day, we have considered 7 days’ moving average of energy consumption, along with the daily variation in the moving average. Use of appliances, newer appliances in households, unexpected situations like COVID 19 can’t be predicted, and hence haven’t been accounted for in the model.