Estimating householdapplianceusepatternthroughsmartmeter dataanalysis

Abstract

The energy sector in India is slated to experience changes in both demand and supply in the coming years. With increased renewable integration and peak demand not coinciding with solar hours, electricity distribution companies will face the challenge of effectively matching demand and supply. While demand flexibility measures can help, there is a need for high-frequency data on the unit level and appliance level demand to formulate better policies. One method to under stand ’behind-the-meter’ activities is through non-intrusive load monitoring. We use the Prayaas eMARCsurveytoconduct non-intrusive load monitoring to predict the air conditioner’s state. Dif ferent models are used: logistic regression, ANN and LSTM. We find the neural networks perform better than the logistic regression. The LSTM model can disaggregate air-conditioner use with 99% accuracy, performing better than the other models.