Module-based explainable predictions of energy consumption (MEPEC)

Energy consumption varies significantly across different areas and is dependent on various factors like weather, type of the consumer, appliances used, etc. If predictable, the data could be used to improvise energy production and develop cost reduction strategies. However, the complicated nature of the dependent parameters makes it difficult to predict the consumption. This submission attempts to address this problem based on historical consumption readings available for a set of consumers along with other related data. The consumers, represented by the meter_id in the dataset, are clustered based on the dwelling type and the number of bedrooms of the representative household. We use linear regression as the primary method to approximate the relationship between the average monthly temperatures observed and the corresponding energy consumption readings for each of the clusters separately and predict the future consumption patterns of the consumers accordingly. The predictions are then explained in natural language based on the features of the consumer (cluster dependent) to make the predictions more comprehensible. These explanations are devised using a template with the predicted data.