Modelling human behaviour in smart home energy management systems via machine learning techniques

Mankind is progressing towards automation solutions that operate via Internet of Things in their everyday lives, which has developed over the years and its demanding nature compels the emerging technologies to smoothen the modern day living. Critical issues like increasing domestic energy usage, if addressed via machine learning techniques in current IOT devices will tremendously help humans in unimaginable ways. The concern of price hikes in energy and the adverse effects on the environment due to inefficient energy usage is a subject that concerns a larger segment of the population. Human behavior plays an important role in energy expenditure. Markets shall never cease to flood with electronic gadgets and humans shall never cease to fancy the upcoming ones that majorly contributes to increasing energy consumption. So we propose a framework with load optimization strategy that shall be responsive to realtime pricing which will exclude frequent human intervention. The load will be optimized during the times of high pricing or when the total power consumption of the household exceeds the maximum constraint. However, Optimal Load Management strategy requires price prediction that can be generated using real-time price algorithm and human activity prediction to ensure discrete decision making. This paper proposes an idea to mitigate energy consumption by modeling human behavior via a combination of Hidden Markov Model and Naive Bayes classifier. Also, we try to minimize the errors generated due to motion/occupancy sensors while detecting a human or an animal(pet). Sometimes activation of motion sensors can result in the wrong detection which can further lead to incorrect activity recognition which then can generate errors in optimizing loads. Therefore, we integrate classification algorithms in our framework which can further aid in correct detection and subsequent strategy generation.