A Context-Aware Hybrid Scheduling Algorithm for Home Energy Management Systems

Sammanfattning

Home Energy Management Systems can be used to control household electricityconsumption by scheduling appliances in a way that reduces high power peaks. This is becoming more relevant as electricity network tariffs move toward models where shortperiods of high power demand may increase households costs and stress on thedistribution grid. Existing scheduling approaches can reduce peaks but many rely on static priorities, predefined operating times, or computationally heavy machine learningmethods. This thesis investigates whether historical household usage patterns can be included in a less computationally heavy context-aware hybrid scheduling algorithm. The proposed algorithm combines Particle Swarm Optimization and Technique for Order Preference by Similarity, where appliance schedules are evaluated using cost, user comfort constraints, and a peak-related penalty. The context-aware strategy uses probability profiles derived from historical appliance usage, while a non-context-aware strategy schedules appliances without this information. The algorithm is evaluated in a controlled simulation experiment using appliance-level electricity data from the Pecan Street dataset. Each household was simulated using three strategies: baseline, non-context-aware, and context-aware scheduling. The results show that both scheduling strategies reduced peak demand, the Peak-to-Average Ratio, and electricity cost compared to the baseline. However the difference between the context-aware and non-context-aware strategies was small and not statistically significant in any of the metrics. This indicates that the main improvement comes from shifting flexible appliances, while the current context-aware component does not provide a clear additional benefit in this experiment. The study demonstrates that historical usagepatterns can be effectively integrated into a hybrid Home Energy Management Systemscheduling algorithm. However further research is required to enhance the utilization of contextual information for peak reduction.