Building Household Electricity Load Profiles Using Deep Learning Technique

Abstract

The research to monitor the variations of electricity usage in the sector of households is becoming extremely popular among the demand side management (DSM). One popular approach is the load profiling of electricity consumption data, recorded on different granularities. With passage of time, data is becoming more sophisticated due to its high resolution. It’s not just the data but the algorithms are more efficient these days. With the help of this high-resolution data and powerful analytical algorithms, researchers are predicting the patterns and behaviors of electricity consumption in a household with high accuracy. One popular approach these days is the use of neural nets to observe these patterns of electricity usage. Although neural nets and their variations are very powerful to learn all the patterns on a temporal level, still the electricity consumption is very dynamic to observe. There are many factors that contribute and affect the usage of electricity by individuals in a household. To solve this issue, and to make our algorithms more accurate in prediction, not just the electricity usage but also the other factors and constraints are important to consider. Our algorithms take the electricity consumption data and incorporate it with other factors including the socio-economic parameters. On this sophisticated and curated data, we applied different neural nets to observe the usage of appliances in a household. We took different datasets and applied these techniques with and without the socio-economic parameters and compared the results. Our research can help to improve and understand the constraints and limitations of demand side flexibility in the residential sector