Privacy preserving data collection for smart grid using self-organizing map

Homomorphic encryption is widely researched in the smart grid area to publish and transfer electricity consumption data between electricity companies. This method makes it feasible to calculate total electricity consumption of neighborhoods without sharing any raw electricity consumption data. In the area of demand response(DR), calculating the total consumption of electricity is important in order to create DR reports which are published by third party to reduce the peak period of electricity usage such as 7 am or 6pm. Nevertheless, the possibility of data exposing or data decryption may lead to individual households private information revealing, for example, the timing of leaving home, timing of arriving home, appliances usage, detailed information of electricity devices. To avoid privacy disclosure, this thesis proposes a new framework based on self-organization map(SOM) which is an unsupervised learning method. The framework can share and publish electricity power consumption data between electricity providers securely and accurately and fulfill DR called SOM with the k-means framework. SOM with the k-means framework enables electricity providers sharing data without raw data published. Meanwhile, nearly 2.5% to 3% error and lower entropy can be achieved, which is a satisfactory result. SOM and k-means framework is a robust and effective approach for DR in the smart grid.