Intelligent Energy Management in Smart Community with Distributed Machine Learning

With the development of electrical technologies, many contemporary buildings, are integrated with local renewable energy resources (RES), e.g., PV: photovoltaic or wind turbine.

Unlike traditional energy consumers, buildings with local RES, i.e., prosumers, can both consume and produce electricity.

The appearance of prosumers, EV, storage and P2P trade dramatically changes the future electricity system and makes building energy demand-side management (DSM) rather complex.

To balance the supply and demand while making full use of the RES, reliable forecasting of load and RES generation is a key requirement in the energy DSM of modern buildings.

However, for smart community with complex grids, accurate forecasting is rather challenging due to the limitations of model overfitting, latency, privacy and high communication costs, which are common in centralized learning.

To address the problem, we will exploit recently proposed distributed machine learning schemes, i.e., federated learning (FL) to address the problems of model accuracy, privacy and communication costs for load and RES forecasting in smart building DSM. We will develop new FL based deep learning algorithms for both load and RES generation forecasting.

We will exploit ADMM based optimization for energy management. Then, our results will be verified in the testbeds of the smart buildings of our partner.

We expect that the proposed technologies can substantially optimize energy usage (thus reduce the greenhouse gas emission) and meanwhile improve the prosumer economy and privacy of users are also protected. The project partners include KTH, SuperGrid AB and Haneberg Lantbruk AB (HLA, building owner and end user).

The research results of the project shall be verified in the smart community of HLA.

The expected results include students and researcher training, new learning algorithms, research articles, testbed, open-source software, improving the electrical usage efficiency and economy in smart buildings.