Data-driven prosumer-centric energy scheduling using convolutional neural networks

The emerging role of energy prosumers (both producers and consumers) enables a more flexible and localised structure of energy markets. However, it leads to challenges for the energy scheduling of individual prosumers in terms of identifying idiosyncratic pricing patterns, cost-effectively predicting power profiles, and scheduling various scales of generation and consumption sources. To overcome these three challenges, this study proposes a novel data-driven energy scheduling model for an individual prosumer. The pricing patterns of a prosumer are represented by three types of dynamic price elasticities, i.e., the price elasticities of the generation, consumption, and carbon emissions. To improve the computational efficiency and scalability, the heuristic algorithms used to solve the optimisation problems is replaced by the convolutional neural networks which map the pricing patterns to scheduling decisions of a prosumer. The variations of uncertainties caused by the intermittency of renewable energy sources, flexible demand, and dynamic prices are predicted by the developed real-time scenarios selection approach, in which each variation is defined as a scenario. Case studies under various IEEE test distribution systems and uncertain scenarios demonstrate the effectiveness of our proposed energy scheduling model in terms of predicting scheduling decisions in microseconds with high accuracy.