Towards a Decision Support System for Real-Time Pricing of Electricity Rates: Design and Application

The share of renewable energy in today’s power grids is continually increasing. However, it is notoriously difficult to accurately forecast renewable electricity sources like wind and solar production with the granularity that energy providers require. To compensate for the fluctuating production and forecast errors, energy providers have to use expensive control energy. This partly negates the positive effect of renewables. Various ideas for load smoothing on the production side have been suggested. Here, we focus on load shifting on the consumer side: electricity rates that may vary in hourly intervals can influence smart devices in private consumer households. With real-time pricing (RTP) the energy provider can send high prices when production is behind forecasts. On the other hand, prices should be cheap when the production exceeds the forecast. Cheap rates would incite electricity consumptions. The challenge is to identify the price signal that will result in the desired load shift at consumers. As the behavior of smart devices is still unknown today we use a simulation prototype and train an artificial neural network with simulation data. As it turns out the neural network leads to good results and achieves hit rates in the task of mapping the desired load shift to a price signal. This hit rate only slightly decreases when we submit the price model to some constraints that increase consumer-friendliness. The advantage of using a neural network is that it can adapt to a slowly changing mix of smart devices in households. By regularly retraining the network we are able to react to the future reality.