Work Detail |
Novel research from Germany and the USA has analyzed the impact of heat pump (HP) integration on the ability of day-ahead load forecasting in energy communities. Using different models, the scientists have also investigated whether HP loads should be forecasted separately from the rest of the household or both together.
An international research group has investigated the impact of heat pumps on energy community load forecasting and has found that using so-called transformer models improves forecasting quality.
“Traditional load patterns will change in many countries by transforming the heating sector towards heat pumps,” explained the academics. “This development has a severe impact on the operators of energy communities. First, it is unclear if the same forecasting methods perform well for traditional household loads and heat pump loads. Second, the potential impact of the aggregation level on energy community load forecasts has not been investigated.”
The group used forecasting models based on machine learning techniques, such as random forests and XGBoost, as well as the recurrent neural networks technique of long-short-term memory networks (LSTMs). In addition, it evaluated the novel neural network architecture of transformers.
In addition to given load and perfect foresight weather data, the academics have proposed additional features that might need to be implemented in each forecasting: the type of day, cyclical calendric features, the rolling average of apparent temperature, the average load at the same time step, and past loads. For each forecasting case and forecasting method, the group used the Bayesian optimization model to identify the most relevant features.
Furthermore, the researchers used the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) method as an extension to each forecasting technique. “An increasing number of load forecasting studies is applying decomposition techniques to improve the model performance,” they explained. “The CEEMDAN algorithm has several advantages over alternative decomposition methods: it exhibits an improved handling of the mode mixing problem, it is more robust to noise, as well as being non-stationary.”
All of the forecasting models, extensions, and aggregation scenarios were applied to a high-quality dataset of household loads in an energy community in Hamelin, Germany. The dataset includes active and reactive power, voltage, and current measurements of 38 households equipped with water-to-water HPs and an additional heating rod as a backup heater. According to the researchers, the installation of HPs, in this case, changed the peak load from 20.1 kW to 80.1 kW.
The model was trained and analyzed to forecast loads at the low-voltage transformer level, to which multiple households of an energy community are connected. Data from 2019 were used for the training and testing of hyperparameter selection, while data from 2020 were used for the actual benchmarking of the different methods.
The results show that the best-performing forecasting methods change after the installation of heat pumps. While random forests or XGBoost deliver reasonable forecasting quality in predicting traditional load, the transformer-based method was more accurate when the data was aggregated with HP.
“The day-ahead energy community load forecasting quality cannot be notably increased by obtaining separate measurements of heat pump loads, which would constitute an additional effort for energy community or distribution grid operators,” the results further showed. “Transformer-based models are also delivering the best performance in a real-world peak reduction battery energy storage systems (BESS) use case for the investigated energy community with heat pumps.”
The research was presented in the paper “The impact of heat pumps on day-ahead energy community load forecasting,” published in Applied Energy.
Researchers from Germany’s Karlsruhe Institute of Technology and Indiana’s Purdue University conducted the research. “We encourage researchers to use our dataset, results, and evaluation pipeline, which we publish open-source, to benchmark novel methods against them to advance accurate forecasting techniques for loads of energy communities with heat pumps and to apply our methodology on alternative datasets,” they highlighted. |