Evaluating Forecasting Techniques for Integrating Household Energy Prosumers into Smart Grids Teodor Petrican, Andreea Valeria Vesa, Marcel Antal, Claudia Pop, Tudor Cioara, Ionut Anghel, Ioan Salomie Email: teodor.petrican@student.utcluj.ro, {andreea.vesa, marcel.antal, claudia.pop}@cs.utcluj.ro, {tudor.cioara, ionut.anghel, ioan.salomie}@cs.utcluj.ro Department of Computer Science, Technical University of Cluj-Napoca Distributed Systems Research Laboratory Abstract—This paper tackles the problem of integrating house- hold energy prosumers in Smart Energy Grids by analyzing a set of state-of-the-art energy forecasting techniques that allow individual or aggregated prosumers to evaluate their future energy demand and inform the Distributed System Operator (DSO) about potential grid imbalances. Thus, the DSO can perform a proactive strategy to manage the grid and avoid problems before they appear. The key element of this approach is the prediction technique, that must be accurate enough such that the resulting grid imbalances can be compensated in real- time. The paper evaluates a set of state-of-the-art statistical and Machine Learning (ML) prediction techniques, such as SARIMA, feed-forward and recurrent neural networks, support vector regression or ensemble prediction models, on real household historical energy demand logs by performing a feature selection process for each ML algorithm as to identify the best elements that influence the energy demand of a house. A set of experiments are performed on the REFIT Electrical Load Measurements data set evaluating each model’s performance with respect to the selected features. Among the evaluated algorithms, the Ensemble Prediction Model gives best prediction accuracy, showing a Mean Absolute Percentage Error (MAPE) of 14.4% followed by the SVM model with a MAPE of 15.4%. Index Terms—Smart Grids, Energy Demand Forecasting, Short-Term Load Forecasting, Multi-layer Neural Network, Re- current Neural Networks, Support Vector Regression I. I NTRODUCTION Energy demand forecasting has become of real interest nowadays, since more and more focus is put on designing efficient systems that are able to control and optimize energy consumption and production worldwide. The key benefits of having such a prediction system are both environmental and economical. During the last decades energy consumption has increased exponentially therefore a proper management has become necessary to keep a sustainable and secure environ- ment. As electric energy cannot be stored for future use, it has become crucial to produce as much energy as needed, and this needs to happen almost in real-time. On the global electricity market, a more accurate hourly/daily prediction is of great interest to obtain in real-time the best purchase prices. Moreover the adoption of different pricing methods is encouraged such as time-of-use billing, demand-based pricing etc. [1] Smart Grids are placing the energy industry in a new era meant to offer a better integration of producer-consumer power generation systems, with major benefits in the eco- nomical, environmental and security directions. Being a two- way communication between the utility (usually called the Distributed System Operator (DSO)) and its consumers, an optimization of the energy consumption is critical for main- taining the power grid reliability and avoiding supply-demand mismatches. As presented in Figure 1 the Smart Grid is a distributed system where each consumer (households in our case) is bidirectionally connected with the DSO. As the available power in the grid varies over time, the DSO may suggest consumers energy curves to reduce their consumption when rates are lower. This is done through a Demand Response (DR) signal. The consumers may accept or not the proposal. In case of acceptance the consumers need to reschedule their consumption according to the DR signal. A new energy curve will be sent to the utility that represents the actual day ahead plan. A crucial aspect of the system described above is the Energy Consumption Prediction Module, as the DSO will make adjustment decisions based on this module’s output, the decisions quality being tightly related to the predictions quality. The focus of our study is to evaluate several models for short-term energy consumption forecasting. The goal is to obtain forecasts of energy consumption for the next day at a granularity of one hour, thus having a 24 steps ahead prediction. The main contribution of this paper is a thorough investiga- tion of state-of-the-art machine learning algorithms used for forecasting and evaluating various instances of their models with different features on the REFIT Electrical Load Mea- surement data set [2] aiming to determine the best algorithm configuration for energy prediction. The rest of the paper is structured as follows: Section II shows related work, Section III presents the models of the machine learning algorithms evaluated for forecasting, Section IV presents the evaluation of the predictions on a real-world dataset, while Section V concludes the paper.