Especially short-term load forecasting, it is important for the operation of an electric energy system. The quality of the load demand forecast is a significant factor for decision making in operation planning, real-time operation, and security analysis. As a result, a forecasting with high precision enables the optimization of the production and the operation of the electrical system. Several short-term load forecasting (STLF) methods including traditional and artificial intelligence-based methods are available.
Historical consumption and its relationship with other relevant variables (i.e climatic, seasonal, economic, demographic indicators etc) define the suitable model of energy consumption. Influence coming from these variables can vary. More in detail climate factors derive from temperature and humidity whereas seasonal factors represent seasonal climate change and load growth year by year. Regarding social factors, it is an indication of human social activities such as work, entertainment etc. Based on the above we conclude that the relationship between electric load and its external factors is complex and nonlinear. For that reason it is quite difficult to model electric load through traditional techniques such as linear or multiple regression, autoregressive moving average (ARMA), exponential smoothing methods, Kalman filtering etc.
The most important management goal is an accurately electric load forecasting. Considering that the electric load often presents nonlinear behavior, as we mentioned, a rigid forecasting approach with strong general nonlinear mapping capabilities is crucial. This is the main reason that nowadays, alternative approaches for load forecasting have had their use improved. Through research on the artificial intelligence field, more precise results come from:
- Artificial Neural Networks (ANN). Main advantage of ANN arises from the fact that the knowledge is extracted from a database and there is no need of previous knowledge about the model.
- Machine Learning algorithms, such as Support Vector Machines (SVMs), either for regression or for classification. In this case the learning strategy is based on the theory of statistical learning, targeting to propose learning techniques that maximize the generalization capacity.
Within global/EU residential and commercial buildings are responsible for about 40% of total energy consumption. Hence building energy modeling is an essential tool in the development of informed decisions, such as identifying energy consumption trade-offs in the building design process, sizing components (e.g., HVAC, Lighting) for a specific building, optimizing control systems and strategies for a building, determining cost-effective retrofit packages for existing buildings. For example at the urban scale, buildings that are close to each other may share many common features such as microclimates, building functions and occupancy patterns. Some energy efficiency measures for a single building could possibly apply to multiple buildings nearby.
In conclusion two components proposed to achieve some good prediction models. Initially some sensor data, about energy consumption and relevant features, collected from residential and commercial buildings. Afterwards an artificial intelligence approach in order to reveal useful correlations on the data and succeed pattern recognition.
Electrical Engineer, Data Scientist