User:FrederikNiklasBach/sandbox

= Load forecasting = Load forecasting (electric load forecasting, electric demand forecasting) refers to all practices, methods, and processes used for forecasting the amount of electricity to be consumed in the future. An accurate prediction increases supply security while at the same time being relevant for capital investment, the environmental quality, the revenue analysis and market research management. Furthermore, efficient electricity management systems are useful for providing a short-term (day-ahead or hour-ahead) energy production plan which can then be utilized for demand response applications like load peak minimization, self-consumption optimization, intelligent energy storage, and predictive control. It has increased in significance due to the energy system transformation. Electricty demand is usually predicted in kW or kWh over different time horizons (such as hourly, daily, or weekly demand) for different regions or geographical zones of interest. The prediction is usually based on parameters that influence residential, commercial and industrial electricity consumption alike. For example:
 * meteorological parameters such as temperature, humidity, sunhours, precipitation
 * Economic variables: such as GDP and unemployment rate and residential space
 * Socio-economic: variables such as national holidays, random events, incidents

Electricity Load forecasting with linear methods
Forecasting hourly electricity demand with a linear model is challenging since it includes several superimposed seasonal parameters and nonlinear effects of exogenous variables such as temperature, humidity and precipitation. A model that incorporates the whole profile therefore fails due to high collinearity between hourly loads. Instead models for each hourly load are used that run regressions separately (e.g 24 regressions for a day). However, these models ignore the interactions between the parameters and therefore only lead to results of „less“ predictive power compared to non-linear methods. Adding on, after removing collinear values, the reduced datasets are prone to overfitting.

Electricity Load forecasting with Artificial neural networks
Artificial Neural Networks (ANN) offer a relatively high accuracy and perform better at solving non-linear and complex problems. Therefore, load predictions with ANNs perform substantially better than linear models because their sensitivity to differing types of variable inputs is  ensured. For example, the electricity demand of households will be affected by a national holiday whereas a hospital might not be influenced. A neuronal network can account for this differing type of input parameter. In many developed countries smart grids are employed or developed which can react to changes in electricity demand based on real-time analyses carried out by neural networks using a digital and connected infrastructure. Even more accurate are Adaptive neuro fuzzy inference systems (ANFIS) which combine neural networks with fuzzy logic.