The objectives of this master's thesis are a comprehensive analysis of the energy consumption of the company AT&S and the development of forecast models to accurately predict future energy needs. This aims to estimate the future energy demand in terms of quantity and associated costs. Based on the company's electricity meter readings and thermal energy requirements, the existing data was prepared for analysis and then used to create forecasts for predicting energy consumption. To achieve these goals, different forecasting methods were initially selected and examined to determine the conditions under which each method is best suited. A thorough literature review was conducted at the beginning of this work, analyzing various forecasting methods. The research revealed that regression methods and neural networks have particularly good properties for forecasting. For the application of regression methods, information about the consumption of individual energy consumers was utilized. The use of neural networks allowed additional information to be linked with the measured values. By comparing consumption time series with additional data, such as outdoor temperatures or maintenance schedules, the accuracy of a forecast could be significantly improved. Regression methods were implemented using the Visplore program. The initially used data was cleansed of measurement errors, and then the data was analyzed for repetitive behavioral patterns. These patterns allowed the data to be divided into sections, each having its own function determined. These functions could then be combined to form a forecast that predicted the energy demand based on past measurements. For the application of the neural network, a Python program code was developed. The neural network had to be trained on the available data, requiring parameters such as the size of test sections and batch sizes to be specified in the program. These parameters had to be adapted to the available data for each forecast. Subsequently, it was possible to generate a forecast based on this learning process. By linking past measurements with additional data, such as maintenance schedules, even more accurate predictions could be made compared to the regression method. For particularly small consumers or those that could not be adequately predicted with the mentioned methods, simpler forecasting methods were applied. The statistical method of linear regression based on consumption averages was suitable for this purpose. If the available data was not precise enough due to missing information or the forecast was too complex for the regression method and neural network, this method could provide an approximation for the energy demand. The company's energy consumption was divided into heat, cold, compressed air, and electrical power, with this thesis focusing on analyzing and forecasting heat, cold, and electricity consumption. The analysis of the company's electricity consumption does not allow for an accurate prediction as a whole, so individual areas were examined separately to determine the most suitable forecasting methods for each.