Reservoir inflow forecasting with artificial neural networks is presented in this paper. Different types of ANN input data were considered such as temperature, precipitation and historical water inflow. Performance of the hourly inflow forecasts was assessed based on a case study of a specific hydropower reservoir in Latvia. The results showed that all the approaches had similar prediction errors implying that for optimal hydropower scheduling uncertainties need to be modelled which is also proposed in this study through generation of several forecast realisations in addition to point predictions.

This paper discusses various forecasting tools that can be used in predicting the thermal load in district heating networks, focusing on day-ahead hourly planning as it is particularly important for cogeneration plants participating in electricity wholesale markets. Forecasts obtained by employing an artificial neural network are compared to a polynomial regression model. Their ability to supplement each other in a combined forecasting tool has been considered as well. Prediction inaccuracy cost is observed and suggested as evaluation criterion. The case studies are based on the district heating network in Riga, Latvia. Recorded data sets of temperature and heat demand are applied for thermal load prediction.

In this paper, we propose a novel stochastic approach to multi-objective optimization of hydroelectric power generation short-term scheduling. Maximization of profit is chosen as the main objective with additional sub-objective-to reduce the number of startups and shutdowns of generating units. The random nature of future electricity prices and river water inflow is taken into account. We use an artificial neural network-based algorithm to forecast market prices and water inflow. Uncertainty modeling is introduced to represent the stochastic nature of parameters and to solve the short-term optimization problem of profit-based unit commitment. A case study is conducted on a real-world hydropower plant to demonstrate the feasibility of the proposed algorithm by providing the power generation company with the day-ahead bidding strategy under market conditions and a Pareto optimal hourly dispatch schedule of the generating units.

Accurate demand forecasting in district heating networks is an essential and imperative task in the everyday operation of both, the network itself and the heating energy suppliers. Multiple regression is one of the possible approaches to solving the forecasting problem with sufficient accuracy and little computational effort. This paper presents a polynomial regression model and offers several additions for its further improvement. It is found that grouping the model residuals by hour-of-day allows notably reducing the forecast error. The value of other modifications and the optimum size of the training set can vary over time, thus an automatic model parameter selection before each new forecast is advised.

This paper discusses the importance of accurate forecasting tools in solving power system planning, modelling and optimization tasks. While artificial neural networks are widely considered to be one of the best prediction methods, their accuracy can vary greatly depending on the network structure and parameters. A method of experimentally finding the best ANN parameters has been offered and tested on heat demand forecasting. Some value of the benefits of increased prediction accuracy on the operation of CHP plants has been identified.