Robert Golob

Neural-network-based water inflow forecasting

Abstract Water inflow forecasting is usually based on precipitation data collected by the ombrometer stations in the river basin. Solution of this problem is rather complex, due to the highly non-linear relation between the amount of precipitation at different locations and the water inflow into the head hydro power plant reservoir. In this paper, a new approach to forecasting water inflow, based on neural networks, is presented. First, selection of input parameters is discussed. Next, the most appropriate architecture of the neural networks, is chosen. Finally, the efficacy of the proposed method is tested for a practical case, and some results are presented.

Intelligent coordinative voltage and reactive power control

A novel local voltage control is proposed where each voltage controller of a generating unit is equipped with additional coordinative voltage controller (CVC), which incorporates intelligent control into the power system. The controllers use artificial neural networks (ANN), trained on a large set of power system states. Optimal power flow (OPF) results serve as the optimal target in the ANN training process. All data the controller needs for operation are acquired locally. The proposed ANN voltage control maintains a sub-optimal power system voltage profile and reduces the power system losses. When equipped with ANN controllers, the system responds satisfactorily to changes in the power system topology, thus improving the power system security.

Fast contingency evaluation by means of the improved adjoint network method

Abstract In this paper, an improvement of the adjoint network (AN) method based on the Tellegen theorem for calculation of power system variables following various contingencies is presented. A new model of PV nodes is developed and the response of PV nodes to line as well as generator outages is modelled accordingly. Furthermore, an improved computational scheme is derived that enables a fast single step evaluation of post contingency power system variables. Example cases of test and practical systems show that the improved adjoint network (IAN) method offers an efficient tool for on-line contingency analysis due to its improved accuracy and computational effort in comparison with the widely accepted 1 PI Q approach based on fast decoupled powerflow (FDPF).

Short and mid term hydro power plant reservoir inflow forecasting

In order to improve water management for hydro cascade systems, a new artificial neural network (ANN) approach to forecasting water inflow into the hydro power plant reservoir based on forecasted precipitation data is presented. Water inflow forecasting into the head hydro power plant (HPP) reservoir is one of most important inputs to the cascade hydro system (CHS) optimization process. Ability to properly forecast the increase of natural inflow can result in increased electric energy production due to enhanced flexibility in stored water management. Due to the Soca river torrential character two separate algorithms for short term and mid term water inflow forecasting are designed. Short term water inflow forecasting is based on precipitation data collected by the ombrometer stations in the river basin and is used for forecasts up to 6 hours ahead. The efficacy of the proposed method is tested for a practical case and some results are presented. Mid term water inflow forecasting is based on the forecasted precipitation data and is capable of predicting water inflows for the next two days. The precipitation forecasts data are obtained with the ALADIN (Aire Limitee Adaptation Dynamique development International) program, which was developed by the Meteo-France in cooperation with Slovenian hydrological institute and other Central European hydrological institutes. The data acquisition system to be implemented as a part of new software in the regional control center is briefly described. Finally, some practical results for both short and mid term water inflow forecasting for Soca river are presented.

Improved adjoint network algorithm for online contingency analyses

Abstract An improved adjoint network based algorithm is proposed for static security assessment. The main feature of the algorithm is the calculation of postcontingency nodal voltages and phase angles in a noniterative fashion. This is achieved by adequate modeling of the generator (PV) nodes to account for the voltage regulators response to line or generator contingencies. The tests show that the improved adjoint network method yields considerable savings in computing time for large power systems while retaining or improving the accuracy of the voltage phasor calculation as compared with the widely used 1P1Q method.

Modelling the Structure of Long-Term Electricity Forward Prices at Nord Pool

This chapter models long-term electricity forward prices with variables that influence the price of electricity. Long-term modelling requires consideration of expected changes in the demand and supply structure. The model combines high-resolution information on fuel costs from financial markets and low-resolution information on the demand/supply structure of the electricity market. We model the latter using consumption and supply capacity and the former with forward prices of fuels, emission allowances and imported electricity. The model is estimated using data from the Nordic electricity market and global long-term forward prices of energy. Owing to a lack of data on consumption and supply capacity, the estimated results provide only the broad influence of these variables on forward prices. Though extrapolation of the prices observed in Nord Pool may suffer from the influence of short-term variables, such as precipitation and temperature, the model yields robust forecasts of the prices of contracts that are not exchange traded.