Wind energy is considered as a clean renewable resource that plays an important role in mitigating greenhouse emissions, which are pivotal in climate change. Therefore, there has been observed a consistent growth of the wind energy sector in the recent past. Unfortunately, despite the undoubted features of being a free and clean energy resource, wind power is generally intermittent and uncontrollable. Market participants as well as power system operators therefore face major challenges, one being unable to control uncertainty and variability of wind power generation.
One main duty of transmission system operators is to maintain the balance of electric power generation and electric load. For wind power farms, maintaining the power balance appears to be quite challenging in view of the fluctuating nature of wind resources as well as the problem of large-scale energy storage. Therefore, transmission systems operators generally schedule an optimal combination of controlled generating units to satisfy a forecasted load, while having an assumed wind generation prediction, power system reserve needs and generation and transmission constraints.
Wind power forecasting therefore becomes an indispensable factor for electricity market players, for instance, energy trading companies and wind energy producers. Therefore, Jacek Wasilewski at PSE Innowacje sp. z.o.o and Dariusz Baczynski from the Warsaw University of Technology Poland formulated an approach allowing for estimating an assembly of prediction models satisfying a number of model learning and testing criteria. They then developed the Pareto-optimization method. They presented a mathematical model and a case study of the wind power-forecasting taking into account the Pareto-optimization of the selected prediction criteria. Their work is published in Renewable and Sustainable Energy Reviews.
The authors discussed and analyzed the application issues of multi-criteria prediction model optimization in the intra- as well as next day wind power prediction. In order to understand the artificial neutral networks, a multi-objective method was developed based on Pareto-optimization. The method takes into account a concept of forecast and enables the analysis of accuracy prediction models implementing a generalized multi-criteria function.
Wasilewski and Baczynski showed that the effectiveness of the ANN-MLP learning reference to a particular criterion could be achieved independently of the algorithms (Back Propagation, Particle Swarm Optimization and hybrid BP+PSO algorithms were used) .The learning process was effective when the learning data set was applied to choose the optimal model. An excess of coordinate of the utopia vector at the NISE procedure occurred only if the ANN process was not effective. However, the utopia vector was as a result of ANN optimization reference to the testing data.
It was observed that no impact of the analyzed wind farm and the ANN-MLP structure on the given outcomes has been evident in most analyzed situations. The ANN-MLP model was a machine learning model with low bias and high variance. The experiment outcomes presented should be confirmed also with the use of other categories of prediction models and different statistical attributes.
Future works by Wasilewski and Baczynski will perhaps focus on developing an interface between a decision making process and the forecasting. This will be aiming at presenting how operators as well as decision makers in energy markets would implement the proposed forecasting method reference on the multi-criteria approach.
Wasilewski and D. Baczynski. Short-term electric energy production forecasting at wind power plants in pareto-optimality context. Renewable and Sustainable Energy Reviews, volume 69 (2017), pages 177–187.Go To Renewable and Sustainable Energy Reviews