An extended new approach for forecasting short-term wind power using modified fuzzy wavelet neural network: A case study in wind power plant

Nowadays, there has been more attention paid to clean energies, especially wind, since there is a shortage of fossil fuel resources, but there is a decrease in pollutants that result from such sources. Such energies play a significant role in generating power. The non-linear nature of wind speed poses challenges and difficulty in exploiting its power. As a result, an accurate and efficient prediction of wind power will serve as a crucial means for solving the system’s planning and operational issues. This article aims to predict a wind power plant’s power output using weather and power plant parameters and employ an extended fuzzy wavelet neural network (FWNN). In the extended method, any fuzzy set rule uses different fuzzy wavelet functions to convert input space into a subspace. A uniform hybrid learning algorithm was used in the extended FWNN method to obtain an optimal proportion of parameters. The method was an optimal combination of the Particle Swarm Optimization (PSO) and Gradient Descent algorithms. The use of optimized PSO is slightly different from the basic PSO in that in this method, two layers of PSO are used within each other. Not only it has high convergence but also higher coordination and adaptability with the gradient descent algorithm. This method was used for the Manjil wind power plant in Iran, with real data being recorded every 10 min. The extended FWNN method was also compared with the conventional prediction methods. The results showed that compared to other methods reported earlier, the proposed method was a more efficient tool and had higher precision for short-term wind power forecasting. •Generator temperature is used as a new and effective parameter for predicting wind power.•An effective and efficient model is extended for Short-term wind power forecasting.•Fuzzy logic is used to overcome the inherent uncertainty of the data.•A hybrid optimization approach is used to increase the accuracy of network training.•The extended FWNN comparison is presented with eight valid prediction methods.