Estimation of natural resources for renewable energy systems

In this chapter, the authors explain a methodology for training a feedforward neural network (FFNN) and a recurrent neural network (RNN) using a genetic algorithm (GA). We use these networks for natural resource estimation. Specific details about the structure and coding of a GA individual are carefully explained, taking advantage of the flexibility offered by GAs. We use this technique to implement the measure-correlate-predict (MCP) method to extrapolate natural resources in a target place with limited data availability. Training and testing errors on wind speed and power for FFNN and RNN are discussed and analyzed. Additionally, the comparison with a reference training method based on the Levenberg–Marquardt algorithm revealed that a GA requires extensive computational resources to reach a reasonable solution.