This paper reviews methods for determining the optimal number of hidden neurons in neural networks over the past 20 years and proposes a new method for Elman networks in wind speed prediction. The random selection of hidden neurons can lead to overfitting or underfitting, affecting the network's performance. The authors test 101 criteria based on statistical errors to find the best model, which improves accuracy and minimizes errors. The proposed model is validated using real-time wind data, showing minimal errors and improved performance. The review covers various approaches, including constructive and pruning methods, and discusses the advantages and disadvantages of each. The proposed method, based on the convergence theorem, is shown to be effective in reducing errors and improving the stability and accuracy of the network.This paper reviews methods for determining the optimal number of hidden neurons in neural networks over the past 20 years and proposes a new method for Elman networks in wind speed prediction. The random selection of hidden neurons can lead to overfitting or underfitting, affecting the network's performance. The authors test 101 criteria based on statistical errors to find the best model, which improves accuracy and minimizes errors. The proposed model is validated using real-time wind data, showing minimal errors and improved performance. The review covers various approaches, including constructive and pruning methods, and discusses the advantages and disadvantages of each. The proposed method, based on the convergence theorem, is shown to be effective in reducing errors and improving the stability and accuracy of the network.