most of the researches on PV power generation forecasting methods have problems such as long time for model training and propose an optimization. Using the BP(backpropagation) neural network, this learning algorithm is mainly applicable to multi-input, multi-output networks. It can rely on ready-made data and input and output without knowing the mathematical relationship between the mapping relationship in which input and output. The mapping relationship is learned and stored. In addition, BP neural networks have great advantages in dealing with non-linear problems and have strong generalization ability.

In this paper, we investigate the representation of wind power forecasting (WPF) uncertainty in the unit commitment (UC) problem. While deterministic approaches use a point forecast of wind power output, WPF uncertainty in the stochastic UC alternative is captured by a number of scenarios that include crosstemporal dependency. A comparison among a diversity of UC strategies (based on a set of realistic experiments) is presented. The results indicate that representing WPF uncertainty with wind power scenarios that rely on stochastic UC has advantages over deterministic approaches that mimic the classical models. Moreover, the stochastic model provides a rational and adaptive way to provide adequate spinning reserves at every hour, as opposed to increasing reserves to predefined, fixed margins that cannot account either for the system’s costs or its assumed risks.