Distributed neural network enhanced power generation strategy of large-scale wind power plant for power expansion

Power generation is an important energy conversion process. A prominent feature of wind power plant compared with traditional power plants is that the equipment utilization has great intermittency and uncertainty. Fortunately, with the popularity of converter-based wind turbines, doubly-fed induction generator wind power plants are able to not only employ wind turbines in converting wind power to electrical power, but also use the converter capacity to produce reactive power. Traditionally, PQ curves considering current constraints are used for active and reactive power distribution, and therein the fixed maximum slip in power control is conservative for the utility of converters. To fully extract the power generation capacity, a novel power control realizing adaptive variation of maximum slip is proposed for each wind turbine to dynamically expand the reactive power capacity in the high active power region. In the meanwhile, the dispatch scheme of the wind power plant based on the conventional PQ curve has to be upgraded accordingly. Given that the unfixed maximum slip makes it impossible to obtain the PQ curve with explicit expression, an online learning method based on neural network is adopted and further developed in a distributed architecture to be consistent with natural distributed characteristics of the large-scale wind power plant. With the help of communication among agents through consensus protocol, the algorithm convergence can be guaranteed even under distinct information fragments. Case studies demonstrate that the proposed scheme is able to achieve up to 30% expansion of reactive power capacity and 16.3% voltage sag alleviation under certain conditions.