Establishing the conditional probability distribution (PD) of wind power forecast error (WFE) is a prerequisite for many stochastic analysis considering wind power integration. However, with the increasingly emergence of new data, the update burden of the conditional PD is getting heavier as the size of training data set grows rapidly. Meanwhile, the centralized training manner of the conditional PD may reveals the data privacy of wind farms (WFs) belonging to different stakeholders. To solve these problems, we propose a distributed privacy-preserving (DPP) incremental update algorithm (DPP-IUA) for updating each WF's conditional PD of WFE considering their correlation. This algorithm consists of two original methods: (1) a DPP incremental Gaussian-mixture-model algorithm (DPP-IGA) for updating the joint PD of the correlated WFs; and (2) a DPP mechanism for deriving each WF's conditional PD of WFE under a given forecast vector. The DPP-IUA keeps each WF's conditional PD of WFE up to date with extremely low update burden. Meanwhile, this algorithm is also fully distributed and strictly protects the data privacy of different WFs. The effectiveness, correctness and efficiency of the proposed DPP-IUA is empirically verified using historical data.
from cs updates on arXiv.org http://bit.ly/2w2jLgP
//
0 comments:
Post a Comment