基于GRU多步预测技术的云储能充放电策略

为解决云储能日前充放电策略预测的问题,提出了一种基于门控循环单元(GRU)多步预测技术的云储能充放电策略形成方法.首先,鉴于云储能优化需求及单用户负荷预测效果不稳定,构建用户聚类后的GRU多步预测方法预测一天的24点功率.然后分析了云储能模式下的用户和云储能提供商两个主体的交互过程,以预测为基础建立了云储能充放电决策滚动优化模型.仿真算例选取实际数据,在预测聚类用户光伏、负荷功率后,滚动优化求解实际值和预测值下的云储能充放电策略.算例通过5种场景验证了在云储能充放电策略中聚类的作用以及GRU多步预测技术的优势,并且证明云储能模式能够进一步削弱光伏、负荷预测误差的影响.

Charging and Discharging Strategy of Cloud Energy Storage Based on GRU Multi-step Prediction Technology

In order to solve the day-ahead charging and discharging strategy prediction problem of cloud energy storage (CES), a formation method of charging and discharging strategy of CES based on multi-step prediction technology for gated recurrent unit (GRU) is proposed. Firstly, in view of the CES optimization demand and the unstable effect of single-user load prediction, after user clustering, a multi-step prediction method of GRU is constructed to predict the power at 24 points of a day. Then, the interaction process between the user and the CES provider in the CES mode is analyzed, and the rolling optimization model of CES charging and discharging decision is established based on the prediction. In the simulation example, the actual data is used, and after the prediction of photovoltaic and load power of cluster users, the CES charging and discharging strategies for the actual data and the predicted data are solved by rolling optimization. Five scenarios are used to verify the role of clustering in CES charging and discharging strategy and the advantages of GRU multi-step prediction technology. It is also proven that cloud energy storage mode can further weaken the influence of photovoltaic and load prediction error.