储能科学与技术 ›› 2021, Vol. 10 ›› Issue (6): 2200-2208.doi: 10.19799/j.cnki.2095-4239.2021.0282

• 储能系统与工程 • 上一篇    下一篇

基于改进K-meansMADDPG算法的风储联合系统日前优化调度方法

蔡新雷1(), 崔艳林1, 董锴1, 孟子杰1, 潘远1, 喻振帆1, 王吉兴2, 孟乡占2, 余洋2()   

  1. 1.广东电网有限责任公司电力调度控制中心,广东 广州  510600
    2.新能源电力系统国家重点 实验室(华北电力大学(保定)),河北 保定  071003
  • 收稿日期:2021-06-22 修回日期:2021-07-08 出版日期:2021-11-05 发布日期:2021-11-03
  • 作者简介:蔡新雷(1986—),男,硕士,高级工程师,研究方向为电力系统运行控制,E-mail:517665114@qq.com|余洋,副教授,从新能源电力系统并网与控制研究;E-mail:yym0401@163.com
  • 基金资助:
    南方电网公司科技项目资助036000KK52190005(GDKJXM20198110)

Day-ahead optimal scheduling approach of wind-storage joint system based on improved K-means and MADDPG algorithm

Xinlei CAI1(), Yanlin Cui1, Kai DONG1, Zijie MENG1, Yuan PAN1, Zhenfan YU1, Jixing WANG2, Xiangzhan MENG2, Yang YU2()   

  1. 1.Electric Power Dispatching Control Center of Guangdong Grid Co. , Ltd. , Guangzhou 510600, Guangdong, China
    2.State Key Laboratory of Alternate Electrical Power System With Renewable Energy Sources (North China Electric Power University), Baoding 071003, Hebei, China
  • Received:2021-06-22 Revised:2021-07-08 Online:2021-11-05 Published:2021-11-03

摘要:

风储联合运行可有效地应对风电出力的不确定性,提高风电竞争力,然而储能与风电联合运行的优化调度问题是一大难点。为保证储能可调控容量的基础上实现风储联合运行收益最大化,提出了基于改进K-means和多智能体深度确定性策略梯度(MADDPG)算法的风储联合系统日前优化调度方法。首先,根据储能特性和运行状态采用萤火虫优化的改进K-means聚类算法实现储能分组;然后,将风电与分组后的储能设备建模为不同的智能体,组成多智能体系统,采用MADDPG算法求解,设计了MADDPG算法的状态空间、动作空间和奖励函数;最后,对算法进行了算例仿真验证。结果表明,所提调度策略能够较好协调风电和储能运行,有效平抑了风电出力波动,与常规深度强化学习相比,提高了风储联合系统的运行收益。

关键词: 风电, 储能系统, 改进K-means, MADDPG

Abstract:

The joint operation of wind power and energy storage can effectively deal with the uncertainty of wind power output and improve the competitiveness of wind power. However, optimizing and dispatching the joint operation of energy storage and wind power is a major difficulty. A day-ahead optimal scheduling method of the wind storage joint system based on improved K-means and multi-agent deep deterministic strategy gradient (MADDPG) algorithm is proposed to maximize the benefits of a wind-storage joint operation by ensuring the adjustable capacity of energy storage. First, the improved K-Means clustering algorithm optimized by the firefly algorithm is used to achieve energy storage grouping; then, the wind power and the grouped energy storage equipment are modeled as different agents to form a multi-agent system. The MADDPG algorithm is used to solve the problem, and the MADDPG algorithm's state space, action space, and reward function are designed. Finally, a simulation example is used to validate the algorithm. The results show that, when compared to conventional deep reinforcement learning, the proposed scheduling strategy can better coordinate the operation of wind power and energy storage, effectively smooth the fluctuation of wind power output, and improve the operating income of the wind storage combined system.

Key words: wind power, energy storage system, K-means, MADDPG

中图分类号: