谷毅,富子豪,王登政,等. 基于机器学习的配用电场景信号覆盖优化技术[J]. 电波科学学报,2024,39(3):1-8. DOI: 10.12265/j.cjors.2023190
      引用本文: 谷毅,富子豪,王登政,等. 基于机器学习的配用电场景信号覆盖优化技术[J]. 电波科学学报,2024,39(3):1-8. DOI: 10.12265/j.cjors.2023190
      GU Y, FU Z H, WANG D Z, et al. Machine learning based signal coverage optimization technology for power distribution and utilization scenarios[J]. Chinese journal of radio science,2024,39(3):1-8. (in Chinese). DOI: 10.12265/j.cjors.2023190
      Citation: GU Y, FU Z H, WANG D Z, et al. Machine learning based signal coverage optimization technology for power distribution and utilization scenarios[J]. Chinese journal of radio science,2024,39(3):1-8. (in Chinese). DOI: 10.12265/j.cjors.2023190

      基于机器学习的配用电场景信号覆盖优化技术

      Machine learning based signal coverage optimization technology for power distribution and utilization scenarios

      • 摘要: 配用电场景是电力系统的重要组成部分,其无线网络设计对于提高配用电终端的信号质量、实现全面监测和管理、提升电力系统效率、支持智能化决策和控制,以及促进能源管理和节能减排具有重要的意义。本文面向配电网典型场景,提出基于机器学习的配电网场景覆盖优化技术。首先,根据实际电力业务终端和需求,建立配用电典型场景环境模型;其次,利用射线追踪技术生成信道大数据,并建立基于广义回归神经网络(general regression neural network, GRNN)的路损预测模型;最后,联合总覆盖率、平均单节点覆盖率、重复覆盖率作为奖励值,利用强化学习优化节点部署位置。仿真结果表明,所提GRNN路损预测模型可精确预测路损,所提联合多维指标的奖励函数计算方法可有效提高网络收敛速率。本文所提算法对配电网无线网络规划和设计具有重要意义。

         

        Abstract: The distribution scenario is an important part of the power system, and its wireless network design is important for improving the signal quality of power terminals, realizing comprehensive monitoring and management, enhancing the efficiency of the power system, supporting intelligent decision-making and control, and promoting energy management and energy conservation. In this paper, we propose a machine learning-based distribution network scenario coverage optimization technique for typical scenarios of distribution networks. Firstly, a distribution typical scenario environment model is established based on actual power business terminals and demands. Secondly, ray-tracing technique is used to generate channel big data and build a generalized regression neural network (GRNN) based pathloss prediction model. Finally, the total coverage, average single node coverage, and repeated coverage are combined as reward values and reinforcement learning is used to optimize node deployment locations. The simulation results show that the proposed GRNN pathloss prediction model can accurately predict the pathloss, and the proposed method of calculating the reward function by combining multi-dimensional indicators can effectively improve the convergence rate of the network. The proposed algorithm is important for the planning and design of wireless networks in distribution networks.

         

      /

      返回文章
      返回