基于深度行为评判策略的5G光伏基站低碳绿能方法

PDF下载HTML

Title:

Low-carbon green energy methods for 5G photovoltaic station based on the deep actor-critic strategy

作者:

邓翠艳¹, 齐小刚²

1. 晋中信息学院, 山西 晋中 030800;
2. 西安电子科技大学 数学与统计学院, 陕西 西安 710071

Author(s):

DENG Cuiyan¹, QI Xiaogang²

1. Jinzhong College of Information, Jinzhong 030800, China;
2. School of Mathematics and Statistics, Xidian University, Xi’an 710071, China

关键词:

5G低碳绿能; 光伏能源; DAC节能策略; 强化学习; 节能策略; 智能体; 电网; 蓄电池

Keywords:

5G low-carbon green energy; photovoltaic; DAC energy-saving strategy; reinforcement learning; power-saving strategy; agent; power grid; battery

分类号:

TP311.13

DOI:

10.11992/tis.202501024

摘要:

5G作为新型信息基础设施，正融入千行百业，超宽带技术及万物互联设备的大规模应用增加了网络能耗、运营成本和碳排放。运营统计5G单站功率已达到2 000 W左右，月均电费可达1 000元左右。随着光伏产业的发展，将清洁能源应用到大功耗5G网络已成为网络低碳绿能发展的一种新途径。本文提出了一种基于深度行为评判(deep actor-critic, DAC)策略的5G光伏基站低碳绿能方法，使用光伏能源代替传统的电网火电能源。建立了光伏、蓄电池和电网一体化储能供能模型；为了最大限度节约碳排放，加大绿能供应效率，设计了一种DAC节能策略；构建5G网络节电智能体，引入A3C(actor-critic algorithm)智能体动作奖励模型来操纵蓄电池节能动作；通过求解更好的奖励值，蓄电池可以找到最佳的充放电策略实现低碳绿能。通过对比仿真结果，在不同季节下，相比Q学习及深度Q网络算法，本文方法具有更好的网络节能表现，提高了低碳绿能效率。

Abstract:

The large-scale application of 5G ultra-broadband technology and IoT devices has significantly increased energy consumption, operational costs, and CO2 emissions. For example, a single 5G base station can consume approximately 2,000 W, incurring monthly electricity costs of about 1,000 yuan. With the development of the photovoltaic industry, integrating clean energy sources into high-power 5G networks has emerged as a promising approach to support low-carbon development. This paper proposes a low-carbon energy management method for 5G networks based on a deep actor-critic (DAC) strategy, which replaces conventional grid power with photovoltaic energy. An integrated energy storage model combining photovoltaic systems, battery storage, and grid connectivity is developed. A DAC-based optimization strategy is then applied to maximize carbon emission reductions and improve the efficiency of green energy supply. To support this strategy, a 5G network power supply agent is constructed using the A3C algorithm in conjunction with a deep Q-network, enabling dynamic energy optimization. By optimizing reward functions, the proposed method achieves high energy efficiency. Simulation results demonstrate that, across different seasons, the proposed method outperforms both Q-learning and the deep Q-network algorithms in improving network energy efficiency, thereby advancing low-carbon and green energy objectives.