Open Access

ARTICLE

Energy System Low-Carbon Transformation Operation Optimization Based on Deep Deterministic Policy Gradient Algorithm

Jing Shi¹, Zesen Li¹, Delv Zhu¹, Bingjie Li¹, Lang Gao^2,*
1 Economic and Technological Research Institute of State Grid Jiangsu Electric Power Co., Ltd., Nanjing, China
2 Sichuan Energy Internet Research Institute Tsinghua University, Chengdu, China
* Corresponding Author: Lang Gao. Email:
(This article belongs to the Special Issue: Advances in Renewable Energy and Storage: Harnessing Hydrocarbon Prediction and Polymetric Materials for Enhanced Efficiency and Sustainability)

Energy Engineering https://doi.org/10.32604/ee.2026.077553

Received 11 December 2025; Accepted 02 February 2026; Published online 11 March 2026

Abstract

In view of the multi-energy subject coupling and operation optimization problems faced by the integrated energy system in the low-carbon transformation, taking a certain city in Jiangsu Province as the experimental object, the research first constructs a city-level low-carbon integrated energy system with hydrogen energy storage as the core hub. Then, the mathematical models of key equipment such as gas turbines, gas boilers, and thermal storage tanks are established. Finally, the multi-energy subject system achieves autonomous optimization and collaborative control through multi-agent DDPG. This method showed good energy utilization efficiency and scheduling flexibility in different seasons. The comprehensive energy utilization rate in heating season, non-heating season and transition season reached 82%, 80% and 78%, respectively. As the photovoltaic coverage rate increased from 5% to 40%, the wind power proportion increased from 10% to 50%, the total system operating cost decreased from US$73.87/MWh to US$49.00/MWh, carbon emissions dropped by approximately 32.4%, and the power curtailment rate decreased by approximately 21.7%. Compared with the single-agent algorithm, the average operating cost was reduced by 9.6%, the carbon emissions were reduced by about 8%, the energy balance error was reduced by about 39.8%, and the convergence speed was reduced by about 28%. These results fully verify the efficient decision-making and dynamic optimization capabilities of the multi-agent DDPG in complex multi-energy interaction environments, providing a feasible technical path for the intelligent operation.

---

Keywords

DDPG; multi-agent; energy system; low-carbon; operation optimization

---