| With the vigorous development of technologies such as artificial intelligence,big data,and cloud computing,data centers,as a critical infrastructure of the digital economy,are rapidly increasing in number and scale.However,the energy consumption of data centers is enormous and growing rapidly,posing a challenge to the "dual-carbon" goal.It is worth noting that the offline computing load processed by data centers has flexible scheduling that can be shifted over time,which also brings potential flexibility in controlling the power load of data centers.Based on the flexible scheduling of computational loads,flexible energy use can be achieved in data centers,which provides the possibility for the coordinated operation and optimization of data centers and the power system.This article starts from the perspective of coordinated optimization of data centers and the power system,using the scheduling of large-scale computational loads as the entry point.It focuses on the energy management model of data centers based on flexible scheduling of computational loads,the energy cost optimization method for large-scale offline load scheduling in data centers,and the real-time balancing market strategy that takes into account the flexible scheduling of computational loads in data centers.The main research work of this article is as follows:Firstly,from the perspective of energy-computing coordinated optimization,this article constructs a model for the scheduling of computational loads and energy management in data centers based on a Markov decision process model,considering technical challenges such as task dependencies,task heterogeneity,and service quality in actual production and operation of data centers.Secondly,this article proposes an optimization method for low-carbon operation of data centers for large-scale offline load scheduling.Considering the massive,online,and real-time characteristics of the computational task scheduling problem in large-scale cloud computing systems,this article constructs an energy-aware computational task scheduling strategy online adaptive algorithm based on deep reinforcement learning.The algorithm introduces reward scaling and baseline function regularization methods to improve the training efficiency and convergence performance of the policy network.Case studies based on actual data show that the proposed algorithm can optimize the energy cost of data centers by 37%and reduce carbon emissions by 14%,while ensuring service quality.Finally,this article proposes a real-time balancing market strategy that takes into account the flexible scheduling of computational loads in data centers.Based on the flexibility of computational load scheduling,data center bidding decisions are optimized to enable data centers to obtain profits in the power real-time balancing market while providing regulation services to the power system.Case studies based on actual data show that the proposed method can optimize the revenue of data centers in the power real-time balancing market and improve the economic operational of data centers. |
