Research On Data Center Energy Storage Scheduling Scheme Based On Reinforcement Learning

In order to meet the ever-increasing demand for network resource services,the data center,as a physical platform for data resource management,which is increasing in scale.The data center is a high energy consumption industry.As the scale of data center expands,the high cost of electricity brings huge expenses to the construction and operation of data center.Therefore,while ensuring the safe and reliable operation of the data center,minimizing the power cost of the data center is very important for the sustainable development of the data center industry.The diversified electricity price market provided by smart grids and the rapid development of data center energy storage battery technology provide opportunities to reduce data center electricity costs.In the context of the diversified electricity price market,through reasonable control of energy storage batteries to " low electricity price to store power,high electricity price to discharge ",the goal of reducing data center power costs can be achieved.The work done in this article is as follows:(1)Aiming at the problem of estimating the remaining power of energy storage batteries,a neural network-based battery remaining power estimation scheme is proposed.The BP neural network is used to approximate the non-linear mapping relationship between the remaining battery capacity and the battery charging and discharging current,terminal voltage,and cycle discharge times.A reasonable excitation function and optimization algorithm are selected,and a neural network-based battery remaining power estimation model is constructed.For the proposed energy storage battery remaining power estimation scheme,the CALCE data set is selected to test the performance of the scheme.The test results show that the remaining power estimation scheme proposed in this paper has a small remaining power estimation error,which meets the error size required by the general battery remaining power estimation.(2)Aiming at the optimization problem of data center energy storage control strategy,a data center energy storage control scheme based on TD(λ)algorithm is proposed.By reasonably modeling the state space,action space and reward function of TD(λ)algorithm,the goal of reinforcement learning is consistent with the goal of minimizing the power cost of data center,and the energy storage control model of data center is constructed.TD(λ)algorithm is used to solve the reasonable energy storage control strategy under different price market environment.For the proposed data center energy storage control scheme,it was verified in the energy storage system simulation environment.The experimental results show that the energy storage control scheme proposed in this paper has a strong power cost saving ability,and the impact of different state space sizes on the performance of the energy storage control scheme has been tested.