Research On Control Strategy Of Unbalance Of Supply Power System For IDC

With demand for information is growing explosively, types of data centerhave become huge electricity users. How to reduce the power consumption of IDC(Internet Data Center), is the issue.Three-phase imbalance is one of the important indicators of power quality.Though the power transport wire is not too long in IDC power system, yet the powerconsumption caused by imbalance is above20%30%because of high current.The cost of this power consumption is paid by customers. So if the imbalance issuewill be resolved, many economic benefits will be obtained. This research is the subkey scientific and technical project in Zhejiang Province.The three-phase unbalance calculation methods are analyzed referring to theresearch of resolving the imbalance home and abroad. The existing technologies aremostly proposed in the perspective of power network. From the characteristics of theIDC room, this essay is researched in the perspective of user load. The main factorsof unbalance are researched. To eliminate the unbalance caused by unbalancedistribution of IT load equipment, the wire power consumption will be lowed down.At last the control strategy based on BP neural network theory is proposed. Themodel is self-learning and adaptive. The working phase current is considered as themain factor. The control system can adjust to the change of current and control theload distribution to balance.At the meantime, the prototype of corresponding hardware control system isdesigned. It is used to control IT equipment to connect the phase. After the testingand verification in a computer room of one university, the hardware system is safeand reliable. And the host computer monitoring software is also designed.Functionally, algorithms off-line training and calculation are on host computer. Therelay of IT load equipment is controlled in underlying controlled. This kind offunction division is useful for using computing resource reasonably and reducing thecost of researching control equipment. After the large scale of off-line training to the neural network, our purpose of design is fulfilled. The system has good relativeadaptation. At last some problems of this project are proposed for the future researchdirection and also for the whole research group.