Research On Distribution Network Reliability Evaluation Technology Based On Distributed Parallel Computing

With the continuous progress of the secure,reliable and intelligent power grid construction,as the last link of the power transmission to the user,the reliability of the distribution network has gradually become the key factor to determine the power supply and distribution capability of the whole power grid.The effective evaluation of reliability can help the power company to find the weak link in distribution network in time.It has important theoretical and practical significance to guide the planning and design of distribution system as well as the upgrading.Therefore,it is very necessary to study a reliability evaluation method with high computation performance and analytical process fitting the actual power grid operation.On the basis of the traditional distribution network reliability evaluation method,this paper combined with the action characteristics of the switch elements,and simplified the network by partitioning the load and virtual nodes.Based on this,it proposes the reliability evaluation process under two scenarios that one is not taken into account of the backup constraint,the other is consider the optimal selection of backup.And in the second scenatio,the fine prediction result of distribution transformer load is achieved to realize the reliability evaluation of load in the future state.The specific work of the paper is as follows:(1)Put forward the prediction technology of distribution transformer load based on Apache Spark computing framework.At first,a dynamic Bayesian network is established as a load forecasting model of transformer areas in this paper,which includes the causal relationship between temperature,humidity,day type and other factors with load and load time series.Next,in order to solve the problem of model parameter calculation based on massive data and the parallelization of lots of transformers' load forecasting,a distributed graph computing platform based on Spark GraphX parallel computing component is built.Then combining Vertex cut and parallel path search of GraphX to extracting probabilistic information fast.Pregel computing model used to parallelizing forward backward algorithm to realize the distributed cluster computing of the prediction method and ensure the parallelization of global prediction tasks.(2)The reliability calculation technology of distribution network based on load block and virtual node partition is proposed,and the equivalent conversion method of topology data of distribution network based on Pandapower is given.Accroding to the operation characteristics of switch elements in distribution network,the network is divided into feeder division,load block and virtual node,thus the complex distribution network with multi load branch is simplified,and the minimum cut set of virtual node is used to replace the cut set of the full network load point,so the calculation is reduced effectively.At the same time,Pandapower's powerful graph search function also provides support for minimum cut set algorithm based on graph structure analysis.(3)Though the analysis of the constraint conditions such as line capacity and reserve margin,the reliability evaluation method of distribution network is put forward in consideration of the optimization of standby power supply.On the basis of the information of the virtual node cut set,the N-1 principle is used to analyze the common influence mode of the switch action under each state of the fault,to determine the optimal backup power and power outage condition of the load,and then get the power outage time.Finally,based on the Transformation-Action operator in Spark,the parallel processing of the reliability calculation process in the future state of the load without the reserve constraints and considering the reserve optimization is completed.According to the parallel implementation of the reliability evaluation method proposed in this paper,the reliability evaluation of the distribution network under two conditions that without the reserve constraint and the reserve optimization is carried out respectively.The rationality and effectiveness of the proposed method are verified by the analysis of the experimental results,and the reliability parallel calculation based on the Spark is indeed carried out.The performance has been greatly improved,and it will have broad application prospects in the face of the increasing distribution network scale.