Research On Intelligent Optimal Control Strategy Of Reactive Power Voltage

The large-scale wind power and other clean energy centralized access to the grid,due to its output volatility,will cause the grid voltage to change many times within a short period of time,increase the transmission costs of power grid enterprise,and affect the quality of power supply.Therefore,with the development of new technologies such as artificial intelligence,it is necessary to propose an optimal control strategy for grid reactive power voltage adaptation to the new situation.At present and in the future,hierarchical voltage control is still an important measure to solve the optimal control of reactive voltage in large power grids.Therefore,a scenario where wind power is centrally connected to the grid is designed and problems such as zoning of large power grids in hierarchical voltage control and selection of leading nodes for secondary voltage control are studied.On this basis,this paper try to use artificial intelligence algorithm to seek a new method of hierarchical voltage control that can quickly form a control strategy.The division of large grids is an important means to reduce the scale of control.In order to solve the problem of how to determine the number of partitions,this paper introduces an AP clustering algorithm that can adapt the data structure and automatically form the number of clusters.Taking the impact of wind power fluctuations in the division process into account,this paper constructs an electrical distance matrix based on the expected wind power output,and implements automatic zoning of PQ nodes based on the AP clustering algorithm to obtain the clustering centers of each region.Considering the control effect of reactive power nodes on the voltage of load nodes,this paper constructs a reactive power control space to obtain the electrical distance from the cluster center,and realizes the zonal merge of reactive power nodes based on the cluster center.Finally,the index to quantify the partition quality is introduced,and the rationality of the proposed method is verified based on the IEEE39 node system.The dominant node can represent the overall voltage level of the area after the partition.The task of secondary voltage control is to maintain the voltage of the dominant node near the target value.The choice of dominant nodes includes both quantity and site selection.In this paper,the electrical distance matrix established in the zoning method is used to determine the number of dominant nodes using principal component analysis.Taking the impact of wind power fluctuations into account,the location of the dominant node takes the centralized access point of the wind power as a dominant node,and the location of the remaining dominant nodes is based on the cluster center obtained after zoning.The calculation example shows that the dominant node selected in this paper can well represent the regional voltage,and the proposed method has a small amount of calculation and wide applicability.Subarea divisions and the selection of dominant nodes are the basis of hierarchical voltage control,and also a key measure to solve the "dimensionality disaster" of the state/action space of reinforcement learning algorithms.The threelevel voltage control takes the minimum loss of the entire network as the optimization goal,the generator terminal voltage as the optimization variable,and uses particle swarm optimization to optimize the entire network to obtain the target node voltage value.The second-level voltage control is based on reinforcement learning algorithm.It builds multiple agents with capacitor bank and transformer tap voltage regulator as the object.It designs a dynamic training environment,state and action set and Q value table that considers various factors.After a lot of training,multi-agents give the best action strategy,and finally make the leading node voltage approach the target value.The examples show that the proposed method can effectively improve the voltage quality of the power grid.The agent has both offline and online training capabilities,and can quickly give action strategies in emergency situations.