基于XGBoost的网格化雷电预测模型

Grid Lightning Prediction Model Based on XGBoost

电网设备智能防雷是人工智能与电力工作很好的结合点。本文采用单一的雷电探测数据,基于XGBoost算法和网格化的思想,建立雷电预测模型。主要结论如下:①模型仅仅依赖雷电探测数据,实现了探测器周围各网格的非定点雷电分级预警,时间提前量为30分钟,网格大小为5 km*5 km,预测效果较好。②模型的预测精确率可达78.6%,预测召回率可达65.6%,F1值可到71.5%。③缩小网格一定程度可提升预测精确率,但漏报率也随之提升,两个指标矛盾又统一;将网格扩充为25宫格时,增加了大量不必要的特征值,导致模型预测效果降低。本文的模型改进了当前雷电预测模型定点预测的局限性,增加了分级预警,在主动防雷模式中具有现实意义。

Intelligent lightning protection for power grid equipment is a good combination of artificial intelligence and power work.This paper uses a single lightning detection data to establish a lightning prediction model based on the XGBoost algorithm and gridding ideas.The main conclusions are as follows:①The model only relies on lightning detection data to achieve non-fixed-point lightning hierarchical early warning for each grid around the detector.The time advance is 30 minutes,the grid size is 5km*5km,and the prediction effect is good.②The prediction accuracy of the model can reach 78.6%,the prediction recall rate can reach 65.6%,and the F1 value can reach 71.5%.③Reducing the grid can improve the prediction accuracy to a certain extent,but the false negative rate will also increase.The two indicators are contradictory and unified;when the grid is expanded to 25 grids,a large number of unnecessary feature values are added,resulting in model prediction effect reduced.The model in this paper improves the limitations of fixed-point prediction of the current lightning prediction model and adds hierarchical early warning,which has practical significance in active lightning protection mode.