基于数据融合的电网强对流临近预警技术研究

Research on severe convection proximity warning technology for power network based on data fusion

强对流天气的破坏力大、预测准确率和实时性较差,且容易对电网造成严重危害,因此构建有效的电网强对流天气临近预警系统具有重要的现实意义。传统的雷达回波技术虽然能够根据云团位置的改变来推断后续天气的变化,但当云团发生融合或分裂时,其预报准确率较低。针对该方法存在的缺陷,文中使用深度学习技术对气象云图数据进行处理,其算法部分将卷积网络与长短期循环神经网络相结合,使其兼具准确性和实时性。在算法对比实验中,该算法的虚警率比卷积法以及LSTM法降低了5.2%和6.0%,说明该算法的性能较为理想。而在测试验证实验中,所提算法能够实现对1 h后强对流天气的预测,并可完成对电网的强对流临近预警,具有良好的工程应用价值。

Severe convective weather has great destructive power and poor prediction accuracy and real⁃time,which is easy to cause serious harm to the power grid.Therefore,it is of great practical significance to build an effective early warning system for severe convective weather in the power grid.Although the traditional radar echo technology can infer the subsequent weather changes according to the position changes of clouds,its prediction accuracy is low when clouds fuse or split.Aiming at the defects of this method,this paper uses deep learning technology to process meteorological cloud map data.The algorithm part combines convolution network and long short⁃term cyclic neural network to make it both accurate and real⁃time.In the algorithm comparison experiment,the false alarm rate of the proposed algorithm is 5.2%and 6.0%lower than that of convolution method and LSTM method,which shows that the performance of the proposed algorithm is ideal.In the test and verification experiment,the proposed algorithm can predict the severe convection weather after 1 h,and can complete the early warning of the approaching severe convection in the power grid,which has good engineering application value.