抄録
For the preventive maintenance of overhead power transmission lines, it is important to predict the possibility of disasters due to snow accretion. The authors propose a new method to estimate the possibility of these disasters for the localized areas covering over Kanto Plain. This method uses neural networks with the input data of ambient temperature, precipitation, wind velocity, and the product of precipitation and wind velocity as time series for 6 hours for each local area meshed by 6km squares. Within Tokyo Electric Power Co., 155 records of the disaster experiences and their meteorological conditions in the recent 13 years are first investigated in detail. Then the relationship between the disasters and the meteorological conditions is analyzed by neural networks. As a result, it was clarified that the neural networks can accurately classify the meteorological conditions by learning the disaster experiences. The simulation results show that the performance of this method exceeds that of conventional method using multiple regression, and that the method can estimate the possibility and the severity of snow disasters with the accuracy of 87% and 76%, respectively.
