The invention discloses a power grid high low temperature refined early warning method in combination with dynamic correction, comprising the steps of: S1, forecasting local weather to generate highest and lowest temperature forecast results and extracting live weather factors influencing local temperature forecast results; S2, establishing highest and lowest temperature forecast equations in dependence on generated temperature forecast results and selected live weather factors; S3, counting highest and lowest temperature forecast errors based on actually measured highest and lowest temperature information and forecast results, and correcting the highest and lowest temperature forecast results based on the temperature forecast errors to obtain final forecast results of highest and lowest temperatures; and S4, performing electric power equipment priority forecasting based on the final forecast results of highest and lowest temperatures and publishing early warning information. The power grid high low temperature refined early warning method can effectively solve the problem of great differences between a power grid system high temperature and low temperature forecast effect and an actually measured result, timely and effectively take corresponding measures to cope with extremely high and low temperatures, and mitigate damage caused by extreme weather.