Extreme weather is a complex phenomenon formed by the spatiotemporal coupling of multiple meteorological parameters. Traditional methods

which are usually based on single factors or simplified assumptions

struggle to accurately characterize the coupling relationships among multiple parameters. This results in an incomplete identification of power system risk scenarios under extreme weather and a lag in assessment results. This limitation makes it difficult for existing planning methods to perform collaborative optimization based on dynamic risks

and it is hard to enhance system resilience while balancing safety and economy throughout the full course of extreme weather events. Therefore

this paper proposes a method of power supply elastic planning based on risk decision. Integrating multi-source meteorological data and historical fault information of power grid

the correlation matrix between meteorological factors and power system state is constructed

and the meteorological power mapping boundary is established by introducing threshold offset coefficient to accurately identify extreme weather risk scenarios. This paper adopts the inverse transformation method based on Copula theory and Monte Carlo sampling

a probability quantization model for the joint distribution of multi-dimensional meteorological factors is constructed to accurately capture the tail dependence between meteorological parameters

and a time-varying failure probability model is constructed based on the service life of the equipment

extreme environment and operation status. With this as the input

a deep active learning model is constructed. The daily load loss is used to dynamically evaluate the risk of supply and demand imbalance of the system

and the flexible planning decision is initiated according to the risk level. Guided by extreme weather risk aversion and flexibility improvement

a two-stage power supply flexibility planning model is built to jointly optimize the system elasticity deficiency rate

cost and load supply ratio. Results show that under a blizzard scenario

this method strictly controls the system's daily load loss within the 150 MW alert threshold

with a maximum of 140 MW; the shares of energy storage and new energy installations increase to 5.61% and 38.90%

respectively; the system resilience insufficiency rate drops to 18%

and the load supply ratio reaches 95%

effectively enhancing the dynamic robustness and power supply resilience of system in response to extreme weather.