Polymer dielectric capacitors capable of stable operation under extreme conditions are essential for advancing electrification, yet conventional trial-and-error approaches are inefficient and severely hinder material discovery. Herein, we propose a machine learning driven high-throughput screening framework to accelerate the development of high performance capacitive energy storage materials for high temperature applications. Our studies demonstrate that introducing alicyclic units modulates the formation of short-range ordered structures, which significantly suppress charge transport via electron localization. Experimental and simulation results demonstrate that the semi-aromatic polyimide film exhibits remarkable energy storage performance, with discharge energy density of 6.74 J cm−3 at 200 °C, and 4.45 J cm−3 at 250 °C while maintaining an efficiency of 90%. Furthermore, the semi-aromatic polyimide film exhibits outstanding self-cleaning capability and cycling reliability, enduring over 105 cycles under harsh conditions of 250 °C and 300 MV m−1. This work illustrates a machine learning assisted strategy for developing high-temperature capacitive energy storage.