The battery system is vital for the safety and durability of a real-world electric vehicle (EV), and the prognosis of battery thermal runaway trigged by various abuse conditions is critical for preventing security incidents, such as spontaneous combustion or explosion. This article presents a real-scenario-based thermal runaway prognosis on a Li(NiCoMn)O ₂ ternary battery in an actual electric bus, deriving historical operation data from the National Monitoring and Management Platform for Electric Vehicles of China. The multiscale detailed features of the fault signals before the accident are extracted using the modified multiscale entropy algorithm, avoiding entropy fluctuations and information redundancy due to the sliding averaging process, as well as overcoming the shortcomings of the single-scale coarse-graining and dimensional disaster. The results show that abnormal cells can be detected as early as one week before the accident. Furthermore, a real-time multilevel prognosis strategy based on the determined anomaly coefficients is proposed, which can highlight the faulty cells and improve the prognosis sensitivity by filtering out the redundant scales. The verification results on the accident vehicle show that the early abnormal cells can be effectively diagnosed, preventing the occurrence of thermal runaway and safeguarding the safety of drivers and passengers in real-world vehicular operation.