Fig. 2: Online data acquisition and dataset analysis.

a EV cloud data flow. Vehicle status data and battery status data are collected by multiple sensors and uploaded to the T-box. These data are then transmitted to the vehicle manufacturer’s basic Telematics Service Provider (TSP) platform via Transmission Control Protocol (TCP)/Internet Protocol (IP), Kafka, and Hypertext Transfer Protocol (HTTP)/HTTPS protocols. It is subsequently sent to the data middle-end platform, where operations such as triage (Kafka), caching (Redis), pre-cleaning (Spark), and archiving (HBase) are performed. The data then flows into a cloud database for visualization (Cloudera) and storage (MySQL/Clickhouse). The proposed network retrieves the resulting data from the cloud database, outputs the results, and interacts with the front-end server through the corresponding API gateway. b Data frame range and vehicle number distribution. This dataset, which includes normal and faulty samples, is derived from three manufacturers: DTI, QAS, and GIS. The graph shows the range of data frames and the distribution of the corresponding number of vehicles. c Statistics on the sample count of different fault types in the database. The four digits of the fault type number represent TR, EL, ISC, and EA, with 1 indicating the presence of a fault. d Network input sample sequences. These include voltage, temperature, and current with on-board SOC for the battery state, and speed and mileage for the vehicle state. The horizontal axis represents the number of sampled frames at 30-s intervals. e Battery state data distribution comparison. This comparison is between normal and faulty samples, including battery voltage, current, and temperature. The samples are taken from a random set of 100,000 data points from 10 normal/faulty vehicles. Source data are provided as a Source Data file.