Low-voltage cables are extensively utilized in residential, commercial, and industrial fields, often arranged in clusters. However, inadequate heat dissipation and high load currents can result in high operating temperatures, accelerating insulation aging and leading to fire hazards. Traditional temperature measurement methods suffer from shortcomings, including contact tests, high cost, uncovered areas, and so on. To overcome these limitations, this article proposes a spatial temperature measurement method based on the electronic nose. The method involves capturing the responses of gas sensors to the decomposition gases emitted by the overheated cables. The response data are then processed using a transformer neural network to detect the temperature rise in the cable. An effective pretraining and fine-tuning paradigm is also employed to optimize the transformer network, enabling the algorithm to resist the influences of the external environment and the sensor drift. The prediction accuracy of this processing algorithm surpassed that of commonly used neural networks and traditional calibration methods. Throughout both the five-month timespan experiment and the high-current overheating experiment, the proposed method consistently demonstrated high-temperature prediction accuracy, thereby affirming its exceptional generalization performance.