Innovation-driven, energy consumption structure optimisation is crucial in developing countries. China, as a typical developing country, has been shifting its economic development model from a traditional approach to innovation-driven, high-quality development. We select nine resource-based regions in China as our research subject and employ a varying coefficient panel model and a Hansen panel threshold model to quantify the marginal and threshold effects of innovation capacity in optimising energy consumption structure. The results indicate that the overall energy consumption structure index for the analysed regions is 0.563, revealing that the structure is rather low and regional economic development relies mainly on low-ranked energy. Astonishingly, the marginal effect reflects obvious heterogeneity: a positive effect in Sichuan (-0.208), Yunnan (-0.207), and Xinjiang (-0.431); nearly no effect in Chongqing, Gansu, and Qinghai; and a negative effect in Guangxi (0.104), Shaanxi (0.244), and Ningxia (1.282). In previous studies, such negative effects have been totally ignored. Further, there are double thresholds and a unique driving mechanism, namely, 'negative driving -> strong positive driving -> weak positive driving'. Interestingly, below the first threshold, improvement in innovation will lead to energy consumption structure degradation. Current policies, such as introducing extensive talent, are not feasible for all resource-based regions. The concerns of policy making for the three different threshold intervals should be: upgrading industrial structures, optimising the structure of human resources, and preventing 'brain drain', respectively.