A concept model of regional risk was constructed for the characteristics of ecosystems alongside the Qinghai-Tibet highway and railway based on the MLP (Multilayer perceptron) model. Seven indices such as snow hazard, drought hazard, and landslide were selected in order to evaluate the integrated ecological risk of the ecosystems along the study area. Results show that the Qaidam montane desert zone had the greatest average risk value (4.26), followed by the Golog-Nagqu high-cold scrub meadow zone (2.80) and the East Qinghai and Qilian montane steppe zone (2.73) among the ecosystems within the six natural zones within the study region. As far as land cover types are concerned, the top three ecological risk values appear in the needle-leaved forest (4.31), desert (4.12), and land without vegetation (3.62), which are higher than those in the other seven types in the study site. Although the risk values are influenced by natural factors and human activities, they are more strongly controlled by natural factors. According to the ecological risk characteristics, the ecosystems within the study area are subdivided into four subregions, including the Qaidam basin region (high risk), the Xidatan to Damxung region (moderate risk), and the Eastern Qinghai-Qilian (slight risk) and Southern Xizang (Tibet) region (slighter risk).

On the basis of Bayesian network, evidence theory is introduced, a kind of distributed power network fault diagnosis method based on Naive Bayesian network and D-S evidence theory is proposed. Firstly, the fault area is determined by the real-time connection method, the fault region is segmented by the butterfly segmentation method. Secondly, according to the historical fault samples, the decision table is established, and knowledge reduction based on Rough Set Theory. The naive Bayesian network model is constructed according the best combination, and the probability of each node is trained. Finally, D-S evidence fusion is performed on the fault component diagnosis information in the overlap between the subnets. The simulation results show that the proposed method can reduce the complexity of modeling and improve the fault tolerance of the system in the condition of incomplete information, and has good diagnosis results.