To achieve the high-value utilization of Dendrobium officinale leaves, a compound gummy candy with desirable sensory quality and flavor characteristics was developed using Dendrobium officinale leaf juice, apple juice, and pear juice as the main raw materials. A four-factor experimental system was established using mixed juice ratio, pectin content, sucrose content, and citric acid content as independent variables. A genetic algorithm-optimized backpropagation neural network (genetic algorithm-backpropagation, GA-BP) combined with response surface methodology was employed to model and optimize the gummy formulation, while an electronic nose and gas chromatography-ion mobility spectrometry (GC-IMS) were applied to characterize its volatile flavor profile. The results showed that the GA-BP model exhibited superior predictive performance compared with the response surface model (R2 = 0.984). The optimal formulation consisted of a mixed juice ratio of 3∶1∶1 (w/w), 6.0% pectin, 38.0% sucrose, and 0.6% citric acid. Electronic nose analysis indicated that the GA-BP group exhibited a richer and more harmonious flavor profile, with the cumulative contribution rates of PC1 and PC2 reaching 95.28%. GC-IMS identified 53 volatile compounds, among which aldehydes, alcohols, and sulfur-containing compounds were identified as key flavor contributors and were significantly enriched in the GA-BP group, resulting in a fresher and more layered aroma profile. This study provides technical support for the formulation optimization of Dendrobium officinale compound gummy candies and further demonstrates the application potential of Dendrobium officinale leaves in functional food development.