Objective To tackle the problem that the Chinese gold futures prices cannot be accurately predicted due to their inherent characteristics of non-linearity instability and high volatility an interval prediction model based on deep learning is proposed. This model aims to more precisely depict and forecast the changing trends of gold futures prices. Methods First three deep-learning models namely the gated recurrent unit GRU long short-term memory LSTM and bidirectional long short-term memory BiLSTM were employed. Independent interval prediction models were constructed using the kernel density estimation KDE and quantile regression QR methods respectively. Then based on these two types of interval prediction m co o n d d e i l t s io n a t c h o e m g b r i i n d ed sea p r r c e h d ic G t S io n a m lg o o d ri e th l m wa w s as fu u r s th e e d r to in o te p g ti r m at i e z d e . th T e a m ki o n d g el th w e ei p gh re t d c ic o t n i f o i n gur in at t i e o r n va t l o c m o i v n e i r m ag iz e e r t a h t e e a a v s era a ge co w n i s d t t r h ain o t f the prediction interval. Results Empirical analysis showed that the combined interval prediction model constructed using the BiLSTM method significantly outperformed other benchmark models in terms of prediction performance. At various confidence levels it exhibited a wider prediction coverage rate and a narrower average interval width. Conclusion The proposed combined interval prediction model can more effectively predict the future fluctuation range of gold futures prices providing more reliable risk assessment and decision-making support for investors.