Brain functional networks play a crucial role in the early diagnosis of neurological or encephaloid diseases， and the estimation of a high-quality brain functional network is one of the most critical challenges. Although numerous brain functional network estimation methods have been proposed， most of them only focus on the correlations among brain regions while ignoring potential dependencies among time points. Recent studies have found that encoding dependencies among time points can effectively improve the discriminative properties of brain functional networks； however， this method only relies on the dependencies between adjacent time points and fails to effectively utilize information from non-adjacent time points， thereby inadequately capturing the temporal characteristics of the brain functional network. To address this limitation， a new brain functional network estimation method was proposed， which introduced a similarity matrix to encode the dependencies among non-adjacent time points， aiming to improve the quality of the estimation. Additionally， an alternating optimization learning algorithm was designed to solve the model quickly. To evaluate the effectiveness of the proposed method， experiments were conducted on three public datasets — ADNI （Alzheimer's Disease Neuroimaging Initiative）， ABIDE （Autism Brain Imaging Data Exchange）， and REST-MDD （REST-meta-MDD Consortium） — for mild cognitive impairment， autism and depression， respectively. Experimental results demonstrate that the brain functional network estimated by the proposed method achieves superior classification performance.