The study proposes a hybrid TFFBN-HDLF deep learning framework to detect epileptic seizures in the elderly based on EEG signals. The framework solves the problems of slow background activities and non-stationary dynamics of brain signals in seniors. The TFFBNC method creates a TFPPNet time-frequency functional brain network by combining the Pearson correlation coefficient and the phase lag index. The SeizureTransNet architecture combines convolutional neural networks with Transformer modules to extract spatiotemporal features. On the CHB-MIT dataset, the framework achieved an accuracy of 98.09% (AUC 99.45%). On the Siena set, the accuracy was 92.49% (AUC 95.64%). These results confirm the higher reliability of seizure monitoring in elderly patients.