The world’s need for green energy is something that cannot be postponed any longer, where the transmission-distribution process requires power distribution in DC voltage. However, currently, the majority use AC voltage, so limited experience and lack of data regarding electrical cable aging under high voltage (HVDC) and their reliability are problems that must be resolved. Crosslinked polyethylene (XLPE) constitutes many insulation cables used today, so estimating the lifetime of DC XLPE cable insulation is urgent research, even though various model-optimization improvements are needed to obtain accurate results. This research begins with pre-processing for the input and output data. These results were then analyzed using two improved model schemes to accommodate the addition of variable space charge and thickness: backpropagation neural network (BPNN) and hybrid BPNN with inverse power model (BPNN-IPM). The learning process uses gradient descent (GD), genetic algorithm (GA), and Levenberg-Marquardt (LM) optimization methods. Finally, the proposed method was verified using experimental data from previous research. The results show that the hybrid BPNN-IPM with LM optimization method is the most accurate: training root mean square error (RMSE) achieved 0 days, and testing RMSE achieved 0.83 days. These results show that the method BPNN-IPM-LM used is most accurate in estimating the lifetime of DC XLPE insulation.