The invention provides a neural-network-based cable hybrid line double-terminal location method, comprising: acquiring training data; extracting training data meeting first preset conditions from the training data, calculating the extracted data in a double-terminal location mode to obtain a first location result formed between an actual fault point and a first busbar, extracting training data meeting second preset conditions from the training data, and calculating the extracted data in a preset neural network simulation location mode to obtain a second location result formed between the actual fault point and the first busbar; judging whether absolute value of a difference between the first location result and the second location result is less than a preset threshold or not; if yes, outputting the first location result as a final result for cable hybrid line fault location; if not, outputting the second location result as a final result for cable hybrid line fault location. The neural-network-based cable hybrid line double-terminal location method according to the invention is applicable to a complex power transmission line with multiple overhead cable lines alternated, can provide location results simply and reliably.