Abstract: OBJECTIVE To recomend the oral dose of levetiracetam for the treatment of benign childhood epilepsy with centro-temporal spikes and assist doctors in making precise individual therapy based on deep Q network. METHODS The follow-up clinical data of 245 cases of benign childhood epilepsy with centro-temporal spiked from Children's Hospital of Chongqing Medical University from 2016 January 1 to 2021 April 29. Leveraging deep reinforcement learning technology, a deep Q neural network model for drug dosage recommendations of childhood epilepsy was build. The professional physician's prescription was compared with the total daily dosage of levetiracetam recommended by the algorithm. RESULTS In terms of the total recommended daily dose distribution, the distribution recommended by the deep Q network was generally similar to that of professional physicians, and it was more inclined to recommend drug doses that were widely distributed and statistically significant in the data set. The accuracy of the intelligent dose recommendation of levetiracetam for benign epilepsy in children with centro-temporal spikes was compared in the deep Q network dose recommendation. The average accuracy rate of the total dose classification of daily medication was 89.7%, and the average error of the classification of recommended drugs was 0.341 3. CONCLUSION The effectiveness of the drug dose recommendation model based on the deep Q network is preliminarily verified, and it provides a reference for the extension of the algorithm model to more anti-epileptic seizure drug dose recommendations.