A 5*5cm square electrode sheet will be placed in the subject's left DLPFC brain area (F3 electrode site) and a 5*5cm square electrode sheet will be placed at the right trapezius muscle at the base of the neck, where anodic polar stimulation will be given at the DLPFC and cathodic stimulation will be given at the trapezius muscle at the base of the neck on the right right side. Subjects will be intervened under a drug cue response task. The whole procedure lasts for 20 minutes. The sham intervention will be in the form of a current that ramps up from 0mA to 2mA and then ramps down to 0mA for the first 20 seconds of stimulation onset. after maintaining a current level of 0mA for 19 minutes and 20 seconds, the current will again ramp up from 0mA to 2mA for the last 20 seconds and then ramp down to 0mA.

The personalised EEG classifier model established through machine learning is used to identify the current game cue response EEG pattern state of the subject during intervention and perform tDCS intervention in real time. The real-time EEG data of the true intervention group was collected for 1 second at a time in real time, and then pre-processed, and feature extraction was performed on the real-time EEG data using the temporal clusters in the first part, and then inputted into the EEG classifier model mentioned above, which in turn calculated the similarity between the current state and the previous game cue response-evoked EEG pattern, i.e., whether or not it was currently in a high craving response state. Once in the high craving state and short-term tDCS intervention, the intervention site is the effective target site found above, and a short-term electrical stimulation intervention of 2 seconds and 2mA is given.

The EEG signal of the subject completing the game cue response task is first acquired, the signal is preprocessed, and then the time-domain voltage feature and the time-frequency domain energy feature of the EEG are obtained by comparing the game cue and the neutral cue condition, and the spatiotemporal clusters of the time-domain feature and the time-frequency domain feature under the above thresholds are obtained by the method based on the cluster statistic in the EEG analysis in the adjacent time window and the adjacent electrode position, respectively. Further, a personalised EEG classifier model is established by machine learning.

The above model is used during intervention to identify the current EEG pattern state of the subject's game cue response and perform tDCS intervention in real time. The sham intervention group used the EEG data of the real intervention group for intervention triggering.

In contrast to the active intervention, the sham intervention consisted of a current ramp from 0mA to 2mA for the first 20 seconds of stimulus onset, followed by a ramp down to 0mA. the current was maintained at 0mA for 19 minutes and 20 seconds, and then the current was ramped from 0mA to 2mA for the last 20 seconds, followed by a ramp down to 0mA. The entire procedure lasted the same as the active intervention, which lasted for 20 minutes.

The severity of the subject's gaming addiction is indicated by the score on the questionnaire (DSM-5)，The DSM-5 scale consists of nine items, each of which has a "yes" or "no" choice, with a "yes" choice counting as one point and a "no" choice being indicated by a score from 1 to 7. "The higher the score, the more severe the online game use disorder.

The severity of the subject's gaming addiction is indicated by the score on the questionnaire (IGD-S9)，The IGD-S9 scale consists of nine items with scores ranging from 9 to 45, with higher scores indicating higher levels of addiction.

Brain waves of subjects in resting or task states, include P300 at Pz electrode, point；time frequency of P300 (alpha, theta, beta, delta).

Under the Iowa Gambling task (IGT), participants were analyzed for the percentage of participants choosing each type of card (high-risk but high-reward, low-risk but low-reward), thus comparing their risky decision-making abilities; under the Signal Stopping task (SST), participants were analyzed for their stop-signal reaction time (SSRT), which is the amount of time it takes an individual to successfully inhibit a response that has been initiated, thus assessing response inhibition; and under the Monetary Incentive Delay task (MID), subjects were analyzed for five conditions (large win, large loss, small win, small loss, neutral ) reaction times and accuracy, assessing participants' expectations of rewards and punishments and outcome feedback.