基于深度强化学习的绿电交易价格预测研究OA

Aiming to enhance the accuracy of green electricity trading predictions,a green electricity price prediction model based on deep reinforcement learning was proposed by integrating graph struc-ture modeling,temporal feature extraction,and a strategy optimization mechanism.First,a graph conv-olutional network(GCN)was employed to model the structural dependencies among trading entities.Second,gated recurrent units(GRUs)were introduced to extract the dynamic evolution patterns of historical prices and external features.Finally,an improved deep Q-network(DQN)was applied to generate an optimal price prediction strategy under state-aware conditions.Experimental results dem-onstrate that the model outperforms comparative models such as LSTM,GCN-LSTM,and standard DQN across multiple evaluation metrics,achieving an average reduction in MAE of 7.5％and a de-crease in ROI prediction error of 13.2％.Furthermore,the model exhibits strong robustness in scenar-ios involving noisy disturbances and sudden policy changes.