This paper focuses on the strategy generation task in high-frequency trading in the electricity market, and explores the application of deep deterministic policy gradient algorithm in continuous bidding decision-making. Based on the information of node marginal electricity price, load deviation, renewable output and offer depth, a transaction dataset consisting of 168,000 market time slices, 42 price variables, six bidding features and 18,400 settlement records is constructed. Through time series feature encoding, market interaction environment modeling and Actor-Critic structure, the continuous bidding actions under revenue and risk constraints are jointly learned. Compared with DQN, PPO and rule-based methods, the proposed framework achieves 18.7% cumulative return rate, 1.64 Sharpe ratio and 9.8% maximum backoff, and the average decision delay remains at 7.8 ms. The results show that DDPG can provide more stable strategy learning ability, more accurate continuous bidding control effect, higher computational execution efficiency and online deployment ability for the power market trading system.
