Research on New Energy Power Generation Prediction and Control Based on Optimized DDPG Algorithm

The high volatility and uncertainty of new energy power generation pose challenges to grid scheduling. This paper employs an optimized DDPG deep reinforcement learning algorithm to build a prediction and control model. The prediction model combines bidirectional LSTM with an attention mechanism, improving average prediction accuracy by 25.6%. The control model adopts a multi-level decision framework, increasing the new energy consumption rate by 11.3 percentage points and reducing system regulation costs by 18.7%. Experimental verification demonstrates that this method offers significant advantages in handling new energy volatility, providing a reliable technical solution for high-proportion new energy grid-connected operation.