考虑预测不确定性的微电网实时控制策略研究

风能、光伏等可再生能源的高比例并网成为了缓解全球能源危机的一项重要措施。然而，可再生能源实时出力中的间歇性和波动性给系统的安全性带来了一定的挑战。为了在提高可再生能源利用率的同时保证系统安全性，提出了一种基于深度强化学习(DRL)算法的运行优化实时调度模型。首先，构建了负荷预测模型实现负荷预测和高斯混合模型拟合预测误差；其次，考虑系统各节点的约束条件，以系统运行成本和安全运行作为优化目标，建立相应优化模型；然后，将优化问题转化为马尔可夫决策过程，并采用双延迟深度确定性策略梯度算法求解；最后，利用DRL算法的环境交互机制和策略自由探索，获得联合调度策略的最优结果。实验结果表明，所提方法具有良好的适应性，并且可以进行在线实时调度。

Real time Control Strategy for Microgrid with Prediction Uncertainty

The high-ratio of renewable energy sources such as wind and PV integrated with power grid has emerged as an essential initiative to mitigate global energy crisis. However, the intermittent and volatility of renewable energy sources present certain challenges to the reliability of the system. A real-time scheduling model for operation optimization based on deep reinforcement learning (DRL) algorithm is introduced to enhance the utilization of renewable energy while guaranteeing system security. Firstly, a load forecasting model is constructed to achieve load forecasting and Gaussian mixture model which is applied to fit forecasting errors. Secondly, considering the constraints of each node of the system, taking system operating cost and safe operation as optimization target, the appropriate optimization model is formulated. And then, the optimization issue is converted into a Markov decision process and addressed by the twin delayed deep deterministic policy gradient (TD3) algorithm. Finally, the optimal joint scheduling strategy is acquired by environment interactive mechanism and policy discretionary exploration of DRL algorithm. The experimental results demonstrate that the proposed method has excellent adaptability and allows for online real-time scheduling.