基于改进DDPG算法的中短期光伏发电功率预测

Study on Short-to-Medium-Term Photovoltaic Power Generation Forecasting Model Based on Improved Deep Deterministic Policy Gradient

摘要:

针对传统仿生智能算法处理异构光伏发电功率预测精确建模问题时存在的线路多阻抗参数约束下方差波动、线损分析易陷入局部极值等不足,提出了一种基于改进深度确定性策略梯度(DDPG)的中短期光伏发电功率预测模型。首先,通过引入多智能体机制,视发电系统涉及到的发电过程参数为独立活性的智能体,构建出具有社会属性的面向发电过程参数信息共享的全局最优协同控制体系。然后,通过改进的DDPG算法实现蓄电池储能功率自主精确调节和发电网输出功率的自动最优预测。最后,基于Tensorflow开源框架在Gym torcs环境下进行模型效能仿真并以某示范性异构光伏发电网为效能评价载体,对模型进行了工程应用合理性验证。

Abstract:

Aiming at the congenital shortcomings of traditional multi-resistance intelligent algorithms to deal with the problems of accurate modeling of heterogeneous photovoltaic power forecasting, such as the line's multi-impedance parameter constraints, lower fluctuations, and line loss analysis easily falling into local extremes, a prediction model for short-to-medium-term photovoltaic power generation is proposed based on improved depth deterministic policy gradient (DDPG). Firstly, by introducing multi-agent mechanism and considering the parameters involved in the power generation system as independent active agents, constructed is a global optimal collaborative control system oriented to power generation process parameter information sharing with social attributes. Then, the battery energy storage power can be adjusted independently and accurately and the power grid output power can be automatically and optimally predicted by using the improved DDPG algorithm. Finally, based on the Tensorflow open source framework, the model efficiency was simulated under the Gym torcs environment, and a model heterogeneous photovoltaic power generation network was used as the performance evaluation carrier to verify the rationality the model.