基于自适应动态规划的售电商长期收益研究

Research on Long-term Revenue of Electricity Retailer Based on Adaptive Dynamic Programming

针对电力现货市场可能出现的电力供需紧张场景下售电商中长期收益最大化问题,提出基于自适应动态规划法(adaptive dynamic programming,ADP)的需求响应动态优化方案。首先以用户负荷的数学期望,建立考虑售电商和用户的需求响应模型,依托用户舒适成本函数的前后联系,构建出售电商中长期动态优化收益函数,利用解析法确定用户参与需求响应的最优响应负荷,并将售电商当前需求响应的收益函数转化为效用函数,采用BP神经网络构建评价网络和执行网络,以离线学习在线使用的框架,根据Bellman最优化原理离线训练评价网络和执行网络。通过7个工业用户的仿真算例,验证所提方法的有效性。

To solve a problem of long-term revenue maximization of the electricity retailers under the condition of probable electricity supply and demand tension in the electricity spot market transactions,this paper proposes a dynamic optimal power demand response scheme based on adaptive dynamic programming(ADP).It firstly establishes a demand response model considering the electricity retailers and users by using the mathematical expectation of the users,and constructs the medium and long term dynamic optimal revenue function of the electricity retailer according to the front and back of the users’comfort cost function.By means of the analytical method,it determine the optimal response load of the users participating in the demand response,and translates the current revenue function of the demand response of the electricity retailer into the utility function.By using the BP neural network,it constructs a critic neural network(CNN)and an action neural network(ANN)for offline studying the online framework.According to Bellman’s principle of optimality,the CNN and ANN are offline trained and evaluated.Finally,simulation studies on seven industrial users validate effectiveness of the proposed method.