基于深度学习的微网需求响应特性封装与配电网优化运行

Deep Learning Based Characteristic Packaging of Demand Response for Microgrids and Optimal Operation of Distribution Network

微网作为新型需求响应资源,并网规模日益增加,但由于微网内部资源组合的不确定性以及源荷随机出力波动性,其参与电网运行的需求响应特性也呈现较大不明朗特征,极大地增加了配电网运行风险。为此,提出一种基于深度学习的微网需求响应特性封装与配电网优化运行新机制,采用数据驱动的方式对微网需求响应特性封装,避免对微网内部模型的解析,充分利用气象信息、价格数据建立微网长短期记忆(LSTM)多维时序需求响应封装模型,在此基础上构造配电网运行优化模型,并提出了基于改进的粒子群优化(IPSO)算法模型求解策略,以最大限度降低配电网电压越限风险与网络损耗。通过含微网群的33节点配电网系统算例进行分析,结果表明了所提微网需求响应封装模型的有效性以及IPSO算法的优越性。

As a new type of demand response resource, the grid connection scale of microgrids is increasing. However, due to the uncertainty of the internal resource combination and the random output fluctuation of the sources and loads in microgrid, the demand response characteristics of its participation in the operation of the power grid also show great uncertainties so that the operation risk of the distribution network is greatly increased. In this regard, this paper proposes a new mechanism for the demand response characteristic package of microgrids based on deep learning and optimal operation of the distribution network. It can package the demand response characteristics of microgrids by a data-driven approach to avoid the analysis of the internal model of microgrids. With fully use of the weather information and price data, a multi-dimensional sequential demand response package model of microgrids based on long short-term memory(LSTM) is established. On this basis, an optimization operation model of distribution network is constructed, and a model solution strategy based on the improved particle swarm optimization(IPSO)algorithm is proposed to minimize the risk of the voltage over-limit and network loss in distribution networks. Through the analysis of the 33-node distribution system with microgrid group, the results show the effectiveness of the proposed demand response package model of microgrids and the superiority of the IPSO algorithm.