基于时变回声状态网的光伏发电在线短期预测方法研究

项目名称： 基于时变回声状态网的光伏发电在线短期预测方法研究

项目编号： No.61573072

项目类型： 面上项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 伦淑娴

作者单位： 渤海大学

项目金额： 16万元

中文摘要： 主要研究时变回声状态网和光伏阵列的发电功率计算模型。①提出具有短期记忆的时变回声状态网。时变回声状态网的储备池由神经元和其延迟单元组成。根据输入时间序列自相关特性和预测性能要求, 时变回声状态网能够动态地调整子储备池神经元的短期记忆能力。提出一个改进的小世界网络模型，并利用它动态指导储备池神经单元的稀疏连接，减小随机稀疏连接的盲目性。另外，研究时变回声状态网的参数优化问题。②建立光伏阵列的电流电压特性显式模型，进而建立光伏阵列的瞬时发电功率计算模型。本项目利用时变回声状态网进行发电预测，具有预测精度高和调节时间短特点。本项目对电网实现新能源电力的最大程度消纳具有重要意义。

中文关键词： 在线预测；数据驱动；多目标优化；光伏发电；神经网络

英文摘要： This project studies a time-varying echo state network and a power model of Photovoltaic array. Firstly, This project proposes a time-varying echo state network with abilities of short-term memory and on-line dynamic self-learning. Reservoir of time-varying echo state network is composed of different backbone neuron and its time-delay units. According to autocorrelation characteristics of input time series and prediction performance requirements, time-varying echo state network can dynamically adjust the short-term memory ability. A improved small world network is developed. And the improved small world network is used to guide the sparse connection of reservoir neural nodes, which reduces the blindness of the random sparse connections. In addition, this project carries out the research on parameter optimization. Secondly, a explicit model for current-voltage characteristic is established and then a power model for PV array is developed. This project use the time-varying echo state network model to predict output of photovoltaic power generation systems. This makes the prediction accuracy higher and settling time shorter. Therefore, this project plays a very important role in grid maximally admitting new energy power.

英文关键词： online prediction；data-driven；multi-objective optimization；photovoltaic power generation；neural network