【作者】 郑超；

【导师】 王少荣；

【作者基本信息】 华中科技大学 ， 电气工程， 2020， 博士

【摘要】 随着可再生能源的广泛接入、电力电子装备的大量应用、新型负荷的不断涌现以及多种异质能源网络的耦合互联,现代电力系统的网络结构和运行特性已经并将持续发生深刻变革。这些变革,既对电力系统中原有技术方法提出了升级和更新的迫切需求,又产生了一批极具复杂度和求解难度的电力系统新问题。因此,当前电力系统面临着如何求解这些复杂新问题的挑战。与此同时,高速网络通信、云计算、人工智能、机器人和无人机等先进技术的飞速发展以及推广运用,使现代电力系统具有数据资源丰富、计算能力强大和信息交互手段便捷等新特性。这些新特性,不但为电力系统中原有技术方法的升级和更新提供了强力的支持,还为研究电力系统复杂问题的求解新方法提供了前所未有的新机遇。鉴于上述背景,本文以抓住机遇解决问题为指导思想,着眼于现代电力系统复杂问题的前沿,紧密结合当前电力系统运行维护方面的迫切需求,充分利用当前电力系统的数据资源,沿着构建模型所基于的有序离散数据集合在时空维度拓展的主线,探索研究现代电力系统新的复杂问题的非线性映射模型的构建方法。本文的主要研究内容如下:(1)针对在非线性映射模型的结构确定和参数求解中存在的不适定问题,根据有序离散数据集合中,或隐含了系统动力学特性,或体现了某种几何特性的事实,提出了基于有序离散数据集合的空间曲线特征正则化方法。该方法仅涉及有序离散数据集合的空间曲线自身的几何特性,能够在对应物理系统特性未知的情况下使用。因而,其具有比较广泛的适用性,尤其适用于数据驱动的建模方法中。(2)在涉及基于单一时间断面有序离散数据集合的非线性映射模型方面,针对变电站指针式仪表示数识别问题,提出了一种指针式仪表图像示数自动识别非线性映射模型。该模型在图像预处理过程中,通过综合地运用带色彩恢复的多尺度Retinex算法和透视变换算法,成功地增强了模型光照度和拍摄倾斜角度的鲁棒性;在仪表示数识别过程中,综合梯度思想和表决法,提出了一种快速可靠的改进Hough圆检测方法,有效地提升了模型的自动识别速度。为电力系统智能运维中的指针式仪表示数自动识别提供了一种鲁棒性强、识别速度快及可靠性高的非线性映射模型。(3)在涉及基于单一时空维度(时间维度)的有序离散数据集合的非线性映射模型方面,针对区域有源配电网动态等值问题,提出了一种基于有记忆人工神经网络的区域有源配电网动态等值非线性映射模型。基于将神经网络映射关系和电路系统的物理特性相结合的方式,在模型结构建立方面,给出用于有源配电网动态等值模型的LSTM深层神经网络的设计准则;在模型参数确定方面,建立了基于曲线曲率特征正则化的目标函数,以及相应的神经网络训练算法。为日趋复杂的区域有源配电网动态等值模型构建提供了一种全新的思路和方法。(4)在涉及基于两个时空维度(时间维度和一个空间维度)的有序离散数据集合的非线性映射模型方面,针对气-电联合系统运行优化问题,提出了一种基于非线性函数空间映射的气-电联合系统运行优化模型。模型建立的过程中,基于函数逼近论,在函数空间中,分别详细地推导了与代数空间微分和积分相对应的操作算子;基于代数空间中的气-电联合系统运行优化模型和函数空间映射,建立了气-电联合系统运行优化的函数空间模型。为考虑天然气系统动态特性的气-电联合系统运行优化模型的构建提供了一种全新的思路和方法。更多还原

【Abstract】 With the wide access of renewable energy,the massive application of power electronics,the continuous development of new loads,and the coupled integration of multi-energy systems,the network structure and operating characteristics of the modern power systems have been deeply changed.The changes not only trigger the urgent need for upgrading and updating the existed technology and methods,but also bring various new problems that are extremely complex and difficult to be solved in the power systems.Therefore,the power systems are facing the challenge of how to address the new complex problems today.At the same time,due to the rapid development and wide application of the advanced technologies,e.g.high-speed networks,cloud computing,artificial intelligence,robots,and drones,there are several new features,such as rich data resources,powerful computing capabilities,and convenient communication ways,in the modern power systems.These new features not only strongly support the upgrade and update of the existed technology and methods,but also bring new opportunities to study the novel methods to address the new complex problems in the power system.In the above background,for studying on the new complex problems in the modern power systems,this thesis regards the way using the opportunity to address problems as a guide,stands on the advances of complex problems of modern power systems,concentrates on the urgent needs from the operation and maintenance,makes full use of the data resources,selects the dimension feature of the sequential data sets in the spatio-temporal coordinate as the clue,and explores the novel nonlinear mapping modeling methods.The main contributions of this thesis are as follow:For the ill-posed problem in the structure construction and parameters estimation of the nonlinear mapping models,since the sequential data sets either contain the dynamic characteristics of the system or reflect a certain geometric characteristic,a regularization method based on the curve geometric characteristics of sequential data sets is proposed.The proposed method only involves the geometric characteristics of the sequential data set,and it could be adopted without knowing the characteristics of the physical system,therefore,it can be widely adopted,especially in data-driven modeling.On the topic of the nonlinear mapping modeling based on the sequential data sets from the snapshot,the study focuses on the reading recognition of analog instruments in substations,a novel nonlinear mapping model of the automatic reading recognition of an analog instrument is proposed.In the image pre-processing,the model has successfully enhanced the robustness of lightness and camera angle by synthetical using the multi-scale Retinex with color restoration and perspective transform.In the reading recognition procedure,according to the gradient approach and voting methods,a fast and reliable improved Hough circle detection method is presented,which effectively improved the automatic recognition speed of the nonlinear mapping model.The proposed model is with strong robustness,fast recognition speed,and high reliability,and can be easily adapted for automatic reading recognition of the analog instruments in the smart operation and maintenance.On the topic of the nonlinear mapping modeling based on the sequential data sets from the temporal dimension in the spatio-temporal coordinate,the study focuses on the dynamic equivalent model of the active distribution networks(ADNs),a dynamic equivalent nonlinear mapping model of ADNs is proposed based on the artificial neural networks(ANN)which have memory capability.For the construction of the model structure,based on the way of combining the mapping relationship of ANN with the physical characteristics of the circuit system,the design guidelines of the LSTM for the dynamic equivalent model of ADNs are given.For the estimation of model parameters,a regularized objective function based on the curvature of the sequential data sets,and the corresponding training algorithm of LSTM are established.The proposed modeling approach provides a novel idea and method for dynamic equivalent modeling of the complicated ADNs.On the topic of the nonlinear mapping modeling based on the sequential data sets from the temporal dimension and a spatial dimension in the spatio-temporal coordinate,the study focuses on the optimal operation of the integrated power and gas energy systems(IPGES),an optimization model for operating IPGES is proposed based on the nonlinear function-space mapping.In the modeling process,according to the function approximation theory,the operation matrices,which are related to the integration and differentiation in the algebraic space,are developed in the function space.Then,based on the optimal energy flow model of IPGES and function-space mapping,an optimization model for operating IPGES is constructed in the function space.The proposed modeling approach gives a novel idea for modeling the optimized operation of IPGES with considering the dynamic characteristics of natural gas systems.更多还原