基于微分代数神经网络的含新能源区域电网端口动态特性学习方法

高比例新能源渗透背景下，建立能够准确描述复杂环境因素影响下含新能源的区域电网端口特性动态模型，对于新型电力系统动态分析至关重要。为此，提出了一种基于微分代数神经网络的含新能源区域电网端口动态特性学习方法。该方法利用微分代数神经网络，基于区域电网接入点的时序量测以及光照强度、温度等环境量测数据，学习以神经网络表达的端口特性模型。所得模型由初始状态提取模块、微分神经网络模块、代数神经网络模块组成，可直接接入电力系统暂态仿真器中，用于分析电力系统整体动态特性。在IEEE-39节点系统中对该方法进行仿真验证，测试结果表明：所得模型能够适应不同环境场景，准确率高，验证了方法的有效性。该建模方法仅依赖端口时序量测，在新型电力系统动态分析中具有较大的应用潜力。

Portal Dynamics Learning Method for Renewable-integrated Regional Power Networks Based on Neural Differential-Algebraic Equations

In the context of high penetration of renewables, it is very important for new power system dynamic analysis to establish a dynamic model that can accurately describe the portal dynamics of renewable-integrated regional power networks under the influence of complex environmental factors. Therefore a neural differential-algebraic equations-based portal dynamics learning method is proposed for renewable-integrated regional power networks. In this method, the differential-algebraic neural network is used to learn the portal dynamics model expressed in the form of neural network based on the time series measurements of the access point of the regional power networks and the environmental measurement data such as the radiation intensity and temperature. The learned model is composed of an initial state extracting block, a neural differential equation block and an algebraic equation block, and can be directly integrated into power system transient simulations to analyze the overall dynamics of power systems. The proposed method is tested through simulation in the IEEE-39 system, and the test results show that the obtained model can adapt to different environmental scenarios with acceptable accuracy, which verifies the effectiveness of the proposed method. The modelling method only needs portal time series measurements and has great application potential in the dynamic analysis of new power systems.