随着电力系统的发展，电力系统中数据的规模快速增长，数据类型也日益复杂。海量的数据能够支撑电力系统的安全稳定运行，但如何高效地使用这些数据，是目前需要重点突破的研究内容。电力系统数据形态比较复杂，按时间维度划分，可分为暂态和稳态数据，其来源主要是电力系统数据采集终端、设备及线路监控系统、新能源控制系统等。采集设备精度不够、传输干扰、通信噪声等原因，所采集电力数据，必须经过数据清洗、质量评估等处理，获得可靠数据集。考虑到电力数据存在时序性等特点，对数据内在规律的挖掘与学习对数据的认知也至关重要。在此基础上，利用各种人工智能技术等手段，针对不同场景的实际需求，将大数据技术真正应用于电力系统。因此，本文从两个方面开展工作，第一，对所采集数据的时序特征进行挖掘，提出数据质量评估方法；第二，将电力系统大数据应用于无功补偿，利用大数据及人工智能手段，实现电力系统的无功功率优化。本文具体的工作内容如下：1、基于电力系统中所采集的数据，分析电力数据的时序特性，在此基础上，提出一种基于ANN预测的电力供应质量数据噪声等级评估方法，模拟历史噪声数据与相应原始数据之间的关系，进一步进行噪声水平估计。该算法在训练模型时只需要知道历史数据的噪声振幅，且可用于大多数噪声类型。通过仿真实验分析了算法性能的影响因素，验证了该算法的效果。2、为了充分利用新型电力系统的大数据和领域知识，基于图卷积神经网络和深度Q网络，提出了一种新型电力系统故障及扰动状态下的电压稳定控制方法。该方法利用新型电力系统拓扑，使用图卷积神经网络对系统节点及其相邻节点的实时数据进行特征提取和融合，然后使用深度Q学习给出无功补偿策略。仿真实验研究表明，该方法可以在新型电力系统发生电压暂降期间实现良好的电压稳定性控制和准确的无功补偿。
With the development of power system, the scale of data in power system is growing rapidly, and the data types are becoming increasingly complex. Massive data can support the safe and stable operation of power system, but how to use these data efficiently is the research content that needs to be broken through. The composition of power system data is complex, which can be divided into transient and steady-state data according to time dimension.These data are mainly collected from power system data acquisition terminal,data monitoring system,wide area measurement system based on vector measurement unit,etc.Due to measurement error,transmission interference,communication noise and other reasons,the collected power data must be processed by data cleaning, quality evaluation,etc.to obtain a reliable data set.The data includes steady-state data and transient data. Due to measurement errors, system disturbances and other reasons, the collected data often have noise, so the data quality needs to be processed and evaluated first. Considering the time sequence and other characteristics of power data, the mining and learning of the internal laws of data is also crucial to the cognition of data. On this basis, various big data means such as machine learning need to be used to apply big data technology to the power system according to the actual needs of different scenarios. Therefore, this paper carries out work from two aspects: first, mining the temporal characteristics of the collected data, and on this basis, proposing a data quality evaluation method; Second, apply big data of power system to voltage stability control scenario, and use big data and artificial intelligence means to realize reactive power optimization of power system. The specific work of this paper is as follows:1. Based on the data collected in the power system, the timing characteristics of power data are analyzed. On this basis, a noise level estimation algorithm based on artificial neural network prediction is proposed to simulate the relationship between historical noise data and corresponding original data, and further estimate the noise level. The algorithm only needs to know the noise amplitude of historical data when training the model, and can be used for most noise types. Through simulation experiments, the factors affecting the performance of the algorithm are analyzed, and the effect of the algorithm is verified.2. To make full use of the big data and domain knowledge of the new power system, based on graph convolution neural network (GCN) and deep Q-learning, a new voltage stability control method under fault and disturbance of power system is proposed. Based on the topology of the new power system, this method uses GCN to extract and fuse the real-time data of power nodes and their adjacent nodes. Then deep Q-learning is used to obtain reactive power compensation strategy. Simulation results show that this method can achieve good voltage stability control and accurate reactive power compensation during voltage sag in the new power system.
