摘 要

随着社会的发展和科技的进步，尤其是信息传输高效化、电力计量仪表智能化、发电输出设备自动化以及电力大数据信息系统等技术的普及与应用，使得电力实时交易成为可能，电力期货市场也应运而生。超短期电力负荷预测是通过将天气、社会、经济等因素充分考虑进来并对未来短期电力负荷进行预测的一种模式。超短期电力负荷预测是实现电力期货交易的重要前提，同时也是电力系统的经济稳定的调度的基础。电网负荷预测的组成包括多个部分，精准的、及时的负荷预测是构建稳定的电网负荷预测的关键所在。超短期电力负荷预测是解决精准电网预测的有力方法。同时，准确的负荷预测也将提高发电设备的综合利用效率。因此，精准的超短期电力负荷预测不仅可以为电力系统的安全、经济、高效运行提供保障，而且可以为调节电力市场供求关系提供依据。

当下，电力期货的市场蓬勃发展，市场已经不满足单纯的精准的负荷预测，为了能够更好的发展电力期货市场，兼具实时、稳定、可靠、智能的新型电力负荷预测方法成为当下研究主流。因此，为保障现代电力系统的经济运行和高效管理，负荷预测已经成为一个重要的研究领域。超短期电力负荷预测研究的历史研究比较丰富，诸多模型也已成熟且大多投入运营。但随着影响负荷变化的因素日益增多，负荷变化曲线随机性强以及传统的预测方法存在诸多局限性，原有的模型在处理速度以及精确度上已无法满足电力需求。针对上述问题，本文的主要内容如下：

第一，分析了新一轮电力改革影响下我国发电行业的发展情况，详细地阐述了电力期货市场的相关概念，同时分析了国内外现有电力期货市场的特点。总结了现有负荷预测模型的特点，分析了不同模型的优点和存在的问题以及当下各类模型的研究进展。

第二，本文基于Python平台提出了一种粒子群算法（PSO）优化RBF神经网络的超短期电力负荷预测方法，通过使用K-Means对数据进行聚类处理，引入PSO算法优化RBF神经网络的正则化参数Gamma和内核参数增量num_neurons，通过建立优化模型进行超短期电力负荷预测。

第三，根据获取的南方电网深圳某售电公司数据，将前42天1104个数据点作为历史数据，后5天120个数据点作为待预测数据。充分考虑天气信息、历史负荷数据、日类型等信息，设置8个输入点，通过选择高斯函数作为径向基函数进行训练。通过实验分析，前文建立的优化模型运行后均方根误差(RMSE)保持在6.5%左右。

关键词：电力大数据、电力期货市场、超短期电力负荷预测、粒子群算法(PSO)、RBF神经网络、K-Means

Abstract

With the development of society and the progress of science and technology, especially the popularization and application of technologies such as high efficiency of information transmission, intelligence of power metering instruments, automation of power output equipment and big data information system of electric power, this makes real-time trading of electric power possible, and the electric power futures market also arises at the historic moment. Ultra-short-term power load forecasting is a method of forecasting short-term power load in the future by fully considering weather, social, economic and other factors. Ultra-short-term power load forecasting is not only an important prerequisite for the realization of power futures trading, but also the basis of economic and stable dispatching of power system. The composition of power grid load forecasting includes many parts, and accurate and timely load forecasting is the key to build stable power grid load forecasting. Ultra-short-term power load forecasting is a powerful method to solve the problem of accurate power grid forecasting. At the same time, accurate load forecasting will also improve the comprehensive utilization efficiency of power generation equipment. Therefore, accurate ultra-short-term power load forecasting can not only provide a guarantee for the safe, economic and efficient operation of the power system, but also provide a basis for regulating the relationship between supply and demand in the power market.

At present, the power futures market is booming, and the market can no longer meet the simple and accurate load forecasting. In order to better develop the power futures market, the new power load forecasting method which is real-time, stable, reliable and intelligent is the mainstream of the current research. Therefore, in order to ensure the economic operation and efficient management of modern power system, load forecasting has become an important research field. The historical research of ultra-short-term load forecasting is rich, and many models have been mature and most of them have been put into operation. However, with the increasing number of factors affecting the load change, the strong randomness of the load change curve and many limitations of the traditional forecasting methods, the original model has been unable to meet the power demand in terms of processing speed and accuracy. In view of the above problems, the main contents of this paper are as follows:

First, it analyzes the development of China's power generation industry under the influence of the new round of power reform, expounds the relevant concepts of the power futures market in detail, and analyzes the characteristics of the existing power futures market at home and abroad. This paper summarizes the characteristics of the existing load forecasting models, and analyzes the advantages and existing problems of different models, as well as the current research progress of all kinds of models.

Secondly, based on the Python platform, this paper proposes an ultra-short-term power load forecasting method based on particle swarm optimization (PSO) neural network. By clustering the data using K-Means, and introducing PSO algorithm to optimize the regularization parameters and kernel parameter increment of RBF neural network, the ultra-short-term power load forecasting is carried out by establishing an optimization model.

Third, according to the data of a power sales company in Shenzhen Southern Power Grid, 1104 data points in the first 46 days are taken as historical data, and 120 data points in the next 5 days are taken as data to be predicted. Fully considering the weather information, historical load data, daily type and other information, eight input points are set, and the Gaussian function is selected as the radial basis function for training. Through experimental analysis, the root mean square error ((RMSE)) of the optimization model established above is kept at about 6.5% after operation.

Keywords: big data of electricity, electricity futures market, ultra short-term power load forecasting, particle swarm algorithm (PSO), RBF neural network, K-Means

目 录

摘 要 3

Abstract 5

第一章 绪论 9

1.1研究背景 9

1.2电力系统负荷预测的发展 13

1.2.1传统负荷预测方法 13

1.2.2现代人工智能预测方法 13

1）人工神经网络 13

2）组合模型预测法 14

3）小波分析法 14

4）极限学习机 14