【作者】 李眉眉；

【导师】 丁晶；

【作者基本信息】 四川大学 ， 水文学及水资源， 2004， 博士

【摘要】 水资源的综合利用和开发与电力负荷预测有着密切的关系，负荷预测的数据作为水资源开发、优化配置、水库调度的重要依据。电力负荷预测对电力系统的安全经济运行也起着十分重要的作用。实践中取得的负荷时间序列包含了诸多影响因素，使得电力负荷表现出复杂性、不确定性、非线性的特点。混沌看作是确定性的非线性系统出现的具有内在随机性的解。混沌时间序列中蕴涵有丰富的动力学信息，研究如何提取这些信息并应用到实际中，是非常有意义的一项工作。本文基于混沌理论，对电力系统负荷演变的混沌变化规律进行分析，在此基础上，研究负荷相空间的非线性预测方案，研究取得了如下具有价值的成果。 从重构相空间理论出发，探讨了相空间参数对重构空间质量的影响，以及确定相空间嵌入维数和延迟时间各种不同的方法。本文对时间序列混沌特性的识别方法，及混沌相空间预测模型进行了详细讨论，为电力系统负荷的混沌分析奠定了基础。 对不同时间尺度的电力负荷：小时负荷、日负荷、月负荷时间序列进行混沌性识别。充分提取电力负荷时间序列蕴涵的混沌特征量：饱和关联维数、Lyapunov指数、Kolmogorov熵，从定量的角度分析电力负荷时间序列的混沌特性。同时，对电力负荷相空间的奇怪吸引子的分形维数也进行了探讨，给出了相空间图。这些研究为进一步电力负荷的混沌相空间预测工作提供保障。 利用电力负荷相轨道的特点，研究混沌相空间的相似点模型、线性回归模型及Lyapunov指数模型对电力负荷的短期预测。实例中分析了相空间嵌入维数和预见期的不同对预测效果的影响。研究表明，几种相空间预测模型对电力负荷短期预测是有效的。 将混沌分析方法成功用于电力负荷多时间尺度的分析中。通过计算长时间尺度的年负荷分解到短时间尺度的月负荷的分解系数，寻找出其混沌变化的特性。在发现分解系数具有混沌性质的基础上，用相空间混沌预测方法进行预测，从而进行降尺度计算分析。 本文将人工神经网络和卡尔曼滤波技术引入到混沌相空间中，提出了基于混沌分析的BP神经网络模型以及混沌相空间的卡尔曼滤波模型。文中详细描述了建模的原理、预测的过程，最后将两个祸合模型用于电力日负荷时间序列的短期预测中，实例应用表明祸合模型的预报精度较高，预报效果是令人满意的。 综上，本文所开展的工作主要在电力负荷的混沌特性分析及其相空间的非线性预测方面，在混沌方法与其他新技术的结合方面做出了探索性研究。该项研究不仅为负荷预测提供了可行的实用方法，而且为负荷预测的进一步研究提供新思路，同时也为其他水文变量的研究工作提供参考方案。关键词:水资源优化利用电力负荷预测混沌理论相空间重构降尺度人工神经网络卡尔曼滤波更多还原

【Abstract】 The utilization of water resources is relative to electric power load forecasting, which could provide useful data for sustainable utilization of water resources, optimum allocation and reservoir dispatch. Electric power load forecasting also plays a very important role in the safe and economic operation of power system. Electric power load is influenced by many factors. So its behavior appears as the characteristics of complexity, uncertainty and non-linearity. Chaos is looked as the solution with internal stochastic property in the nonlinear deterministic systems. It’s very significant work to research on how to obtain and use colorful dynamical information hidden in chaotic time series. Based on the chaos theory, chaotic characteristic of power load time series is analyzed and its forecasting methods are studied in this paper. Some research achievements have been obtained as following.The influence of phase space parameters on phase space quality and the methods for determining delay time & embedding dimension are discussed on the reconstruction theory. Then identification and prediction approaches for chaotic time series are sum up.For electric power load of different time scale including hourly load, daily load and monthly load, quantitative calculation about saturation correlation dimension, maximum Lyapunov exponent and Kolmogorov entropy of power load is used to identify their chaotic characteristics, and conclude that power load time series belong to chaotic series. Moreover, the fractal dimensions of strange attractor in load phase space are estimated, and their phase diagrams are presented.Further work is studying short term load forecasting using neighbor model, linear regression model and Lyapunov exponent model in the phase spaces. At the same time the influence about different embedding dimensions and prediction time on forecasting result is considered. The prediction results indicated that the chaotic phase space model is effective for short term load forecasting.A new chaotic method is successful used for multiple time scale analysis of electric power load. After researching the chaotic characteristic of decomposition coefficient that annual load with long time scale was decomposed to monthly load with short time scale, it shows that decomposition coefficient series is a chaotic one. Phase space model is used for forecasting decomposition coefficient, and prediction result is used for calculating monthly load. The study proves feasibility for applying chaotic analysis on downscaling calculation.Leading artificial neural networks and Kalman filter technique into chaotic phase space, this paper presents two combined models. One is a BP neural networks model based on chaotic analysis. The other is a chaotic Kalman filter model combined the chaotic method with real-time adjustment technique. The principle of building model and forecasting steps about new method are explored in detail. Then application on short term forecasting of daily load based on the combined models is discussed. The prediction result shows that the new combined models could get high precision.Above all, the main study for electric power load focuses on three aspects: about chaotic characteristic analysis, about nonlinear forecasting, and about combined chaotic models. Not only can the research provide practicable method for load forecasting, but also it suggests a new idea for further work. Moreover, it provides a valuable scheme for studying other hydrologic variable.更多还原