【作者】 尹新；

【导师】 何怡刚；

【作者基本信息】 湖南大学 ， 电工理论与新技术， 2011， 博士

【摘要】 电力负荷预测对于电力系统的建设和运行均有着极其重要的意义和作用。预测精度的高低直接影响着电力系统的安全运行和经济效益。本文从近年来广泛应用于电力负荷预测的群智能算法出发,研究和分析现有的群智能算法以及与之结合的预测模型。通过对算法和模型的改进,建立新的预测模型,并应用于实际的电力负荷预测,得到了预期的效果。本文主要研究的内容如下:1.研究两种典型的群智能算法:粒子群优化算法(PSO)和蚁群优化算法(ACO)。通过对不同函数优化的分析,对比了两者的优点和不足。以粒子群算法为基础,并吸收模拟退火算法以及蚁群算法的思想对其进行改进,提出了两种改进群智能算法:IPSO和HA算法。比较改进的群智能算法、标准的粒子群算法和蚁群算法对不同函数的优化,表明改进的群智能算法具有一定的优越性和有效性。2.将群智能算法和神经网络相结合建立新的预测模型。将群智能算法分别和BP、Elman神经网络结合,利用群智能算法代替BP算法优化神经网络的权值和阈值,建立基于群智能算法优化的神经网络预测模型:PSOBPNN,PSOENN,IPSOBPNN,IPSOENN,HABPNN和HAENN。通过对经典的混沌序列进行模拟仿真,验证了基于群智能算法的神经网络预测模型比传统的神经网络模型具有更快的收敛速度以及更高的预测精度。3.中长期负荷预测对电力系统规划和建设具有重要的意义。将群智能算法和灰色预测模型相结合建立基于群智能算法的灰色预测模型。利用群智能算法代替传统的最小二乘法直接求解灰色模型参数,避免了由于灰色模型在求解过程中因背景值的取值不当而带来的误差。通过对中长期电力负荷的预测,验证了基于群智能算法的灰色预测模型比传统的灰色预测模型具有更高的预测精度和更加广泛的适用范围。4.电力负荷短期预测方法是当前研究的热点问题,训练样本的建立方法与预测精度密切相关。分析和比较了多步预测和分步预测的优点和不足,结合两种方法的优点提出一种新训练样本建立方法。利用PSO,IPSO和HA算法优化的神经网络预测模型对某市实际的电力系统进行短期电力负荷预测,结果表明HA算法与新预测方法结合可获得更高的预测精度。5.分析和比较了秋季负荷和夏季负荷的特点,验证了秋季温度变化不大的情况下,负荷和温度之间无明显的相关性;而在夏季温度变化比较大的情况下负荷和温度之间具有明显的相关性。由此确定对秋季负荷进行预测时不引入温度因子,而对夏季负荷预测时则引入温度因子。对实际的电力系统进行短期电力负荷预测,对于秋季负荷不引入温度因子也可达到很高的预测精度,而对夏季负荷引入温度因子具有更高的预测精度,说明温度对夏季电力系统负荷的预测是不可忽视的因素。6.采样信号的预处理对于提高预测精度具有特殊的意义。通过最优小波包分解,将复杂的电力负荷序列分解成多个简单的、规律性强的负荷分量。分别对这些分量运用新的预测方法进行预测,在对夏季负荷低频分量进行预测时,引入实时温度因子,最后重构负荷序列得到最终预测结果,提高了预测精度。通过实例仿真,验证了该方法的可行性。更多还原

【Abstract】 Load forecasting is very significant and useful in the construction and operation of power systems. Prediction accuracy directly affects the safety operation of power systems and economic benefits .In this paper, first of all, swarm intelligence algorithms was be studied. In recent years it has been widely used in load forecasting. Next, a predictive model that combined with swarm intelligence algorithm was studied. Through existing algorithms and models which have been improved, the new prediction model will be obtained and used in practical load forecasting, getting good results. The main content of this article is as follow:1. Research on two typical swarm intelligence algorithms: particle swarm optimization(PSO) and ant colony optimization algorithm(ACO). Both the advantages and disadvantages are gained from analysis and comparison of their results for different function optimization,. Simulated annealing algorithm and absorb advantages of ant colony algorithm,Improved particle swarm optimization (IPSO)and hybrid algorithm(HA) have been proposed. Improved Swarm Intelligence Algorithm, the standard particle swarm algorithm and ant colony algorithms, would be compared for different function optimization. The results show that the improved swarm intelligence algorithm has more advantages and effectiveness.2. Combined the swarm intelligence algorithms and neural network to get a establishment of new forecasting mode. Respectively, the improved swarm intelligence algorithm combined with BP & Elman neural network. The improved swarm intelligence algorithms optimized neural network weights and thresholds established optimization algorithm based on swarm intelligence neural network prediction model: PSOBPNN,PSOENN,IPSOBPNN,IPSOENN,HABPNN and HAENN. Through the classical chaotic sequence simulation, it has been verified that the algorithm based on swarm intelligence neural network prediction model has faster convergence rate and higher forecast accuracy than the traditional one.3. Using swarm intelligence algorithm instead of the traditional least squares method to directly solve the gray model parameters. This avoids the background value of the Gray model solution process due to the error caused by improper values. Through the middle and long term load forecast, it has been verified that the gray prediction model based on swarm intelligence algorithm has better prediction accuracy and the broader scope of application than the traditional one.4. Short-term load forecasting method is a hot topic of current research, establishing methods of training samples is closely related with the prediction accuracy. Step multi-step prediction and forecasting would be studied. According to the analysis of multi-step prediction step prediction, a new prediction method combined the advantages of the two methods would be proposed. Under the new forecasting method, training samples would be established. Neural network model is used to complete the city’s actual short-term load forecasting and the results show that the new forecasting method has better prediction accuracy while it’s combined with HA.5. Analysis and comparison of fall and summer load characteristics , prove such a rule: the small correlations of power load and temperature in the fall and crucial relevance of power load and temperature in the summer. Power load forecasting during the fall do not need to use the temperature factor but get good results. Power load forecasting during the summer, the temperature factor need to be used in order to obtain better results.6. Sampling signal preprocessing for improving prediction accuracy is of special significance.Through best wavelet packet decomposition, the complex load sequence into several simple regularity of the load weight. Each of these components using new forecasting method to predict the load in the low frequency components of the summer forecast, the introduction of real-time temperature factor, the final reconstructed sequence of the final load forecast results, to improve the prediction accuracy. An example of simulation would verify the feasibility of the method.更多还原