基于粗神经网络的电力系统短期负荷预测研究

发布时间：2018-02-27 01:18

本文关键词： 电力系统粗神经元 RBF神经网络小波神经网络萤火虫算法短期负荷预测　出处：《南京邮电大学》2017年硕士论文　论文类型：学位论文

【摘要】：在整个电力系统的运作和规划中,短期负荷预测一直扮演着重要的角色。电力负荷变化主要与负荷的组成成分、天气、时间以及人们社会活动等因素有关,这些因素造成负荷变化量具有多元非线性和时变性,其变化规律无法用数学公式来准确表达。为了适应电力系统短期负荷对其预测精度越来越高的要求,本文将粗糙集理论与神经网络相结合,建立了基于粗神经网络的短期负荷预测模型。具体研究如下:(1)针对短期负荷多元非线性和时变性的特征,本文将粗糙集理论与RBF神经网络相结合,给出了一种基于粗神经元RBF(Radial basis function)神经网络的预测模型。粗集理论在分析和处理不精确、不一致、不完整的信息方面优势明显。而RBF神经网络以非线性神经元来协同处理信息,能够大规模并行处理数据,具有较强的自学习能力。在神经网络中引入粗神经元可以很好地将粗糙集和神经网络的优势相结合。这种粗神经网络预测模型的输入由传统的单一值变为双值,在保持较强学习能力的同时,可反映一段时间内负荷数据的波动情况,能很好地处理不精确的信息。由于RBF神经网络的初始参数通常都设为随机数,导致预测结果不确定,预测精度具有随机性。因此本文提出遗传算法优化粗神经元RBF神经网络,利用遗传算法搜索网络的最优初始权值,以此作为网络训练的初始参数,从而提高网络的预测精度。(2)由于小波神经网络在分析信号方面具有优势,其学习能力强,训练速度较快,在预测领域应用较为广泛。因此本文在建立基于粗RBF神经网络预测模型的基础上,还将粗神经元引入小波神经网络中,构建了一种“紧致性”的粗小波神经网络,给出了基于粗小波神经网络的短期负荷预测模型。这种网络模型训练速度比粗RBF神经网络快,泛化能力更强。(3)由于粗小波神经网络中需要寻优的参数较多,在网络训练的过程中,计算也会变得复杂。而萤火虫算法计算效率高,需要设置的参数少,其收敛速度和寻优精度都较高。因此本文采用改进的萤火虫算法对粗小波神经网络的相关参数进行优化,在全局内找出网络的最佳参数。将优化后的粗神经网络用于短期负荷预测,有利于加快网络的收敛速度,提高网络的泛化能力。实例仿真结果表明,本文所提出的方法能有效反映负荷的波动信息,改进后的预测模型训练速度快、非线性拟合能力较强、预测精度较高,在实际运用中具有重要意义。
[Abstract]:Short-term load forecasting has always played an important role in the operation and planning of the whole power system. The variation of power load is mainly related to the composition of the load, weather, time and social activities of people. These factors result in multivariate nonlinear and time-varying load variation, which can not be accurately expressed by mathematical formula. In order to meet the demand of higher and higher forecasting accuracy for short-term load in power system, In this paper, a short-term load forecasting model based on rough neural network is established by combining rough set theory with neural network. In this paper, a prediction model based on rough neuron RBF(Radial basis function neural network is presented by combining rough set theory with RBF neural network. The advantage of incomplete information is obvious, and RBF neural network can process data in large scale parallel by using nonlinear neurons to process information together. The introduction of rough neurons into neural networks can combine the advantages of rough sets and neural networks well. The input of this rough neural network prediction model changes from a traditional single value to a double value. While maintaining strong learning ability, it can reflect the fluctuation of load data over a period of time, and can deal with imprecise information well. Because the initial parameters of RBF neural network are usually set as random numbers, the prediction results are uncertain. Therefore, genetic algorithm is proposed to optimize the rough neuron RBF neural network, and the genetic algorithm is used to search the optimal initial weight of the network, which is regarded as the initial parameter of network training. In order to improve the prediction accuracy of the network, the wavelet neural network has advantages in signal analysis, its learning ability is strong, and its training speed is relatively fast. It is widely used in the field of prediction. On the basis of establishing the prediction model based on rough RBF neural network, the rough neuron is also introduced into the wavelet neural network to construct a kind of rough wavelet neural network with "compactness". A short-term load forecasting model based on rough wavelet neural network is presented. The training speed of this network model is faster than that of rough RBF neural network, and the generalization ability is stronger. In the course of network training, the calculation becomes more complicated. The firefly algorithm is very efficient and requires less parameters. Therefore, the improved firefly algorithm is used to optimize the parameters of rough wavelet neural network. The optimal parameters of the network are found in the whole world. The application of the optimized rough neural network to short-term load forecasting is beneficial to speeding up the convergence speed of the network and improving the generalization ability of the network. The simulation results show that, The method presented in this paper can reflect the fluctuation information of load effectively. The improved forecasting model has the advantages of fast training speed, strong nonlinear fitting ability and high prediction accuracy. It is of great significance in practical application.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;TM715

【相似文献】