基于人工神经网络的短期电力负荷预测研究

发布时间：2018-03-28 04:30

本文选题：短期电力负荷预测　切入点：人工神经网络　出处：《浙江大学》2017年硕士论文

【摘要】：短期负荷预测是电力系统运行和控制的重要基础性工作,一直是学术研究的热点问题。由于电力负荷历史数据本质上是一个随机非平稳序列,完全无误差的预测目前是不可能的,因此研究者们一直致力于提升预测的精度。人工神经网络具有自学习、泛化能力强等优点,已被广泛应用于短期电力负荷预测中,取得了较为理想的效果。近年来,人工神经网络领域又取得了可喜的突破,出现了深度学习这一新的研究领域。本文基于人工神经网络的最新发展成果,结合实际数据对短期电力负荷预测问题进行了相关研究。主要内容包括:(1)建立了基于改进粒子群算法优化极限学习机的短期负荷点预测模型。该模型将改进粒子群算法与极限学习机结合,利用改进粒子群算法强大的全局搜索能力对极限学习机的输入权值及隐含层偏置矩阵进行寻优。基于用户实际负荷数据得到的仿真结果验证了该模型的有效性。(2)基于深度学习领域的神经网络结构,分别建立了带词嵌入层的多层长短期记忆网络短期点预测模型和带词嵌入层和卷积层的长短期记忆网络点预测模型。基于用户实际负荷数据,验证了上述模型的有效性,并与基于改进粒子群优化极限学习机的模型进行了对比。(3)针对电力负荷预测中的不确定性,建立了基于改进粒子群优化极限学习机的短期区间预测模型;同时,提出了一种改进的比例系数法,能够在点预测的基础上生成更合理的预测区间,基于用户实际负荷数据的算例表明,该方法可以得到比较理想的区间预测结果。
[Abstract]:Short-term load forecasting is an important basic work in the operation and control of power system and has been a hot topic in academic research. Because the historical data of power load is essentially a random non-stationary sequence. At present, it is impossible to predict without error, so researchers have been working to improve the accuracy of prediction. Artificial neural network has been widely used in short-term power load forecasting because of its advantages of self-learning and strong generalization ability. In recent years, the field of artificial neural network has made a gratifying breakthrough, and the new research field of deep learning has emerged. This paper based on the latest development of artificial neural network, Based on the actual data, this paper studies the short-term power load forecasting problem. The main contents include: 1) A short-term load point forecasting model based on improved particle swarm optimization algorithm for extreme learning machine is established. The model will improve particle size. Group algorithm and extreme learning machine, Using the powerful global search ability of improved particle swarm optimization (PSO) algorithm, the input weights and hidden layer bias matrices of LLMs are optimized. The simulation results based on the actual load data of users verify the validity of the model. Neural network structure based on deep learning domain, The short and short term point prediction models with word embedding layer and long and short term memory network with word embedding layer and convolution layer are established respectively. Based on the actual load data of users, the validity of the above model is verified. Compared with the model based on improved particle swarm optimization (PSO) limit learning machine, a short-term interval prediction model based on improved particle swarm optimization (PSO) limit learning machine is established in view of the uncertainty in power load forecasting. An improved proportional coefficient method is proposed, which can generate a more reasonable prediction interval based on the point prediction. An example based on the actual load data shows that the method can get a better interval prediction result.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM715

【参考文献】