神经网络方法在光伏监控中的若干应用

发布时间：2018-02-23 21:13

本文关键词： BP神经网络 L-M算法学习率自适应传感器辨识故障诊断　出处：《杭州电子科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：能源问题一直受到人们的广泛关注,以光伏发电为代表的新能源技术得到了重视,光伏监控是利用好太阳能的重要手段。但是光伏监控中存在着复杂非线性问题,如何利用人工智能技术将光伏监控中存在的难题朝着智能化方向发展是本文的主要研究动机。本文基于神经网络方法从光角度传感器模型辨识以及组件故障诊断方面做了应用研究,具体内容如下:第一部分:主要介绍神经网络的理论基础。首先,从生物神经元描述出发,对其抽象出的人工神经元原理以及工作特点进行了介绍。然后,基于环境提供信息的多少对神经网络的三种学习方式进行了总结与描述。最后,对于神经网络学习过程中所采用的三种常用算法原理以及典型网络结构进行了总结与归纳。第二部分:选择了BP神经网络模型作为本文主要研究对象。首先,对于BP神经网络模型的算法原理进行了简单概述,对于该模型工作过程中的信号前馈计算和误差反向传播分别进行了理论推导,获取到最终的误差迭代公式。然后,在网络结构设计方面进行了研究,关于输入输出样本的选择和处理做了深入讨论。最后,主要对于网络训练中的参数选择进行了研究与分析。第三部分:针对光伏监控中使用的一种特殊光角度传感器存在模型难以辨识的问题,提出了L-M算法优化的BP神经网络系统辨识模型。首先,针对光角度传感器中存在的制造或安装误差而无法准确测量角度的非线性问题,建立了用于模型辨识的BP神经网络模型。然后,研制了一套基于光角度传感器的实验数据采集系统,获取到训练神经网络的输入输出实验数据。最后,对于训练完成且达到目标设定值的神经网络,可以有效地根据光角度传感器测得的电流数据预测出输出角度值,该方法可以有效地应用于机理建模中存在参数未知甚至难以建模的数学问题。第四部分:针对光伏组件存在运行状态与运行参数之间的复杂非线性问题,提出了学习率自适应的BP神经网络故障诊断模型。首先,采用数学语言分析了运行参数与环境条件之间的复杂非线性问题,建立了学习率自适应的BP神经网络故障诊断模型。然后,为了验证模型的有效性,采用的是光伏组件本构方程进行实验数据采集,获取到用于训练网络的输入输出数据。最后,对于训练完成且达到目标设定值的神经网络,采用一组新的运行数据用来验证神经网络,结果表明该网络可以有效地对组件运行状态进行识别与分类,验证了方法的有效性。
[Abstract]:The energy problem has been paid more and more attention, and the new energy technology, represented by photovoltaic power generation, has been paid attention to. Photovoltaic monitoring is an important means to make good use of solar energy. However, there are complex nonlinear problems in photovoltaic monitoring. How to make use of artificial intelligence technology to develop the problem of photovoltaic monitoring towards the direction of intelligence is the main motivation of this paper. This paper based on the neural network method from the perspective of light sensor model identification and component fault diagnosis. Has done the applied research in the aspect, The main contents are as follows: the first part mainly introduces the theoretical basis of neural network. Firstly, from the description of biological neurons, the abstract principle and working characteristics of artificial neurons are introduced. The three learning methods of neural network are summarized and described based on the amount of information provided by the environment. Finally, Three common algorithms and typical network structure used in the learning process of neural network are summarized and summarized. The second part: the BP neural network model is selected as the main research object of this paper. The algorithm principle of BP neural network model is briefly summarized. The signal feedforward calculation and error back propagation in the working process of the model are derived theoretically, and the final error iterative formula is obtained. In the aspect of network structure design, the selection and processing of input and output samples are discussed. Finally, This paper mainly studies and analyzes the parameter selection in network training. Part three: aiming at the problem that a special optical angle sensor used in photovoltaic monitoring is difficult to identify. The identification model of BP neural network system optimized by L-M algorithm is proposed. Firstly, aiming at the nonlinear problem of manufacturing or installing errors in optical angle sensor, it can not accurately measure the angle. A BP neural network model for model identification is established. Then, a set of experimental data acquisition system based on optical angle sensor is developed to obtain the input and output experimental data of the training neural network. Finally, For the neural network which has completed the training and reached the target set value, it can effectively predict the output angle value based on the current data measured by the optical angle sensor. This method can be effectively applied to the mathematical problems where the parameters are unknown or even difficult to model in the mechanism modeling. Part 4th: aiming at the complex nonlinear problem between the running state and the operating parameters of the photovoltaic module, A BP neural network fault diagnosis model with adaptive learning rate is proposed. Firstly, the complex nonlinear problems between operating parameters and environmental conditions are analyzed by mathematical language. A BP neural network fault diagnosis model with adaptive learning rate is established. Then, in order to verify the validity of the model, the photovoltaic component constitutive equation is used to collect the experimental data, and the input and output data for the training network are obtained. A new set of running data is used to verify the neural network for the neural network which has completed the training and reached the target set value. The results show that the neural network can effectively identify and classify the running states of the components and verify the effectiveness of the method.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;TM615

【参考文献】