Python基于RNN的温室温度预测研究+源代码
时间:2019-04-27 15:27 来源:毕业论文 作者:毕业论文 点击:次
| 摘要：能否对温室温度进行合理预测是衡量温室智能化控制程度的一个重要依据。本设计将温室里湿度、室外温度、光照强度等因素作为神经网络的输入数据，对神经网络进行训练，从而对下一时刻的温室温度进行预测。由于传统的时间序列预测分析具有很多问题，使得人们难以快速、方便的建立合适的模型，而适合处理序列数据的RNN则可以解决这个问题。除此之外，RNN的一个模型，LSTM，能够解决RNN的容易出现的梯度消失、梯度爆炸的问题，所以本设计选择LSTM进行温室温度预测。首先，根据LSTM的定义，使用Python和基于Python的开源的数值计算扩展NumPy实现LSTM的网络结构，接着对输入数据进行处理，最后将预测结果误差控制在较小范围内。在实验中，不同的权值初始化、数据的处理方式对结果影响较大和学习率对预测结果影响比较明显。34941 毕业论文关键词：RNN；长短时记忆网络 ；温室 ；NumPy Research on the Prediction of Temperature in Greenhouse Based on RNN Abstract: Whether the reasonable forecast of greenhouse temperature is an important basis for measuring the degree of intelligent control of greenhouse. The design uses greenhouse humidity, outdoor temperature, light intensity and other factors as the input data of the neural network to train the neural network, so the next time of the greenhouse temperature is able to be predicted. Because the traditional analysis of time series prediction has many problems, it is difficult to quickly and easily establish the appropriate model. While RNN, suitable for dealing with the serial data, can solve this problem. In addition, the RNN model, LSTM, can solve the problem of RNN prone gradient disappearance and gradient explosion, so this design chooses LSTM to carry out greenhouse temperature prediction. Firstly, according to the definition of LSTM, the NumPy and Python were used to implement the LSTM network structure. The NumPy is a Python-based open source numerical calculation package. Secondly the input data was processed. Finally, the error of prediction was controlled to a small extent. In the experiment, the different value for weights initialized and the different methods for data processing lead to a great influence on the result and the learning rate had obvious influence on the prediction result. Keywords: RNN; LSTM; Greenhouse; NumPy 目 录 摘要： 1 关键词： 1 Abstract 1 Keywords 1 1. 绪论 2 1.1. 研究背景及意义 2 1.2. 国内外研究现状 2 1.3. 研究内容与技术路线 3 1.3.1. 研究内容 3 1.3.2. 技术路线 3 2. RNN的介绍和分析 3 2.1 RNN简介 3 2.2 RNNs的扩展和改进模型 8 2.2.1 Simple RNN（SRN） 8 2.2.2 Bidirectional RNN（双向网络） 9 2.2.3 Echo State Network（回声状态网络） 9 3. 长短时记忆网络（Long Short-Term Memory） 10 3.1. RNN 和LSTM 的比较 10 3.2. LSTM 的特点 11 3.3. LSTM的变体 12 4. LSTM 实现 13 4.1. LSTM 前向传播 14 4.2. 反向传播 15 4.3. LSTM 实现 16 (责任编辑：qin) |