基于数据挖掘的玉米市场价格预测

发布时间：2018-01-07 23:19

本文关键词：基于数据挖掘的玉米市场价格预测　出处：《青岛大学》2017年硕士论文　论文类型：学位论文

【摘要】：农产品的价格波动及其将来的走势与人们的生活息息相关,而玉米是我国的主要粮食产量之一,生产量很大,种植广泛,而且用于粮食作物、饲料作物、燃料作物等方面的消费量也很大。所以近年来,该粮食的产量、价格、需求量等备受关注,尤其是价格是人们最关注的。多年来,国内外很多学者对玉米市场做了研究,有的将时间序列的各种模型应用到价格的分析与预测上,效果比最初的线性模型好。而现在使用的各种算法准确度高、效果好,通常不用受很多条件的限制,而且运算速度也是比较快的。所以,很多学者在考虑农产品分析问题上也会使用数据挖掘的算法。本文致力于解决实际问题,以农产品价格--玉米价格为研究对象(选取的数据为2005年1月到2016年11月的价格),在各位学者研究的基础上,综合考虑并选择合适的方法,根据实际问题,分析借助R工具建立模型,选取最优模型进行最终的预测。本文分析实例问题时借助R工具,通过分析建立时间序列模型、神经网络模型:对于时间序列模型,选用AR(1)、ARIMA(1,1,1)、ARIMA(2,1,3)、多重季节MA模型ARIMA(2,1,3)、季节模型ARMA(2,0,2)这5个模型,选取出最优的模型;对于神经网络模型,选用前五个月、六个月、十二个月的月度数据分别作为输入神经元,输出神经元为未来一个月、两个月、三个月、四个月、未来五个月的月度数据,通过设置隐藏层的单元个数、反向传播速率、最大迭代次数等指标,共建立45个模型,通过比较选出最优模型;最后将两种方法进行比较,最终选出最优模型,进行预测。通过对这个具体问题的分析,可以看出:(1)价格波动的影响因素有很多,多种因素结合起来很难用线性的模型表示,用传统的模型很大局限性,精度也不高。而神经网络可以很好地解决非线性问题,因为神经网络可以任意近似任何非线性函数。神经网络可以任意精度的逼近任何非线性函数。本文最终选取的预测方法是神经网络建模方法。(2)通过影响因素的分析,可以看出有很多指标无法量化,所以从价格趋势角度进行分析。(3)通过各种模型的结果的比较,可以看出,神经网络算法用于市场价格预测,得到的预测结果与真实值得误差是比较小的。综合分析,最终选取神经网络模型进行预测,并且根据预测未来时间的长短,选取不同的模型进行预测,得出预测结果。最后,将最终得出的结论进行延伸,分析未来更长时间的价格。
[Abstract]:The price fluctuation of agricultural products and its future trend are closely related to people's daily life, and corn is one of the main grain production in China, the production is very large, widely planted, and used for food crops, feed crops. Fuel crops and other aspects of consumption is also very large. So in recent years, the grain production, price, demand and so on, especially the price is the most concerned. For many years. Many scholars at home and abroad have done research on the corn market, some of the time series models are applied to price analysis and prediction, the effect is better than the original linear model. The effect is good, usually do not have to be limited by a lot of conditions, and the computation speed is also quite fast. Many scholars will also use data mining algorithms to consider the problem of agricultural product analysis. This paper is devoted to solving practical problems. Taking the price of agricultural products-corn price as the research object (the selected data is from January 2005 to November 2016), on the basis of the research by the scholars, the appropriate method is comprehensively considered and selected. According to the actual problem, the model is built with the help of R tool, and the optimal model is selected for the final prediction. In this paper, the time series model is established by means of R tool when analyzing the case problem. Neural network model: for the time series model, the ARIMA1 / ARIMA1 / ARIMA1 / ARIMA / AIMA / 1 / 1 / 3 / 1 and the multiple seasonal MA model ARIMA / 2 / 1 / 1 are selected. (3) the 5 models are selected as the best model. For the neural network model, the first five months, six months, twelve months of monthly data as input neurons, output neurons for the next one month, two months, three months, four months. In the next five months, 45 models are built by setting up the number of hidden layer units, the rate of back propagation, the maximum number of iterations, and so on, and the optimal model is selected by comparison. Finally, the two methods are compared, the optimal model is selected and predicted. Through the analysis of this specific problem, we can see that there are many factors that affect the price fluctuation. The combination of many factors is difficult to be expressed by linear model, and the traditional model is very limited, and the accuracy is not high, and the neural network can solve the nonlinear problem well. Because the neural network can approximate any nonlinear function arbitrarily and the neural network can approach any nonlinear function with arbitrary precision. The final prediction method chosen in this paper is the neural network modeling method. Through the analysis of influencing factors. We can see that there are many indicators can not be quantified, so from the point of view of price trend analysis. 3) through the comparison of the results of various models, we can see that the neural network algorithm is used for market price prediction. The prediction results and the real value of error is relatively small. Comprehensive analysis, the final selection of neural network model for prediction, and according to the length of the prediction of the future, select different models to predict. Finally, the final conclusion is extended to analyze future prices for a longer period of time.
【学位授予单位】：青岛大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP311.13;F323.7

【参考文献】