APSO_SVR模型在我国大豆价格预测的应用研究

Predicting Chinese Soybean Price Based on APSO_SVR

国内大豆价格受到多重因素影响,如大豆进口量、国内大豆供给量、中国居民消费价格指数等,因此呈现非线性等特点。大豆价格的剧烈波动会对农户的种植结构和国家政策产生影响,因此准确预测大豆价格具有重要意义。支持向量回归(SVR)因其优越的寻优能力和较高的预测精确度而被广泛应用于非线性时间序列回归中。本文提出一种自适应粒子群算法(APSO)优化的SVR模型来预测我国大豆价格,该模型通过将现实空间内的数据映射到高维空间内,在高维空间内构造线性回归函数,从而判别原有空间内数据之间的关系。在SVR模型参数优化时,由于粒子群算法易陷入局部最优解,因此采用惯性权重更新和适应度变异的粒子群算法(APSO)对预测模型参数进行优化。采用2009年1月-2016年12月的国内大豆价格月度数据进行预测,结果表明APSO优化的SVR模型在大豆价格预测中精度较高,且能准确反应大豆价格的未来趋势,为从事大豆种植者及经营者提供决策依据。

Soybean price was influenced by many factors, such as soybean imports, domestic soybean outputs, consumer price index etc. The characteristics of soybean price is non-linearity, randomness etc. The fluctuation of soybean price would influ- ence farming structure and national policy of soybean. Exact predicting soybean price is significant for farmers and soybean policy. Support vector machine was widely used in nonlinear time series because of its superior search capability and high accuracy. In this paper, SVR model optimized with adaptive particle swarm optimization （APSO） was used to predict soybean price. In this model, the data was mapped to high-dimensional space from real space. The linear regression function was con- structed in the high dimensional space to distinguish the data relations in the real space. The parameters of SVR model was op- timized with particle swarm optimization （PSO）, but the PSO was usually trapped local optimization results. Therefore, adaptive strategy of fitness mutation and inertia weight updated Was used to structure APSO. The data of soybean price from Jan. 2009 to Dec. 2016 were used to forecast. The results indicated that APSO and SVR model was accurate and effective. The SVR model can accurately reflect future trend of soybean price and provide decision basis for soybean farmers and soybean businessman.