基于FSA-LSSVR模型的安徽省耕地变化预测

针对耕地变化内部规律及其外部驱动因子的特点，利用鱼群算法优化最小二乘支持向量机回归（FSA-LSSVR）模型，探讨耕地变化预测模型，为耕地保护政策制定提供参考依据。结果表明:鱼群算法的全局搜索能力能使支持向量机算法有效地收敛到参数γ和σ的全局最优解；FSA-LSSVR模型的预测精度指标远高于多元线性、GM（1，1）和BP神经网络模型，且优于FSA-SVM，求解速度明显优于SVM。FSA-LSSVR模型可以解决SVM内部参数难以确定的问题，适用于多因素参与的高维非线性的耕地变化预测，而且速度快、精度高，具有推广价值。

Prediction of Cultivated Land Change of Anhui Province Based on FSA-LSSVR Model

With respect to accordance with the internal law of cultivated land change and the characteristics of external driving factors, the change predictive model of cultivated land was developed by using the method of FSA (Fish Swarm Algorithm) optimize Least Square Support Vector Regression (FSA-LSSVR) model. It could provide reference for cultivated land protection policy. The results indicated that the global search capability of FSA could make the support vector machine algorithm effectively converge to global optimal solution of the parameters γ and σ. Furthermore, the FSA-LSSVR model prediction accuracy index was much higher than the multiple, GM (1,1) and BP neural network model, and it was superior to FSA-SVM, the speed of processing data is obviously better than the SVM. Therefore, it concluded that the FSA-LSSVR model could solve the problem of SVM internal parameters which were difficult to be determined. It was applicable to many factors involved in high-dimensional nonlinear prediction of cultivated land change. Moreover, it was high speed, high precision and worthy of promotion.