采用优化特征子集选取和改进SVR的养殖禽舍温度预测算法

Temperature prediction algorithm for poultry house based on optimized feature subset selection and improved SVR

为提高养殖禽舍温度预测算精度,降低数据冗余度和差异性对预测结果的影响,提出一种基于智能优化特征子集选取和模糊聚类改进SVR(Support Vector Regression)的温度预测模型。首先,构建最优特征子集选取模型,通过设计最优特征子集选取指标,以降低特征之间冗余度和数据维度;采用改进的离散灰狼算法对特征子集选取模型进行求解,以实现最优特征子集选取。其次,建立模糊聚类改进SVR预测机制,通过设计多度量核FCM(Fuzzy C-means)算法,以实现数据样本自动分类;提出与数据样本分类相对应的SVR预测算法,并采用灰狼算法对SVR参数进行优化,最大程度降低样本数据差异性对预测精度的影响。最后,融合最优特征子集选取和模糊聚类改进SVR预测机制,以实现养殖禽舍温度高精度预测。仿真结果表明,该算法实现不同季节条件下养殖禽舍温度的高精度预测,相比于其他预测算法,预测精度提高约23.7%~37.8%。所提养殖禽舍温度预测算法具有良好的预测性能,具有一定的推广应用价值。

In order to improve the accuracy of temperature prediction for poultry houses and reduce the impact of data redundancy and difference on the prediction results, a temperature prediction model based on intelligent optimization feature subset selection and fuzzy clustering improved SVR(Support Vector Regression) is proposed. Firstly, the optimal feature subset selection model is constructed, and the optimal feature subset selection index is designed to reduce the redundancy and data dimension between features;The improved discrete Gray Wolf algorithm is used to solve the feature subset selection model to realize the optimal feature subset selection. Secondly, the fuzzy clustering improved SVR prediction mechanism is established, and the multi-core FCM(Fuzzy C-means) algorithm is designed to realize the automatic classification of data samples;A SVR prediction algorithm corresponding to data sample classification is proposed, and the Gray Wolf algorithm is used to optimize the SVR parameters to minimize the impact of sample data differences on prediction accuracy. Finally, the optimal feature subset selection and fuzzy clustering are combined to improve the SVR prediction mechanism to realize the high-precision prediction of poultry house temperature. The simulation results show that the algorithm realizes the high-precision prediction of poultry house temperature in different seasons and different climatic conditions, and the prediction accuracy is improved by about 23.7%-37.8% compared with other prediction algorithms.