【摘要】 针对传统森林火灾检测算法精度低，以及大规模、多特征的火灾数据存在冗余信息等问题，该文提出了一种基于改进邻域粗糙集的优化反向传播神经网络(BPNN)火灾预测方法。首先，考虑到数据集具有高维特征空间和高度特征冗余等特点，设计出一种基于混沌反学习蝙蝠(BA)算法的邻域粗糙集特征选择算法，对火灾原始数据集进行特征寻优，得到约简属性子集；然后，构建BA算法优化的BPNN预测模型，将约简属性子集输入该模型中，得到火灾预测的结果；最后，通过平均分类准确度、F1值、精确度、曲线面积、召回率、平均误差率这6种评价指标，在UCI公开森林火灾数据集上分析和检验模型的分类性能。在2个数据集上的实验结果显示，基于混沌反学习策略的算法准确率为94.3%和52.7%,与邻域粗糙集结合后准确率达到98.1%和59.6%,证明了该文算法具备较高的检测精度。更多还原

【Abstract】 In response to the low accuracy of traditional forest fire detection algorithms and the redundancy in large-scale, multi-feature fire data, this paper proposes a fire prediction method based on an optimized back propagation neural network(BPNN)with an improved neighborhood rough set. Firstly, considering the characteristics of the dataset such as high-dimensional feature space and high feature redundancy, a neighborhood rough set feature selection algorithm based on the chaotic anti-learning bat algorithm(BA)is designed to optimize the features of the original fire dataset, obtaining a reduced attribute subset. Then, a BPNN prediction model optimized by the BA is constructed, into which the reduced attribute subset is fed to obtain fire prediction results. Finally, the classification performance of the model is analyzed and tested on the UCI public forest fire dataset through six evaluation metrics: average classification accuracy, F1 score, precision, area under the curve, recall, and average error rate. Experimental results on 2 datasets show that the accuracy of the algorithm based on the chaotic anti-learning strategy is 94.3% and 52.7%,and after combining with the neighborhood rough set, the accuracy reaches 98.1% and 59.6%,proving that the proposed algorithm possesses high detection accuracy.更多还原