残差修正模型在森林火灾预测中的应用

基于1990-2013年福建省森林火灾发生次数建立残差修正模型,并与BP神经网络模型、马尔科夫链模型、赋权组合预测模型进行比较.结果表明:残差修正预测模型的预测精度达到95.33%,而BP神经网络模型预测精度是87.77%,马尔科夫链模型预测精度为74.85%,赋权组合预测模型预测精度为88.3%,残差修正模型预测效果优于其他3个模型,说明使用其对离散的森林火灾数据进行短期预测是有效可行的.

Residual correction model to predict the number of forest fire

To better predict occurrence of forest fire, residual correction model was established to predict the number of forest fire in Fujian Province from 1990 to 2013, comparing with data derived from back propagation (BP) neural network, Markov chain model, weighted hybrid model and compiled data. Results showed that residual correction model demonstrated the highest accuracy up to 95.33%, with accuracy of BP neural network model and weighted hybrid model being 87.77% and 88.3%, respectively, and Markov chain model being the least accurate model (74.85%). To summarize, residual correction mode was likely to be a feasible and effec-tive model on discrete data prediction in short term, such as forest fire.