关键词: 林业害虫, 美国白蛾, 鲸鱼算法, 长短时记忆网络, 注意力机制

Abstract:  It is of great significance to accurately predict  the occurrence of forest pests in China for improving the level of forest resource risk management and control as well as the early warning of forest pests. The occurrence of forest insect infestation is not only related to temperature and humidity, but also complicated with other meteorological factors. In order to achieve accurate prediction of forest insect infestation, meteorological data and insect infestation data are transformed into a time series prediction problem in this study. In this paper, the relationship between the occurrence of American white moth in the “Millennium Forest” of Xiongan New Area  and the meteorological environment at that time was studied, combining swarm intelligent optimization algorithm and deep learning algorithm, a forest pest prediction model based on WOA-BiLSTM-BA algorithm was proposed. Firstly, WOA was used to continuously search for the optimal parameter combination of BiLSTM through iterative optimization to avoid the subjectivity of manual parameter selection and high training cost. Secondly, the Bahdanau Attention module BA was introduced into BiLSTM network to dynamically allocate weight information, and finally the prediction results were output through the fully connected layer.  By comparing the proposed model with the traditional BP prediction model, LSTM prediction model and BiLSTM prediction model, the results showed that the effect of WOA-BiLSTM-BA model was better than that of other control prediction models, with R² reaching 0.989 1, RMSE only 0.073, MAPE 0.227 5 and MAE 0.056 4.

Key words: forest pest, American white moth, whale algorithm, long?short?term memory network, attention mechanism