基于BP神经网络模型的福建海域赤潮预报方法研究

赤潮往往给渔业生产和人类的生命安全造成极大的危害,但由于赤潮的成因十分复杂,对其进行预报非常困难。本研究收集了福建海区2000年至2016年发生的219个赤潮案例有效数据,应用BP神经网络人工智能模型建立了其与气温、降水、风速、气压和日照5个气象因子的非线性关系,并将这些赤潮案例数据与相应的气象指标按闽东、闽中和闽南3个海区,分别输入模型进行学习、训练与预测。结果显示:1)闽东海区53个训练样本45个预测正确,正确率达84.91%,3个模拟预测样本全部正确;2)闽中海区69个训练样本58个预测正确,正确率达84.06%,4个模拟预测样本全部正确;3)闽南海区85个训练样本的运算预测结果63个正确,正确率74.12%,5个模拟预测样本全部正确,达到预期的结果。研究表明,以气象因子为自变量采用BP神经网络模型对赤潮的发生进行预测是可行的,该方法可为赤潮的预测提供新的途径。

Red tide forecasting model based on BP neural network in Fujian sea area

Red tide is one of marine disasters. It often causes great harm to fishery production and human life. Therefore, it is necessary to strengthen the early warning and forecast of red tide. However because the formation of red tide is very complex, it is very difficult to predict red tide. At home and abroad, there have been a lot of reports about the prediction and forecast of red tide. Different scholars have discussed the reasons for the formation of red tide using different research methods. In this study, 219 red tides data were collected in Fujian sea area from 2000 to 2016. The nonlinear relationship between the 5 meteorological factors, such as temperature, precipitation, wind speed, air pressure and sunshine, was established by using the BP neural network model. First of all, the total collected data of red tide and the corresponding meteorological data were divided into 3 sea areas data called Eastern, Central and South Fujian sea areas, according to their geographical locations, then the three groups of data were input into the model for it to learn and train. The results show that: 1) the 53 training samples in eastern Fujian sea area gave 45 correct predictions, the correct rate was 84.91%, and the 3 simulated prediction samples in the same area were all correct. 2) in 69 training samples of central Fujian sea area, 58 predictions were correct, the accuracy rate was 84.06%, and the 4 simulation predictions were all correct. 3) in 85 training samples in south Fujian sea area, 63 prediction results were correct, and the correct rate was 74.12%, and the 5 simulation samples were all correct. All the expected prediction results achieved the desired goals. Therefore, it is feasible to predict the occurrence of red tide based on the BP neural network model, which can provide a new way to forecast the red tide.