【作者】 张映梅；

【导师】 赵惠艳； 李修炼；

【作者基本信息】 西北农林科技大学 ， 农业昆虫与害虫防治， 2003， 硕士

【摘要】 小麦吸浆虫是北半球间隙性猖獗危害小麦的重要害虫，也是我国麦区的重要害虫之一，隶属于双翅目（Diptera）瘿蚊科（Cecidomyiidae）。危害小麦的吸浆虫有两个种，一种是麦黄吸浆虫Contarinia tritici（Kirby），另一种是麦红吸浆虫Sitodiplosis mosellana（Gehin）。小麦吸浆虫对小麦可造成毁灭性灾害，在我国历史上曾多次爆发，本世纪50年代、80年代猖獗危害。90年代以来，每年仍在部分地区猖獗危害，对小麦生产造成了较大的影响。因此，做好小麦吸浆虫发生的预测预报工作，对指导小麦吸浆虫的防治工作有着十分重要的作用。 本论文利用关中地区1933～2000年小麦吸浆虫的发生资料以及气象资料，通过逐步回归法筛选出了影响小麦吸浆虫发生的主要气象因子，即一月份平均温度（X₁）、三月份平均温度（X₃）、前一年八月份平均温度（X₅）、二月份平均降雨量（X₈）。 将筛选出的主要气象因子作为预测因子，分别用单因子回归预测法、逐步回归预测法、判别分析预测法、马尔柯夫链预测法、模糊数学预测法与神经网络预测法建立了关中地区小麦吸浆虫的发生程度的预测模型。 用所建立的预测模型对1933～1995年的资料进行回测，历史符合率分别是：71.4％、82.5％、74.6％、77.8％、69.8％、100％；并对1996～2000年的小麦吸浆虫的发生程度进行预测，预测符合率分别为：60％、60％、40％、40％、60％、100％。从以上数字可以看出，前五种预测方法的回测准确率都比预测准确率高，但回测准确率也不十分理想，说明这些预测方法对小麦吸浆虫的回测和预测效果都不如神经网络方法。尤其是判别分析预测法和马尔柯夫链预测法，这两种预测方法的预测准确率和回测准确率相差较大，说明预测效果不稳定，对小麦吸浆虫的预测实际应用价值不大。只有神经网络预测法的回测和预测准确率均为100％，说明这种预测方法的预测效果稳定，有很好的应用前景。此外，本文还应用列联表预测法和周期预测法对小麦吸浆虫的发生程度进行了预测实验，结果也表明这两种方法在小麦吸浆虫预测中的实际应用价值不大。 人工神经网络是近年来人工智能领域中发展迅速的信息处理技术，它能够处理高度非线性问题，具有跟踪性能好、适用面广、容错能力强等优点。它主要用于模式识别和图像处理，预测和管理等方面，近年来也逐渐开始应用于病虫害预测预报，但在小麦吸浆虫的预测预报却未见报道。 本研究选用Sigmoid函数为结点作用函数，用Visual Basic 6.0语言编写了通用的神经网络预测模型，并对小麦吸浆虫的发生进行了预测，取得满意的效果，此模型也可应用于其它作物病虫害的预测预报。更多还原

【Abstract】 Wheat midge is the important pest on wheat in Northern Hemisphere, and it is also one of important pests on wheat in our country. It belongs to the Diptera Cecidomyiidae. There are two kinds midges on wheat, one is Contarinia tritici (Kirby) and another is Sitodiplosis mosellana (Gehin). Wheat midge can result in the destructive disaster to wheat. It had broken out many times on our country history, and has great impact on wheat yield. Therefore, the forecasting of wheat midge is very important to the control of wheat midge.Data of 14 weather factors during 1933~2000 are studied by using stepwise regression analysis. Key factors affected wheat midge occurrence of Guanzhong district are the average daily temperature in January (Xi), the average daily temperature in March (X3), the average daily temperature in August of last year (X5) and precipitation in February (X8).Six prediction methods are selected to predict occurrence level of wheat midge. To make occurrence degree prediction, single factor regression algorithms, stepwise regression algorithms, discriminatory analysis, Markovian model, fuzzy mathematics theory and Back propagation neural network algorithms are studied and models are builded.Back prediction accurate ratio of six methods are 71. 4%, 82.5%, 74.6%, 77.8%, 69.8%, 100% respectively. Prediction accurate ratio are 60%, 60%, 40%, 40%, 60%, 100% respectively. The accurate rate of return of the previous five methods are higher than their prediction, but their accurate rates of return are also not good, all these show that these predicting methods are not good in the forecasting of wheat midge than neural network. And particularly is the discriminatory analysis and Markovian model, their accurate rate of return are more higher than their prediction, this show that their prediction resolutions are unsteady and they have good application value in the prediction of wheat midge. Only can the result of back prediction and prediction under back propagation neural network reach to 100%. This method has good prediction result, and its prediction is stable. So it has the good applied prospect.Artificial neural network (ANN) is one of the quick developing processing techniques in artificial intelligence. ANN has convergent speed, perfect trace performance, and high tolerant errors as well. Specially, it is suitable for complex non-linear prediction. The neural network is primarily using on patternrecognition, photo treated, economic prediction and management. In recent year, it alse is applied in the prediction of plant disease and pest, but there are still no reports on wheat midge prediction. In this paper, ANN was applied in the prediction of wheat midge.The algorithm is used to imitate the brain’ s ability to make decisions and draw conclusions when presented with complex, noisy and/or partial information. It has three layers: input layer, hidden layer and output layer. This prediction model has been passed in Visual Basic 6. 0. This model has in common use, and can be used in the prediction of other diseases and pest insect.更多还原