南水北调与水利科技（中英文）2024，Vol.22Issue(4)：672-683,12.DOI:10.13476/j.cnki.nsbdqk.2024.0069
基于GA-CNN的农业旱灾恢复力及影响因素
Agricultural drought resilience and influencing factors based on optimized convolutional neural network of genetic algorithm
摘要
Abstract
In the case of global warming,it was established that the amount of uncertainty is growing with regard to droughts.Due to this,different parts of the world experience different types of drought and different measures when it comes to the recovery process of drought,leading to great losses in agriculture.Hence,it is momentous for agricultural droughts to be explored in more depth and to look for proper ways to handle them. Convolutional neural networks(CNN)provide high stability and generalization capability because of the parameter sharing and sparse connection,which decrease the number of parameters for weights and bias respectively.However,they highly depend on the selection of the learning rate(η)to decide their efficiency.Genetic algorithms(GA)have a strong attribute of global search and therefore can be used to optimize functions that are nonlinear and unbalanced and those that comprise of multi-peak;the results obtained in practice have proven to be very efficient.Hence,the use of a CNN model optimized by a GA-CNN to assess the population's drought resilience.The conceptual framework and analysis of the distribution of water resources in relation to the agricultural economic development of the studied region allowed choosing the following 11 indicators to estimate the level of agricultural drought risk.Thus,applying the principles of the GA-CNN model,the level of drought resilience of the study area was determined for 2010-2021.To find the driving forces of the time evolution of resilience,the entropy method was applied.The results show that during the study period,the agricultural drought resilience of Nehe City was on a rising-triangle course.The temporal change in the agricultural drought resilience in the study area was affected by forest coverage,grain yield per unit area,per capita water resources.With reference to the benchmarks using CNN and SVM,the GA-CNN model offered a decrease in the value of EMA by 23.51％ and 32.36％ ,ERMS by 14.42％ and 25.32％ ,and increase in R2 by 0.08％ and 1.08％ ,respectively.This means that in the areas of fit,ability to adapt,stability and reliability,and the assessment of the model,GA-CNN performs better as compared to others. Based on the main constraints of drought resilience in the study area mentioned above,future research and development strategies should target the reduction of available water supply,the improvement of food productivity,and the rise of forest cover.To sum up,proper management of agricultural water resources,increasing the production capacity of food,effective protection of forests,as well as the greatest possible use of forest resource potentiality,is critical for increasing stable and promising agricultural drought resistance.These measures will also be of help in the diffusion of improvement to neighboring areas and assisting in a mutually beneficial augmentation of the agricultural regions for drought incidences.关键词
农业旱灾恢复力/遗传算法/卷积神经网络/熵值法/影响因素Key words
agricultural drought disaster resilience/genetic algorithm/convolutional neural network/entropy method/influencing factor分类
建筑与水利引用本文复制引用
刘东,姜宸一,张亮亮,李佳民..基于GA-CNN的农业旱灾恢复力及影响因素[J].南水北调与水利科技（中英文）,2024,22(4):672-683,12.基金项目
国家自然科学基金项目(52309012 （）
52179008 （）
51579044 （）
41071053) （）
国家自然科学基金联合基金项目(U20A20318) （U20A20318）
清华大学水圈科学与水利工程全国重点实验室开放基金项目(sklhse-2023-A-04) （sklhse-2023-A-04）
水利部水圈科学重点实验室基金项目(mklhs-2023-03) （mklhs-2023-03）
黑龙江省自然科学基金联合引导项目(LH2023E003 （）
LH2021E007) （）
