为合理高效地利用流域水资源，本论文将雅砻江流域划分为5个区域，对其未来规划年在不确定条件下的需水预测和水资源配置进行了研究。通过采用灰色关联度分析和极端梯度提升的特征选择算法分别选取了相应经济社会指标并组合形成神经网络模型输入数据，在此基础上构建了耦合无偏灰色-神经网络的需水预测模型，并采用马尔科夫链方法修正了需水预测结果。明确了在流域水资源配置过程中存在的可利用水量等不确定因素并利用区间规划、随机规划和模糊规划方法进行了表征，构建了基于可信度模糊约束分析的区间两阶段水资源优化配置模型，明晰了在系统收益最大化的目标函数下，不同规划年不同可信度及不同可获得水量概率下的流域水资源配置情况。主要研究结论如下：

（1）区域需水量一定程度上反映了经济社会发展的程度，经济社会发展水平较高的凉山区和攀枝花区为主要用水区域。通过计算需水影响因素数列与对应需水量数列的关联度，并在此基础上剔除了与需水量的关联度在平均值以下的经济社会指标，保证了神经网络模型输入数据的有效性。利用特征平均覆盖率得到各经济社会指标对于需水量的不同贡献度和重要性排序，筛选保留重要性排序即贡献度前50%的需水影响因子。结合关联度分析和重要性排序能够较好地识别筛选出相关的经济社会指标作为对应的需水影响因素。

（2）无偏灰色模型能够很好地预测规划年流域5个区域的经济社会发展数据，并对建立的20个需水预测神经网络模型的检验样本进行了误差检验，结果表明构建的无偏灰色-神经网络模型能够有效预测5个区域的未来需水量。利用马尔科夫链分析法对神经网络得到的预测值进行修正，得到预测区间值。比较神经网络预测值和用马尔科夫链修正过的神经网络预测值与实际值的平均绝对百分比误差，20个利用马尔科夫链模型的修正后的模型平均绝对百分比误差值更小，表明改进后模型精确度更高，并在此基础上对2019-2030年的神经网络需水预测结果进行了修正。

（3）水资源配置量和短缺量受到水资源可获得量水平和可信度水平的影响，水资源可获得量水平高，可配置水量越高，短缺量越少；可信度水平越高，可配置水量越少，短缺量越大。不同可信度水平会导致系统收益不同，同一规划年内，可信度水平增加，系统收益将会减少。流域2030年缺水量较大，尤其是农业部门用水，在未来应加大供水水利设施开发并提高用水节水效率。构建的水资源不确定规划模型可以很好处理水资源系统中的众多不确定性信息，将系统约束条件的可信度模糊关系量化，为流域水资源管理者做出科学决策提供支持，达到系统收益最大化的目标。

In this paper, in order to utilize water resources reasonably and efficiently,the Yalong River Basin is divided into five regions, and the water demand prediction and water resource allocation in the future planning year under the uncertain conditions are studied. Based on the corresponding economic and social indicators, the input data of neural network model is selected using the feature selection algorithm of grey correlation analysis and extreme gradient lifting methods. Further the water demand prediction model of coupled unbiased grey and neural network is constructed. The water demand predictions modelled are modified using Markov chain method in the end. The uncertain factors such as available water quantity in the process of basin water resources allocation are defined and characterized by interval planning, stochastic planning and fuzzy planning methods. A two-stage interval water resources optimal allocation model based on reliability fuzzy constraint analysis is constructed. With the goal of system revenue maximization, different reliability and availability in different planning years are clarified. The main conclusions are as follows:

(1) To some extent, regional water demand reflects the degree of economic and social development in the Yalong River Basin. Liangshan district and Panzhihua district with high level of economic and social development are the main water consumption areas. By calculating the correlation degree between the influencing factors of water demand and the corresponding water demand series. Based on the results, the economic and social indicators with the relative low correlation degree (lower than the average correlation degree) are eliminated to ensure the validity of the input data of the neural network model. In addition, the different contribution degree and importance order of each economic and social index to water demand are obtained by using the characteristic average coverage rate, and the influence factors of the first 50% of contribution degree are screened. Combined with correlation analysis and importance ranking, we can better identify and select relevant economic and social indicators as the corresponding influencing factors of water demand.

(2) Unbiased grey model can predict the economic and social development data of five regions in the planning year, and test the error of 20 test samples of water demand prediction neural network model. The results show that the unbiased grey and neural network model can effectively predict the future water demand of five regions. Markov chain analysis method is used to modify the prediction value of neural network and get the prediction interval value. Compared with the average percentage error between the predicted value of neural network and the predicted value of neural network modified by Markov chain and the actual value, the average percentage error of 20 models modified by Markov chain model is smaller，it shows that the improved model is more accurate. On this basis, the prediction results of water demand of neural network in 2019-2030 are modified.

(3) The allocation and shortage of water resources are affected by the availability level and reliability level of water resources. The higher the availability level of water resources, the less the shortage; the higher the reliability level, the less the configurable water, and the greater the shortage. Different confidence levels will lead to different system benefits. In the same planning year, the increase of confidence level will reduce the system benefits. In 2030, there will be a large water shortage in the basin, especially in the agricultural sector. In the future, we should increase the development of water conservancy facilities and improve the efficiency of water conservation. The uncertain planning model of water resources can deal with many uncertain information in water resources system, quantify the reliability fuzzy relationship of system constraints, provide support for basin water resource managers to make scientific decisions, and achieve the goal of maximizing system revenue.