【作者】 郑玉胜；

【导师】 黄介生；

【作者基本信息】 武汉大学 ， 系统工程， 2004， 硕士

【摘要】 本文以东风渠灌区为背景，对灌溉用水量预测的方法进行了探讨。 首先简要分析并总结了现有灌溉用水量预测方法，指出了现有方法的局限性。 然后介绍了时间序列中的线性随机模型，并用AR模型对季节性的时间序列进行了分析和预测，模型先对季节性的时间序列进行了预处理，使之成为稳定的序列，再通过AR模型分析求解。通过分析，发现系列本身的相关性较好，但由于模型本身的局限性，预测结果不够理想，其根本原因在于模型本身的线性性质。 由于上述线性随机模型的局限性，本文引入神经网络模型对灌溉用水量进行了预测。神经网络模型的一个主要优点是其强大的非线性映射能力。在时间序列的基础上，本文建立了BP网络模型。由于BP网络的传统算法不够理想，本文又引入了非线性最小二乘法中的LM算法，不仅提高了网络的精度，而且缩短了训练的时间。考虑到遗传算法具有良好的全局搜索特性，搜索到全局最优解的概率比起神经网络的算法来说要大得多，而在局部搜索方面则不如神经网络算法，本文最后尝试先用遗传算法优化初始权重、再用LM算法进行修正的方法建立了灌溉用水量的预报模型。采用该模型对东风渠灌区灌溉用水量进行了预测，得到了较好的结果。更多还原

【Abstract】 The forecast methods for irrigation water use were studied and a case study was carried out in Dongfenqu Irrigation District, Hubei Province. The methods used for the forecast of irrigation water use were first reviewed and the limits were analyzed.Then the linearity stochastic model was introduced and an AR model was used to forecast the monthly irrigation water use. Firstly, the historical data of irrigation water use were pretreated and the time series was made as steady series. Then the forecasting was carried out with AR model. The result showed that the AR model is not conpetent for the forecast of irrigation water use although the time series showed reasonable correlation. Linearity of the model may be the probable matter.Therefore the non-linear model was considered and a BP artificial neural network model was developed. Levenberg-Marquardt (LM) algorithm was used for reducing the training time and combined with genetic algorithm to search the globel optimization points. Using this model, a case study was carried out for Dongfengqu Irrigaiton District and the results were compared with the observed data. The results showed that the model could forecast the irrigation water use reasonably.更多还原