【摘要】：大伙房水庫作為渾河中上游上的控制性骨干工程,其防洪意義尤為重要。大伙房水庫現(xiàn)行的洪水預(yù)報模型為大伙房模型(DHF),該模型為集總式模型,由于集總式模型的特性其不考慮水文過程,因此存在對各支流的流量及水位等不能有效描述的問題,而且該模型的參數(shù)在選擇上是通過優(yōu)選法選定或人工試錯法確定的,并需要實時校正,因此對率定工作者的要求高且率定工作較為繁瑣。鑒于以上不足本文綜合考慮資料數(shù)據(jù)的限制、輸入數(shù)據(jù)、邊界條件和流域特征等客觀因素,建立了一種半分布式BP神經(jīng)網(wǎng)絡(luò)洪水預(yù)報模型,實際應(yīng)用中觀察發(fā)現(xiàn)其易陷入局部最小點,且預(yù)報時長較短,因此隨后在此基礎(chǔ)上對其做出改進,建立了DHF-GA-BP神經(jīng)網(wǎng)絡(luò)耦合洪水預(yù)報模型,并將建立的模型應(yīng)用于大伙房壩址以上流域進行洪水預(yù)報,該改進模型結(jié)合了大伙房模型與BP神經(jīng)網(wǎng)絡(luò)的優(yōu)勢,并引入遺傳算法,彌補了半分布式BP神經(jīng)網(wǎng)絡(luò)模型的不足。本文研究的主要內(nèi)容及相應(yīng)結(jié)果如下：(1)建立模型前對壩址以上流域進行分區(qū)。分區(qū)時以距離各個水文站最近的自然流域分水線為界,采用DEM數(shù)據(jù)與ArcGIS軟件最終劃分出Ⅰ、Ⅱ和Ⅲ三個區(qū)域,北口前水文站所在區(qū)為Ⅰ區(qū),占貝水文描述站所在區(qū)為Ⅱ區(qū),南章黨水文站所在區(qū)為Ⅲ區(qū)。(2)對將要輸入模型中的現(xiàn)有數(shù)據(jù)進行預(yù)處理。利用逐步回歸分析法對初始數(shù)據(jù)進行篩選,保證輸入的數(shù)據(jù)對輸出結(jié)果有顯著影響。篩選結(jié)果Ⅰ、Ⅱ區(qū)均為8個顯著因子、Ⅲ區(qū)為6個顯著因子、入庫斷面即全流域為11個顯著因子。(3)建立半分布式BP神經(jīng)網(wǎng)絡(luò)模型對大伙房水庫進行實時洪水預(yù)報。預(yù)報模型分為兩部分,第一部分為子流域流量預(yù)報,第二部分為入庫斷面流量預(yù)報,第一部分的輸出結(jié)果為第二部分的輸入數(shù)據(jù)。結(jié)果顯示,子流域預(yù)報和入庫斷面預(yù)報的結(jié)果較好且該模型對子流域內(nèi)的洪水調(diào)度也可起到輔助作用,但在預(yù)報精度、模型效率以及預(yù)報時長方面仍有待提高,應(yīng)對其改進。(4)改進模型建立前預(yù)先應(yīng)用遺傳算法率定大伙房模型參數(shù),使其可以分別應(yīng)用于各子流域的洪水預(yù)報。檢驗期預(yù)報結(jié)果顯示,預(yù)報效果較好,部分場次雖有不合格現(xiàn)象但整體預(yù)報水平仍可滿足正式預(yù)報需求。(5)針對半分布式BP神經(jīng)網(wǎng)絡(luò)模型的缺陷進行改進。應(yīng)用遺傳算法優(yōu)化原BP神經(jīng)網(wǎng)絡(luò)的初始權(quán)重及閾值,結(jié)合遺傳算法率定參數(shù)后的大伙房模型的預(yù)報成果,形成改進的半分布式BP神經(jīng)網(wǎng)絡(luò)模型——DHF-GA-BP神經(jīng)網(wǎng)絡(luò)耦合模型。預(yù)報結(jié)果顯示,改進后的洪水預(yù)報模型在運行效率和預(yù)報精度上都較原模型有所提高,延長了預(yù)報時長,且在一定程度上避免了遺傳算法率定參數(shù)后的大伙房模型預(yù)報誤差的累積。
[Abstract]:Dahuofang Reservoir is a controlled backbone project in the middle and upper reaches of Hunhe River, and its flood control significance is particularly important. The current flood forecasting model of Dahuofang Reservoir is the Dahuofang model (DHF), which is a lumped model. Because of the characteristics of the lumped model, it does not consider the hydrological process. Therefore, there is a problem that the flow and water level of each tributary can not be effectively described, and the parameters of the model are determined by optimal selection or artificial trial and error method, and need to be corrected in real time. Therefore, the requirements for calibration workers are high and the calibration work is more tedious. In view of the above shortcomings, considering the limitations of data, input data, boundary conditions and watershed characteristics, a semi-distributed BP neural network flood forecasting model is established in this paper. In practical application, it is found that it is easy to fall into the local minimum point and the forecast time is short. Therefore, on this basis, the DHF-GA-BP neural network coupling flood forecasting model is established. The model is applied to flood forecasting above Dahuofang dam site. The improved model combines the advantages of Dahuofang model and BP neural network, and introduces genetic algorithm to make up for the shortcomings of semi-distributed BP neural network model. The main contents and corresponding results of this paper are as follows: (1) before the establishment of the model, the watershed above the dam site is divided. Taking the natural watershed waterline nearest to each hydrologic station as the boundary, the DEM data and ArcGIS software are used to divide the three areas of 鈪，

本文編號：2472963