水電站群中長期徑流預(yù)報及發(fā)電優(yōu)化調(diào)度的智能方法應(yīng)用研究
發(fā)布時間:2018-07-17 19:21
【摘要】:近些年來,我國的水電建設(shè)事業(yè)取得了突飛猛進的發(fā)展,形成了一批具有電站級數(shù)多、裝機規(guī)模大等特點的梯級水電站群。大規(guī)模水電站群的不斷投入運行,對水電系統(tǒng)運行管理問題構(gòu)成了嚴峻挑戰(zhàn)。預(yù)報和調(diào)度是水電系統(tǒng)運行管理中的兩大核心問題,其中,及時可靠的徑流預(yù)報信息是科學(xué)制定水電調(diào)度方案的重要依據(jù),而合理的調(diào)度方案則是充分利用水能資源的重要保障。對于徑流預(yù)報問題,模型參數(shù)率定效率及精度是評價模型性能的關(guān)鍵指標,也是當(dāng)前研究徑流預(yù)報所面臨的難點;對于調(diào)度問題,隨著梯級電站級數(shù)的不斷增多,精細化調(diào)度對于梯級水電站群的意義越來越大,梯級間的水流滯時對中期發(fā)電調(diào)度的影響不容忽視。因此,研究如何提高徑流預(yù)報精度和發(fā)電優(yōu)化調(diào)度的求解效率以及合理評價因水流滯時產(chǎn)生的滯時電量問題,對于提升水電系統(tǒng)運行管理水平具有重要意義。本文以我國南方的水電站群為工程背景,針對中長期徑流預(yù)報問題,深入研究了具有較高預(yù)報精度的人工智能技術(shù)建模方法:針對水電站群調(diào)度問題,研究了并行智能算法求解技術(shù),同時研究了考慮水流滯時的水電系統(tǒng)中期發(fā)電優(yōu)化調(diào)度問題。主要成果如下:(1)針對前饋神經(jīng)網(wǎng)絡(luò)在徑流預(yù)報中常采用梯度下降的參數(shù)率定方法,存在計算耗時長和容易陷入局部最優(yōu)等缺陷,建立了基于極端學(xué)習(xí)機算法(ELM)的小波神經(jīng)網(wǎng)絡(luò)(WNN-ELM)中長期徑流預(yù)報模型。該模型結(jié)合小波分析強大的數(shù)學(xué)分析功能,首先利用atrous小波變換對原數(shù)據(jù)序列進行分解,然后將小波分解系數(shù)作為單隱層前饋神經(jīng)網(wǎng)絡(luò)模型(SLFNs)的輸入,實際徑流數(shù)據(jù)作為輸出建立模型;同時充分利用ELM算法極快的學(xué)習(xí)速度和良好的泛化能力,將其應(yīng)用于小波神經(jīng)網(wǎng)絡(luò)模型參數(shù)率定。以我國西南地區(qū)的漫灣和洪家渡兩座水電站的月徑流預(yù)報為例進行驗證,并與SLFNs-ELM模型和支持向量機(SVM)進行比較,結(jié)果表明,SLFNs-ELM模型峰值預(yù)測精度略優(yōu)于SVM,而WNN-ELM模型的預(yù)報精度明顯優(yōu)于SLFNs-ELM 和 SVM模型。(2)針對回聲狀態(tài)網(wǎng)絡(luò)(ESN)在徑流預(yù)報中因采用線性回歸率定模型參數(shù)容易出現(xiàn)過擬合問題,建立了基于貝葉斯回聲狀態(tài)網(wǎng)絡(luò)(BESN)的日徑流預(yù)報模型。該模型將貝葉斯理論與ESN模型相結(jié)合,通過權(quán)重后驗概率密度最大化而獲得最優(yōu)輸出權(quán)重,提高了模型的泛化能力。通過安砂和新豐江兩座水庫日徑流預(yù)測實例表明,BESN模型是一種有效、可行的預(yù)測方法,與傳統(tǒng)BP神經(jīng)網(wǎng)絡(luò)和ESN模型對比,進一步表明BESN模型具有更好的預(yù)測精度。(3)針對差分演化算法(DE)求解水電站群優(yōu)化調(diào)度問題時易陷入局部最優(yōu)問題,同時為提高DE算法計算效率,建立了水電站群長期優(yōu)化調(diào)度并行混合差分演化算法(PCSADE)。首先,結(jié)合混沌理論隨機性和遍歷性強等優(yōu)點,利用tent映射生成DE算法的初始種群并實現(xiàn)對控制參數(shù)(縮放因子和交叉因子)進行動態(tài)調(diào)整;其次,利用模擬退火算法,依據(jù)Metropolis準則改進DE算法的局部搜索能力:最后,基于Fork/Join并行框架技術(shù),實現(xiàn)了對改進算法(CSADE)的并行化計算,并測試了不同種群規(guī)模下PCSADE算法的性能。以紅水河流域梯級水電站群長期優(yōu)化調(diào)度為應(yīng)用實例,表明所提混合算法CSADE具有較好的尋優(yōu)能力,而所提出的并行混合算法PCSADE不僅能大幅度提高求解效率,而且提高了求解質(zhì)量。(4)針對梯級電站因最上游和最下游距離較遠而存在中期調(diào)度水流滯時問題,建立了考慮滯時電量的期末蓄能最大模型,并采用基于兩階段次梯度法乘子更新策略的拉格朗日松弛法進行模型求解。首先利用逐步優(yōu)化算法求解發(fā)電量最大模型,以確定調(diào)度期內(nèi)的系統(tǒng)出力過程;然后建立期末蓄能最大模型,根據(jù)系統(tǒng)負荷需求,利用拉格朗日松弛法進行模型求解,并在建立對偶問題的基礎(chǔ)上,以逐次逼近算法求解對偶問題,以兩階段次梯度法進行乘子更新。以瀾滄江中下游梯級水電站群中期優(yōu)化調(diào)度為例,將所提模型與未考慮滯時電量的模型進行對比,結(jié)果表明,中期滯時電量對計算結(jié)果有一定影響,中期調(diào)度需充分考慮調(diào)度結(jié)果的后效性。最后對全文做了總結(jié),并對有待深入研究的問題作出了展望。
[Abstract]:In recent years , China ' s hydropower construction has made a rapid development , and has formed a series of cascade hydropower stations with the characteristics of large number of power stations and large installed scale . The continuous operation of large - scale hydropower stations has posed a serious challenge to the operation and management of hydropower systems .
In this paper , with the increasing number of cascade power stations and the increasing number of cascade hydropower stations , the effect of refined scheduling on the cascade hydropower station is more and more important . In this paper , we study how to improve the accuracy of runoff forecast and solve the problem of long - term power generation scheduling in hydropower system .
The results show that the forecasting accuracy of the model is better than that of the SLFNs - ELM model and the support vector machine ( SVM ) .
Secondly , by using simulated annealing algorithm , the local searching ability of the DE algorithm is improved according to the Metropolis criterion : finally , the parallel computation of the improved algorithm ( CSADE ) is realized on the basis of the parallel frame technique .
Then the maximum model of the end - cycle energy storage is established . Based on the system load demand , the Lagrangian relaxation method is used to solve the dual problem , and the dual - phase gradient method is used to solve the dual problem .
【學(xué)位授予單位】:大連理工大學(xué)
【學(xué)位級別】:博士
【學(xué)位授予年份】:2015
【分類號】:TV737
[Abstract]:In recent years , China ' s hydropower construction has made a rapid development , and has formed a series of cascade hydropower stations with the characteristics of large number of power stations and large installed scale . The continuous operation of large - scale hydropower stations has posed a serious challenge to the operation and management of hydropower systems .
In this paper , with the increasing number of cascade power stations and the increasing number of cascade hydropower stations , the effect of refined scheduling on the cascade hydropower station is more and more important . In this paper , we study how to improve the accuracy of runoff forecast and solve the problem of long - term power generation scheduling in hydropower system .
The results show that the forecasting accuracy of the model is better than that of the SLFNs - ELM model and the support vector machine ( SVM ) .
Secondly , by using simulated annealing algorithm , the local searching ability of the DE algorithm is improved according to the Metropolis criterion : finally , the parallel computation of the improved algorithm ( CSADE ) is realized on the basis of the parallel frame technique .
Then the maximum model of the end - cycle energy storage is established . Based on the system load demand , the Lagrangian relaxation method is used to solve the dual problem , and the dual - phase gradient method is used to solve the dual problem .
【學(xué)位授予單位】:大連理工大學(xué)
【學(xué)位級別】:博士
【學(xué)位授予年份】:2015
【分類號】:TV737
【相似文獻】
相關(guān)期刊論文 前10條
1 崔清溪,王智中;用矩陣運算求解水電站群最優(yōu)裝機容量[J];長江水利教育;1993年02期
2 馬元s,
本文編號:2130678
本文鏈接:http://sikaile.net/kejilunwen/shuiwenshuili/2130678.html
最近更新
教材專著