據中(zhong)科(ke)院工程熱物理所消息,該單位研究人(ren)員應用機器學習在光伏熱電混合系統結(jie)構參數優化研究中(zhong)取得新進展。
熱電轉換(huan)(huan)技(ji)術和(he)光(guang)電轉換(huan)(huan)技(ji)術是(shi)利用能(neng)(neng)源轉換(huan)(huan)材料將太陽能(neng)(neng)直接(jie)轉換(huan)(huan)為電能(neng)(neng)的(de)兩種(zhong)主(zhu)要技(ji)術形式。二(er)者在提高太陽能(neng)(neng)利用率的(de)技(ji)術研(yan)究(jiu)發展方面的(de)關(guan)鍵點(dian),一(yi)是(shi)研(yan)究(jiu)和(he)開發高效熱電材料和(he)光(guang)伏材料及其組件(jian),二(er)是(shi)發展聚光(guang)型光(guang)伏熱電技(ji)術。
太陽能(neng)熱電(dian)(dian)(dian)(dian)(dian)光(guang)電(dian)(dian)(dian)(dian)(dian)復(fu)(fu)合(he)發電(dian)(dian)(dian)(dian)(dian)系統的核心(xin)是熱電(dian)(dian)(dian)(dian)(dian)光(guang)伏(fu)復(fu)(fu)合(he)發電(dian)(dian)(dian)(dian)(dian)單元,聚集太陽光(guang)入射后(hou),光(guang)伏(fu)電(dian)(dian)(dian)(dian)(dian)池首先利(li)(li)用光(guang)譜響應的波長(chang)產生(sheng)(sheng)電(dian)(dian)(dian)(dian)(dian)能(neng),不能(neng)被光(guang)伏(fu)電(dian)(dian)(dian)(dian)(dian)池利(li)(li)用的光(guang)譜能(neng)量通(tong)過熱電(dian)(dian)(dian)(dian)(dian)器件進行熱電(dian)(dian)(dian)(dian)(dian)轉(zhuan)換。復(fu)(fu)合(he)發電(dian)(dian)(dian)(dian)(dian)系統能(neng)夠更充分地(di)利(li)(li)用太陽能(neng)全光(guang)譜,并且將光(guang)伏(fu)電(dian)(dian)(dian)(dian)(dian)池不能(neng)利(li)(li)用的能(neng)量轉(zhuan)入熱電(dian)(dian)(dian)(dian)(dian)器件,不僅可(ke)以降低光(guang)伏(fu)電(dian)(dian)(dian)(dian)(dian)池工(gong)作時的溫度(du),還能(neng)產生(sheng)(sheng)額(e)外的電(dian)(dian)(dian)(dian)(dian)能(neng)。
光伏(fu)熱(re)(re)電(dian)復(fu)合系統(tong)的(de)發電(dian)效率不僅受太陽(yang)光聚光比、系統(tong)工作時溫度、外接(jie)負載大小等因素影響,光伏(fu)電(dian)池(chi)與熱(re)(re)電(dian)器件的(de)結構參數(shu)也會顯著影響混(hun)合系統(tong)的(de)輸出性(xing)能。為了研究這一參數(shu)的(de)具體影響,中科院工程熱(re)(re)物理(li)所新(xin)技術實驗室新(xin)能源材料(liao)與設備團隊研究人員選取(qu)單晶硅、砷化鎵太陽(yang)能電(dian)池(chi)和不同尺寸(cun)的(de)熱(re)(re)電(dian)器件進(jin)行(xing)了研究。
圖(tu)1 熱電器(qi)件(jian)輸出功率隨n值(zhi)的(de)變化(hua)
如圖1所示,在(zai)不同光伏電(dian)池與熱(re)(re)電(dian)器(qi)(qi)件(jian)面積比值(n值)下,熱(re)(re)電(dian)器(qi)(qi)件(jian)輸出(chu)功率會產生明(ming)顯的差異。經(jing)過(guo)優化研(yan)究人員(yuan)發現,出(chu)當二者面積比為4時(shi),熱(re)(re)電(dian)器(qi)(qi)件(jian)獲得(de)最大(da)的功率輸出(chu)。在(zai)混(hun)合(he)動力系(xi)(xi)統(tong)中(zhong),光伏電(dian)池和(he)熱(re)(re)電(dian)器(qi)(qi)件(jian)輸出(chu)的電(dian)能相互(hu)隔離(li)的情況下,熱(re)(re)電(dian)器(qi)(qi)件(jian)獲得(de)最大(da)輸出(chu)功率,混(hun)合(he)系(xi)(xi)統(tong)的輸出(chu)性能最佳。
圖2 基(ji)于機器學習對系統組(zu)件結構尺寸(cun)的優化
圖3 機器學習預測結果的實(shi)驗驗證
獲得(de)實驗數據(ju)后,研(yan)究人(ren)員選用機器(qi)學習方(fang)法,對混合系統光伏(fu)與熱(re)電器(qi)件的結構參(can)數再次(ci)進(jin)行(xing)優化(hua),通過對實驗數據(ju)的整理(li)、訓(xun)練、學習,建立DNN、LSTM、LSTMA三種模(mo)型(xing)進(jin)行(xing)預測,經與實驗結果(guo)對比,LSTM模(mo)型(xing)的準確率(lv)更高。訓(xun)練出的模(mo)型(xing)進(jin)一步預測n值為4.41時可獲得(de)最優輸出性能(neng)。
這項(xiang)研(yan)究(jiu)通過對光伏熱電混(hun)合系(xi)統的(de)優化,探討(tao)了(le)機器學習在結構參數優化方(fang)面的(de)應(ying)用(yong)(yong),為(wei)其(qi)在能量轉換系(xi)統性能提(ti)升的(de)應(ying)用(yong)(yong)方(fang)面提(ti)供了(le)實例和(he)參考。該項(xiang)研(yan)究(jiu)成果在Engineered Science期刊(kan)上(shang)發表,被選為(wei)封面論(lun)文(wen),張航研(yan)究(jiu)員(yuan)為(wei)論(lun)文(wen)通訊作(zuo)者,博士生(sheng)何澤明和(he)副(fu)研(yan)究(jiu)員(yuan)楊明為(wei)共同第一作(zuo)者。