張晨晨 叢意林 田野 郭安柱 劉濤 馬永志



摘要: ?針對單一的預測方法難以綜合描述冷負荷變化的規律性問題,本文以初投入使用的青島市某自習室空調系統為研究對象,對基于改進算法的空調冷負荷組合預測進行研究。為獲得動態負荷數據,搭建了TRNSYS模擬仿真平臺,對擾動因子經平均影響值(mean impact value,MIV)和Spearman相關性分析及特征變量篩選后,對預測算法進行優化。通過引入隨機粒子和混沌算法,建立基于標準粒子群算法的組合粒子群算法(combined particle swarm optimization, CPSO),得到組合粒子群優化后向傳播網絡(back propagation, BP)負荷預測模型CPSOBP,并引布谷鳥搜索(cuckoo search,CS),確立布谷鳥搜索支持向量回歸(support vector regression,SVR)負荷預測模型CSSVR,建立基于遺傳尋優的灰色預測模型GAGM(1,N)。同時,將各模型的負荷預測值帶入模糊系統中,建立實時模糊組合預測模型(fuzzy combination,FC),并采用Markov(M)對組合誤差進行修正。結果表明,基于Markov的模糊組合預測算法FCM優于CPSOBP、CSSVR和FC,組合精度與3個優化模型相比分別提高了26.32%,62.16%,94.68%,說明基于馬爾可夫的模糊組合預測算法FCM可以彌補各算法的不足,降低了預測誤差,提高了預測準確率。該研究為空調節能運行策略的制定提供了理論參考。
關鍵詞: ?模糊系統; GAGM(1,N); CPSOBP; CSSVR; Markov
中圖分類號: TP391.9; TU831.6 文獻標識碼: A
世界能源需求的增加帶來了能源消耗的激增[1]。由于建筑工程約占世界能源消耗的30%[2],而采暖、通風和空調系統(heating ventilation and air conditioning,HVAC)在建筑能耗中所占比例最大[3],且與其他部分能源消耗相比,通過將能源供應與實際負荷需求相匹配[4],減少能源消耗具有更大的潛力。因此,提高暖通空調的運行效率對降低能耗至關重要,而準確預測冷卻負荷在此意義重大[5]。全球氣候和人類生命行為的復雜性,導致空調負荷呈現非線性、多變性和動態性的特點[3],這對空調負荷的預測精度提出了更高的要求。負荷預測的方法包括物理建模法、參數模型和非參數模型法。物理建模是利用傳熱機制搭建模擬平臺,但是其無法保證實時性[6-9];參數模型是通過分析影響因素與冷負荷之間的關系,建立數學模型或統計模型,統計模型的方法主要包括統計回歸[10]和時間序列,統計回歸算法結構簡單,但評價指標難以確定,時間序列通過分析歷史負荷的規律性以預測冷負荷[11],其只用于負荷均勻變化的系統[12]。非參數模型因其囊括智能算法而受到廣泛關注,主要有決策樹[13]、灰色預測[14]、遺傳算法[15]、粒子群算法[16]、布谷鳥算法[17]、神經網絡[18]和支持向量機[19]。決策樹算法又稱判定樹,是多分枝有向、無環的樹狀結構,算法效率高,計算量小,但處理不好時間序列與非線性數據;灰色預測(Grey)在訓練參數較少時,可得到較為準確的預測結果,但對隨機性強,離散度大的建筑負荷,預測精度低[20];支持向量機(support vector machine, SVM)泛化能力強,在解決維度災難問題和局部最小問題上有天然的優勢,結構簡單,魯棒性強,但不適用于大量樣本;BP神經網絡具有強大的非線性映射能力和自學習能力,但其容易陷入局部極小,收斂速度慢,對網絡初值樣本數量較為敏感,對復雜的非線性問題預測精度低。因此,許多學者提出了改進算法。Wei L Y等人[21]提出了自適應期望遺傳算法,優化自適應網絡模糊推理系統,并通過對比證實了模型的有效性;D. Sedighizadeh等人[22]提出了一種結合隨機最優粒子的廣義粒子群優化算法,與其他混合粒子群算法在均值和標準差方面均體現了優越性;N. Kumar等人[16]提出了一種基于改進布谷鳥搜索(cuckoo search)算法和自適應高斯量子行為粒子群優化算法的混合算法;Li D L等人[23]采用自適應PSOSVM方法,建立新的自適應短期負荷預測模型,自適應PSOSVM方法預測精度高,泛化能力強,可行性強;D. Tien Bui等人[24]建立了遺傳算法和帝國主義競爭算法,優化人工神經網絡在節能住宅熱負荷和冷負荷估算中的權值和偏差,取得了較好的預測精度。而單一的混合算法很難表現出優化模型的全部信息,單一的預測方法難以綜合描述冷負荷變化的規律性。因此,本文采用多個混合算法,分別優化各個預測模型的參數,再將各預測模型放入模糊推理系統,分段動態地提取組合權重,并將各預測模型組合起來,同時考慮到模擬負荷過程中產生的隨機誤差,采用馬爾科夫鏈對誤差進行了修正,降低了預測誤差,提高了預測準確率。
1數據來源與處理
1.1TRNSYS模擬平臺
本文以初投入使用的青島市某自習室空調系統為研究對象,基于Trnsys動態仿真平臺,獲得了動態逐時負荷,并對多功能自習室負荷模擬參數進行設置。青島市多功能自習室負荷模擬參數如表1所示。
1.2輸入變量篩選
本文采用MIV與spearman系數結合的方式,提取外擾和內擾特征變量因素反復計算,取MIV均值絕對值,選擇貢獻率大的成分,再充分考慮自習室內的負荷,呈周期性變化的歷史負荷對當前時刻t負荷的影響,以及內擾和外擾的延遲作用,經過試錯法反復比較,進而計算不同時刻每個成分的spearman系數,最終選擇確定度大于0.6的成分作為輸入。
2組合預測模型
2.1CPSOBP預測
粒子群搜索BP網絡最優的閾值和權值初值,以提高BP對初值的敏感度。針對標準粒子群收斂慢、易早熟的問題,引入改進算法。本文首先改進速度更新公式,再引進混沌算法流程,形成組合算法。
1)改進粒子群。改進的粒子群為
2)混沌算法。混沌映射具有隨機性和遍歷性的特點,將最優解映射到logistic方程的定義域[0,1]中,經過有限次迭代得到混沌序列后,將其逆映射到原解空間,計算得到混沌序列可行解的適應度值,保留混沌最優可行解。CPSOBP結構流程圖如圖1所示。
2.2CSSVR預測
核函數參數和正則化系數是控制SVR預測精度的關鍵。CS算法具有搜索能力強和搜索路徑優的特點,對SVR的核參數和正則系數尋優能夠有效的提高精度。CS算法通過維持Levy飛行產生隨機解[16],即
2.3GAGrey預測
灰色模型通過將原始序列轉變為規律性,弱化原數據的隨機性,深入挖掘預測對象的演化規律。參數a和參數b影響灰色預測結果,當矩陣接近退化時,最小二乘法求參預測精度低。本文采用遺傳算法代替最小二乘法求解參數優化模型。
GA通過選擇、交叉和變異完成進化過程,是一種高效的全局優化算法。采用遺傳算法優化灰色模型參數,GAGrey結構流程圖如圖3所示。
2.4組合預測
不同偏差的預測模型反應不同信息,組合預測將各模型的有效信息整合優化,得到最優解的近似解。傳統的權重分配未考慮權重的動態特性,不同段各優化預測模型的有效信息不同,因此將權重分段分配,分段提取有效信息。將預測結果模糊化,并根據模糊規則建立自適應模糊組合預測模型。模糊推理數據列表如表2所示。
3馬爾可夫鏈誤差修正
4案例分析
4.1組合預測
模糊系統組合優化模型,馬爾可夫修正組合結果算法流程如圖4所示,各模型相對誤差分布如圖5所示,兩種組合預測方式的相對誤差分布如圖6所示。由圖5可以看出,各優化和組合后的預測模型,其性能更佳,比PSOBP模型精度提高26.32%,比CSSVR的預測精度提高62.16%,比GAGrey預測精度提高94.68%,且優于線性組合模型;由圖6可以看出,各優化和組合后的預測模型依舊存在峰值誤差,因此馬爾科夫系統可以對誤差進行修正。
4.2誤差修正
按照聚類原理,將45個時間點相對誤差數據確定為6個中心,根據中心劃分成6個狀態區間,Kmeans計算聚類中心和狀態區間劃分結果如表3所示,各點所屬狀態區間分布如圖7所示,修正前后相對誤差對比如圖8所示。
由圖8可以看出,馬爾可夫修正后,在7月27日~29日這3天中,每天分別有76.47%,92.86%,85.71%個時刻的預測性能均有所提高,誤差峰值大大降低,修正后的模型FCM比組合模型FC的預測精度提高57.14%。
采用平均絕對誤差(mean absolute error,MAE)和均方根誤差(root mean aquare error,RMSE)對優化預測和修正結果進行綜合評價。修正前后性能對比如圖9所示。由圖9可知,通過修正前后性能對比,FCM預測模型的RMSE和MAE均小于各優化預測模型。
5結束語
本文以初投入使用的青島市某自習室空調系統為研究對象,主要對基于改進算法的空調冷負荷組合預測進行研究。以自然啟發的CS,CPSO,GA全局優化算法為基礎,以神經網絡BP,SVR,Grey為主體,分別建立了CPSOBP優化預測模型、CSSVR優化預測模型和GAGrey優化預測模型,基于模糊理論將3個優化預測模型帶入模糊系統中,從而建立了動態馬爾可夫組合預測模型FCM,最后將組合預測模型應用于空調系統的冷負荷預測案例中,修正后的模型FCM比組合模型FC的預測精度提高了57.14%,驗證了本文所提出算法的有效性,由預測誤差分析可知,本文預測算法精度較高。該研究為空調的節能運行策略提供了具有實際意義的參考。
參考文獻:
[1]Sarbu I, Adam M. Experimental and numerical investigations of the energy efficiency of conventional air conditioning systems in cooling mode and comfort assurance in office buildings[J]. Energy and Buildings, 2014, 85: 4558.
[2]Kilic U, Tugrul A B. Identification and analysis of risks associated with gas supply security of turkey[C]∥Energy Systems and Management. Istanbul: Springer International Publishing, 2015: 241252.
[3]Shi H, Xu M H, Li R. Deep learning for household load forecastinga novel pooling deep RNN[J]. IEEE Transactions on Smart Grid, 2018, 9(5): 52715280.
[4]Jing Z, Liu X. A hybrid method of dynamic cooling and heating load forecasting for office buildings based on artificial intelligence and regression analysis[J]. Energy and Buildings, 2018, 174: 293398.
[5]Fan C, Xiao F, Zhao Y. A shortterm building cooling load prediction method using deep learning algorithms[J]. Applied Energy, 2017, 195: 222233.
[6]Li N, Wang K, Cheng J. A research on a following day load simulation method based on weather forecast parameters[J]. Energy Conversion and Management, 2015, 103(6): 691704.
[7]Li Z, Huang G. Reevaluation of building cooling load prediction models for use in humid subtropical area[J]. Energy and Buildings, 2013, 62: 442449.
[8]Dahanayake K W D K C, Chow C L. Studying the potential of energy saving through vertical greenery systems: Using EnergyPlus simulation program[J]. Energy and Buildings, 2017, 138: 4759.
[9]Lim H S, Kim G. Prediction model of cooling load considering timelag for preemptive action in buildings[J]. Energy and Buildings, 2017, 151: 5365.
[10]Asadi S, Hassan M, Beheshti A. Development and validation of a simple estimating tool to predict heating and cooling energy demand for attics of residential buildings[J]. Energy and Buildings, 2012, 54: 1221.
[11]Zhou C G, Fang Z S, Xu X N, et al. Using long shortterm memory networks to predict energy consumption of airconditioning systems[J]. Sustainable Cities and Society, 2020, 55: 1020.
[12]Brockwell P J, Davis R A, Berger J O, et al. Time series: theory and methods[M]. Berlin: SpringerVerlag, 1987.
[13]Yu Z, Haghighat F, Fung B C M, et al. A decision tree method for building energy demand modeling[J]. Energy and Buildings, 2010, 42(10): 16371646.
[14]Harb H, Boyanov N, Hernandez L, et al. Development and validation of greybox models for forecasting the thermal response of occupied buildings[J]. Energy and Buildings, 2016, 117: 199207.
[15]Mauro C, Ales P, Leonardo T, et al. Prediction of energy performance of residential buildings: A genetic programming approach[J]. Energy and Buildings, 2015, 102: 6774.
[16]Kumar N, Shaikh A A, Mahato S K, et al. Applications of new hybrid algorithm based on advanced cuckoo search and adaptive Gaussian quantum behaved particle swarm optimization in solving ordinary differential equations[J]. Expert Systems with Applications, 2021, 172: 118.
[17]Yang X S, Deb S. Cuckoo search: recent advances and applications[J]. Neural Computing and Applications, 2014, 24(1): 169174.
[18]Deb C, Eang L S, Yang J J, et al. Forecasting diurnal cooling energy load for institutional buildings using Artificial Neural Networks[J]. Energy and Buildings, 2016, 121: 284297.
[19]Jain R K, Smith K M, Culligan P J, et al. Forecasting energy consumption of multifamily residential buildings using support vector regression: Investigating the impact of temporal and spatial monitoring granularity on performance accuracy[J]. Applied Energy, 2014, 123: 168178.
[20]Zhang Y, Fang C, Li T J, et al. Research on shortterm load forecast of shopping mall based on similar day selection and bp neural network[C]∥Proceedings of 2018 3rd International Conference on Automation, Mechanical and Electrical Engineering. Shanghai: DEStech, 2018: 289295.
[21]Wei L Y. A hybrid model based on ANFIS and adaptive expectation genetic algorithm to forecast TAIEX[J]. Economic Modelling, 2013, 33: 893899.
[22]Sedighizadeh D, Masehian E, Sedighizadeh M, et al. GEPSO: A new generalized particle swarm optimization algorithm[J]. Mathematics and Computers in Simulation, 2021, 179: 194212.
[23]Li D L, Zhang X F, Qiao M Z, et al. A shortterm load forecasting method of warship based on PSOSVM method[C]∥Applied Mechanics and Materials. Wuhan: Trans Technology Publications LTD, 2012: 569574.
[24]Tien D, Bui H, Moayedi D, et al. Predicting heating and cooling loads in energyefficient buildings using two hybrid intelligent models[J]. Applied Sciences, 2019, 9(17): 35433568.
作者簡介: ?張晨晨(1994),女,碩士研究生,主要研究方向為優化算法對空調負荷的預測。
通信作者: ?馬永志(1972),男,博士,副教授,主要研究方向為大數據與云計算技術。 Email: hiking@126.com
Research on Combined Forecasting of Air Conditioning Cooling Load Based on Improved Algorithm
ZHANG Chenchen, CONG Yilin, TIAN Ye, GUO Anzhu, LIU Tao, MA Yongzhi
(College of Mechanical and Electrical Engineering, Qingdao University, Qingdao 266071, China)
Abstract: ?In order to solve the problem that it is difficult to comprehensively describe the regularity of cooling load change with a single forecasting method, this article takes the cooling load in a study room in Qingdao, China, which has been put into use for the first time, as the research object, and establishes a TRNSYS simulation platform to obtain sufficient dynamic load data. After using the mean influence value (MIV) and Spearman correlation coefficient to screen the characteristic variables, the prediction models are optimized: the random particle and chaos algorithm are introduced to establish the combined particle swarm optimization (CPSO) algorithm based on standard particle swarm optimization (PSO) algorithm. This is done in order to optimize back propagation (BP) and establish CPSOBP forecasting model;The cuckoo search support vector regression (CSSVR) forecasting model is established by introducing cuckoo search (CS);The grey prediction model GAgrey (1, N) based on genetic optimization(GA) is established; Load prediction values of each model are brought into the fuzzy system to establish the realtime fuzzy combination (FC) model. Finally, Markov(M) is used to correct the combination error. The results show that FCM is superior to CPSOBP, CSSVR and FC, and accuracy is respectively, 26.32%, 62.16%, 94.68% higher than the three optimization models. It gives full play to the advantages of each algorithm, makes up for the shortcomings of each algorithm, and greatly reduces the prediction error, increases the reliability of forecasting system. This study provides a theoretical reference for the formulation of energysaving operation strategy of air conditioning.
Key words: fuzzy system; GAGM(1, N); CPSOBP; CSSVR; Markov