【正文】 t of social economy. At present, it is very wide that studies on methods of short term forecast for urban water consumption based on stationary time series, and they have gained better effects. Because time series of urban middlelong term water consumption that is affected by lots of social factors, are characterized with tendency and randomness, forecast methods are more plicated and fewer are researched. This paper researched the method of middlelong term forecast for urban water consumption based on nonstationary time series.Methods: In terms of concerned theories of time series analysis, . nonstationary time series can be divided into the certain part and stochastic part, a forecast model of integrating Artificial Neural Network (ANN) with Auto Regressive Moving Average (ARMA) is presented. Aiming at nonstationary time series of urban middlelong term water consumption, their certain part denoting nonstationary trend can be fitted and forecasted by Momentum Back Propagation model of ANN, and their stochastic part denoting stationary ponent can be fitted and forecasted by ARMA model. The sum of their forecast values is considered as the forecast value of urban middlelong term water consumption. Results: Through ANNARMA model being applied to simulate and forecast middlelong term water consumption in one city, research results indicate: (1) relative errors between forecast values and practical values are all less than 6%。曾鳳章(1943)，女，漢族，北京人，北京理工大學教授，博士生導師，研究方向：工業(yè)工程。 middlelong term forecast。確定項可用與時間有關的確定性函數(shù)（如多項式、指數(shù)或正弦函數(shù)）擬合，表示時序的非平穩(wěn)趨向；隨機項表示平穩(wěn)的隨機成分，可用ARMA模型擬合。訓練后的BP網(wǎng)絡可以預測非平穩(wěn)時序未來的確定項。估計方法有矩估計、最小二乘估計、極大似然估計等。需要指出的是，ANNARMA模型具有一定的局限性。從表1可以看出：（1）ANNARMA模型的總預測值與實際值的相對誤差不超過6%。方程式（9）成為具有階自回歸部分、階滑動平均部分的模型。當或超過15時，可用正態(tài)分布來近似，并構造如下統(tǒng)計量進行檢驗： （1）式中，為游程的期望數(shù)， （2）為游程的標準差， （3）對于的顯著水平，若（按原則），則可接受（平穩(wěn)性假設）；否則，拒絕（接受非平穩(wěn)性）。1 基本思路目前，許多文獻的時間序列分析缺乏平穩(wěn)性檢驗，常常以平穩(wěn)假設為前提去應用ARMA模型；這雖然能夠降低問題的難度，但也因簡化了具有決定性影響的非線性因素，從而導致錯誤的結論[8]。因此，該集成模型應用于基于非平穩(wěn)時間序列的城市用水量中長期預測，具有科學性和可行性。 聯(lián)系方式：0106891 6003(固定)；136 8350 4662 Email：caifengbit caifengbit，曾鳳章(導師)(1北京理工大學 北京 100081)文摘:目的：面對城市水資源供需矛盾日益加劇的現(xiàn)狀，城市用水量預測已成為城市建設與水資源規(guī)劃工作的重要內容之一。 Artificial Neural Network (ANN)。由于函數(shù)擬合確定項的方法存在著函數(shù)選取較難、人為因素干擾較大等突出缺點，文獻[1213]等研究表明，ANN模型可以較好地擬合確定項，ARMA模型可以擬合殘差項，兩者預測的疊