【正文】 偏置），中間是通過指數(shù)激活函數(shù)來傳遞。 這個模型包括了先前所以數(shù)據(jù)高達 p 的延遲，如上所示，這些數(shù)據(jù)不是獨立的，它與負荷有不用程度的相關性。 圖 一天中負荷變化的示例 混合神經(jīng)網(wǎng)絡 我們所研究的負荷預測系統(tǒng)由兩類網(wǎng)絡組成： ARMA 模型的線性神經(jīng)網(wǎng)絡和前饋非線性神經(jīng)網(wǎng)絡。例如：溫度變化對民用和商業(yè)負荷的影響大于它對工業(yè)負荷的影響。 為了說明這個基于實際情況的負荷預測系統(tǒng)的神經(jīng)網(wǎng)絡的性能，我們采用一個公共機構提供的實際需求數(shù)據(jù)來訓練系統(tǒng)，利用三年（ 1989， 1990， 1991）中每小時的數(shù)據(jù)來訓練這個神經(jīng)網(wǎng)絡，用 1992 年每小時的實際需求數(shù)據(jù)用來驗證整個系統(tǒng)。要用大量的歷史數(shù)據(jù)來訓練神經(jīng)網(wǎng)絡，以減小平均絕對誤 差百分比 (MAPE)。非線性神經(jīng)網(wǎng)絡常用來捕獲負荷與各種輸入?yún)?shù)（如歷史負荷值、氣象溫度、相關濕度等）間的高度 非線性關系。由于電力負荷和各種參數(shù)（天氣的溫度，濕度，風速等）之間的高度非線性的關系，無論在電力負荷預測建?；蛟陬A測中都有重要的作用。準確的負 荷預測對于高效的發(fā)電調(diào)度，開停機計劃，需求方的管理，短時維護安排或其他目的等是很必要的。非線性神經(jīng)網(wǎng)絡是用來捕獲負荷和各種輸入?yún)?shù)之間的高度非線性關系。 (2) Fast to retrain the system。a crossover point in both strings is selected randomly。 this is known as underfitting. Conversely, if the neural work is too large, then it can fit not only the underlying signal but also the noise in the training set。 finally, we describe the enhancement to our system in Section VII. Variables Afferting ShortTerm Load Some of the variables affecting shortterm electxical load are: Temperature Humidity Wind speed Cloud cover Length of daylight Geographical region Holidays Economic factors Clearly, the impacts of these variables depend on the type of load: variations in temperature, for example, have a larger effect on residential and mercial loads than on industrial load. Regions with relatively high residential loads will have higher variations in shortterm load due to weather conditions than regions with relatively high industrial loads. Industrial regions, however, will have a greater variation due to economic factors, such as holidays. As an example, Figure shows the loadvariation over one day, starting at midnight. Figure Example of load variation during one day Hybrid Neurak Networks Our shortterm load forecasting system consists of two types of works:linear neural work ARMA model and feedforward .Nonlinear neural nonlinear neural work is used to capture the highly nonlinear relation between the load and various input use the linear neural work to generate an ARMA model which will be mainly used to capture the load variation over a very short time period(one hour). Linear Neutal Networks The general multivariate linear model of order p with independent x,is tptpititttptpitittt uxcxcxc xcxczazazazaz ???? ????????????????? ? ??221102211 Where: tz electrical load at time t tx independent variable at time t tu random disturbance at time t ii ca, coefficients Linear neural works can successfully learn the coefficient and from the historrcal load data,and the independent variables,WidrowHoff has been used to determine the coefficient. This model includes all the previous data up to lag shown above ,these data are not independent ,and have varying degrees of correlation with the studies can be used to determine the most significant parameters to be includes in the model,allowing many to be reduces the size and puter time for a model of given accuracy,or increases the accuracy for a model of given size. NonLinear Neural Networks For nonlinear forecasting,a nonlinear model analogous to the linear model is: ? ? tptittttptitttt uxxxxxzzzzfz ?? ???????? , 2121 ??? where:f(.) is a nonlinear function determined by the artificial neural work. Layered, feedforward neural works are used, typically with one hidden layer (although in some cases with two). The layers are fully connected, with one bias unit in each layer (except the output layer). The output of each unit is the slum of the weighted inputs (including the bias), passed through an exponential activation fiinction. Our modiked backpropagation method is applied. The errors are defined to be the sum of the squares of the deviations between the puted values at the output units and the actual or desired values。 in Section III, we present the hybrid neural work used in our system。 Section I1 describes the variables sigdicantly affecting short term load forecasting。 and in Section VI, some simulation result is given。 the effects of these weights cancel somewhere downstream. The same is true for the hidden units. Therefore, in conventional backpropagation for nonlinear neural works, there is no automatic elimination of extraneous input nodes or hidden nodes. However, in practical forecasting it is necessary to achieve a parsimonious model, one which is neither too simple nor too plex for the problem at hand. If the neural work is chosen to be too small (to have too few input or hidden units), then it will not be flexible enough to capture ithe dynamics of the electrical demand system。 the putations in each step are greater but the number of iterations is greatly reduced. Reductions in training time are desirable not only to reduce putation costs, but to allow more alternative input variables to be investigated, and hence to optimize forecast accuracy. Determination of Network Structure As we stated above, the neural work used in load forecasting tends to be large in size, which results in longer training time. By carefully choosing work structure (., input nodes, output nodes), one will be able to build a relatively sma