پیش بینی خنک کننده بار توسط ترکیبی از تئوری مجموعه راف و شبکه عصبی مصنوعی بر اساس تکنیک همجوشی داده
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|29495||2006||14 صفحه PDF||سفارش دهید||4654 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Applied Energy, Volume 83, Issue 9, September 2006, Pages 1033–1046
A novel method integrating rough sets (RS) theory and an artificial neural network (ANN) based on data-fusion technique is presented to forecast an air-conditioning load. Data-fusion technique is the process of combining multiple sensors data or related information to estimate or predict entity states. In this paper, RS theory is applied to find relevant factors to the load, which are used as inputs of an artificial neural-network to predict the cooling load. To improve the accuracy and enhance the robustness of load forecasting results, a general load-prediction model, by synthesizing multi-RSAN (MRAN), is presented so as to make full use of redundant information. The optimum principle is employed to deduce the weights of each RSAN model. Actual prediction results from a real air-conditioning system show that, the MRAN forecasting model is better than the individual RSAN and moving average (AMIMA) ones, whose relative error is within 4%. In addition, individual RSAN forecasting results are better than that of ARIMA.
Because of significant amounts of energy consumed in heating, ventilating and air-conditioning (HVAC) systems, there have been efforts to study energy use performance and promote good operational practice in buildings. In order to improve the operation of HVAC systems, it is necessary to have reliable optimization routines . Accurate prediction of the dynamic air-conditioning load in a building is a key for the optimal control of a HVAC system. It is vital for adjusting the starting time of cooling to meet start-up loads, so minimizing or limiting the electric on-peak demand. Many techniques have been proposed in the last few decades for short-term load forecasting (STLF) . The traditional techniques that have been used for the STLF include autoregressive integrated moving-average (ARIMA) model , linear regression (LR) technique , neural-network (ANN) model , and grey model . However, no single one has performed well enough because each model can take a few or usually only one relevant factor into consideration. For the combined forecasting method , it cannot make use of the full capability of residual subspace. In addition, using all data would compromise the robustness of the prediction scheme. Recently, with the developments of artificial intelligence, alternative solutions to the STLF problem have been proposed. Expert systems have been successfully applied to STLF . For an accurate forecasting model, it is important to understand which factors influence the load level most. Such knowledge is often acquired from experienced operators. However, manual selection of factors pertinent to prediction task would not guarantee an optimal solution: the inaccurate estimations result in inaccurate predictions. Hence, there is a need for a reliable model that can select relevant factors automatically from historical data. As a typical data-mining method, rough-set theory, which can be used for attribute reduction, provides a solvable method for this problem. By attribute reduction, irrelevant factors to the tasks can be identified and removed. ANN, on the other hand, is a more promising area of artificial intelligence since it does not rely on human experience but attempt to learn by itself the functional relationship between system inputs and outputs. But ANN has two obvious shortcomings when applied to a large number of data  and . The first is that ANN requires a long time to train the huge amount of data of large data-bases. The second is that ANN lacks explanation facilities for their knowledge. The combination of rough sets and neural networks is a natural choice because of their complementary features. Therefore, a method of integrating RS theory and ANN (RSAN) is presented in the paper. The method consists of three stages. In the first procedure, RS will be applied to find relevant factors for the load. Then, relevant information will be used as input to the ANN. Finally, synthesizing multi-RSAN model (MRAN), using the data-fusion technique will be used to predict the load.
نتیجه گیری انگلیسی
A novel method of integrating RS and ANN models based on a data information technique was applied to the forecasting of the air conditioning cooling load, which is used in cooling-load estimation for the first time. Some conclusions are:- • RS can remove redundant attributes without any classification information loss: real test results show that only lesser measurements of many AHUs and FHUs can work well to forecast the cooling load. • It is easy to realize because the RS can relax, the burden of training and optimum principle is employed to deduce the weights of each RSAN model. • Results shows that general load-prediction model by synthesizing multi-RSAN can improve the accuracy and enhance the robustness of load-forecasting results because more effective information is used. • The MRAN forecasting model may be better than the individual RSAN and ARIMA ones, whose relative error is within 4%. In addition, individual RSAN forecasting results are also better than that of the ARIMA.