پيش بيني بارش روزانه استان كرمان با شبكه هاي عصبي مصنوعي (مطالعه موردي: كرمان، بافت و ميانده جيرفت)

Daily Precipitation Forcasting Using Artifical Neural Networks in the Province of Kerman:A Case Study of Stations of Kerman, Baft, and Miandeh Jiroft

اهداف: هدف اين پژوهش، پيش بيني بارش روزانه با استفاده از آمار روزانه هواشناسي ايستگاه هاي كرمان، بافت و ميانده جيرفت، طي دوره مشترك آماري 23 ساله (2012-1989) مي باشد. روش: براي دست يافتن به هدف تحقيق، به آموزش شبكه هاي عصبي مصنوعي با ساختار پرسپترون چند لايه و شبكه عصبي تابع پايه شعاعي پرداخته شد. تركيب هاي مختلف پارامترهاي كمينه دما، بيشينه دما، ميانگين دما، رطوبت نسبي، سرعت و جهت باد و نيز ميانگين فشار، به عنوان ورودي هاي شبكه هاي عصبي مصنوعي و بارش روزانه به عنوان خروجي شبكه درنظر گرفته شدند. يافته ها/ نتايج: نتايج نشان داد شبكه هاي عصبي مصنوعي پايه تابع شعاعي از دقت بسيار بيشتر و خطاي كمتري نسبت به شبكه عصبي پرسپترون، براي تخمين بارش روزانه در هر سه ايستگاه برخوردار هستند. نتيجه گيري: در بهترين تركيب با پارامتر هاي كمينه و بيشينه و حداقل دما و رطوبت نسبي، سرعت و جهت باد و نيز ميانگين فشار در ايستگاه كرمان با ضريب همبستگي 907/0 و جذر ميانگين مربعات خطاي 014/0، بهترين مدل پيش بيني بارش در اين تحقيق شناخته شدند.

1- INTRODUCTION Precipitation is one of the important parameters of climatology and atmospheric science that has high importance in human life. Precipitation forecasting has an important role in management and warning of this problem. Due to the interaction of various meteorological parameters in the calculation of rain, it leads it to a very irregular and chaotic process. Today, Artificial Neural Networks is a developed method that is applied for estimatationand prediction of parameters using intrinsic relations among data. 2- THEORETICAL FRAMEWORK Reliable forecasts form the basis of any warning system used as a non-structural means of Flood and Drought disaster mitigation. Many of the techniques used in the past for Daily Precipitation Forecasting are based on some assumed daily precipitation. Most of such applications have used the Multi Layer Perceptron (MLP) type ANN models coupled with the error Back Propagation (BP) algorithm, butthePropagation algorithm may beconverging to local minimum points in the parameter space.An alternative to the MLP is the Radial Basis Function (RBF) network which has linear parameters . An RBF network is a two-layer feed-forward type network in which the input is transformed by the basis functions at the hidden layer. At the output layer, linear combinations of the hidden layer node responses are added to form the output. The name RBF comes from the fact that the basis functions in the hidden layer nodes are radially symmetric. 3- METHODOLOGY In this paper Artificial Neural Networks (ANNs) as a reliable method for forecasting precipitation was used. For predication and analysis of effective factors in precipitation, Matlab7 software was employed. The inputs of ANNs model are; Minimum , Maximum , Mean Temperature , Relative Humidity , Direct ,Spread Wind and pressureat statistical periods,23 years (1989-2012). 4- RESULTS & DISCUSSION This paper presents a comparison between two Artificial Neural Network (ANN) approaches, namely, Multilayer Perceptron (MLP) and Radial Basis Function (RBF) networks, in Daily Precipitation Forecasting and precipitation was considered as the output of the Network. 80 percent of data were used for training the network and the 20 percent of rest for testing the model. The best network used in this paper is a "feed forward" type with sigmoid algorithm and Radial Basis Function with "spread 4" and "Goal 0" and precipitation data of Kerman, Baft ,Miandeh Jiroft stations used as the output of ANN model . The analysis of output resulting from Radial Basis Function Neural Networks shows that these models have better accuracy and a high ability to forecast precipitation than other Neural Networks. 5- CONCLUSION& SUGGESTIONS In this paper, the results of the analyses have shown that,with minimum and maximum parameters, minimum temperature and relative humidity, the wind speed and direction and mean of pressure in Kerman station with correlation coefficient 0.907 and root of mean square error of 0.014 are known as the best model for predicting rainfall.