بررسي تاثير تعديل دامنه تغييرات داده ها بر كارايي مدل شبكه عصبي مصنوعي و درخت تصميم رگرسيوني در پيش بيني خشكسالي

Evaluation of the Effects of Data range Modification on Efficiency of Regression Decision Tree and Artificial Neural Networks for Drought Prediction

خشكسالي يكي از اثرات تغيير سامانه اقليمي است. پيش بيني خشكسالي نقش مهمي در اعمال روش هاي موثر مديريت منابع آب ايفا مي كند. روش هاي مختلفي براي مطالعه خشكسالي وجود دارد. روش تحليل داده هاي بارندگي، جز عمومي روش‌هاي تحليل خشكسالي به شمار مي‌رود. لذا پيش بيني دقيق و پيش از وقوع بارش مي تواند شرايط را براي ارزيابي وضعيت خشكسالي فراهم نمايد. هدف از اين پژوهش، بررسي تا ثير پيش پردازش داده ها بر عملكرد دو مدل داده كاوي در پيش بيني خشكسالي در ايستگاه سينوپتيك يزد مي باشد. در اين رابطه از دو روش شبكه عصبي مصنوعي و درخت تصميم رگرسيوني كه از انواع روش هاي داده كاوي محسوب مي شود استفاده شد و شبيه سازي ها در دو حالت كلي صورت گرفت. در حالت اول، از مقادير اصلي برخي پارامترهاي اقليمي استفاده و ميزان بارش 12 ماه پيش از وقوع پيش بيني گرديد. در حالت دوم ميانيگن متحرك سه ساله همان داده ها به مدل معرفي و پيش بيني بر همين اساس انجام شد. در پايان براي ارزيابي دقت و درستي دو روش مورد استفاده، معيارهاي آماريRMSE ، R مورد استفاده قرار گرفت. يافته ها نشان داد كه اعمال ميانگين لغزان روي داده هاي اصلي به نحو چشمگيري در بهبود كارايي هر دو مدل موثر مي باشد و در اين شرايط هر دو روش درخت تصميم رگرسيوني و شبكه عصبي مصنوعي در ايستگاه يزد قادرند با ضريب اطمينان بالايي ميزان بارش را 12 ماه پيش از وقوع برآورد نمايند.

One of the effects of climate system change is the occurrence and intensification of drought phenomenon. Prediction of drought condition can play an important role in mitigation of its effects as well as effective management of the available water during the drought periods. Different approaches have been presented for evaluation of drought. Analysis of precipitation data is the general method for drought evaluation, as acceptable prediction of precipitation before its occurrence, would be necessary and effective for analysis of drought. The purpose of this research is the evaluation of the effect of data processing on applicability of two data mining models on drought prediction in Yazd station. In addition, during the recent decades some new computer based models have been developed for drought prediction and in most of the cases they have presented quite satisfactory results. In this research, prediction of precipitation, which is the main component on drought occurrence, has been carried out in Yazd synoptic meteorological station. Therefore, two data mining methods including Regression Decision Tree (RDT) and Artificial Neural Networks (ANN) have been used, and simulations were carried out in two different conditions. In the first condition, the measured values of some meteorological variables were used as inputs and the amount of precipitation was predicted 12 months in advance. However, In the second condition, 3-year moving average of data were the inputs of the models for prediction of precipitation amount 12 months before its occurrence. Finally for evaluation of the model performance in different conditions, statistical criterion including R and RMSE were employed. Results indicated that using moving average of data as inputs of the models has considerably improved the performance of the models. Both RDT and ANN methods are able to predict the amount of precipitation in Yazd station 12 months before its occurrence.