بررسی روند و پیش بینی بارش با استفاده از شبکه عصبی مصنوعی در کاشان (مقاله علمی وزارت علوم)
درجه علمی: نشریه علمی (وزارت علوم)
آرشیو
چکیده
بارش باران جز مهم ترین پدیده های جوّی است که بر زندگی بشر، پوشش گیاهی و جانوری تأثیر می گذارد. پیش بینی بارش باران برای اهداف مختلفی مانند فعالیت های کشاورزی، پیش بینی سیلاب، تأمین آب شرب و بسیاری از موارد از اهمیت بسیار بالایی برخوردار است. هدف این پژوهش بررسی روند و پیش بینی بارش ایستگاه کاشان طی دوره 49 ساله (1350-1398) است. بنابراین ابتدا داده ها بارش گردآوری و سپس به صورت میانگین فصلی و سالانه تنظیم شدند. در ادامه با استفاده از روش من-کندال معنی داری روند بارش و با استفاده از روش برآورد کننده شیب خط سنس، میزان شیب خط روند، آزمون شد. طبق نتایج در سری های زم انی می انگین بارش کاشان روند معنی داری در سطوح اطمینان 99% و 95% مشاهده نگردید؛ اما بااین حال میانگین بارش کاشان به طور متوسط در هرسال حدود 60/0 میلی متر کاهش یافته است. همچنین با استفاده از روش شبکه عصبی مصنوعی داده های بارش جهت پیش بینی بررسی شد. طبق نتایج بعد از آزمون شبکه 2 لایه پنهان و 10 نرون در لایه های میانی مدل نسبتاً بهتری را ارائه کرد. با بررسی و تطبیق مقادیر نمودار همبستگی مشخص گردید پیش بینی بارش برای ایستگاه کاشان با نتایج واقعی ایستگاه مطابقت کاملی نداشته است. همبستگی بین مقادیر واقعی و پیش بینی شده توسط شبکه برابر با 47/0 می باشد. همچنین ثابت شد مقادیر پیش بینی شده بارش توسط شبکه عصبی در ترکیب با الگوریتم ژنتیک نزدیک تر به داده های واقعی بارش و داده های پیش بینی شده توسط شبکه عصبی بدون ترکیب با الگوریتم ژنتیک از مقدار واقعی دورتر بوده و روند غیرخطی دارد. بنابراین بین میانگین های شبیه سازی شده بارش با مقدار واقعی در ایستگاه کاشان اختلافی فروانی وجود ندارد.Investigation of trend and precipitation forecast using artificial neural network during 49 years (Case study in Kashan station)
Introduction Rainfall is one of the most complex and accidental natural phenomena. The impact of tangible or intangible factors is so effective in creating an accident Which has led this process from a clear legal system to a complex and chaotic one. In general, understanding how the climate is changing. In particular, the trend of precipitation changes is one of the issues that have been considered by atmospheric and hydrological researchers in recent years. Neural networks are simple computational tools for testing data and creating models of data structures Reduction of rainfall in Kashan station is one of the problems that has caused many agricultural lands to be taken out of the agricultural cycle due to quantitative and quantitative water loss. And it has led to many economic and social problems, such as immigration. Therefore, studying, detecting, and forecasting rainfall in this region can be useful for the present and future of decision-makers to solve the problems of the region. Materials and methods This research is of applied type and its method is descriptive-analytical and the study area in Kashan city. Required data during the statistical period of the Kashan synoptic station is obtained from the Meteorological Organization to determine the trend of precipitation changes The Man Kendall method is used the Mann-Kendall test is one of the most common and widely used non-parametric methods of time series trend analysis. Using the Man-Kendall method, data changes are identified, and their type and time are determined. In this research, to implement different neural network models and determine the optimal structure, the neural network toolbox in MATLAB software has been used. One of the most widely used neural network models is the multilayer perceptron model, which is based on a post-diffusion algorithm. In this type of network, forward input data is processed and the processing path does not return to the neurons of the previous layer, and the output of each layer will only affect the next layer. Results and discussion In order to investigate the trend of increasing or decreasing the amount of precipitation in Kashan station, the Sense slope method was used, and to evaluate the significance of the precipitation trend, the Mekendal method was used. The results of the annual analysis showed that in general, no significant trend was observed in the average rainfall time series of Kashan. Therefore, no natural jump has been observed in the average rainfall of the Kashan station. Sense test statistics at 99% and 95% confidence levels confirm the decrease in rainfall at Kashan station. Thus, the rainfall of the studied station during the study time series has decreased by -0.6 mm each year. The greatest impact of reduced rainfall is related to winter. The most important advantage of the neural network over other intelligent systems is the ability of the network to learn from its surroundings. First, the data for entering the network was divided into three categories, 70% of which were for network training, and 30% of the data was allocated for testing and validation. Finally, due to repeated trials and errors to build a suitable network, a network with two hidden layers and 10 neurons had the highest accuracy for prediction. Optimization diagram for artificial neural network Kashan station precipitation data showed that 110 artificial neural networks were created in repetition. In this research, the network was able to predict the amount of precipitation using the introduced variable of correlation rate of 0.47 According to the results obtained from the values predicted by the network in combination with genetic algorithms such as actual precipitation data and predicted data It has a nonlinear process by the network without combination with genetic algorithm. But the difference in results for the simulated precipitation averages at Kashan station is still clear. Conclusion n the present study, the precipitation data of the Kashan synoptic station were evaluated seasonally and annually using the Menkendal method and Sense slope. The results of the Menkendal test trend in seasonal and annual average rainfall time series showed that there is no significant trend in 99% and 95% levels. On average, Kashan station has had a decrease of -0.60 mm of precipitation every year. At first glance, it seems that this decrease (0.6 mm) in Kashan's very dry station and low rainfall is a small amount. But for a period of 50 years, this reduction reaches 30 mm, if this trend continues in the long run, it will have very catastrophic consequences for the region's groundwater, agriculture, and drinking water basins. Also, an artificial neural network method was used to predict monthly rainfall. According to the results of using the multilayer perceptron model using the error propagation algorithms and the number of 10 neurons in the middle layers and two hidden layers had less error than other structures. Therefore, the variable introduced in the study with a correlation rate of 0.47 was able to predict the amount of precipitation. By reviewing and adapting the values of real data, the forecast, and the results of the correlation diagram, it was found that the precipitation forecast for Kashan station was to a reasonable extent consistent with the actual results of the station. According to the results, the error obtained from simulating precipitation data of Kashan station using neural network performance was very low, which indicates the proper performance of the network in the testing and validation stage Finally, the ability to estimate and predict precipitation using artificial neural network (ANN) at Kashan station has performed better in the months of low rainfall and the dry period when rainfall is minimized. But in general, the artificial neural network for predicting rainfall in Kashan station has shown high performance
تبلیغات
