مدل‏سازي جريان روزانه رودخانه با استفاده از برنامه ‏ريزي ژنتيك و شبكه عصبي (مطالعه موردي: حوضه آبخيز معرّف امامه)

Daily River Flood Modeling using Genetic Programming and Artificial Neural Network (Case Study: Amameh representative watershed)

فرايند بارش- رواناب پيچيده و غيرخطي است و مدل‏سازي آن به دليل عدم قطعيت‏هاي زياد يكي از مهم‏ترين دغدغه‏هاي پژوهشگران در حيطه مسايل منابع آب به‌شمار مي‌رود. از بين روش‏هاي مورد استفاده، مدل‏هاي هوشمند در پيش‏بيني چنين فرايندهايي مفيد و موثرند. بنابراين، به منظور مدل‏سازي جريان رودخانه از روش‏هاي شبكه عصبي مصنوعي و همچنين برنامه‏ريزي ژنتيك به منزله روشي صريح‌ـ كه جزو الگوريتم‏هاي تكاملي به‌شمار مي‌رود‌ـ در حوضه آبخيز معرّف امامه و در دوره آماري 1349 - 1350 تا 1390 - 1391 (42‌ساله) استفاده شد. بدين منظور، از داده‏هاي هواشناسي و آب‏سنجي در مقياس روزانه و در قالب 62 مدل پيشنهادي استفاده شد. نتايج نشان داد برنامه‏ريزي ژنتيكي، از ميان مدل‏هاي فراوان، خطاي كمتري داشت. خطاي مدل‏ها نيز وقتي كه فقط از عملگرهاي اصلي رياضي و توان استفاده شد به‌مراتب كمتر بود. سرانجام، با توجه به معيارهاي ارزيابي مورد استفاده در اين تحقيق، ساختار پيشنهادي با ورودي‏هاي (مدل 54) دما، باران، و تاخيرهاي باران تا دو روز، رطوبت ‏نسبي و تبخير و تعرق و تاخير جريان تا دو روز به عنوان بهترين مدل با خطاي 001/0، 031/0، و 009/0 در مرحله آموزش و 001/0، 032/0، و 009/0 در مرحله آزمايش به‌دست آمد.

Introduction Rainfall-runoff relationship is one the most complicated issues in hydrological cycle and its accurate estimation is one the most important concerns in water resources engineering and management. In addition, Rainfall-runoff modeling process is complex and non-linear due to the large uncertainties in the field of water resources is considered. None of the statistical and conceptual models are able to be a better and capable model for that. But today using nonlinear networks as intelligent system for forecasting such complicated event can be efficient and effective in many problems of ecology. Materials and methods Furtherefore Geneexpression programming, a branch of evolutionary algorithms, is able to optimize the model structure and its componentsTherefore, for modeling river flow was used artificial neural network and as well as Genetic programming an explicit method that is considered part of evolutionary algorithms in Amame apprehensive watershed is situated in the Northern Slope of the Alborz Range in Mazandaran province in IRAN country for a period of fifty five years from 1970-1971 to 2011-2012 periods (42 years). For this purpose, the meteorological and hydrometric data in the form of 62 proposed models were used. Every ANN is interconnected network of many processing units which called neuron. Neurons are the smallest unit in artificial neural network. These neurons are very similar to biological neuron and the cell of human brain. Whereas the speed of these neurons is more than biological neurons, their ability and capacity are less than them. Neuron in every layer is connected through weights to next layer of neuron. The associated parameters with each of these connections are called weights. These weights represent information which is being used by the net to solve a problem. During the training network these weights, constant amount of that assemble with them, and bias are changed consecutively until the target function reach to favorite amount. We used activation functions (sometimes called a transfer function or threshold function) for transfer output from every layer to next layer.These activation functions may be logistic sigmoid, linear, threshold, Gaussian or hyperbolic tangent functions, depending on the type of network and employed training algorithm.On the other hand, the method which used for achieving weights and biases are learning rule for favorite and terminal amount. In fact, this rule is a complex mathematical algorithm. Every network needs two group data to create and be acceptable: training series and testing series. About 80 percentages out of data is belonging to training and the rest of it is used for testing. Duration the learning time, amount learning of network is evaluated continually by target function. In all cases, a multi-layer perceptron (MLP) ANN was employed for rainfall–runoff modeling, with the weights determined by error back-propagation. Sigmoid activation functions were used at all nodes in the hidden and output layers. For ANN method it was used Muti Layer Perceptron with Back Propagation algorithm and one upto three hidden layer and one upto thirty neurons. In spite of statistical methods such as ANN, decision tree etc., GP is self-parameterizing that build models without any user tuning. A GP method is a member of the Evolutionary Algorithm family, which are based upon concept of natural section and genetic. In fact the basic search strategy behind GP is a genetic algorithm (GA) that was created by Holland (1975), although GP was developed and introduced much later by Koza (1992). This method has many similarities with genetic algorithm such as GP works with a number of solution sets that known collectively as a population rather than a single solution at any time therefore the possibility of getting trapped in a local optimum is avoided. As you know it is one the most important problems in ANN. But GP is different from traditional genetic algorithm In that it typically operates on parse tree instead of bit string. A parse tree is built up from a terminal set (the input variables in the problem and randomly generated constants, i.e. empirical model coefficients) and a function set hence the basic operators used to form the GP model. Results and discussion The results showed that in a MLP method with two hidden layer has best function. Furthermore increase the number of neurons in the hidden layer can somewhat reduce errors, but then increase the number of neurons not only increases efficiency but also cause errors of models increase. So that, the results showed Amameh in the hidden layer neurons is used as a hidden layer neurons is between 1 and 16 while the hidden layer neurons are used between 1 and 12. In addition out of many models first genetic programming has less errorS and second when used mathematical function and power has less errors. Conclusion With regard to the evaluation criteria used in this study such as MSE, RMSE and MAE the proposed structure of inputs (model number 54) meteorological variables such as temperature, rain and rain delays up to two days, relative humidity and evapotranspiration and delays during the two days as the best the model was obtained. So that the errors of this model for MSE, RMSE and MSE were 0.001, 0.031 0.009 and 0.001, 0.032, 0.009 in modeling and testing phase respectively.