Hybrid Approach of Prediction Daily Maximum and Minimum Air Temperature for Baghdad City by Used Artificial Neural Network and Simulated Annealing

Abstract

Temperature predicting is the utilization to forecast the condition of the temperature for an upcoming date for a given area. Temperature predictions are done by gathering quantitative data in regard to the current state of the atmosphere. In this study, a proposed hybrid method to predication the daily maximum and minimum air temperature of Baghdad city which combines standard backpropagation with simulated annealing (SA). Simulated Annealing Algorithm are used for weights optimization for recurrent multi-layer neural network system. Experimental tests had been implemented using the data of maximum and minimum air temperature for month of July of Baghdad city that got from local records of Iraqi Meteorological Organization and Seismology (IMOS) in period between 2010 to 2016. The results show that the proposed hybrid method got a high accuracy prediction results that reach nearly from real temperature records of desired year. Science, 2018, Vol. 59, No.1C, pp: 591-599 592 Keywords IntroductionArtificial neural networks (ANN) are one of the best learning methods known in the present day to solve the certain types of problems, that gives a powerful strategy to estimating vector-valued, discrete-valued and real-valued target functions. Backpropagation neural network (BPNN) is the most common method of ANN technologies. BPNN is a type of multi-layer feed forward neural network model, will be able to learn a wide range of model mapping relationship, which simulate the intelligent behavior of human brain, and is commonly used in the area of classification and pattern recognition, as well as prediction and many other various fields, this because it has strong adaptive ability. Nevertheless, the BPNN is dependent upon the error gradient descent process that the weight inevitably is categorized as local minimum points, besides slow convergence speed and may easy to cause faults for example shock. Simulated Annealing (SA) is great at global searching, and search for accuracy seems to be partial capacity Limited. SA offers a significant advantage over alternative methods is the possibility to avoid being trapped at local minimum. SA algorithm uses a random search that not only allows changes that reduce objective function, as well as several changes that raise it. The SA algorithm works effectively on a neighborhood search within solution space, acceptance probability, and inferior ways to avoid trap (i.e., local optimal solution) [ [1, 2].In this study, a temperature predicting models has been designed for predicting the minimum and maximum air temperature for month of Juley by using a combination of artificial neural network (ANN) and Simulated Annealing (SA). Minimum and maximum temperature for lead seven days were taken into account in this analysis Iraq Meteorological Organization and Seismology (IMOS) provides forecast of deviation of minimum and maximum temperature for one week and forecast of minimum and maximum temperature are very essential whenever there was a high or low temperature epoch. Statemen...