The Research On ATM Cash Flow Forecasting

The issue how much money the bank ATM machines need to put every day is directly related to the vital profit of the banks. If the money that put in is much more than the ATM machine output, there will be funding for banks idle issue, idle issue can not brought the profit to the bank; if money that put into the ATM machine is less than the money output, banks will increase the workload and inconvenience to users. Thus, the ATM to pay the value of the daily budget or forecast has become very important.With three index, gray theory and BP neural network technology used to the predict field theory, forecasting technology has been developed rapidly. This article first application of gray theory prediction model for monthly pay of the value of a bank branch ATM machines were predicted; the application of three indices and BP neural network, respectively, on the day to pay the value of a bank branch ATM machines were separate forecast. Every single model and have carried out the optimal parameter selection. The use of a single model theory to predict the error between the predicted value and the original data are generally larger, often predict the results are not satisfactory. Combination forecasting technique is to give several different weights prediction methods to form an integrated combination of predictive models. Combination forecasting model prediction accuracy is greater than the single model prediction accuracy, so that the results are more consistent with the data to predict trends. Due to a combination of predictive technology matures gradually get more and more widely used, the paper carried a combined neural network to predict the three indices. Forecasting techniques were used over a bank ATM cash flows were predicted given period, showed a combination of forecasting results than a single model to predict more accurately, the error is smaller.What we do in this thesis is that improving based on classic gray model, with a linear function respectively, logarithmic function to process the raw data series constructed two models to predict the results of these two models is better than the traditional; on the basis of the traditional neural network models performed using LM optimization algorithm has been improved, the establishment of a new forecasting model, the results show improved results and better forecasting models; 9 cubic exponential model parameters to optimize the choice to make the prediction more quickly accurate.