A robust firefly algorithm with backpropagation neural networks for solving hydrogeneration prediction
Hydrogeneration prediction typically has composite structures such as nonlinearity, non-stationarity, and fluctuation, which converts its predicting to be very tough. The applications of backpropagation neural network (BPNN) are very various and saturated. The linear threshold part of the BPNN produ...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer
2018
|
| Subjects: | |
| Online Access: | http://umpir.ump.edu.my/id/eprint/23312/ http://umpir.ump.edu.my/id/eprint/23312/ http://umpir.ump.edu.my/id/eprint/23312/ http://umpir.ump.edu.my/id/eprint/23312/1/A%20robust%20firefly%20algorithm%20with%20backpropagation%20neural%20networks%20for%20solving%20hydrogeneration%20prediction%20-%20s00202-018-0732-6.pdf |
| Summary: | Hydrogeneration prediction typically has composite structures such as nonlinearity, non-stationarity, and fluctuation, which converts its predicting to be very tough. The applications of backpropagation neural network (BPNN) are very various and saturated. The linear threshold part of the BPNN produces rapid learning with bounded abilities, also the procedure of BPNN causes the slow speed of training. The objective of this study, first, a firefly algorithm (FA) based on the k-fold cross-validation of BPNN has been suggested to predict data for keeping rapid learning and prevents the exponential increase in operating parts. Second, it is to construct on this method to improve an efficient process for prediction problems that can discover efficient solutions at a high speed of convergence. For this purpose, the suggested approach that makes a hybridizing the FA with the robust algorithm (RA), where RA is used to control the steps of randomness for the FA while optimizing the weights of the standard BPNN model. The algorithms were verified on an original dataset of the Himreen Lake Dam. The results display that the regression coefficient, root-mean-square error, mean absolute error, and mean bias error values of the
suggested model are 99.86%, 1.87%, 0.91%, and 0.31%, respectively. Furthermore, the performance of the suggested robust
firefly algorithm model is better than previously mentioned models in terms of speed and accuracy of prediction. |
|---|