Suppose i have three inputs and one output. How can I use ANN series for forecasting with my data.
Hi Isam,
I attached a link to one of my works in which I compare 6 different predictive models for forecasting hourly water demand. One of the models I explained is based on ANNs. All the models are implemented by using R libraries also briefly explained in the paper. Hope this help.
Cheers!
Article Predictive models for forecasting hourly urban water demand
These references may help you:
http://www.sciencedirect.com/science/article/pii/S1364815299000079
http://link.springer.com/article/10.1023/A:1014415503476
Also, this is a general reference about forecasting using ANN
http://www.sciencedirect.com/science/article/pii/S0169207097000447
All the best.
Thanks to all it is a rich information.
Taher.
I want to use want to commodity seasonal prices in June, July and August to forecast manufactured product. I can use other ANN. But how can i use ANN time series.
I think, these references are very useful:
http://www.sciencedirect.com/science/article/pii/0925231295000208
http://www.sciencedirect.com/science/article/pii/S0957417409004850
Have a look at them.
All the best.
Dear isam,
First you should define input variables, e.g. lagged prices in your case.
Second you can preprocess data to remove trends and seasonal variations (this depends on the time series features).
Then you use NN to build model: lagged prices -> future price.
See examples of using NNs for forecasting:
https://www.researchgate.net/publication/278716545_Forecasting_Time_Series_with_Multiple_Seasonal_Cycles_Using_Neural_Networks_with_Local_Learning
https://www.researchgate.net/publication/280735662_Short-Term_Load_Cross-Forecasting_using_Pattern-Based_Neural_Models
Conference Paper Forecasting Time Series with Multiple Seasonal Cycles Using ...
Conference Paper Short-Term Load Cross-Forecasting using Pattern-Based Neural Models
The numerous literature links given should get you through, however, I think it depends how you formulate your problem. If you make your training data such that, the input leads the output by certain period interval. Then given a training set, your model will be predicting future quantities right.
You may consider using an hybrid approach such as the one in this paper:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.462.3756&rep=rep1&type=pdf
hybrid approaches often give better results.
yes, it can be done but the network design depends on the time of time series applied
- Badges
- Science topic