Info
此问题已关闭。 请重新打开它进行编辑或回答。
Neural Networks. General approach to predict nearest future value (recognise incomplete pattern)
1 次查看(过去 30 天)
显示 更早的评论
I need a general idea (and learn a bit of terminology as well) on how to approach the following problem:
1. I have data coming in real-time but in uniform intervals (1s). each portion can have 1 or more data points.
2. values are between 1000 and -1000 but most of the time they oscillate between 500 and -500.
3. I would like to model this in Matlab to train an NN to learn patterns that precedes events when data is crossing 1000 or -1000
4. there is strong suggestion that only about 200 recent data points are relevant to such event.
-- given 1 and 4, the task is basically to process a snapshot of last 200 data points every second and train NN to predict an 'event' in nearest future
My immediate idea is to have a NN with 2000 x 200 neurons which will 'memorize' what happened before each event (using last 200 data points)(2000 is on/off for each 1 of the value between 1000 and -1000).
And then try to generalize them together into few dozens of 'patterns' each need to have a probability attached (eg pattern#1 is likely to produce event with probability 75% compared to pattern#13 with probability 20%)
What I really need at this point is to get an outlook on NN architecture/types with clear understanding what Matlab can help to implement quickly (using toolboxes) rather than programming this from the scratch.
Also suggestions to use another approach (different from NN) are welcome.
0 个评论
回答(0 个)
此问题已关闭。
另请参阅
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!