[time-nuts] Neural net to control oven temperature ?

[Glen English VK1XX]

Hi Chase

thanks for the email. thanks for the tip on use of logistical classifiers.

Agreed the PID (and  variations ) is a seemingly perfect fit , at least 
at the top level.. My guess is that the type of disturbance the 'the 
system' (affecting, ultimately, the set temperature) (the device) could 
be classified (in real time) as a cause of different mechanisms, and for 
a specific mechanism, there might be a more optimal solution to minimize 
error.

My primary intrest in these things looking at new ways to do old things 
better..  I like systems that predict the error that is coming, before 
it occurs...so I like adaptive filter driven control systems . I am 
slowly getting my head around alpha-beta and Kalmans as time permits. 
The most popular neural net function is of course computers playing 
games- feed it the history of 10,000 games and as Chase says, it figures 
out the patterns of Y  in a sea of X

If anyone is interested in this stuff, you dont need to buy a dev kit. 
You can do it all in Python. Or C . Once you understand the basics , it 
is easy enough to program. If you dont understand the basics, you might 
not be able to acheive a desired outcome.

There are quite a few good books on these subjects for Python for those 
interested.

I wish I could go back to school and do a year or two on this stuff...

glen





On 10/07/2019 12:23 PM, Chase Turner wrote:
> Hi Glen,
>
> This is actually something I know a little about.
>
> Neural nets are most useful for feature selection, that is, finding the
> important x that is a function of y, in a very large sea of x variables. In
> this case, we already know what's important, which is temperature
> stability. So, a neural net would be a bit much when we already know what
> feature is important for function. Additionally, unless I'm mistaken, oven
> control is probably a linear relationship of some sort or another, and
> neural nets are much better suited for examining and revealing insights
> about non-linear data.
>
> If you have a method by which you can collect the necessary data that has a
> bearing on the oven functionality, you'd probably be better off training a
> logistic classifier, and using it instead. That said, both methods would be
> overkill, imo- I'd use a PID instead.
>
> Best,
> Chase
>
> On Tue, Jul 9, 2019 at 10:00 PM Glen English VK1XX <
> glenlist at pacificmedia.com.au> wrote:
>
>> Has anyone tried to use a Neural net to control oven tmep, rather than
>> the ye olde PID ?
>>
>> IE the algorithm learns from previous beheviour and successfully
>> predicts behaviour (or not).
>>
>> I'm sure there are a few out there proficient with machine learning
>> algorithms.
>>
>> Might make a good masters thesis I bet.
>>
>> Given that oven control based on inputs and whatever is not random,
>> unlike say flicker etc.
>>
>> glen
>>
>>