Idea!

Found an idea and it came before Sunday! It came from a site I was reading over that reminded me of a gripe I had when I first started my current job. Predicting elevator demands. I can't tell you how many times the elevators in the office do seemingly ridiculous things leaving me waiting, and I couldn't help thinking that there could be a better way of doing it. 

There are a few things I would like to see it do. I want to try preemptively positioning elevators and determining when and which elevator to send to a given floor when one is requested based on what the neural net has learned about daily trends.

I think this will be a lot more work than I was thinking of originally, but it is a concept I find very interesting. Got a few components to it.

 - An Elevator Control Module (ECM) that can handle any number of elevators on a building with any number of floors. Its primary tasks will be to take requests from virtual users, record data about usage, take directions to move the elevators from an external system, request advice from an external system when user input comes in and lastly to be able to operate in the absence of the external advice with some preset logic.

 - An artificial neural network (ANN) that will connect to the ECM and provide advice on which elevator to send when a user requests an elevator and to drive preemptive positioning of elevators in anticipation of upcoming demands.

 - A logic system to train the ANN. The logic system will have 3 potential algorithms it can use to affect the weights throughout the ANN, each designed for different environments. A decision tree, a fuzzy logic algorithm and a genetic algorithm implementation. The more good decisions an algorithm makes the more likely it is to be used in training the ANN.

 - A simulation application including a UI for the building, elevators and to drive the virtual people. Each floor in a build will have a configurable average and maximum occupancy, there will be a number of different shifts people will work as well as different behaviours (some employees take regular smoke breaks, some people eat lunch at their desk, etc...), as well as virtual users which aren't "regulars".

Not sure which task I will tackle first. Probably the ECM, UI, AI Algorithms and then the ANN. But we shall see.