Neural Networks in Accelev (AI) explained.
Deep neural networks (AI) used by us for new Grid Monitoring (from v2.61) solve difficult to define and isolate algorithmic problems, like daily load waving and random spikes caused by induction motors or spark jumps (for example – bimetals at electric heaters).
It is challenging to find optimal coefficients, that represent the influence of various voltage drops or power delivery waving. I tried to solve that using semi-AI methods, like heuristic algorithms.
Honestly, such methodology works in 90% of possible variations, detecting electric cookers, other cars being charged, electric heaters etc. But it fails (mostly) while induction engines are used, especially at its start, where two negative situations occur: variance of the phase of voltage to the phase of current (caused by induction) and huge spike-like voltage drop at the start of an electric engine.
I started to find a more innovative method, capable of detecting and memorizing such random and unusual behaviour and finally – correct car charging to avoid fuses being tripped off.
I personally work for years with neural networks. I decide to build a pre-learned network; local circumstances can teach that. Nobody can predict real electricity usage and homeowner behaviour or hobby ????
The neural network has an input and output, similar to the standard equation.
Our inputs (X1, X2, X3 etc) are actual voltage, average voltage during the last 30 min, time of the day (as voltage waves due to two higher load times – morning and evening). Spikes detected a variance of phase between voltage and current seen (thus – induction power units detected) are also taken into consideration.
As an output (Y)
we have the current we should deliver to the car.
If there are no suspicious behaviours detected – in most cases, power will be full. It may be reduced constantly only when the grid is weak (thin cables are seen) for safety reasons.
By combining all input parameters, our AI tries to predict proper current for EV (car)
If it fails (breakers tripped) it has enough energy in condensers to memorize inputs and last decision to self – learn, that such decision was wrong and it tries to correct whole „brain” to avoid such wrong decision (proven by tripped off breakers).
In most cases such tripped off breakers will never occur, as we pre-learn our AI. But if such an accident occurs, your Accelev would learn from that how to avoid that situation next time.
I hope I have explained the basis of AI artificial intelligence used in Accelev. If you have more technical questions - do not hesitate and ask.