networking - Is Backpropagation okay for this or should i try another approach? -

i'm making kind of "the life game" creatures , food on world. creatures eat food in order gain energy , when have enough energy reproduce. energy food provides based on time food has been on world, longer better. i'm using genetic algorithm creatures want them learn what's best through generations because thing have 4 ways search food:

1.- nearest one

2.- 1 provides more energy

3.- 1 provides best % mutate genes , gain cool new things

4.- 1 represents lower danger reach

i have neural network input of 100 genes, 4 hidden layers , 1 output layer it's vector 4 components, each of them indicating search option should go based on genes.

network doing great initial input, thing want keep feeding network on generations can things example creatures lazy genes evolved searching nearest food 1 provides more energy can "do nothing" longer periods of time (because yea, they're lazy).

seems have pass in inputs everytime train it, impossible because new inputs generate on generations , can't training around 100 inputs every generation due high cpu ussage.

my question: propagation okay particular case or should use different approach need?

thanks in advance

edit: neural network takes around 20-60 seconds converge , every crossover might occur once every 0.2 - 2 seconds. first implementation.