Benchmarking a Recurrent Linear GP Model on Prediction and Control Problems

In this work, a recurrent linear GP model is designed by introducing the concept of internal state to the standard linear Genetic Programming (GP), so that it has the capacity of working on temporal sequence data. We benchmarked this model over four standard prediction and control problems, which include generic even parity problem, sun spot series prediction, Lorenz Chaotic time series prediction and pole balance control problem. From the experimental results, the recurrent linear GP model appears to be very competitive compared to those algorithms relying on spatial reasoning of the temporal problem.