Solving the Constrained Problem in Model Predictive Control Based Motion Cueing Algorithm with a Neural Network Approach

Abstract

Because of the critical timing requirement, one major issue regarding model predictive control-based motion cueing algorithms is the calculation of real-time optimal solutions. In this paper, a continuous-time recurrent neural network-based gradient method is applied to compute the optimal control action in real time for an MPCbased MCA.We demonstrate that by implementing a saturation function for the constraints in the decision variables and a regulation for the energy function in the network, a constrained optimization problem can be solved without using any penalty function. Simulation results are included to compare the proposed approach and substantiate the applicability of recurrent neural networks as a quadratic programming solver. A comparison with another QP solver shows that our method can find an optimal solution much faster and with the same precision.