An Adaptive Sampling Algorithm with Dynamic Iterative Probability Adjustment Incorporating Positional Information.

Physics-informed neural networks (PINNs) have garnered widespread use for solving a variety of complex partial differential equations (PDEs). Nevertheless, when addressing certain specific problem types, traditional sampling algorithms still reveal deficiencies in efficiency and precision. In response, this paper builds upon the progress of adaptive sampling techniques, addressing the inadequacy of existing algorithms to fully leverage the spatial location information of sample points, and introduces an innovative adaptive sampling method. This approach incorporates the Dual Inverse Distance Weighting (DIDW) algorithm, embedding the spatial characteristics of sampling points within the probability sampling process. Furthermore, it introduces reward factors derived from reinforcement learning principles to dynamically refine the probability sampling formula. This strategy more effectively captures the essential characteristics of PDEs with each iteration. We utilize sparsely connected networks and have adjusted the sampling process, which has proven to effectively reduce the training time. In numerical experiments on fluid mechanics problems, such as the two-dimensional Burgers' equation with sharp solutions, pipe flow, flow around a circular cylinder, lid-driven cavity flow, and Kovasznay flow, our proposed adaptive sampling algorithm markedly enhances accuracy over conventional PINN methods, validating the algorithm's efficacy.

Optimal control of a black-box system based on surrogate models by spatial adaptive partitioning method.

When the dynamic model of a classical optimal control problem is explicit, we can transform this problem into a nonlinear programming problem and solve it by employing a traditional method. However, in some cases, no mathematical model of state equations is provided explicitly except for input-output data obtained from a simulation model. The hybrid model composed of functional mockup unit blocks generated in multiple platforms is a typical example. In this work, we regard these blocks as black-box models and use hierarchical neural network model to surrogate right-hand-side derivative functions of state equations. Specifically, to obtain highly accurate hierarchical neural network model, we explore a spatial adaptive partitioning criterion combining global sensitivity indices and interval length of local spaces based on the input-output data. Compared with models trained by several other partition criteria, numerical results verify that surrogate models obtained by the spatial adaptive partitioning method have higher accuracy. A mathematical example and a trajectory optimization problem of the black-box industrial robot Manutec r3 indicate the effectiveness of our proposed strategy.