Neural Solution to Dynamic Overdetermined System With Applications to Data Fitting and Parameters Estimation
提出一种积分神经解法求解动态超定系统，利用饱和或非连续投影函数和噪声抑制能力，获得具有优越收敛性的最小二乘解，并通过数据拟合和参数估计验证其有效性。
In recent years, dynamic overdetermined systems have sprung up and been broadly employed for handling different problems in real time. This article makes improvements in this direction by proposing, investigating, and analyzing an integral neural solution (INS) to solve the dynamic overdetermined system. Notably, an error function is constructed in the first place. Then, aided with a generated neural dynamic framework, an INS model is devised, which exploits not only saturated or even noncontinuous projection functions but also possesses noise-suppression ability with integral enhancement information. Theoretical analyses and computer simulations manifest that the proposed INS model is able to acquire the least-squares (LS) solution with superior convergence property, contrasted with the existing methods, e.g., zeroing neural network (ZNN). On top of that, applications to data fitting as well as parameters estimation ulteriorly validate the feasibility and effectiveness of the proposed INS model for handling the dynamic overdetermined system.