Plant Estimation and Feedback Control for Online Advertising Campaigns

Nov 6, 2018
[Work published prior to Yahoo]

A feedback control system is proposed for online advertising campaign control. The proposed system consists of a forecasting system for estimating the plant model and a Proportional-Integral (PI) controller for closed-loop feedback control. The forecasting system uses a temporal filter to reduce the volatility in the raw data and a smoothing spline regression algorithm to estimate an analytical model for the temporally filtered forecasting data. Based on the analytical model, the plant gain at the current operating point is estimated and is used to determine the PI controller gain to achieve a constant loop gain for the desired consistent behavior of the closed-loop system. Initial experiment in the ONE-by-AOLTM DSP demonstrated improvement on the cost performance compared to existing campaign controllers.


  • International Symposium on Computational Intelligence and Industrial Applications (ISCIIA2018)
  • Conference/Workshop Paper