Application of selective ensemble learning soft sensor modeling based on GPR in the solar thermal power collection system

In order to achieve zero carbon emissions, new concepts and algorithms are needed in the control of renewable energy to ensure the safe and reliable operation of the grid under the trend of zero carbonization in the future. In this paper, the ensemble learning soft sensor model is applied to the outlet temperature control of the solar thermal power collection system. Due to the lack of accuracy and diversity of base learners in traditional ensemble methods, ensemble pruning is ignored, and the lack of adaptability. So, the estimated performance falls short of the expected value. Therefore, a novel predictive control method for the outlet temperature of the collector system is proposed based on the Gaussian process regression selective ensemble learning soft sensing model (SESMGPR). Firstly, the collecting heat field temperature values were extracted from the Dunhuang Dacheng solar photothermal power plant as historical data, and the subsets and subspaces of the historical data were obtained by bootstrap and partial least squares regression (PLS), respectively. Secondly, the local domain (LTs) was identified by Gaussian mixture model (GMM) clustering, and the local GPR model was established for each LT to obtain the base model (MGPR). Finally, the set pruning strategy based on a genetic algorithm (GA) was used to select the MGPR model with high estimation performance, and the base models were fused by the Stacking algorithm to obtain the SESMGPR model. Besides, adaptive control is used to alleviate the performance degradation of SESMGPR due to the randomness of solar radiation, ambient temperature, wind speed, etc. Experimental results show that the SESMGPR method can effectively handle time-varying variations of the solar thermal power collection system and maintain high prediction accuracy.