A robust transfer deep stochastic configuration network for industrial data modeling is proposed to address challenging problems such as the presence of outliers (or noise) and conditional drift of the data model due to changes in working conditions. Assuming that outliers follow the t-distribution, the maximum a posteriori estimation is employed to evaluate the read-out weights, and the expectation maximization algorithm is used to iteratively optimize the hyperparameters of the distribution. Moreover, the knowledge contained in the data are expressed in the form of the model structure, connection weights and outlier distribution, and a knowledge-data-based robust transfer strategy is developed to offset the impact of insufficient training data on the learning performance of a deep stochastic configuration network with a new working condition. Comparative experiments are carried out using the historical furnace temperature data of a municipal solid waste incineration plant in China. The results show that the proposed method performs more favorably in robust data modeling and mitigates the impact of changes in working conditions on the applicability and accuracy.