Due to the limitation of network structure and training mode for shallow neural network, network learning ability and generalization ability are weaker than deep learning network under large sample conditions. Therefore, a data fusion algorithm of water quality sensor based on stacked hybrid encoder is proposed. The algorithm forms a deep learning network model by stacking automatic encoder and sparse automatic encoder, which realizes the feature mining and sparse representation of sample data. The network model can fit the complex nonlinear function after large-scale sample training, and has generalization ability for low quality sample data. It can improve the accuracy of prediction classification. Simulation results show that the proposed algorithm achieves higher classification accuracy.