The present invention discloses an optimization design method for structural parameters of biomass boiler economizers and belongs to the field of big data learning models. In the present invention, a sample database is established by utilizing historical operating big data of biomass boiler economizers, a heat exchanger residual self-attention convolution model is established based on a CNN and a self-attention mechanism, a plurality of target parameters to be optimized are quickly predicted through machine learning, and multi-target optimization of structural parameters to be optimized in the economizers can be performed in combination with an iterative optimization algorithm. Compared with traditional optimization for all variables of a biomass boiler economizer, the self-attention mechanism can automatically focus on features with high importance, to better optimize variables with high importance, making the subsequent optimization and adjustment convenient and quick, and greatly reducing the optimization cost.