Against the backdrop of "dual-carbon" goals, growing installed capacity of new energy, changes in user load characteristics and increased load demand have intensified pressure on grid supply-demand balance. To maintain grid stability and fully tap the adjustable load potential of industrial users, an industrial user load potential assessment strategy based on a hybrid MPA-CNN-LSTM model combined with confidence interval correction is proposed. First, building on existing load characteristics, load reduction characteristics are introduced—describing the types and methods of load reduction among different users in the same industry—as inputs to the MPA-CNN-LSTM prediction model. Second, the MPA-optimized CNN-LSTM neural network is trained using actual adjustable potential data from responsive users to predict industrial users' adjustable potential. Finally, the confidence interval correction method is applied to refine the predicted adjustable potential, enhancing accuracy.