With the continuous expansion of offshore wind power installed capacity

the number of diverse tasks derived from homogeneous data has significantly increased. To facilitate information sharing and mutual enhancement across tasks

this paper proposes a symbolic sequence generative framework (SSGF) that uses a unified modeling paradigm to represent and model multiple tasks. The framework designs a universal symbolic system

encoding task types

input data

and target outputs into standardized token sequences

and introduces a “next token prediction” training mechanism to transform diverse tasks into a unified sequence generation problem. Based on this

a three-layer decoder-only model is constructed as the underlying architecture

featuring fully shared parameters and consistent inference paths

enabling multi-task collaborative training and inference within the same neural network. By applying the proposed method to offshore wind power data

this paper achieves unified modeling for both wind power prediction and abnormal data imputation tasks

. The mean squared error of the wind power prediction task is decreased by 3.18% compared to training separately

and the imputation accuracy of the abnormal data imputation task is increased by approximately 1.72%. This method can be used to not only eliminate structural fragmentation issues in conventional methods

but also significantly reduce model design and deployment costs

providing a feasible pathway for building a unified

efficient

and scalable intelligent sensing framework.