To tackle the issue of interval sensitivity in the application of WSN data collection based on the prediction mechanism of fitting curve, this paper proposes fitting curve similarity sequence based on time periods to transform data collection problem based on prediction into the similarity estimation under certain accuracy. Feature-Based similarity is assumed to obey the Gaussian distribution. By studying the maximum probability of accurate prediction of perceived similarity, the proposed method uses Greedy algorithm to dynamically adjust the predicted similarity. Finally, it uses PSO algorithm to achieve inference of the predicted data based on predicted similarity. Simulation results show that this algorithm has achieved the desired results, and also provides great improvement in terms of energy consumption.