Aiming at the problems of data redundancy and difficulty in capturing dynamic information in IoT time series data, this paper proposes a data-driven dynamic time series classification algorithm. The dynamic information in the time series collected by sensing devices is extracted by DiPCA (dynamic internal principal component analysis) to realize the role of dimensionality reduction and refining dynamic information; the parameters of the classification algorithm are optimized by using the sparrow search algorithm to enhance the performance of the SVM algorithm and make it model the temporal features containing shapelet local features, which finally constitutes a two-way evolutionary algorithm framework to realize the temporal classification function. The performance of the algorithm is examined using UCR temporal data set and edge computing simulation data, and the results show that the comprehensive performance of the algorithm is significantly improved compared with the basic algorithm, and the effectiveness and superiority of the classification function of the algorithm in the simulation environment is verified.