Abstract
With the development of big data and data stream processing technology, the research of load predicting algorithm has gradually become the research hotspot in this field. Nevertheless, due to the complexity of data stream processing system, the accuracy and speed of current load predicting algorithms are not meet the requirements. In this paper, a load predicting algorithm based on improved Growing Self-Organizing Map (GSOM) model is proposed. The algorithm clusters the input modes of the data stream processing system by neural network, and then predicts the load according to its historical load information, optimizes it according to the characteristics of stream processing system, and a variety of strategies are introduced to better meet the load predicting needs of stream processing systems. Based on experimental results, the proposed algorithm achieved higher prediction accuracy rate and speed significantly compared to other prediction algorithms.