关键词: 数据生成, 数据安全, 时序数据, 短序列, 生成对抗网络

Abstract: Longterm time series data is difficult to obtain in reality, which seriously restricts the development of applications such as situational awareness and threat analysis in cyberspace security. Deep learningdriven data generation methods can effectively protect the privacy of original data, where ensuring the high utility and diversity of generated data is crucial. However, existing methods used random splicing of shortterm data to construct training data, which cannot ensure that the distribution of generated data meets expectations, affecting the availability of generated data. To address the above problems, this paper proposes a highutility time series generation method combining sequence correlation graph and generative adversarial network. By constructing sequence correlation graph and probability weighted generative adversarial network, the original data distribution is accurately fitted. Experimental results on multiple real data sets show that the method can generate longterm time series data with high utility and diversity based on shortterm original data, showing its great potential in practical applications.

Key words: data generation, data security, time series data, short sequence, GAN(generative adversarial network)