Abstract: Sensors in cloud control systems generate a large amount of time series data, which need to be classified to make automated decisions. As artificial intelligence algorithms are widely applied in cloud control systems, their security issues are particularly important. To study the security problems of deep models in cloud control systems, a new adversarial attack algorithm for deep time series models is proposed, and a new performance index for adversarial samples of time series data is defined. The UCR dataset is used to conduct experiments,the superior performance of the algorithm is verified and the vulnerability of deep models in cloud control systems is shown.