EPSTO-ARIMA: Electric Power Stochastic Optimization Predicting Based on ARIMA
Yuqing Xu, Guangxia Xu, Zeliang An, Yanbin Liu · 2021
With the advance of the energy industry and the Internet, electricity data sharing has unleashed the full potential for social production during the past decade. However, electricity data inference attack will incur the disclosure of private information and the unavailability of valuable data, thus deteriorating national security and affecting the conversion of energy big data to social public service value.To cope with this challenge in electricity shared data, EPSTO-ARIMA (Electric Power Stochastic optimization Predicting Model Based on Autoregressive Integrated Moving Average) was proposed to increase prediction error of attackers by utilizing the concept of stochastic sampling, optimization, data poisoning and adversarial examples. The generation of adversarial examples is interfered with the prediction effect of EPSTO-ARIMA. The model was validated by seven sets of data from three datasets. Experimental results indicate that EPSTO-ARIMA could increase prediction error. For publicly dataset “Column2”, the proposed EPSTO-ARIMA achieves 61.31% higher prediction error than ARIMA (Autoregressive Integrated Moving Average model), respectively. Simultaneously, the terrific results in other datasets have also been ascertained the viability and generalization ability of our proposed EPSTO-ARIMA.