Abstract: In power grid operation, it is able to effectively prevent the violation of not checking the power according to the regulations due to reasons such as omission or intentionality through detecting whether the operator has performed electricity testing operation. Therefore, this study proposes a method for detecting electricity testing behavior based on the fusion of back propagation(BP) neural network and multiple acceleration sensors. The proposed method uses acceleration sensors to collect the acceleration, angular velocity, and attitude angle data at the constructor's arm and the electricity testing pole respectively, and after feature extraction, the BP neural network recognizes five key actions in electricity testing operation, such as pulling away the electricity testing pole, putting away the electricity testing pole, performing electricity testing, carrying the electricity testing pole to walk, and climbing up and down the pole to carry the electricity testing pole. The experimental results show that the classification accuracy of the five key actions in electricity testing operation reaches 97.4%, which is better than the common recognition methods such as support vector machines(SVM), decision tree(DT) and K-nearest neighbor(KNN) in the existing research, and can satisfy the purpose of the detection of electricity testing behaviors. At the same time, it can be complemented with the vision-based power checking behaviors, so that it can be adapted to the detection of the unchecked illegal behaviors in the more complex and diversified power grid operation scenarios.