An LSTM-based Motion Recognition Method for Power Operation and Maintenance

doi:10.16182/j.issn1004731x.joss.19-FZ0336

Abstract

Abstract: The security of power operation and maintenance has always been a subject of great social concern. In order to avoid serious consequences caused by the fault of staffs, a motion recognition method for power operation and maintenance jobs based on LSTM (Long Short-Term Memory) is proposed, which covers the whole process from data collection, data processing to motion classification and recognition, then it can recognise and supervise the behavior of staffs who are at work. In addition, a simulation experiment is conducted between the deep learning algorithm LSTM and the traditional machine learning algorithm KNN based on the newly constructed data set of power operation and maintenance jobs. The results show that LSTM achieves better performance than KNN. When the time window is 120 frames, the accuracy based on LSTM reaches 91.32%, which is about 2 percentage points higher than KNN.

Key words: power operation and maintenance security, LSTM, motion recognition method, simulation experiment

CLC Number:

TP391.9

Liu Peizhen, Jia Yuxiang, Xia Shihong. An LSTM-based Motion Recognition Method for Power Operation and Maintenance[J]. Journal of System Simulation, 2019, 31(12): 2837-2844.

References

[1] Gao L, Bourke A, Nelson J.Evaluation of accelerometer based multi-sensor versus single-sensor activity recognition systems[J]. Medical Engineering & Physics (S1350-4533), 2014, 36(6): 779-785.
[2] Massé F, Gonzenbach R R, Arami A, et al.Improving activity recognition using a wearable barometric pressure sensor in mobility-impaired stroke patients[J]. Journal of NeuroEngineering and Rehabilitation (S1743-0003), 2015, 12(1): 72.
[3] Salarian A, Russmann H, Vingerhoets F J, et al.Ambulatory monitoring of physical activities in patients with Parkinson's disease[J]. IEEE Transactions on Biomedical Engineering (S0018-9294), 2007, 54(12): 2296-2299.
[4] Bonomi A G, Goris A, Yin B, et al.Detection of type, duration, and intensity of physical activity using an accelerometer[J]. Medicine & Science in Sports & Exercise (S0195-9131), 2009, 41(9): 1770-1777.
[5] Yang J Y, Wang J S, Chen Y P.Using acceleration measurements for activity recognition: An effective learning algorithm for constructing neural classifiers[J]. Pattern Recog Lett (S0167-8655), 2008, 29: 2213-2220.
[6] Pirttikangas S, Fujinami K, Nakajima T.Feature Selection and Activity Recognition from Wearable Sensors. In Ubiquitous Computing Systems[C]. Berlin: Springer. 2006, 516-527.
[7] Vapnik V.The nature of statistical learning theory[M]. New York: Springer Science & Business Media, 2013.
[8] Junker H, Amft O, Lukowicz P, et al.Gesture spotting with body-worn inertial sensors to detect user activities[J]. Pattern Recognition (S0031-3203), 2008, 41: 2010-2024.
[9] Mathie M, Celle B G, Lovell N H, et al.Classification of basic daily movements using a triaxial accelerometer[J]. Medical and Biological Engineering and Computing (S0140-0118), 2004, 42(5): 679-687.
[10] Garcia-Ceja E, Brena R F, Carrasco-Jimenez J C, et al. Long-Term Activity Recognition from Wristwatch Accelerometer Data[J]. Sensors (S1424-8220), 2014, 14(12): 22500-22524.
[11] Berchtold M, Budde M, Schmidtke H, et al.An extensible modular recognition concept that makes activity recognition practical[C]. Annual Conference on Artificial Intelligence. Berlin, Heidelberg: Springer, 2010: 400-409.
[12] Reddy S, Mun M, Burke J, et al.Using mobile phones to determine transportation modes[J]. ACM Transactions on Sensor Networks (TOSN)(S1550-4859), 2010, 6(2): 13.
[13] Minnen D, Westeyn T, Ashbrook D, et al.Recognizing soldier activities in the field[C]. 4^th International workshop on wearable and implantable body sensor networks. Berlin: Springer, 2007: 236-241.
[14] Altun K, Barshan B.Human activity recognition using inertial/magnetic sensor units[C]. International workshop on human behavior understanding. Berlin, Heidelberg: Springer, 2010: 38-51.
[15] McGlynn D, Madden M G. An ensemble dynamic time warping classifier with application to activity recognition[C]. International Conference on Innovative Techniques and Applications of Artificial Intelligence. London: Springer, 2010: 339-352.
[16] Tapia E M, Intille S S, Haskell W, et al.Real-time recognition of physical activities and their intensities using wireless accelerometers and a heart monitor[C]. Proc. Int. Symp. on Wearable Comp. New York: IEEE, 2007.