Meter Detection

Detecting malfunctional smart meters based on electricity usage and targeting them for replacement can save significant resources. For this purpose, we developed a novel deep-learning method for malfunctional smart meter detection based on long short-term memory (LSTM) and a modified convolutional neural network (CNN). Our method uses LSTM to predict the reading of a master meter based on data collected from submeters. If the predicted value is significantly different from master meter reading data over a period of time, the diagnosis part will be activated, classifying every submeter to identify the malfunctional submeter based on CNN. We propose a time series-recurrence plot (TS-RP) CNN, by combining the sequential raw data of electricity and its recurrence plots in the phase space as dual input branches of CNN.

For more details, please refer to the paper.

If you are using our work in your research, please cite us as

@ARTICLE{9300285,

  author={Liu, Ming and Liu, Dongpeng and Sun, Guangyu and Zhao, Yi and Wang, Duolin and Liu, Fangxing and Fang, Xiang and He, Qing and Xu, Dong},

  journal={IEEE Industrial Electronics Magazine}, 

  title={Deep Learning Detection of Inaccurate Smart Electricity Meters: A Case Study}, 

  year={2020},

  volume={14},

  number={4},

  pages={79-90},

  doi={10.1109/MIE.2020.3026197}}

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes. See deployment for notes on how to deploy the project on a live system.

Prerequisites

Keras=2.2
tensorflow=1.9

Explanations for each file

Data

Our raw data is in fodler sitaiqu including the usage(kilowatt_everyday_2year.xlsx), the current(electriccurrent_hours_2year.xlsx) and the voltage(voltage_hours_2year.xlsx).

Data processing is accomplished in data_processing0.py

Residential Area’s Error Prediction Task

input.py will generate the input for lstm.

more_lstm.py is used to compare the result in different sequence length. Hence, in order to exlude the contingency, we choose to predict 10 times for each sequence length in k_lstm.py and draw_k_lstm.py.

The comparision of classical methods is accomplished in svr.py.

Malfunction-injected Residential Area Detection Task

We generated our data of residential area with malfunctional meters in bomb.py.

The detection task is finished in check.py.

Malfunctional Submeter Classification Task

We generated our data in samples.py, which imported single_bomb_wave.py and single_input_wave.py.

The classification task is accomplished in combine_model.py.

To test the performance of different proportions of malfunctional meters, we did some comparision in change_bome_rate.py.