Smart meter based on time series modify and constructive backpropagation neural network

This paper proposed new technique for improving smart meter. This technique is built based on Non-Intrusive Load Model (NILM) which is combined with time series modify (lag-1). This technique is employing modification of time series data to predict the operating of electrical appliances which is operated simultaneously. The advantage of the technique is capable to identify using of energy consumption of appliances without adding sensor in each appliances. Another advantage of this method is the simplification of the Neural Network (NN) training process, because with the concept of lag-1, each appliance requires only one data record. Signals from current sensors are processed by the microprocessor to identify the type of appliance currently operating by using NN. Data resulted by NN is sent to the display and also sent to the SD Card which can show the bill of each electrical equipment in detail. From the experiment result, it can be proof that smart meter capable to identify the use of appliances and also capable to monitor the use of energy consumption real time with 5% error tolerance in averages. With this performance, the smart meter has big chance to implement in the real systems and mass production.