Data Analysis Based On Smart Meters

Data analysis based on smart meters that is non-intrusive load monitoring without adding additional equipment.The target intends to deduce the individual energy consumption of appliances used in the house,based on measuring the aggregate power consumption with single smart meter.The electric appliance's historical data,current states power consumption are used to do the model's characteristic,our goal is to get more electric device information.At the same time,comprehensive device consumption information can help users improve their energy usage habits and save energy.The thesis analyses advantages and disadvantages of some disaggregation models form domestic and overseas.Combined with the historical information of the equipment,two kinds of energy disaggregation prediction models based on smart meters are proposed.Firstly,the paper proposes an improved factorial hidden markov model based on the state of the appliance.This method introduces the state power consumption of the electrical appliances for energy disaggregation.The traditional model based on a single appliance or the states of device has the same performance,but in multi-state electric appliances disaggregation study,the model based on the state of the device has a higher energy distribution accuracy rate.Secondly,the thesis proposes a new association rule learning model based on time partition.The method searches interrelationship between electric appliances and uses the habit of device to generate frequent itemsets,we use the frequent itemsets training strong association rules.This model reduces the data complexity of the model without affecting the accuracy of the energy disaggregation.The model reduces the idle time period aggregation power data and apriori information that without affecting the accuracy of the energy disaggregation,and then,it is cut down the data complexity of calculation.In this paper,we compare the energy disaggregation model with the public three datasets and adopt the universal evaluation criteria.The experimental results demonstrate the factorial hidden markov model based on the state of the appliance has more accuracy rate than traditional model.In addition,the proposed association rule learning model based on temporal partition has the smallest prediction error,the highest energy distribution rate and the highest energy disaggregation accuracy rate.This model has the same performance in the small power appliance and the same power appliances.