The Research On Algorithm Of The NIALM Of Residential Appliances

For alleviating the energy supply shortages and power system, it has significant meaning to strengthen energy consumption monitoring in buildings. Traditional monitoring method is restricted in application because of its costs and efficiency, the NILM(Non-Intrusive load Monitoring) emerged in recent years, this method only need to deploy data sampling equipment(namely electric meter) at power entrance, it can not only reduce monitoring costs, but also confer electric meter powerful function. With this technology, electric meter is not simply to implemention of measuring the total power consumption at aggregate-level, it’s able to estimate the power consumption at appliance-level.The NILM system is based on events in this paper, events refers to ON/OFF operation of appliances. The steady features of loads are used to load identification. On the one hand, the steady features of loads in normal operation obey statistical regularity. On the other hand, it requires larger storage and faster computing capacity to capture transient features and is not beneficial to practical application. This paper is based on previous studies of power change, current harmonic ratio and V-I trajectory, use them as steady features. It firstly presents definitions of each feature, then highlights how to extract the feature of event load from aggregate sampling data as well as event detection algorithm. This step is followed by verifing the extraction method though experiments.This paper then presents the method to built database with features of low-voltage residential appliances, put forward a way to combine decision tree algorithm and database, which can implement load identification and load classification from electrical data caused by different combinations of loads, the validity is verified by simulation and experiments. different features, this algorithm can avert weakness of features, it can recognise load in a simple and efficient way. Load identification algorithm compare extracted features of unknown load with features in database, then select the one having minimum difference as identification result, and load classification monitor those appliances having similar front-end circuit, it’s less dependent on completeness of database as well as load features. The essence of decision tree algorithm is dividing and conquering, it can combine different features in a simple and efficient way.