Non-Intrusive Appliances Load Monitoring System Using Neural Networks

A non-intrusive appliances load monitoring system has been developed to ascertain the behavior of each electrical appliance in a household by disaggregating the total household load demand. This system does not need to intrude into a house when metering power consumption of each appliance. Therefore, the system has significant cost advantages and is less troublesome to the customers. This paper describes a non-intrusive monitoring system which is especially useful for inverter-driven appliances that frequently change their operational state. Conventional non-intrusive load monitoring systems cannot treat these inverter-driven appliances easily because of their complicated operation. In the developed system, the load consumption of household appliances is identified by the pattern recognition ability of a Neural Network (NN), which perceives the pattern of harmonics flowing out of the house. The patterns of current and phase at each harmonic order depend on what kind of appliances are in on or off. An air conditioner, a refrigerator, an incandescent lamp, a fluorescent lamp and a television set were used as household appliances to obtain training data for NN’s learning. The inference ability of the NN from unknown data was simulated. From the simulation result, it was verified that the NN could identify the operating status and load consumption of an inverter-driven air conditioner from the pattern of harmonics of the total household load. For this reason, the non-intrusive appliances load monitoring system for inverter-driven appliances such as air conditioners is viewed as reliable. Introduction For efficient management of electrical power companies, the prediction of peak load is one of the important factors. Precise prediction requires ascertainment of what and how household electric appliances are used, especially in the residential sector that has not yet been well investigated. In Japan, power companies monitor the usage of appliances in most case studies; however, such monitoring is not conducted very broadly. The reason is that the number of residential customers is enormous and the monitoring tools are less cost-effective. Therefore, a simple and convenient monitoring system is required for this purpose. Information gathered by the system can be used not only for efficient management but also for general promotion of energy conservation and energy efficiency. Figure 1 shows the trend of electricity consumption in Japan ([MITI] 1998). Because the industrial sector has made extraordinary efforts in energy-saving since the oil crises, their electricity consumption growth has been slower. Meanwhile, the residential and commercial sectors both show rapid growth. Accordingly, energy conservation measures in residential and commercial sectors are marked nowadays. If a simple and convenient monitoring system becomes available, energy efficiency in these sectors can be promoted. Information and Electronic Technologies 7.183 This paper proposes a new Non-Intrusive Appliances Load Monitoring System (NIALMS) using a Neural Network (NN) (Bishop 1995) for residential customers. The system disaggregates the total electric load to individual loads of each appliance by perceiving the pattern of harmonics flowing out of a house. It is especially useful for inverter-driven appliances that frequently change their operation modes. Pattern recognition capabilities of NN have been used to identify the operating state of the appliances. Such a system can lighten the burdens and the costs concerned with metering. 0 50 100 150 200 250 300 350 400 450 500 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 Year 10 3 G W ‚ ˆ 0 0.5 1 1.5 2 2.5 3 3.5 4 G ro w th R at e commercial residential industrial commercial growth residential growth industrial growth Figure 1. Annual Electricity Consumption in Japan The New NIALMS There are two general types of monitoring systems: intrusive and non-intrusive. The former is a common metering system that measures consumption load by connecting power meters to each appliance or feeder line inside a house. Since it is necessary to enter the house, this system is called “intrusive”. It can yield very accurate data but imposes costs and the burden of setting up sensors and data storage units on the households concerned. Therefore, there is an advantage to monitoring appliances from outside the home. This type of monitoring is called “non-intrusive” because there is no need to enter the house. This type infers the load of appliances using only information gathered at a revenue meter. Figure 2 shows an outline of the NIALMS. A conventional NIALMS (Hart 1992; Norford and Leeb 1996) infers individual loads of ON/OFF appliances by checking “signatures” that provide information on their activities. A signature is described as signature space, which consists of real and reactive power found from the difference between ON and OFF activity. In other words, if a step change is detected in total load, it means that an appliance has changed its activity and can be identified by checking the signature space. However, the conventional NIALMS encounters identification difficulty when there are inverter-driven appliances in a house. Inverter-driven appliances are difficult to discern as to their ON/OFF activity because their loads involve many usage conditions and can change smoothly and frequently, not just ON/OFF. Therefore, it is difficult to detect step changes and to check the signature space. Figure 3 shows a daily load changing of an inverter-driven air conditioner. Here, an inverter-driven appliance is one in which power is supplied through inverter circuits to make it easy and controllable.