Research On Algorithms Of Non-Intrusive Load Decomposition

With the rapid development of smart grid, load monitoring of power system has higher require of intelligentized level than before. Most of the current monitor methods are intrusive which need monitoring every load of the system. Those methods require a large number of hardware devices. So it costs too much to purchase, install, maintain those devices. Moreover, each monitoring device is a data source. So it will be difficult to process and analyze the data when a power system has lots of loads. To solve the above problems, non-intrusive load decomposition emerged. Non-intrusive load decomposition is a method of obtaining the amounts and status of all loads in the power system by installing monitoring devices on system entrance and analyze the data it produces. So non-intrusive load decomposition can also realize the monitoring of the loads inside of the power system.At present, the non-intrusive load decomposition research is still in its infancy stage which has not yet form complete theoretical system and mature algorithms. So this paper studies the non-intrusive load decomposition of small power system steady state and transient state and proposing a series of algorithms with robustness and completeness that have been tested by lots of examples. The main results of this paper are as follows:(1)Establishing steady state current decomposition model and its improved form. Proposing steady state current decomposition algorithms based on this model. Then using the algorithms for decomposing steady state examples. The results show that the algorithms can obtaining the amounts and status of the load inside of power system which prove that the algorithms could realize the non-intrusive load decomposition of power system steady state. At last, two methods are presented for the applications of those algorithms.(2) Proposing a series of algorithms of transient state current’s detection, separation, feature extraction, feature space dimension reduction, transient state distiguishment, result verification, ect. Then using those algorithms for decomposing a large number of transient state examples. The results show that the algorithms can detect the linearly separable transient state of power system effectively and judge the type of transient state. So the algorithms realize the non-intrusive load decomposition of the power system’s linearly separable transient state.(3) Considering power system has linear non-separable transient state, we increase the dimension of feature space which learning form theory of pattern recognition. Then distinguishing the transient state in higher dimensional space. At last, proposing an efficient algorithm. The results of examples analysis show that the algorithm can decompose the linear non-separable transient state with high-efficient. Linear non-separable algorithms and linear separable algorithms constitute the complete transient state decomposition algorithm together.