Research On Perception Method And Impact Assessment Of Non-intrusive Load Cluster

With the promotion of the strategy of carbon peak and carbon neutralization,China is speeding up the adjustment of energy production and consumption structure,and the importance of the new power system as the key carrier to achieve the goal of double carbon has become increasingly prominent.Under this background,it is urgent to give full play to the important role of power demand side management.As a key technology to realize power demand side management,load monitoring can provide a basis for intelligent and refined power load management.In recent years,the nonintrusive load monitoring method has received widespread attention.This method only needs to install monitoring equipment at the power entrance,which does not affect the daily life of users and has low economic cost.Considering that the current nonintrusive load monitoring methods are mostly focused on the perception of different types of individual load states,which are not suitable for load clusters composed of a great quantity of similar loads with similar parameters and consistent operating conditions in commercial buildings,schools and other scenarios.In addition,the impact of the access of a large number of load clusters on the power system cannot be ignored.In view of the above problems,this thesis proposes a non-intrusive load cluster sensing method and its impact assessment method.The main research content of this thesis is divided into the following four parts:(1)The domestic and foreign research status of non-intrusive load monitoring technology and load cluster perception are summarized.Combined with the functional structure and operation characteristics of load cluster,the basic framework of nonintrusive load cluster perception is constructed.The preprocessing method of load cluster operation data is introduced,and the typical random characteristics and waveform characteristics of load cluster are analyzed,which provides data support for subsequent experiments.(2)According to the random characteristics and data distribution characteristics of load cluster,a load cluster clustering algorithm based on a two-layer Gaussian mixture model is studied.Firstly,K-means algorithm is used for pre-clustering to quickly determine the number of clustering centers,and then two-layer Gaussian mixture model is used to achieve fine clustering of load cluster categories.This algorithm is compared with the K-means algorithm in the commonly used load cluster data,which verifies the effectiveness of the clustering algorithm.(3)Using the relevant theory of knowledge graph,knowledge is extracted from heterogeneous and multi-source load cluster operation data,and the feature graph of load cluster is constructed through feature optimization fusion and processing.The underlying load cluster operation data with random characteristics is abstracted into high-dimensional information suitable for classification and decision-making.The combined support vector machine models corresponding to each type of load cluster are constructed respectively,which realizes the accurate identification of the type of load cluster.The universality and effectiveness of the proposed algorithm are verified using load cluster data from measured data and Combed dataset.(4)According to the different computing resources required by various parts in the process of non-intrusive load cluster perception,a non-intrusive load cluster sensing system based on end-cloud cooperation is proposed.The load cluster category identification process in the actual scene is implemented on the elastic cloud server provided by Huawei Cloud.The comprehensive evaluation method of the power consumption impact of the load cluster connected to the power system is proposed,which combines subjective and objective weighting to complete the comprehensive evaluation of each load cluster in different periods of time.