Research On Non-intrusive Load Feature Extraction And Load Recognition

With the sustained growth of electricity consumption,the peak-to-valley difference of the power grid gradually expands,and requirement for the flexible control of demand side also increases.The demand-side flexible loads show considerable controllable potential and have become the main target of demand response.In recent years,concepts such as energy-saving and emission-reduction and energy efficiency optimization have been rapidly popularized,which play a role in promoting the optimization of user consumption management.Load monitoring provides basic support for the above-mentioned issues.The development of smart grid,smart home and advanced metering infrastructure technologies provide a good platform for load monitoring applications.Researchs on load monitoring and decomposition are with broad prospects.The load monitoring data contains valuable user energy consumption information,which can be used to abstract internal details through data mining analysis.Those details are meaningful to the application of the two-way interaction strategy in the smart grid,demand side management,power grid construction,load optimization management,etc.Besides,they can also provide auxiliary references for electric power companies to develop work plans and for users to grasp their own power consumption so as to actively participate in grid interaction.In view of the impact of users’ power consumption patterns on load characteristics,this paper proposes a new non-intrusive load monitoring method,uses clustering analysis techniques to construct load signature database of steady-state features based on the differences in users’ power consumption patterns,combines multi-signatures with conditional judgment to build the non-intrusive load decomposition model,finally works out load recognition through genetic optimization.Firstly,this paper describes the current development of non-intrusive load monitoring,then analyzes the typical load signatures and the corresponding extraction techniques from the perspectives of time domain,frequency domain and graphics,including the PQ characteristics,current characteristics,VI characteristics,harmonic characteristics,instantaneous characteristics and transient characteristics.Furthermore,the necessity of load signatures combination is analyzed,and the significance of establishing a load signature database is discussed due to the essential needs of load recognition.Secondly,this paper foucuses on the influence of users’ power consumption patterns on load characteristics.According to the correlation between users’ electrical bahavior and the load operating strategy,this paper analyzes the users’ power consumption pattern from two aspects of different usage time and different behavioral habits,then proposes a method of subdividing the consumption patterns by weekdays and weekends through clustering analysis so as to construct load power signature data sets.By comparing typical clustering analysis algorithms,this paper proposes a power partitioning process based on affininity propagation clustering analysis,and realizes power partitioning of air conditioning load at different usage periods.Furthermoer,combining the specific operating mode resulted from the user’s electrical behavioral habit,a signature database of sample loads is constructed,which provides support for load decomposition.Finally,this paper constructs a load decomposition model based on steady-state power signatures,then proposes a non-intrusive load recognition method based on conditional judgment and genetic optimization.This paper designs conditional judgment to working state rationality and auxiliary conditional judgment to user’s specific electrical behavioral habit to realize decomposition of different electrical appliances and the corresponding operating states from the total load signal.Then multi-scenes of load recognition are carried out and the recognition results are evaluated through index parameters.The case study shows that the proposed method realizes effective identification of various electrical appliances and their working conditions.The proposed method focuses on easily accessible steady state power signatures as main recognition basis and helps to reduce hardware costs to some extent.Besides,the proposed method considers the impacts of user’s power consumption pattern and improves the load recognition effects,which is of practical significance to some extent.