Research On Power Data Mining In The Smart Grid

For ensuring the reliability of power grid operation,the optimal allocation of energy resources and the demand response for electricity users,the current research has focus on the smart grid.During the process of real-time acquisition,transmission and storage,large massive power data collected from generation-side to demand-side can be categorized into power supply,user demand,equipment status,etc.These data contain the evolution rules of user electricity usage,user demand behavior and equipment fault so that how to adopt data mining and data analysis in finding these rules is the key problem for smart grid.Electrical equipment fault would threaten the power grid operation.Thus,this paper selects transformer and high voltage circuit breaker for studying the intelligent fault diagnosis.Through the automatic analysis and self-study,artificial intelligence could improve the reliability and accuracy of equipment fault diagnosis and promote the intelligence of fault diagnosis for power grid.Load management on demand-side can guarantee the reliability and economy of power grid.The commercial load plays a significant role on the power grid.There are lots of electrical appliances and distributed energy resources in the commercial buildings.Thus,the data analysis for commercial buildings can be used in constructing load interval forecast,load scheduling strategy and the demand response program,which can promote the intelligent allocation of power resources.The followings are the contents in this paper:In order to solve the limitations of IEC three ratio method when applied in the transformer fault diagnosis,this paper proposes a fault diagnosis method based on SVM multi-class probability output and evidence theory.With the help of probabilistic output,SVM output turns from hard decision to soft decision,providing evidential theory with basic probability assignment.Evidence theory is suitable for uncertainty analysis,fusing various criterions into one uniform conclusion by means of evidence fusion formula.Case study based on dissolved gas validates the reliability,universality and accuracy of the proposed method.The current research on high voltage circuit break fault diagnosis lacks of unknown fault detection and real-time model classification updating.This paper proposes an adaptive fault diagnosis based on particle swarm optimization(PSO),support vector domain(SVDD)and fuzzy kernel clustering algorithm(KFCM).In the proposed method,P-SVDD can detect the unknown fault sample while P-KFCM overcomes the dependence on the initial parameters for recognizing the known samples.Combined with P-KFCM,normalized partition coefficient cluster validity analysis is used in learning new fault.The simulation results based on coil current demonstrate the effectiveness of the proposed method,compared with the famous algorithms.At present,the most of short term load forecast belong to point load forecast,which cannot give the fluctuation range of predicted load.Meanwhile,there is little related research on residential,commercial and industrial load.This paper proposes a day-ahead commercial load interval forecast based on similar day and kernel density estimation,which provide the decision-making basis for the load management and demand response in the regional grid.Based on the data analysis of commercial load characteristics,similar day unit decomposes point load forecast into daily average load prediction and pre-unit curve search and provides each time node with multiple predictive values.Then,these predictive values are synthesized into load interval by kernel density estimation unit.The reason for selecting kernel density estimation is due to its free-form and independence of the sample distribution,which is more fit for the randomness and volatility of commercial load.The simulation results collected from North Carolina State University,which is used in the commercial case study,demonstrate the universality and strong robustness of the proposed method.Few load scheduling strategies focus on minimizing the monthly electrical bill so that we present a plug-and-play demand response(DR)algorithm for economic monthly load dispatching in the commercial buildings on the basis of distributed energy resources and building end-uses.Based on the monthly load analysis,the monthly load dispatching consists of three modes(minimizing peak load,minimizing daily energy charge and balancing the minimization of peak load and daily energy charge).Then,two-stage nesting optimization algorithm is built for the optimal monthly electricity fee of demand charge and energy charge.In the outer layer,the daily optimal mode which firstly meets the constraints with a feasible solution is selected for the targeted day.Combined with PV,energy storage and plug-and-play DR,the inner layer optimizes the feasible solution of the selected optimal mode.Moreover,plug-and-play DR uses a penalty function for adjusting the amount of building end-use loads to be shifted before or after the targeted hours.The proposed DR,which is resource agnostic,can be as a generic control method for enhancing building energy management systems.Data collected from the North Carolina State University and the Duke Energy commercial TOU rate are used for the case studies.The simulation results show that the proposed method is satisfactory and robust.