Voltage Stability Margin Assessment Based On Decision Tree And Random Forest

With the development of society,the power load is increasing.Due to the constraints of environmental conditions,the power load center is far away from the power center,the operation of the power system is close to the limit,and the voltage stability problem is increasingly prominent.To prevent voltage instability,voltage stability analysis of the power system is required.Ways to analyze voltage stability include off-line analysis of online applications and direct online analysis.When analyzing the voltage stability,it is necessary to calculate the different operation modes and grid structure of the power system.As the size of the power grid increases and the interconnection between the regional power grids,the complexity of the network and the diversity of the operation modes increase continuously.The access to large-scale renewable energy has added more uncertainties to the operation of the power system.Grid operators need to analyze more system topologies and operating conditions to obtain voltage stability assessment rules.It takes a lot of calculation and a long time to obtain voltage stability judgment rules under various operating conditions.Direct on-line analysis of voltage stability can't meet the time requirement.Manual or semi-automatic method is very time-consuming to analyze offline data and obtain voltage stability evaluation rules.The analysis results depend on personal experience and knowledge.Because the experience and knowledge of each grid operator are limited,the analysis results are not comprehensive.Therefore,it is urgent to extract voltage stability evaluation rules automatically and effectively.This paper proposes a method to automatically extract voltage stability assessment rules by data mining technology.The core idea is to extract voltage stability determination rules from massive data by data mining technology on the basis of massive data obtained by offline simulation,and the obtained rules can judge voltage stability online.First,a large amount of offline data is obtained through offline simulation.Grid dispatchers usually divide the grid operation status into:normal state,warning state and emergency state.This paper divides the grid operation state into the above three states according to the voltage stability margin value,performs PV curve analysis on the power system,and calculates the PV curve.The voltage stability margin values at each operating point are obtained,and the sample set required by the data mining algorithm is obtained.Secondly,on the basis of obtaining the voltage stable sample set,due to the high dimensionality of the power system,the input samples of the data mining algorithm contain more training attributes,which makes the model training time longer.In order to remove redundant attributes and improve the computational efficiency of data mining algorithms,based on the consideration of voltage instability,this paper selects the training attributes closely related to voltage stability.Firstly,based on modal analysis,the voltage stability affecting power system is determined initially.The key attributes are further optimized according to the correlation coefficient method,and finally the best data mining algorithm training attribute set is obtained,which reduces the dimension of the training attribute and improves the computational efficiency of the data mining algorithm.Then,a decision tree algorithm is used to establish a static voltage stability margin assessment classification model for the power system.This classification model has certain generalization ability and high accuracy.From the model,a set of rules capable of judging the steady state of the power system is obtained,and the running state of the power system can be judged in real time.Finally,in order to improve the evaluation accuracy of the voltage stability margin assessment model a random forest algorithm is used to establish a voltage stability margin assessment model for the power system.Considering the grid structure of different N-K faults,different load growth directions are considered when generating PV curves,and a more comprehensive training sample set is considered to evaluate the static voltage stability margin online.The grid dispatcher can use the data mining model to quickly determine the grid stability state based on the data collected in real time,and formulate corresponding control measures according to the decision rules in the decision tree and the random forest model.