Research On Power Load Modeling Method Based On Component-Based Modeling Approach

The mathematical model of power load is an important part of power system digital simulation,and its accuracy will directly affect the results of system simulation.Because of its non-linearity,time-varying,decentralization and other characteristics,the research on load modeling has been slow.Especially,the research on load modeling based on statistical synthesis method has obviously lagged behind.With the economic development and social progress of our country,the scale of power system is getting larger and larger,and the requirement for stable operation and reliability of power grid is getting higher and higher.The capacity and composition of power load are also undergoing tremendous changes.Under such a background,the research and establishment of a model that can truly and accurately reflect the actual load of the power grid has become a key issue in the field of load modeling research.Based on the load data of typical induction motor,this thesis uses the algorithm of machine learning as the technical support,and carries out the load modeling research based on Component-Based Modeling Approach.This thesis firstly improves the shortcomings of the traditional Component-Based Modeling Approach,and introduces the hierarchical agglomeration clustering algorithm in machine learning to study the classification of electric power load represented by induction motor.The results show that the algorithm can effectively classify electric power load according to its characteristics.The results were good and accorded with the expected results.Then,the application of artificial neural network to load modeling is studied,and an improved BP algorithm is proposed to optimize the network training process.There are two main improvements.One is to change the fixed learning rate to the adaptive learning rate,so that the algorithm can adjust the learning rate adaptively according to the learning situation,so as to improve the learning efficiency.The other is to introduce a new neuron activation function to improve the network performance.The test results show that the improved strategy effectively improves the computational speed and accuracy of the algorithm.This thesis also analyses the characteristics of load static model in polynomial form and power function form,identifies the parameters of the same static load curve,and finds that power function form has better advantages and is more suitable for static load model.Based on the current popular TensorFlow framework,an artificial neural network is compiled and implemented to build a dynamic load model.When building the model,the dynamic characteristic data of the load is calculated by using the statistical load data,and then the neural network is trained with these data as the training samples.After the training,the neural network can be used as the model of real load for other related digital simulation calculation.The results of validation calculation show that the improved BP neural network proposed in this thesis can complete the dynamic load modeling well,and the dynamic characteristics of the model obtained are very close to the original load.It is a feasible load modeling method.