Line Loss Calculation Based On Machine Learning

As an extremely important economic and technical indicator in the power field,the line loss rate can comprehensively reflect the planning and design of the power system,production and operation status,and the management and operation level of the power enterprise.Scientific and reasonable theoretical line loss calculation is beneficial to the relevant departments to formulate loss reduction measures to improve economic efficiency.The line loss of the distribution network accounts for a relatively large proportion of the power grid loss and has a huge loss reduction space.However,due to its complex structure,numerous loads,difficult maintenance of the distribution network structure diagram,and difficulty in obtaining complete operating data,the traditional line loss calculation method requires a lot of manpower and material resources in practical application,and the calculation is difficult.A large amount of line loss related data has been accumulated in years of line loss calculation work.Machine learning algorithms can use historical data to establish the relationship between electrical characteristics and line loss,thereby simplifying the line loss calculation process.Therefore,this paper proposes a method for calculating the line loss of distribution network based on machine learning.Firstly,preprocess the historical data of the distribution network to eliminate the influence of outliers on the calculation results and improve the quality of the data set.Secondly,this article fully considers the influence of the power load of the distribution network and the first-end power supply curve on the line loss,and extract the characteristics of the active power of the load based on the actual physical meaning and mathematical statistics.The active power of 24 points in the power supply curve has the characteristics of high correlation and large data scale.Therefore,the principal component analysis method is used to extract features,which not only retains most of the information in the original data,but also achieves the purpose of streamlining the data set size and reducing the correlation.Then combining the distribution network structure and grid operation data,the electrical characteristic system is established.Thirdly,the BP neural network model is established.The neural network is trained using historical operating data of the distribution network,and the complex function relationship between the multiple electrical characteristics and the line loss results is fitted.BP neural network has several problems such as easy to fall into local optimum,slow convergence and difficult to determine hyperparameters.Optimization algorithms such as Nesterov momentum stochastic gradient descent,RMSProp and Adam are used to optimize the neural network model,and the Bayesian hyperparameter tuning algorithm is used to determine the hyperparameters,the problem of low efficiency of manual parameter adjustment is solved.Finally,according to the actual power grid data in a certain area,the influence of different electrical characteristic systems on the line loss calculation results is analyzed.The results show that the proposed method can perform accurate and effective line loss calculation,and the required data is easy to obtain,which has certain practical significance.