Studies On Algorithms For Power Grid Operation Section Estimation Based On Pattern Recognition And Machine Learning

Security check is a class of applications for security and stability checking of dispatching plan and operation,which is of great significance to the safety and economic operation of power grid.The generation technology of schedule power flow is the basis of security check analysis and calculation.The generation technology of schedule power flow should have the characteristics of good convergence,high calculation efficiency and high accuracy.In this paper,the generation technology of schedule power flow for security check is studied,and an algorithm for power grid operation section estimation is proposed.Main tasks are as follows:A similar section search algorithm based on pattern recognition is proposed.In the process of schedule power flow generation,the lack of planned data needs to be supplement first.The algorithm draws on the concept of similar days,and searches similar sections in historical sections based on pattern recognition.The algorithm chooses partial power flow data of similar section as supplements of plan data of the section to be checked to satisfy the data requirement of power grid operation section estimation algorithm.A schedule power flow generating algorithm based on state estimation is proposed.Aiming at the problems of poor convergence performance and low computational efficiency commonly existing in the traditional methods of schedule power flow generating based on power flow calculation,state estimation algorithm is used to calculate the schedule power flow in the paper.Drawing on the concept of correntropy in informatics,the state estimation algorithm based on maximum correntropy criterion is used to calculate the schedule power flow of the section to be checked,which is balanced power flow satisfying power constraints of transmission sections and various planning data.Because the state estimation algorithm uses redundant measurements and does not need step-by-step power flow adjustment,the schedule power flow generating algorithm has good convergence and high computational efficiency.An unbalanced power adjusting algorithm based on machine learning is proposed.There is unbalanced power between the calculated value of schedule power flow and the real power flow of the section to be checked.The traditional method of unbalanced power allocation is to balance regional power flow by adjusting the generation plan of designated generations.An unbalanced power adjusting algorithm based on neural network with single hidden layer and extreme learning machine is proposed to post-analyze the historical calculation results of schedule power flow and gain regionalized and accurate unbalanced power allocation characteristics.The measurement values of state estimation algorithm are corrected to improve the accuracy of schedule power flow calculation.An example of IEEE 9 nodes system and an actual example of Northeast Power Grid are designed to verify the effectiveness of the algorithms proposed in the paper.By using different construction methods of historical power flow data and different matching methods of similar sections,the functions of three modules including similar section search,schedule power flow generation and unbalanced power adjustment are validated in different examples.It is proved that the accuracy of the proposed algorithm is high.The advantages in convergence and computational efficiency are demonstrated in the paper.