基于机器学习的新能源场站实时异常检测模型研究OA

In order to solve the problem of high cost of grid-connected assessment caused by the lag of automatic generation control(AGC)and automatic voltage control(AVC)data monitoring in new energy stations,and promote the transformation of industry supervision mode from"passive assessment"to"active control",this study constructs a real-time anomaly detection model based on machine learning,focusing on breaking through the problem of cross-security zone data fusion and dynamic index calculation.By designing the standardized ana-lytical framework of IEC 60870-5-104 protocol(hereinafter referred to as 104 protocol)and power E lan-guage,efficient data processing of 120 000 records per second is realized.Combined with the Kalman filter cali-bration mechanism,the deviation rate of key indicators such as AGC operation rate is ensured to be less than 3%.At the same time,based on the stream parallel computing framework and the double-set power prediction optimization model,it supports real-time processing of massive data and prediction accuracy is excellent,pro-viding a technical basis for active management and control.During the trial operation of the model,the results are remarkable,the assessment cost is reduced by 286 000 yuan,the number of passive assessments is reduced by 73%,and the accuracy of anomaly detection is 96.7%.The real-time performance of data processing is ex-cellent,and the average delay is less than 100 ms.The root mean square error of the power prediction is re-duced by 25%after optimization,and the deviation rate of the main station is optimized from 11.8%to 2.7%,which fully verifies the practicability and reliability of the model.The real-time anomaly detection model con-structed in this paper effectively solves the problem of timeliness and accuracy of new energy station supervi-sion,provides technical support for the assessment of"two rules",and promotes the industry to achieve refined and intelligent management.The real-time,stability and scalability of the model meet the needs of large-scale supervision.It has the value of popularization and application.