基于CNN-GAN数据增强网络的电厂锅炉管道温度压力及健康状态预测

Prediction of Temperature,Pressure and Health Status of Power Plant Boiler Pipeline Based on CNN-GAN Data Augmentation Network

电厂锅炉管道系统随着机组参与深度调峰而承受的交变载荷不断增加,对管道温度、压力预测和健康状态评估提出了新的要求。基于电厂真实管道温度、压力数据,提出了多阶管道健康评价算法,搭建了CNN-GAN数据增强网络。通过CNN-GAN数据增强网络与管道真实运行数据,生成带管道健康评价函数的合成数据集。预测模型采用双层LSTM结构,对管道未来30步的温度、压力和健康状态进行预测。所提出的多阶管道健康评价算法能够准确判定温度、压力超额定和短期波动较大等非正常工况,对指导运维人员检修以及提前预警具有一定的价值。CNN-GAN算法比原始GAN能够生成更加真实的管道温度压力仿真数据集。未来30步的预测步长中,不同工况条件下,建立的预测网络最大绝对误差控制在6%以内。

With the unit participating in deep peak regulation,the alternating load of boiler pipe system in power plant is increasing con-tinuously,which puts forward new requirements for pipe temperature and pressure prediction and health state assessment.Based on the real pipeline temperature and pressure data of the power plant,a multi-stage pipeline health evaluation algorithm is proposed and a CNN-GAN data augmentation network is built.Through the CNN-GAN data augmentation network and the real operation data of the pipeline,a synthetic dataset with a pipeline health evaluation function is generated.The prediction model uses a two-layer LSTM structure to predict the temperature,pressure,and health status of the pipeline for the next 30 steps.The proposed multi-stage pipeline health evaluation algo-rithm can accurately determine abnormal working conditions such as excessive temperature and pressure and large short-term fluctuations,which has certain value for guiding operation and maintenance personnel to maintenance and early warning.The CNN-GAN algorithm can generate a more realistic pipeline temperature and pressure simulation dataset than the original GAN.In the next 30 prediction steps,the maximum absolute error of the established prediction network is controlled within 6%under different working conditions.