Prediction for water level of steam generator based on multi-signal fusion by support vector regression

During change of marine nuclear power plant load, the water level of steam generator often waved heavily and even more false water level occurred. In the mean- time, water level measurement channels failed once in a while. All problems affected heavily self-regulation of water level and operator's correct judgment on the operation states. Therefore, a water level predicting method based on multi-signal reconstruction was brought forward. This method added relevant signals which may affect water level change as input information for prediction. Comparing with predicting method based on analyzing the history of water level change only, this method improves accuracy, stability and reliability of prediction, and supports predicting in longer term. (authors)