Synthesis report on soft sensor activities at Hammarby Sjöstadsverk
Many process parameters at a wastewater treatment plant are expensive, difficult or even impossible to measure online, limiting the possibilities for efficient process monitoring and control. One way to provide wastewater treatment plants with online process information is so-called soft sensors. A soft sensor is a virtual sensor in the form of a mathematical model that estimates the value of a parameter whose value is unknown, e.g. a parameter that is hard to measure online, solely based on values of other parameters whose values are known, e.g. parameters that are easier to measure online.
This report summarizes results from IVL's soft sensor related activities at Hammarby Sjöstadsverk. It very briefly mentions two previous projects and focuses on the results from two master theses and two more recent soft sensor projects.
The parameters for which soft sensors were developed within the projects were typically different fractions of phosphorous, nitrogen, organic matter and suspended solids in various process steps. All soft sensor models were PLS-models calculated on laboratory data as y-values and online process data from the control system as x-values. The most recent projects also included data from acoustic sensors.
The performance of the soft sensors varied significantly and some of them showed promising results. The soft sensors that were based on acoustic data had in most cases comparable or better performance than corresponding models based on process data, suggesting that acoustic measurements is a promising approach. Furthermore, it was concluded that a crucial factor for successful soft sensor model development was access to large data sets from reliable online sensors and laboratory analyses. The data should represent a wide range of water characteristics and process conditions and there must also be enough for external validation of the models. It was also pointed out that well-maintained online sensors, automatic monitoring of model validity and re-calibration of models when necessary is important for well-functioning soft sensors when they are implemented in the process.
Future considerations such as stricter effluent regulations, more extreme weather conditions and a change of focus from just treating the wastewater to viewing it as a resource are predicted to further increase the need for better monitoring and control of the wastewater treatment processes. The rapid progress of information technology and further improvements of both acoustic measurements and model development will probably facilitate the development of reliable soft sensors and make it a potential approach to meet wastewater treatment plant's current and future needs for process monitoring.