Introduction
The influent flow to most waste water treatment plants (WWTPs) is currently
controlled by only the level in the sewerage. This level-based control of the
influent flow causes peak discharges at a WWTP after rainfall events, even if there is enough storage
available in the sewerage to discharge the precipitation in a more gradual
way. As a result both hydraulic and biological peak loads are considerable.
This affects the performance of the WWTP. Furthermore, the capacity of the
post-treatment is in general smaller than the maximum hydraulic capacity of
the WWTP. This results in a significant bypass of the post-treatment during
peak discharge.
Outline of the predictive control for the influent flow of WWTP
Bennekom .
![]()
Therefore, reduction of peak discharge by optimising the influent flow
control is expected to be effective. Predictive control with dry-weather
forecast can be applied to limit the influent flow at the end of (or even
during) rainfall events. Where level-based control is characterised by a
reactive response, predictive control anticipates changing circumstances. In
this way the available storage of the sewerage can be used without causing
extra combined sewer overflow (CSO) or water on street (WOS). This increases
the treatment efficiency of the whole water cycle, which benefits the surface
water quality.
For WWTP Woudenberg preliminary study was carried out on using predictive
control. It was shown that the amount of bypass can be reduced by
approximately 65 % . The study used a
conservative approach, in which the discharge is limited after the
rainfall event. It was suggested that the amount of bypass could be even more
reduced in case the discharge is limited during the rainfall event.
However, this could lead to more CSO or WOS if amounts of rainfall are
predicted incorrectly. To verify the results of the preliminary study, a
pilot project was started for four catchment areas, to investigate the true
reduction of bypass using predictive control. This paper describes the
methodology, implementation and results of one of the four pilots
illustrating the effectiveness of predictive control in reducing peak
discharges to the WWTP.
Material and methods
A predictive controller for influent flow is implemented in this pilot
project using the Aquasuite® Advanced
Monitoring and Control platform . The controller is
used to limit the flow at pumping stations discharging from the sewerage to
the WWTP. Rainfall and dry-weather
flow (DWF) predictions are applied as the basis for the predictive control.
The rainfall prediction is obtained from the High Resolution Limited Area
Model (HIRLAM) of the Royal Netherlands Meteorological Institute (KNMI). The
DWF prediction is obtained by a measurement data driven technique
. Based on real-time measured water levels in the
sewerage and both rainfall and DWF predictions, a discharge limitation is
determined by a volume optimisation technique . This
is shown in Fig. .
During a rainfall event, the discharge limitation is at maximum capacity.
After the rainfall event, when levels are below a defined critical level, the
sewerage is emptied with a limited discharge. In Fig. an example
with a discharge limitation at 70 % of the maximum capacity is
shown. Based on the time until the next rainfall event and the maximum time
to empty the sewerage, the influent flow is optimised between the maximum
capacity and the capacity of the post-treatment.
Rainfall forecast
As travelling times of the sewerage can take up to 24 h ,
precipitation predictions with similar periods are required. Optimal
utilisation of the storage of the sewerage is only possible with information
of subsequent rainfall peaks. Therefore, at the beginning of this project,
HIRLAM of the KNMI was selected as the numerical weather prediction (NWP)
forecast system from which the precipitation data were obtained. This
rainfall forecast has raster cell sizes of 11.0 by 7.0 km, a forecast
horizon of 48 h and a refresh rate of circa 6 h. A single time series is
obtained for each specified area by transforming the information of the
raster cells within the polygon by application of a geostatistical method.
Dry-weather flow prediction
An accurate prediction of the dry-weather flow is essential to determine the
total predicted flow to the WWTP. The influent flow has, at WWTPs with mainly
domestic waste water, a typical day pattern. Both (time-dependent) water
demand of inhabitants and contribution of industrial discharges, as well as
properties of the sewerage like travelling time, influence the exact
characteristics of this pattern.
A data driven technique is applied to obtain this DWF prediction which is
self-learning based on real-time flow measurements. Based on previously
occurring patterns of each specific day of the week, a daily curve is
automatically obtained, except for those with deviations due to peak flows
caused by precipitation. Combination of the prediction of the total daily
volume with this predicted curve determines a DWF prediction. For each
specific catchment area, a distinct DWF prediction is obtained. The DWF
prediction has a 48 h forecast horizon, just like the rainfall forecast.
This particular technique is based on the fully adaptive forecasting model
for short-term drinking water demand . It is so
generic that it can be applied to waste water discharge with some small
adjustments of the settings. Before this pilot project, the application of
this technique to waste water discharge had already successfully been
implemented, yet with a rather different objective. DWF prediction is an
accurate measure for the load of biological oxygen demand (BOD) prediction.
Therefore, it can be applied within a predictive controller for aeration
.
Predictive control
A volume optimisation technique is applied to determine predicted storage and
discharge. This technique is in principle equal to the predictive control
applied in drinking water supply . For that specific
purpose, the outflow of a reservoir is determined by the water demand
prediction of the supply area, and the inflow of the reservoir is kept as
constant as possible within the constraints. In this pilot project, this
technique is applied to reduce peak flow to the WWTP after rainfall using the
available storage of the sewer. The outflow of the reservoir is kept limited
within the constraints.
The predictive controller for influent flow control as implemented for this
pilot project consists of the following components. Firstly, actual utilised
storage of the sewerage is derived from real-time water level measurements
and relationships between level and volume. These relationships are obtained
from storage curves derived from sewer models and are verified with measured
values for level and discharge. Subsequently, the inflow to the sewerage is
predicted. The sum of runoff, DWF and discharge predictions of connected
upstream catchments determines the inflow prediction. The runoff is based on
the connected paved area and the precipitation prediction. In the third step,
the sewerage is modelled as a single reservoir for each catchment, which is
optimally used within the imposed constraints. This results in the optimal
outflow prediction.
Only if the actual level in the sewerage is lower than a critical level and
the predicted precipitation is below a threshold, is discharge limitation
turned on. The discharge is limited to an optimal capacity meeting the
imposed constraints. In case a subsequent significant rainfall event is
forecasted at a moment before the sewerage can be completely emptied at
post-treatment capacity, discharge will be optimised. Depending on the local
configuration, the discharge can either be gradually limited between maximum
and post-treatment capacity or kept at maximum capacity for a calculated
period after the rainfall event. In this way, the sewerage is emptied before
the next rainfall event, and bypass of the post-treatment is minimised.
Implementation
For the pilot project this control is implemented for the WWTPs Bennekom,
Ede, Woudenberg and Harderwijk (Netherlands). These are conventional WWTPs
with a post-treatment. The allocated hydraulic capacity of the post-treatment
is for all these locations lower than the hydraulic capacity of the WWTP. The
implementation is executed in close cooperation between water authority
Vallei and Veluwe and the connected municipalities. The predictive controller
is implemented for all four locations. This study analyses the implementation
of WWTP Bennekom in detail.
Before implementation, influent flow at WWTP Bennekom took place at three
different stages: 200, 500 and 1000 m3h-1. Only switching
between fixed stages is possible, since none of the screw pumps are provided
with variable speed drives. However, above 700 m3h-1 silting of
the sand filters occurs. Therefore, an extra fixed stage of
700 m3h-1 was configured for this project. In case of suitable
conditions discharge is limited from 1000 to 700 m3h-1 by the
predictive control.
Performance of the predictive control at WWTP Bennekom for the
period February–September 2016.
![]()
Example of bypass reduction with a conservative approach at WWTP
Bennekom for the event of 4 March
.
![]()
Phasing and monitoring
The predictive control was first implemented in advisory mode to check the
results in the real-time, full-scale situation. After this period the control
is switched on if the results from the advisory period were satisfactory. The
control is continuously monitored by the application of key performance
indicators (KPIs) for both sewerage and post-treatment. Storage utilisation
and travelling time were defined as the main KPIs for the sewerage. The
efficiency of the predictive control itself was defined as a major KPI for
the post-treatment. The efficiency of the predictive control is determined as
the ratio of the bypass of the post-treatment prevented and the amount of
bypass for the situation without predictive control .
Results and discussion
In advisory mode the results prove that most of the rainfall events could
have been discharged to the WWTP with a reduced capacity and therefore the
amount of bypass can be reduced without extra CSO. At the beginning of
February 2016 the predictive control for WWTP Bennekom was activated. The
predictive control has been continuously activated, apart from the period
between 10 and 12 February and 13 and 14 March due to maintenance activities.
During the analysed period (February–September) several precipitation events
which could cause bypass occurred.
Performance control total period
The performance of the predictive control at WWTP Bennekom for the analysed
period is shown in Fig. . Both the amount of bypass that occurred
and the amount that was prevented are presented for each day. For those days
where the predictive control was (partially) deactivated due to maintenance
activities, the prevented bypass was predicted.
For the analysed period (February–September 2016) the results at WWTP
Bennekom show that 51 % of bypass volume has been prevented with
predictive control compared to the original situation with level-based
control. The results also show that big differences occur between the months
regarding the effect of the predictive control. The analysed period covers
two-thirds of the year, including the whole summer period.
During the first 3 months (February–April) 67 % of the bypass could
have been prevented by predictive control; however, this was in reality
12 % lower due to maintenance activities. The effects of subsequent
rainfall events were limited, since the intervals between distinct predicted
events were in general longer than the travelling time with reduced capacity
. Maximum travelling times did not exceed 12 h and the
reduction of bypass occurred without extra CSO. The precipitation during this
(early) spring period can be characterised as gradual.
Example of bypass reduction with a conservative approach at WWTP
Bennekom for the event of 16
September.
![]()
During the next 2 months (May and June) 33 % of the bypass was
prevented by predictive control. Especially June was a month with severe
thunderstorms. The precipitation during this (early) summer period can be
characterised as difficult to predict and with large amounts within short
time periods. The effects of subsequent rainfall events were large. Discharge
limitation of the predictive control can only occur below critical levels and
without significant rainfall in the near future. Finally, during the last 3
months (July–September) 70 % of the bypass was prevented by
predictive control.
Performance: single rainfall events
In this pilot project, a conservative approach is used in which the
predictive control is limited after the rainfall event. In
Fig. bypass reduction after a rainfall event of 4 March
is illustrated. The efficiency of the predictive control for this single
event accounts for 47 %. This is entirely due to the chosen approach
of applying discharge limitation after the rainfall event, not a
subsequent rainfall event. If the progressive approach had been applied,
allowing discharge limitation during the rainfall, bypass could be
completely prevented for this specific event. This holds for the majority of
the events in the spring period February–April .
The progressive approach could also be of interest for the summer period. In
Fig. bypass reduction after a rainfall event of 16
September is illustrated. Also for this thunderstorm, bypass could be
completely prevented with the progressive approach. Although the progressive
approach is riskier regarding CSO, as mentioned before, it is worth
considering from the perspective of bypass prevention. Risk reduction can be
obtained by increasing the accuracy of the precipitation prediction.
Precipitation prediction
The results show that most rainfall events are adequately predicted with
respect to appearance and timing, and it is shown that this prediction can be
used for control. However, further research can show whether the shape and
volume of rainfall forecast might be improved by application of more advanced
techniques, or a combination of techniques. Especially for the summer period,
Hirlam Aladin Regional on Mesoscale Operational NWP in Euromed (HARMONIE)
precipitation prediction performs better than HIRLAM . In
addition, HARMONIE has raster cell sizes of 2.5 by 2.5 km ,
which results in a more distinctive selection of cells for each catchment.
Nowcasting extrapolates the initial condition, determined by detailed
observed data, to a forecast with a horizon between 0 and 6 h. It offers
more accurate information for the very short term, but it loses its value for
the long term . Although the forecast horizon of
nowcasting is too short in comparison to travelling times in the sewerage, it
might be meaningful to combine it with HARMONIE to use the best of both
worlds. NWP unpredicted events could be covered with nowcasting. Also, the
usage of the spread of ensemble forecasts accounting for uncertainty is
considered. Discharge limitation could be disabled in case of unpredictable
weather.