Dissolved Oxygen Control in Biological Wastewater Treatments with Non-Ideal Sensors and Actuators

Abstract

The improvement of the dissolved oxygen control is one of the main objectives in the research works on control of wastewater treatment plants. In the research literature, most of the works are based on benchmark simulation models, where ideal sensors and ideal actuators are commonly considered. However, it is important to note that the main difficulty of the dissolved oxygen control is due to noise and delay in the sensors and actuators. These are taken into account in this article with the aim of dissolved oxygen control improvement using the benchmark simulation model no. 1. The main purpose of this work is to highlight the need to take them into account and conduct a first step in analyzing how they affect the usually considered dissolved oxygen control approaches. The work proposes an approach for dissolved oxygen control improvement within non-ideal sensors and actuators using the benchmark simulation model no. 1, where a precise catalog and characterization of sensors and actuators are also provided (although not used). Filters are used to reduce the noise of the sensors. Artificial neural networks are designed to predict the value of dissolved oxygen, to compensate the delay produced by filters and sensors, as well as to anticipate the time needed by the actuator to obtain the desired value. The artificial neural networks take into account the microorganisms present in the wastewater, as well as their food and energy source, to predict the value of dissolved oxygen. The article shows different options of artificial neural networks for dry weather, rain, storm, and variable set-point. The results show meaningful integral of square error improvements, around 80% in dry weather and greater than 50% with rain and storm influents, as well as a significant reduction of abrupt changes in the actuator.

Industrial Control under Non-Ideal Measurements: Data-Based Signal Processing as an Alternative to Controller Retuning

“…Applying this, we try to assure the decoupling of the delay correction from the considered control strategy. Results will show that the proposed system is able to yield similar and even better results than the system proposed in [36] where similar approaches have been designed and optimized towards the scenario.…”

“…Besides, the scalability is increased since ANNs can be applied in different industrial scenarios (not constrained to WWTP ones): the knowledge of the controlled scenario will be directly derived by ANNs if they are trained with proper data [10] (Chapters 1, 2, 5 and 17). Results in Section 4 show that our data-based enhanced control approach is able to offer similar and even better results in the control performance than the ones obtained in [36], a control approach specially designed and optimised to work in WWTP scenarios.…”

“…Now, the best improvement in terms of the IAE, a 41.05%, is offered by the ANNconf2 configuration when it is compared to the Default PI structure and the dry weather is simulated. When the performance of the proposed approaches is compared to the structure presented in [36], the improvement is reduced, however, results are still better. The structure performing better in the case of the dry weather is the ANNcon f 2, which is able to offer IAE and ISE values equal to 0.336 and 0.035, respectively.…”

“…In addition, it is important to highlight that if measurements are denoised and their delays corrected, the gain of the controller can be increased and therefore, an improvement in the control performance achieved. This principle was implemented in [36] and therefore, it has been considered in the implementation of this proposal. In terms of the control strategies where denoised and delay-corrected measurements are considered, one can observed that the best performance is not always given by the same configuration.…”

“…In terms of the input measurements, both architectures consider inputs obtained from the mass balance equation of the S O,5 variable, its conversion rate and the biological processes described in [36,37]. These variables are: The grid search methodology has been considered to determine the internal structure of both denoising approaches, i.e., their hyperparameters.…”