Archive | 2021
Network-wide sewer odour and corrosion management by model predictive control
Abstract
Sewer networks are one of the most critical urban water infrastructures of modern societies. However, they suffer from serious corrosion and odour problems due to the production of hydrogen sulfide (H2S), which cost several billion dollars every year globally for sewer replacement and rehabilitation. Chemical dosing is commonly applied to reduce H2S through chemical reactions. However, the dosing rates typically follow deterministic guidelines rather than being controlled online. In addition, sewer networks are hybrid systems consisting of discrete pump operations and continuous dynamic sewage flow and compositions, leading to extreme difficulties to achieve desired distributions of the dosed chemical across the network. No effective strategy for network-wide sulfide control is currently available, leaving the water industry with no other options than over-dosing, accompanied by large costs and still unsatisfactory performance. The overall aim of this thesis is to develop and demonstrate smart real-time control (RTC) strategies to achieve network-wide control of H2S through cost-effective use of chemicals.In this study, a novel model predictive control (MPC) approach is developed, which integrates three predictive models and one operation optimization algorithm to control sewer odour and corrosion problems. The three mathematical models are used to predict the sewage flow rates entering the network, the hydraulic flow in the network and the sewage characteristics in the network, respectively. The optimization algorithm then searches for optimal operation of the chemical dosing stations and selected sewage pumping stations (SPS) to have hydrogen sulfide controlled across the whole network.First, in Chapter 4, an autoregressive with exogenous inputs (ARX) model was developed for real-time prediction of the inflow into a network with a prediction horizon of a few hours. It was calibrated through regressing the historical flow measured by supervisory control and data acquisition (SCADA) system, and the rainfall data from a local rain gauge. The approach was validated at two SPSs with different hydraulic characteristics and catchment features, both achieved good prediction performance during both dry weather and wet weather. It was then used for supporting online Mg(OH)2 dosing in a rising main targeting a pH of 9 at the end of the pipe, and maintained pH at 9.09 ± 0.04 during dry weather, and 9.00 ± 0.08 in wet weather with storms. Compared to flow-paced dosing (an industrial dosing method), the chemical saving was between 10% to 50% under varying weather conditions.In Chapter 5, with the ability to predict future inflows into a network with the ARX model, an online optimization algorithm was developed to search for the optimal control sequence (chemical dosing rates and scheduling of selected pumps over a pre-selected horizon) to maintain the dosed-chemical concentration at the target location(s) at pre-selected levels. First, optimal chemical delivery was formulated as a dynamic constrained multi-objective optimization problem. An improved elephant herding optimization (iEHO) algorithm was established to select an appropriate control sequence with both the control performance and chemical costs consider. The algorithm presented satisfactory convergence rates in an astronomical searching space. For large networks, the high computational load was further reduced using a newly proposed optional-event-driven strategy, with which optimisation is not triggered by less significant pumping events. The holistic method was verified through simulating Mg(OH)2 dosing in a real-life sewer network, involving 14 SPSs, with real flow data as inputs. The proposed optimisation algorithm outperformed industry standard methods (flow-paced dosing without pump scheduling) by reducing the under-dosing periods by 66%, and the over-dosing periods by 61%. It also outperformed the generic optimisation algorithm, reducing under-dosing periods by 39% and over-dosing periods by 48%.The optimisation method was further developed in Chapter 6. In Chapter 5, the control objective was set as satisfactory distribution of the dosed chemical at the targeted locations, independent of sulfide levels. However, sulfide production in sewers is highly dynamic, affected by many physical factors and in-sewer biochemical processes. Thus, in Chapter 6, the control objective of the real-time dosing was improved to control sulfide at the target points at or below a pre-specified level, rather than maintaining a constant chemical concentration. A simplified model for predicting sulfide production in the sewer network was developed. This model was then incorporated into the iEHO algorithm. The model served to predict sulfide distribution across the sewer network for each control sequence generated by the iEHO in search for the optimal control sequence.\xa0 This method was validated via a simulating study of a real-life sewer network with FeCl2 dosing. The required amount of Fe2+ at each control step could be adaptively determined based on the sulfide to be precipitated. With the aim of reducing the total dissolved sulfide (TDS) at the rising main network discharge point to below 0.5 mg S/L, the controlled TDS was 0.03 ± 0.01 and 0.02 ± 0.01 mg S/L during dry and wet weather, respectively.The networks investigated in both Chapter 5 and Chapter 6 were both rising main networks. However, large-scale sewer networks generally consist of both rising mains and gravity mains. Unlike flows in rising mains, which is readily known based on SPS operations, flows in gravity mains need to be predicted with hydraulic models. The well-known Saint-Venant equations can calculate gravity flows accurately, however the computational demand is too high to be acceptable for the optimisation algorithm, which needs to call the model for numerous times during each step of optimisation. In Chapter 7, two simplified hydraulic models, namely a first order plus dead time (FOPDT) model and an exponential weighted moving average (EWMA) model, were proposed to predict gravity flows in real-time. The models were calibrated using the outputs of the full Saint-Venant equations. While achieving excellent fit with the full hydraulic model (r2>0.98), these models improved the execution speed by four orders of magnitude enabling their use in online optimisation. The simplified models were employed in a simulation study of real-time control of NaOH dosing in a real-life sewer network comprising both rising mains and gravity mains, which achieved robust control performance.Collectively, this thesis developed and demonstrated a versatile MPC methodology for network-wide sulfide control in the sewer networks through cost-effective use of chemicals. This method is applicable for controlling dosing of different chemicals in sewer networks comprising both rising mains and gravity sewers, to realize robust sulfide control in varying weather conditions. The developed control strategy, predictive models, and optimisation algorithm can potentially be applied to RTC of combined sewer overflow minimization, integrated control of urban water systems, and other prospects of smart cities.