Patent Application
20220228174
Anaerobic digestion (AD) is a mature, efficient, and renewable biotechnology for organic waste removal/stabilization and/or energy recovery that has been successfully and widely implemented for treatment of a variety of substrates. AD is a biological process in which organic matter is degraded in absence of oxygen to generate methane (CH4) and CO2 (biogas) as the major end-products. However, AD involves a complex network of interactions between different groups of micro-organisms that need specific conditions to survive and remain active since they are sensitive to changes in process conditions. For any given AD system, start-up can be a crucial phase as it can determine the entire progression of the system and an ineffective start-up can lead to inefficient process onsets (e.g. sub-optimal or unstable performance in terms of organic matter removal and biogas production). AD start-up can be one of the major operational obstacles owing to the slow growth rate of the key AD microorganisms (particularly methanogens) and the adaptation requirements of the micro-organisms towards the new conditions. Owing to these limitations and in absence of a well-adapted inoculum, AD systems could require long times to start-up (e.g. 2-8 months) before reaching full performance capacity (de Lemos Chernicharo, 2007; Lier et al., 2008; Puñal et al., 2000). For AD systems, reduction of start-up times can be an important factor to increase efficiency and technical competitiveness (de Lemos Chernicharo, 2007) and to minimize costs (Holubar et al., 2003). In this regard, efficient operational control and optimization during AD process start-up can be beneficial and economical in safely driving the system towards an optimal operation (Sbarciog and Vande Wouwer, 2014).
On the practical control side, effective and sufficient monitoring procedures are important for successful operation of an AD process. However, the few number of variables monitored on-line in commercial scale AD systems presents a limitation on the development of effective control strategies. Equipment availability, skilled personnel availability, monitoring costs, and difference in opinions of technical consultants are the major factors affecting the extent of monitoring (e.g. number of variables measured and frequency) implemented in real life AD plants (Drosg, 2013). However, when looking at monitoring costs, one should also evaluate the economic losses resulting from insufficient monitoring (Drosg, 2013). In case of achieving an effective AD start-up, monitoring efforts should be particularly highest since this is a very sensitive phase (Drosg, 2013).
Considering the start-up challenges and limited progress in designing controllers for effective AD start-up management, described herein are improved and practically feasible control processes for automatic start-up of an AD system.
Provided herein are systems, computer-readable media, and methods for controlling the start-up phase of anaerobic digestion reactors.
In one aspect, the system comprises:
one or more input devices configured at least to receive one or more input signals corresponding to one or more sensors connected with an anaerobic digestion reactor;
one or more output devices configured at least to transmit one or more output signals corresponding to one or more actuators connected with the anaerobic digestion reactor; and
a computing system communicatively connected with the one or more input devices and the one or more output devices, the computing system configured to, at least:
determine, based at least in part on the one or more input signals, one or more values of one or more input variables of a nonlinear model predictive controller, the nonlinear model predictive controller being configured with a nonlinear model of anaerobic digestion having a reduced number of model state variables based at least in part on the one or more input variables that are available due to the one or more input signals;
update the nonlinear model predictive controller based at least in part on the one or more values of the one or more input variables; and
cause the one or more output signals to be generated based at least in part on one or more values of one or more output variables of the nonlinear model predictive controller.
In some embodiments, the nonlinear model of anaerobic digestion comprises a system of ordinary differential equations. In some embodiments, the system of ordinary differential equations is based at least in part on variables corresponding to: an effluent concentration of total acetate from the anaerobic digestion reactor, a concentration of aceticlastic methanogens in the anaerobic digestion reactor, a total alkalinity in the anaerobic digestion reactor, a methane production rate of the anaerobic digestion reactor, an effluent concentration of total inorganic carbon from the anaerobic digestion reactor, an effluent concentration of organic substrate from the anaerobic digestion reactor, and a partial pressure of carbon dioxide in an output of the anaerobic digestion reactor.
In some embodiments, the nonlinear model of anaerobic digestion includes methane production occurring through (i) an aceticlastic methanogenesis pathway and (ii) a hydrogenotrophic methanogenesis pathway.
In some embodiments, the nonlinear model of anaerobic digestion determines total alkalinity from acetate (dissociated), bicarbonate, and hydroxide ions alone.
In some embodiments, the one or more output variables of the nonlinear model predictive controller comprise: a volumetric inflow rate of organic substrate to the anaerobic digestion reactor, a volumetric inflow rate of dilution water to the anaerobic digestion reactor, and a volumetric inflow rate of alkali addition to the anaerobic digestion reactor.
In some embodiments, an objective function of the nonlinear model predictive controller is based at least in part on: an effluent concentration of volatile fatty acids as acetate from the anaerobic digestion reactor, a concentration of aceticlastic methanogens in the anaerobic digestion reactor, a total alkalinity in the anaerobic digestion reactor, a methane production rate of the anaerobic digestion reactor, and a cost term penalizing an amount of alkali added proportional to a volumetric inflow rate of alkali addition to the anaerobic digestion reactor.
In some embodiments, the anaerobic digestion reactor comprises a continuous anaerobic digestion reactor with solids retention.
In some embodiments, the reduced number of model state variables comprises an effluent concentration of total acetate from the anaerobic digestion reactor, a concentration of aceticlastic methanogens in the anaerobic digestion reactor, an effluent concentration of total inorganic carbon from the anaerobic digestion reactor, and a total alkalinity in the anaerobic digestion reactor effluent.
In another aspect, provided are one or more computer-readable media collectively having stored thereon computer-executable instructions.
In some embodiments, the computer-executable instructions, when executed with one or more computing systems, collectively at least:
receive one or more input signals corresponding to one or more sensors connected with an anaerobic digestion reactor;
determine, based at least in part on the one or more input signals, one or more values of one or more input variables of a nonlinear model predictive controller, the nonlinear model predictive controller being configured with a nonlinear model of anaerobic digestion having a reduced number of model state variables based at least in part on the one or more input variables that are available due to the one or more input signals;
update the nonlinear model predictive controller based at least in part on the one or more values of the one or more input variables; and
cause one or more output signals to be generated based at least in part on one or more values of one or more output variables of the nonlinear model predictive controller, the one or more output signals corresponding to one or more actuators connected with the anaerobic digestion reactor
In some embodiments, the nonlinear model of anaerobic digestion comprises a system of ordinary differential equations. In some embodiments, the system of ordinary differential equations is based at least in part on variables corresponding to: an effluent concentration of total acetate from the anaerobic digestion reactor, a concentration of aceticlastic methanogens in the anaerobic digestion reactor, a total alkalinity in the anaerobic digestion reactor, a methane production rate of the anaerobic digestion reactor, an effluent concentration of total inorganic carbon from the anaerobic digestion reactor, an effluent concentration of organic substrate from the anaerobic digestion reactor, and a partial pressure of carbon dioxide in an output of the anaerobic digestion reactor.
In some embodiments, the nonlinear model of anaerobic digestion includes methane production occurring through (i) an aceticlastic methanogenesis pathway and (ii) a hydrogenotrophic methanogenesis pathway.
In some embodiments, the nonlinear model of anaerobic digestion determines total alkalinity from acetate (dissociated), bicarbonate, and hydroxide ions alone.
In some embodiments, the one or more output variables of the nonlinear model predictive controller comprise: a volumetric inflow rate of organic substrate to the anaerobic digestion reactor, a volumetric inflow rate of dilution water to the anaerobic digestion reactor, and a volumetric inflow rate of alkali addition to the anaerobic digestion reactor.
In some embodiments, an objective function of the nonlinear model predictive controller is based at least in part on: an effluent concentration of volatile fatty acids as acetate from the anaerobic digestion reactor, a concentration of aceticlastic methanogens in the anaerobic digestion reactor, a methane production rate of the anaerobic digestion reactor, and a cost term penalizing an amount of alkali added proportional to a volumetric inflow rate of alkali addition to the anaerobic digestion reactor.
In some embodiments, the anaerobic digestion reactor comprises a continuous anaerobic digestion reactor with solids retention.
In some embodiments, the reduced number of model state variables comprises an effluent concentration of total acetate from the anaerobic digestion reactor, a concentration of aceticlastic methanogens in the anaerobic digestion reactor, an effluent concentration of total inorganic carbon from the anaerobic digestion reactor, and a total alkalinity in the anaerobic digestion reactor effluent.
In another aspect, a method for controlling a start-up phase of anaerobic digestion reactor operation is provided, the method comprising:
receiving, with one or more input devices, one or more input signals corresponding to one or more sensors connected with an anaerobic digestion reactor;
determining, with a computing system, based at least in part on the one or more input signals, one or more values of one or more input variables of a nonlinear model predictive controller, the nonlinear model predictive controller being configured with a nonlinear model of anaerobic digestion having a reduced number of model state variables based at least in part on the one or more input variables that are available due to the one or more input signals;
updating, with the computing system, the nonlinear model predictive controller based at least in part on the one or more values of the one or more input variables; and
causing, with the computing system, one or more output signals to be generated based at least in part on one or more values of one or more output variables of the nonlinear model predictive controller, the one or more output signals corresponding to one or more actuators connected with the anaerobic digestion reactor.
In some embodiments of the method, the nonlinear model of anaerobic digestion comprises a system of ordinary differential equations. In some embodiments, the system of ordinary differential equations is based at least in part on variables corresponding to: an effluent concentration of total acetate from the anaerobic digestion reactor, a concentration of aceticlastic methanogens in the anaerobic digestion reactor, a total alkalinity in the anaerobic digestion reactor, a methane production rate of the anaerobic digestion reactor, an effluent concentration of total inorganic carbon from the anaerobic digestion reactor, an effluent concentration of organic substrate from the anaerobic digestion reactor, and a partial pressure of carbon dioxide in an output of the anaerobic digestion reactor.
In some embodiments, the nonlinear model of anaerobic digestion includes methane production occurring through (i) an aceticlastic methanogenesis pathway and (ii) a hydrogenotrophic methanogenesis pathway.
In some embodiments, wherein the nonlinear model of anaerobic digestion determines total alkalinity from acetate (dissociated), bicarbonate, and hydroxide ions alone.
In some embodiments, the one or more output variables of the nonlinear model predictive controller comprise: a volumetric inflow rate of organic substrate to the anaerobic digestion reactor, a volumetric inflow rate of dilution water to the anaerobic digestion reactor, and a volumetric inflow rate of alkali addition to the anaerobic digestion reactor.
In some embodiments, an objective function of the nonlinear model predictive controller is based at least in part on: an effluent concentration of volatile fatty acids as acetate from the anaerobic digestion reactor, a concentration of aceticlastic methanogens in the anaerobic digestion reactor, a methane production rate of the anaerobic digestion reactor, and a cost term penalizing an amount of alkali added proportional to a volumetric inflow rate of alkali addition to the anaerobic digestion reactor.
In some embodiments, the anaerobic digestion reactor comprises a continuous anaerobic digestion reactor with solids retention.
In some embodiments, the reduced number of model state variables comprises an effluent concentration of total acetate in the anaerobic digestion reactor, a concentration of aceticlastic methanogens from the anaerobic digestion reactor, an effluent concentration of total inorganic carbon from the anaerobic digestion reactor, and a total alkalinity in the anaerobic digestion reactor effluent.
Greek letters
Subscript
Superscript
As used herein, “methanogenesis” refers to the formation of methane by microbes known as methanogens.
As used herein, a “methanogen” refers to a microorganisms that produces methane as a metabolic byproduct under anaerobic conditions.
As used herein, “acetoclastic methanogenesis” refers to methanogenesis from acetate.
As used herein, “acetoclastic methanogen” refers to a microorganism that produces methane from acetate, such as methanogenic euryarchaea of the order Methanosarcinales.
Described herein is a nonlinear model predictive control (NMPC) scheme for automatic (and ideally optimal) start-up of an anaerobic digestion (AD) system. The control scheme is designed with its practical implementation in mind, based on the use of feasible-to-measure process variables and a very condensed simple AD model (as a prediction model for NMPC optimizations). An overall comprehensive control framework is provided as a practically feasible tool for operators and interested users to implement. The simple AD model (SADM) implemented in the NMPC scheme provides the advantage of predictions of the complex AD process with practical usability and minimal structural complexities. The SADM described herein provides advantages over existing models. Example advantages include decreasing the number of model states and incorporating both methanogenesis pathways for a justified and more accurate prediction of CH4 output. As another example, the SADM also provides the advantage of interfacing with an existing AD plant system to use the available measurements and system updates for estimation of the main model variables and rates, such that the number of model states and complexity remain small and simple. The model thus provides a practical feature that is useful for AD plant operators and engineers. This can enable future operations of the AD plant to be modeled and thereby help predict future operational issues and direct early corrective action using variables provided by the existing AD plant.
In one aspect, the model is useful for AD of readily biodegradable substrates in continuous reactor systems with solids retention. The results of NMPC simulations show that the SADM predictions are generally in good agreement with virtual AD plant behavior (e.g., modelled with ADM1 which is one of the most comprehensive AD models and a widely known and accepted model for AD processes).
There are a number of factors affecting the AD start-up phase such as feed characteristics (type, composition, and concentration), seed sludge or inoculum characteristics (quality, microbial community profile, and relative proportion of micro-organisms with respect to each other), feed loading rates, hydraulic retention time (HRT), and solids retention time (SRT) (Goberna et al., 2015; Janke et al., 2016; Pandey et al., 2011; Vadlani and Ramachandran, 2008). Several experimental studies (focusing on operational aspects), ranging from lab scale to full scale systems testing, exist in literature that have focused on one or more of these factors to propose strategies for effective and/or rapid start-up of AD systems (Alvarado-Lassman et al., 2010; Angelidaki et al., 2006; Angenent et al., 2002; Borzacconi et al., 2006; Dong et al., 2010; Fdéz.-Güelfo et al., 2010; Goux et al., 2016; Janke et al., 2016; Lagerkvist et al., 2015; Liang et al., 2017; Lloret et al., 2013; Meng et al., 2014; Pandey et al., 2011; Puñal et al., 2000; Vadlani and Ramachandran, 2008; Williams et al., 2013). Based on these studies, AD start-up phase has been typically considered complete after achievement of one or more of the following: design organic loading rates (OLR); stable and non-inhibitory levels of volatile fatty acids (VFAs) and/or pH; adequate substrate removal efficiencies; stable biogas production (in terms of flow, CH4 content, and/or composition).
In terms of operational strategy, feeding regime (e.g. OLR and nutrients supply) and inoculation procedures (e.g. inoculum availability, characteristics, and amount) have been major factors affecting the AD performance during start-up according to the existing experimental studies on AD start-up. The feeding regime during AD start-up allows the sensitive micro-organisms to adapt to the different or changing environments and also to avoid any destabilizations in case of feed overloads and/or acidification inside the reactor. As such, it is commonly observed across the existing studies that the AD systems are started up, after carrying out any necessary inoculation procedures, by feeding in a gradual manner i.e. the OLR is low initially and then gradually increased according to the status of the digester (e.g. VFA levels, pH, and biogas production). Although the results have been successful, such manual operational feeding strategies provide no assurance of optimality and are likely suboptimal. Fearing AD process destabilization, some plant operators reject valuable feed (Holubar et al., 2003) during start-up phase. Also, according to some technical reports and guides on biogas plant monitoring and operation, starting up with significantly low organic loads is not advised since it can lower AD plant productivity and negatively impact microbial growth due to lack of adequate food supply (Drosg, 2013; Esteves et al., 2012).
Aiming for optimal operation during AD start-up can help in increasing the practical feasibility of AD technology at large scale. Model predictive control (MPC) provides a strategy for improving AD start-up, where model predictions and latest measurements are used to determine the set of control input or manipulated variable (MV) changes that can optimally drive the control outputs or variables (CVs) to their desired targets (set-points) (Seborg et al., 2011). A characteristic feature of MPC contributing to its success is the consideration of constraints while carrying out the control optimizations (Grune and Pannek, 2011). Very few studies exist in literature on MPC implementation for AD systems (Gaida et al., 2012, 2011; Haugen et al., 2014; Kil et al., 2017; Mauky et al., 2016; Ordace et al., 2012; Xue et al., 2015), and the inventors are unaware of MPC studies particularly focusing on AD start-up. Nevertheless, these studies have shown the effectiveness of MPC strategy for controlling AD systems and its flexibility in implementing different control designs and scenarios.
A factor in the success of model based controllers including MPC is the availability of an accurate and reliable model. For virtual testing scenarios prior to experimental testing, using reliably structured and complex AD models as virtual plants is important for accurate representation of the process plant behavior. With accurate virtual plant models, the control schemes can be more accurately evaluated and assessed. In some embodiments, the comprehensive and widely accepted Anaerobic Model No. 1 (ADM1) by Batstone et al. (2002) is used as the virtual plant.
The main objective of conventional MPC is to determine a set of manipulated input changes (i.e. MVs) to optimally drive the selected process outputs (i.e. CVs) to their desired set points (Seborg et al., 2011) while considering constraints (if any). These optimization calculations incorporate process plant measurements and model predicted future outlook of the outputs.
At each sampling time instant,
The adopted process model can be utilized in parallel with MPC calculations. While a linear or linearized model based MPC (linear MPC) is common, using a nonlinear model can be advantageous in case of highly nonlinear systems (Seborg et al., 2011). Thus, in some embodiments, the control scheme is a nonlinear MPC (NMPC) scheme.
An exemplary NMPC scheme for optimal start-up control is depicted in
By incorporating the CVs in the NMPC scheme (
In some embodiments, aceticlastic methanogenic biomass is an effective target group for favoring methanogenic biomass capacity build-up in the reactor and provides a novel CV in the NMPC control scheme for AD start-up described herein. Aceticlastic methanogenic micro-organisms have slow growth rates (maximum specific growth rates of 0.12-2.85 1/d (Lier et al., 2008)), are highly sensitive to a number of conditions (e.g. pH, temperature, ammonia levels), and are major CH4 producers amongst the methanogenic groups.
Production of CH4, together with the levels of VFA intermediates, confirms the conversion and removal of COD from the system. In some embodiments, methane production is one of the common variables measured and monitored for assessing AD output performance. In some embodiments, methane production can be used to indirectly assess stability in AD systems.
In some embodiments, adjusting the organic substrate flow rate during start-up is a manipulated variable (MV). Dilution water can be used to reduce process acidification by diluting the accumulating VFAs. Also, such dilution can help to temporarily reduce organic levels in case of unexpected organic feed overloads. In some embodiments, additional alkali is added to increase the buffering capacity of the AD system and avoid acidification. Acid overload can be a common problem during start-up (e.g., see Lagerkvist et al., 2015), typically when treating readily biodegradable substrates.
In one aspect, a nonlinear, condensed AD model (SADM in
The resulting SADM is a semi-mechanistic model and is composed of the set of Eqs. (1)-(10) for well-mixed AD systems with solids retention (e.g. high rate AD reactor configurations with recycles and continuously stirred tank reactors (CSTRs) with solids retention):
Ordinary Differential Equations (ODEs)
Total Methane Production Rate
Uptake Rates
Eq. (10) is solved using the SADM state variables (Z, Sac, and SIC) together with expressions for ionic speciation equilibrium for Sac and SIC to find pH and corresponding ionic species concentrations:
The notations for the different variables and parameters in Eqs. (1)-(10) are listed in the nomenclature section. By considering assumption 2, the organic substrate utilization rate in Eq. (7) is simply equal to the net accumulation of the organic substrate by convective transport (linear expression). Also, assuming rapid consumption of H2 during hydrogenotrophic methanogenesis reaction (assumption 5), the rate of hydrogen consumption via methanogenesis in Eq. (9) is correlated directly with the organic substrate utilization rate (Eq. (7)). For aceticlastic methanogenesis, the acetate uptake rate kinetics model in Eq. (8) incorporating pH inhibition is adopted from ADM1 (Batstone et al., 2002) but without the inorganic nitrogen limitation and ammonia inhibition terms. For estimation of liquid to gas transfer rate of the inorganic carbon state (last term in Eq. (2)), the gas phase partial pressure of CO2 (PCO
As shown in
The biochemical processes considered for the simple AD model (SADM) are applicable for any suitable process involving AD of readily biodegradable substrates. Some of the terms in the example ordinary differential equations (ODEs) described above may be dependent on particular design and physical characteristics of the reactor system being implemented. In accordance with at least one embodiment, the NMPC scheme involves a high rate, continuous AD system (e.g., an upflow anaerobic sludge bed reactor). Such high rate AD reactor configurations are typically characterized by biomass/solids retention. Hence, the example SADM ODEs described above may be particularly effective for ‘continuous’ AD reactor systems with ‘solids retention’.
As will be apparent to one of skill in the art, when considering models for a real-time NMPC scheme, a lower number of model state variables can provide advantages including efficiency (e.g., computational) and practical effectiveness. The example SADM described herein includes just four such state variables, namely Sac(M), Xac(M), SIC(M) and Z, an improvement over previous models (which, for example, had 6, 13, 35 and more states). As recognized by the inventors, this particular subset of possible model state variables are sufficient for describing AD of readily biodegradable substrates in a continuous AD reactor system with solids retention. In accordance with at least one embodiment, keeping the number of state variables low involves interfacing with an operating AD plant/system to use the available measurements/updates for estimation of model variables and rates.
In one aspect, the overall NPMC objective function consists of set-point error tracking terms for the control variables (concentration of effluent volatile fatty acids as acetate, concentration of aceticlastic methanogens in reactor, and total methane production rate) and a cost term penalizing the amount of alkali added (proportional to the volumetric flow rate of the alkali input which is one of the manipulated variables). Also, a number of constraints have been considered for minimization of the NMPC objective function. As an example, for one of the NMPC designs considered (denoted as NMPC Base), three control variables (Xacsp, Sacsp, rCH
In another example (denoted as NMPC No rCH4), the methane production rate CV is not considered (i.e. only 2 CVs are targeted) and the resulting NMPC objective function is given by Eq. (19) with the constraints being the same as those defined above (Eqs. (12)-(17)).
The proposed objective function (JP) for Eqs. (11) and (19) is optimized over the prediction horizon (P) and the number of unknowns or decision variables for the optimization is equal to the product of the number of MV variables and the control horizon (M).
The constraint in Eq. (12) is the classical optimal control constraint where the optimization has to follow the process dynamics, modelled currently with the proposed SADM. The constraints in Eqs. (13)-(14) place limits on the total influent volumetric flows, ensuring that the total influent volumetric flow is at least being utilized at the minimum design flow but not exceeding the maximum threshold design flow (to avoid washout of biomass). These constraints are specifically applicable for high rate AD systems (case scenario considered in the current work), where upflow liquid velocity (vup) is an important design parameter since it affects the hydraulic flow rate and plays an important role in granulation (de Lemos Chernicharo, 2007). To address acid accumulation problems during start-up, the stability constraint given by Eq. (15) is defined to maintain stability indicators within safe limits. In some embodiments, the intermediate alkalinity to total alkalinity ratio is a stability diagnostic indicator and can be below a certain value (e.g., 0.3) to ensure non-inhibitory (towards the AD micro-organisms, particularly methanogens) or neutral pH levels inside the reactor. To avoid organic overloading and to operate the AD process within the practical ranges of OLR, a maximum allowable OLR restriction (Eq. (16)) is considered. Finally, Eq. (17) places lower bound on the individual influent volumetric flows (MVs).
AD Start-Up Performance with the Proposed NMPC Designs
The list of implemented process conditions, NMPC tuning parameters, CV set-points, and other information for testing the proposed control scheme are shown in Table 1. The tuning parameters (M and P), considered as the base settings, were arbitrarily chosen but determined during preliminary simulations to achieve satisfactory performance. In general, the values of CV set-points to use depend on the desired performance and objectives of the implemented process. In some embodiments, the set-points listed in Table 1 were determined from preliminary simulations at steady state using ADM1, with organic substrate feeding at constant feed rate (OLR=15 kg/m3·d) and sufficient alkali addition. The value of maximum permissible OLR listed in Table 1 was selected within the practical range of OLRs for the high rate AD systems. The weights for the NMPC objective function (w1, w2, w3, w4) were arbitrarily chosen but with higher importance given to the set-point tracking of Xac and rCH
In general, all the start-up strategies tested (NMPC and manual) yield stable and satisfactory performance by successfully achieving or nearly achieving CV set-points (
NMPC variations in which CH4 production is one of the priorities during start-up (i.e. NMPC Base), lead to a start-up strategy with a very aggressive initial feeding at very high OLRs (contrary to conventional practices) as seen in
The values of the overall as well as of individual terms contributions to the NMPC objective functions (Eqs. (11) and (19)) after optimisations are shown in
Model inaccuracy issues can be partly overcome in MPC by implementing output feedback strategies which use an output bias correction (Seborg et al., 2011), where the errors or biases between the latest plant measurements and the (k+1)th sampling time instant model predictions (from previous sampling instant) can be added as corrections to the model predictions when conducting control calculations at the latest sampling instant. The NMPC controllers achieved successful AD performance targets during start-up using the simple prediction model (sADM) considering constraints.
A number of quantitative assessments were conducted for a clearer comparison and additional analyses of the tested startup operation and control strategies. Table 7 summarizes the results obtained using the different quantitative indicators. The ITAE criterion (an integral error criterion used in controller design or tuning) gives an indication of the overall performance along the process operation duration in terms of the time-weighted errors between the process outputs and the set-points. For the steady state offset errors (as percentages) reported in Table 7, a zero or negative value for the Sac CV and a zero or positive value for the Xac or rCH4 CV corresponds to achievement of the corresponding set-point at steady state. In terms of achieving the nominal OLR of 15 kgCOD/m3·d, the manual strategy with 50% pAB set sate for OLR and the NMPC design without the methane production rate CV (NMPC No rCH4) was the fastest. However, the manual strategy with 50% pAB set rate affects the AD performance output of the CVs when looking at the values of integral of the time-weighted absolute error (between set-point and the CV) (ITAE) criterion and the CV offset errors at steady state reported in Table 7. Comparing the ITAE and steady state offset values of the Xac and rCH4 CVs across the tested strategies, the NMPC designs provide a significant advantage over the manual strategies in general.
Referring to
A number of practical case studies and scenarios were evaluated to assess the two NMPC designs (NMPC Base and NMPCNo rCH4) for robustness and possible improvements. The details on these scenarios and the results obtained have been provided in Appendix E. In brief, the following cases were simulated:
1. Model (sADM) mismatch/bias correction, in which at each sampling event, the differences between the plant outputs and sADM predictions are added as corrections to the sADM predictions for the next NMPC optimisation step.
2. Assumption of online Xac measurement availability without any delays (i.e. no Xac estimator) in order to assess the impact of the state estimator.
3. Instrument error via random noise ranging between −10% and 10% change relative to actual measurements of the four sADM states and CH4 production rate.
4. Measured disturbances of substrate influent COD (2 tests: 75% decrease and 75% increase in the base influent COD of 2.75 gCOD/L at 10th day for 15 days)
5. Unmeasured disturbances of substrate influent COD (4 tests: +5% and +75% random changes in influent COD of 2.75 gCOD/L; 80% decrease and 80% increase in influent COD of 2.75 gCOD/L at 10th day for 15 days)
The results (figures and quantitative assessments) for the above cases are provided in Appendix E (Table E and
For the scenario involving model bias correction, both NMPC designs yield faster performance with respect to switching to manual operation at constant nominal OLR of 15 kgCOD/m3·d (Table E) compared to the results in Table 6. However, the NMPCBase design yields constant CV offset errors relative to setpoints (Table E and
With the assumption of Xac online measurement availability, the NMPCBase design yields stable performance but is unable to drive the process towards the CV set-points (
Despite the instrument errors occurring throughout the process operation time, both NMPC designs yield satisfactory results as the process is driven towards the CV set-points without any failures (Table E and
With measured disturbances of the influent COD of the substrate, the NMPCBase design yields stable performance but with significant intermediate variations of CVs (
Under unmeasured disturbances in the influent COD (i.e. the sADM is unaware of the changes in substrate influent COD concentration), both NMPC designs are able to drive the process towards the set-points without any stability issues. However, intermediate variations in CVs are observed in general (
In summary, the NMPCNo rCH4 configuration appears to provide a more robust and superior performance compared to the NMPCBase alternative configuration against the different scenarios tested. The NMPCBase fails to drive the process towards the set-points for a number of the scenarios considered. This indicates that the NMPC objective function formulation (or in other words, the process variables targeted for optimisation) plays a key role in the level of success of the NMPC scheme for start-up. Also, these results could indicate that the unconventional method of starting with a high OLR at the beginning of operation during start-up (according to NMPCBase design) may not be an effective strategy despite its possible apparent optimality respect to the conventional approach of low to high OLR.
For accurate representation of the process plant or system against which the control scheme is being designed and tested virtually, the ADM1 model described in Batstone et al. (2002) can be used to simulate the AD process dynamics. In some embodiments, ADM1 is implemented as the virtual AD plant for evaluating the NMPC scheme. An empirical correlation for solids retention time (SRT) with the liquid upflow velocity (vup) was also incorporated (relevant for high rate AD systems) in both the virtual plant model (ADM1) and the SADM, based on the solids retention formulations of Fedorovich et al. (2003).
On the practical side, the available measurements for AD process monitoring are not enough to evaluate and assess the state of the process (Jimenez et al., 2015). Although a variety of variables are monitored in commercial scale AD plants, the number of variables measured online is quite limited (pH, temperature, biogas flow rate, biogas composition, biogas yield, and partial pressures) while a majority are measured off-line and less frequently. However, at the level of research developments, advanced and novel sensors have been successfully developed in existing studies for online monitoring of key variables (Jimenez et al., 2015). As such, a number of existing studies have successfully developed online measurement tools for VFA levels, alkalinities (partial and total), COD, and volatile suspended solids (VSS) (Boe & Angelidaki, 2012; Falk et al., 2015; Molina et al., 2009; Morel et al., 2005; Nielsen et al., 2007; Steyer et al., 2002).
Considering typical practical limitations in AD instrumentation, in some embodiments the measurements in the NMPC scheme (including SADM) for direct determination or estimation of process variables and parameters is shown in Table 2. The significant issue is the lack of online estimation of microbial biomass concentrations. Soft sensors or state estimators (e.g. Kalman filter) can be used for slow, delayed offline measurements (e.g. methanogenic biomass activity tests for Xac). In the current work, the SADM equations (Eqs. (1)-(10)) were used (in support of having a semi-mechanistic approach) to estimate the aceticlastic methanogenic biomass concentration at every sampling event. In some instances, for estimation via SADM, the set of initial model states are determined from the last available measurements. Also, a 48 h delay in methanogenic activity measurement update was considered when utilizing measurements as initial points for the SADM state estimation of Xac (
In some embodiments, a simple case study involving optimal start-up control of an AD system treating readily biodegradable substrates is provided for implementing and virtually testing the NMPC scheme (
For virtual plant simulation with ADM1 predictions, benchmark ADM1 parameter values reported in Batstone et al. (2002) for mesophilic high rate systems have been used. The parameters in the SADM equations listed in Eqs. (1)-(10) (νac/s, νCO
In one aspect, the interfacing of the virtual plant (ADM1) outputs into NMPC and SADM compatible variables is shown in Table 4. The effluent organic substrate concentration (Ss) and CO2 partial pressure (PCO
The proposed NMPC scheme was implemented in MATLAB® R2015b. MATLAB function files and a Simulink® file were created for achieving the tasks in the control scheme implementation (
In one aspect, AD process monitoring is accomplished using online measurements, off-line measurements, or both. An extensive review of instrumentation and control in AD processes is provided by Jimenez et al. (2015, Reviews in Environmental Science andBio Technology, Vol. 14, pp. 615-648), which is incorporated by reference herein. In addition, the report by Kock & Eberlein (2012, see the internet at portal.ea-stmk.at/documents/20181/25550/D_5_1_Best_Practice_Monitoring.pdf/13b61dbe-a886-454d-97e4-24aeada73cab) provides a list of variables monitored in some of the existing AD/biogas plants in Europe. Table 5 provides non-limiting examples of conventional variables and parameters and sensors or equipment used to detect or measure the variables or parameters in existing commercial AD systems. The preference of having one specific sensor or measurement technique over another may depend on the equipment availability and monitoring cost constraints of the AD plant being implemented.
Table 6 provides non-limiting examples of online variables and parameters, and sensors or equipment used to measure or detect the variables or parameters, in research facilities. Some of these sensors have been validated in full scale AD systems.
The models described herein can be used in AD start-up processes. The process of AD occurs in several steps and typically involves a community of micro-organisms that interact together in the absence of oxygen to decompose organic matter into biogas as the main end product. Biogas is mostly methane (CH4) and carbon dioxide (CO2), with very small amounts of water vapor and other gases. The carbon dioxide and other gases can be removed, leaving only the methane. Biogas is a renewable energy source that can be used in a variety of ways. For example, biogas can be valorised in a combined heat and power (CHP) unit to produce electricity and heat. Alternatively, biogas can be upgraded to biomethane to reach the purity of natural gas and be injected into the municipal gas grid or be used as transportation fuel. The AD process also generates stable residue called “digestate” as the other main end-product. Digestate is a wet mixture that is usually separated into a solid and a liquid (dewatering). In some cases, the digestate can be rich in nutrients and hence, used as fertilizer for crops.
Methanogenic archaea (micro-organisms involved in the terminal bioconversion reaction step of the overall AD process) are considered rate-limiting key-players of the AD process due to their slow growth rates and high sensitivity to different environmental conditions (Weiland, 2010). Thus, process inhibition can often be encountered due to an imbalance between the VFAs and other precursor-producing bacteria and methanogenic archaea (Ahring et al., 1995). Moreover, many environmental parameters such as the digestion temperature (Luo et al., 2015), the organic loading rate (OLR) (Goux et al., 2015) or even the substrate type (Westerholm et al., 2016) greatly influence the microbial community development and structure.
The AD process is typically divided into four main stages (hydrolysis, acidogenesis, acetogenesis and methanogenesis), each involving different microbial communities (Weiland, 2010). Hydrolysis refers to the break down of large, complex polymers like carbohydrates, cellulose, proteins and fats by hydrolytic enzymes into simple sugars, amino acids, and fatty acids. During acidogenesis, simple monomers are broken down into volatile fatty acids. During acetogenesis, the products of acidogenesis are broken down into acetic acid, releasing hydrogen and carbon dioxide. Methanogenesis occurs when methanogenic bacteria produce methane by cleaving two acetic acid molecules to form carbon dioxide and methane (aceticlastic methanogenesis), and/or by reduction of carbon dioxide with hydrogen (hydrogenotrophic methanogenesis).
AD systems adhere to the same basic principles whether the feedstock is food waste, animal manures or wastewater sludge. It will be understood that even though the AD process is the same, the design and management of AD systems will vary based on the feedstock, desired products, and economic/practical constraints. Examples include stand-alone digesters, on-farm digesters, and wastewater treatment plant digesters. The different types and operational modes of digestion systems are described below.
Digesters can be designed to run at different temperature ranges. The temperature ranges are typically 86-1000 F (25-45° C.) for mesophillic systems and 122-140° F. (50-60° C. or above) for thermophilic systems. The advantages of thermophilic systems include faster throughput and biogas production per unit of feedstock and/or volume of digester, and the high temperatures kill higher numbers of pathogenic microorganisms. The disadvantages of thermophilic systems include higher capital costs, higher energy costs for heating, and they generally involve more hands on management. Mesophillic digesters have the advantages of being easier to operate and maintain, but have the disadvantage that they do not result in high pathogen kill.
Digesters can be designed to process one type of feedstock or to process multiple feedstocks. Digesters can also be designed for co-digestion of more than one type of organic feedstock at the same time. In some embodiment, the feedstock is pre-treated before digestion (e.g., blended, screened, thermally conditioned, etc.).
A wet digestion system, also referred to as a low solids AD system, typically processes feedstock with less than 15 percent solids content. Because of the relatively high liquid content, the feedstock for a wet digester is often a slurry that can be mixed or pumped. A dry digester system, also referred to as a high solids AD system, generally processes feedstock with greater than 15 percent solids content. The feedstocks for a dry digester system can be stacked. Dry AD systems are generally less expensive to operate because there is less water to heat and there is more gas production per unit feedstock. However, wet AD systems generally have a lower set-up capital cost.
AD systems can be operated in batch or continuous flow mode. In batch mode, the feedstock is added to the digester all at the same time and there is a set period of time for digestions to occur. Following digestion, the digester is emptied and reloaded with new feedstock. In a continuous flow system, the feedstock is constantly fed into the digester and digested material is continuously removed. Batch digestion systems have the disadvantage of having to open the digester and restart the system every few weeks, which can create operational challenges. Thus, most digesters are continuous flow systems, which also provide more biogas per unit feedstock and lower operating costs. However, dry systems can be operated in batch mode, and highs and lows in gas production can be normalized by using multiple batch digesters with staggered changeover times.
AD systems can include multiple digesters to ensure each stage occurs sequentially and is as efficient as possible. Multiple digesters can produce more biogas per unit feedstock but at a higher capital cost, higher operating cost, and greater operational management.
Some AD systems comprise one or more vertical tanks that input feedstock through a pipe on one side while digestate overflows through a pipe on the other side. Horizontal plug-flow systems use a solid feedstock (called a ‘plug’) that flows through a horizontal digester at the same rate it is fed into the digester. Vertical tanks have the advantage of being simpler and cheaper to operate, but the disadvantage that feedstock may not reside in the digester for the optimum period of time to produce the most biogas. Horizontal tank systems are more expensive to build and operate, but provide the advantage that the feedstock will reside in the digester for an optimum period of time to produce the most biogas.
The anaerobic digestion reactor may include any suitable anaerobic digestion equipment. The sensors may include any suitable sensor of anaerobic digestion reactor attributes. Sensors may utilize any suitable signaling protocol to communicate with the computing system. The computing system may utilize any suitable input and/or output devices to receive and/or transmit signals. The computing system may transform and/or decode received signals into nonlinear model predictive controller input variables, and similarly transform and/or encode nonlinear model predictive controller output variables into signals suitable for transmission. The network may include any suitable number of additional control layers (including zero).
The nonlinear model predictive controller may have any suitable number of input and/or output variables. Nonlinear model predictive controller input variables may include input variables from the nonlinear model of anaerobic digestion. Examples of input variables include control variables, manipulated variables, variables corresponding to sensor outputs, variables corresponding to estimated reactor attributes, and variables corresponding to actuator inputs. Examples of output variables include control variables, manipulated variables, and variables corresponding to actuator inputs. The nonlinear model predictive controller may maintain a history of input and/or output variables as well as of derived state variables. The nonlinear model predictive controller may be configured with any suitable nonlinear model (preferably simple in terms of computational efforts and/or of low computational complexity) of anaerobic digestion (e.g. SADM).
In accordance with at least some embodiments, the system, apparatus, methods, processes and/or operations for message coding may be wholly or partially implemented in the form of a set of instructions executed by one or more programmed computer processors such as a central processing unit (CPU) or microprocessor. Such processors may be incorporated in an apparatus, server, client or other computing device operated by, or in communication with, other components of the system. As an example,
and/or the fixed disk, as well as the exchange of information between subsystems. The system memory and/or the fixed disk may embody a tangible computer-readable medium.
It should be understood that one or more of the embodiments described herein can be implemented in the form of control logic using computer software in a modular or integrated manner. Alternatively, or in addition, embodiments may be implemented partially or entirely in hardware, for example, with one or more circuits such as electronic circuits, optical circuits, analog circuits, digital circuits, integrated circuits (“IC”, sometimes called a “chip”) including application-specific ICs (“ASICs”) and field-programmable gate arrays (“FPGAs”), and suitable combinations thereof. As will be apparent to one of skill in the art, notions of computational complexity and computational efficiency may be applied mutatis mutandis to circuits and/or circuitry that implement computations and/or algorithms. Based on the disclosure and teachings provided herein, a person of ordinary skill in the art will know and appreciate other ways and/or methods to implement one or more embodiments described herein using hardware and/or a combination of hardware and software.
Any of the software components, processes or functions described in this application may be implemented as software code to be executed by a processor using any suitable computer language such as, for example, Java, C++ or Perl using, for example, conventional or object-oriented techniques. The software code may be stored as a series of instructions, or commands on a computer readable medium, such as a random access memory (RAM), a read only memory (ROM), a magnetic medium such as a hard-drive or a floppy disk, or an optical medium such as a CD-ROM. Any such computer readable medium may reside on or within a single computational apparatus, and may be present on or within different computational apparatuses within a system or network.
aAngelidaki et al. (2006) Method.
bCalculation of the integral error criteria has been conducted for a time duration of 120 days. The integrals were calculated using numerical integration in MATLAB ® R2015b.
cITAE - Integral of the time-weighted absolute error = ∫0t t . | setpoint (t) − measurement(t) | . dt.
The main objective of conventional MPC is to determine a set of manipulated input changes (i.e. MVs) to optimally drive the selected process outputs (i.e. CVs) to their desired set points (Seborg et al., 2011) while considering constraints (if any). These optimisation calculations incorporate process plant measurements and model predicted future outlook of the outputs.
At each sampling time instant,
The adopted process model plays a critical role in the success of MPC since the model is utilized in parallel with MPC calculations. While a linear or linearized model based predictive control (linear MPC) is common, using a nonlinear model is advantageous in case of highly nonlinear systems (Seborg et al., 2011). In this case, the control scheme is said to be a nonlinear MPC (NMPC) scheme.
Practically, microbial biomass measurements require long times (days) and hence, the levels of the microbial biomass cannot be monitored online. For control purposes, state estimation is required to estimate the levels of aceticlastic methanogens (Xac); one of the key control variables in the proposed NMPC scheme. For the estimation of Xac, the simple AD model (sADM) had been implemented (same as the model used during NMPC optimizations). The estimation methodology is illustrated in
To evaluate and compare (with each other) the performance of the implemented NMPC designs (NMPC Base and NMPCNo rCH4) for robustness and possible improvements, different scenarios were considered during AD start-up control:
Biasvariable=Plant measurementvariable−sADM predictionvariable; at a given sampling instant
Corrected sADM predictionvariable=sADM predictionvariabie+Biasvariable; for next NMPC optimization
Table E and
Table E summarizes the AD start-up performance with the NMPC designs tested under the different scenarios in terms of quantitative assessments.
aCalculation of the integral error criteria has been conducted for a time duration of 120 days The integrals were calculated using numerical integration in MATLAB ® R2015b
bITAE - Integral of the time-weighted absolute error = ∫0t t . |setpoint(t) − measurement (t) | . dt
dAverage value from the last set of results (towards end of simulation time) showing similar values when corrected to 3 significant figures.
It is understood that the examples and embodiments described herein are for illustrative purposes only and that various modifications or changes in light thereof will be suggested to persons skilled in the art and are to be included within the spirit and purview of this application and scope of the appended claims. All publications, patents, and patent applications cited herein are hereby incorporated by reference in their entirety for all purposes.
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/IB2020/055985 | 6/24/2020 | WO | 00 |
| Number | Date | Country | |
|---|---|---|---|
| 62865875 | Jun 2019 | US |
