Metaheuristic optimization of data preparation and machine learning hyperparameters for prediction of dynamic methane production.

Machine learning algorithms provide detailed description of the anaerobic digestion process, but the impact of data preparation procedures and hyperparameter optimization has rarely been investigated. A genetic algorithm was developed for optimizing data preparation and model hyperparameters to simulate dynamic methane production from steady-state anaerobic digestion of agricultural residues at full-scale. A long short-term memory neural network was used as prediction model. Results indicate that batch size, learning rate and number of neurons are the most important model parameters for accurate description of methane production rates, whereas combination of hyperparameter and data preparation optimization shows best model efficiencies, with a root mean square scaled error of 76.5 %. Mass of solid feed, time and mass of volatile solids are the most relevant input features. This study provides fundamental steps for optimal prediction of dynamic biomethane production, as a reliable basis for improving bioconversion efficiency during anaerobic digestion of agricultural residues.

Demand-oriented biogas production to cover residual load of an electricity self-sufficient community using a simple kinetic model.

Flexible biogas production can enable demand-oriented energy supply without the need for expensive gas storage expansions, but poses challenges to the stability of the anaerobic digestion (AD) process. In this work, biogas production of laboratory-scale AD of maize silage and sugar beets was optimized to cover the residual load of an electricity self-sufficient community using a simple process model based on first-order kinetics. Experiments show a good agreement between biogas demand, predicted, and measured biogas production. By optimizing biogas conversion schedules based on the measured gas production, a gas storage capacity of 7-8 h was identified for maximum flexibility, which corresponds to typical gas storage sizes at industrial biogas plants in Germany. Various stability indicators were continuously monitored and proved resilient process conditions. These results demonstrate that demand-oriented biogas production using model predictive control is a promising approach to enable existing biogas plants to provide balancing energy.

Systematic simplification of the Anaerobic Digestion Model No. 1 (ADM1) - Model development and stoichiometric analysis.

Rigorous process models provide a reliable basis for model-based monitoring and control of anaerobic digestion plants. Due to the complex model structure and non-linear system characteristics, the established Anaerobic Digestion Model No. 1 (ADM1) is rarely applied in industrial plant operation. The present investigation proposes a systematic procedure for successive model simplification and presents the description of five model variants of a mass-based ADM1. Individual model structures greatly differ in their number of implemented process phases, characteristic components and required parameters. Simplified model variants combine nutrient degradation and biogas formation based on first-order sum reactions, whereas complex model structures describe individual degradation pathways and intermediates during acido- and acetogenesis. Characteristic features of the derived model structures as well as the stoichiometric methane potentials and microbial biomass yields of the underlying degradation pathways of individual model variations are evaluated and discussed in detail.

Systematic simplification of the Anaerobic Digestion Model No. 1 (ADM1) - Laboratory experiments and model application.

Due to a limited number of available measurements on agricultural biogas plants, established process models, such as the Anaerobic Digestion Model No. 1 (ADM1), are rarely applied in practise. To provide a reliable basis for model-based monitoring and control, different model simplifications of the ADM1 were implemented for process simulation of semi-continuous anaerobic digestion experiments using agricultural substrates (maize silage, sugar beet silage, rye grain and cattle manure) and industrial residues (grain stillage). Individual model structures enable a close depiction of biogas production rates and characteristic intermediates (ammonium nitrogen, propionic and acetic acid) with equal accuracy as the original ADM1. The impact of different objective functions and standard parameter values on parameter estimates of first-order hydrolysis constants and microbial growth rates were evaluated. Due to the small number of required model parameters and suitable system characteristics, simplified model structures show clear advantages for practical application on agricultural biogas plants.

Towards a standardization of biomethane potential tests: a commentary.

Inter-laboratory reproducibility of biomethane potential (BMP) is dismal, with differences in BMP values for the same sample exceeding a factor of two in some cases. A large group of BMP researchers directly addressed this problem during a workshop held in Leysin, Switzerland, in June 2015. The workshop resulted in a new set of guidelines for BMP tests published in 2016, which is the subject of the present commentary. The work has continued with two international inter-laboratory studies and one additional workshop held in Freising, Germany, in 2018. The dataset generated by the two inter-laboratory studies were used to refine the validation criteria for BMP tests. Based on these new results an update to the original guidelines is proposed here.

Methane emissions from the storage of liquid dairy manure: Influences of season, temperature and storage duration.

Methane emissions from livestock manure are primary contributors to GHG emissions from agriculture and options for their mitigation must be found. This paper presents the results of a study on methane emissions from stored liquid dairy cow manure during summer and winter storage periods. Manure from the summer and winter season was stored under controlled conditions in barrels at ambient temperature to simulate manure storage conditions. Methane emissions from the manure samples from the winter season were measured in two time periods: 0 to 69 and 0 to 139 days. For the summer storage period, the experiments covered four time periods: from 0 to 70, 0 to 138, 0 to 209, and 0 to 279 continuous days, with probing every 10 weeks. Additionally, at the end of all storage experiments, samples were placed into eudiometer batch digesters, and their methane emissions were measured at 20 °C for another 60 days to investigate the potential effect of the aging of the liquid manure on its methane emissions. The experiment showed that the methane emissions from manure stored in summer were considerably higher than those from manure stored in winter. CH4 production started after approximately one month, reaching values of 0.061 kg CH4 kg-1 Volatile Solid (VS) and achieving high total emissions of 0.148 kg CH4 kg-1 VS (40 weeks). In winter, the highest emissions level was 0.0011 kg CH4 kg-1 VS (20 weeks). The outcomes of these experimental measurements can be used to suggest strategies for mitigating methane emissions from manure storage.

Augmenting Biogas Process Modeling by Resolving Intracellular Metabolic Activity.

The process of anaerobic digestion in which waste biomass is transformed to methane by complex microbial communities has been modeled for more than 16 years by parametric gray box approaches that simplify process biology and do not resolve intracellular microbial activity. Information on such activity, however, has become available in unprecedented detail by recent experimental advances in metatranscriptomics and metaproteomics. The inclusion of such data could lead to more powerful process models of anaerobic digestion that more faithfully represent the activity of microbial communities. We augmented the Anaerobic Digestion Model No. 1 (ADM1) as the standard kinetic model of anaerobic digestion by coupling it to Flux-Balance-Analysis (FBA) models of methanogenic species. Steady-state results of coupled models are comparable to standard ADM1 simulations if the energy demand for non-growth associated maintenance (NGAM) is chosen adequately. When changing a constant feed of maize silage from continuous to pulsed feeding, the final average methane production remains very similar for both standard and coupled models, while both the initial response of the methanogenic population at the onset of pulsed feeding as well as its dynamics between pulses deviates considerably. In contrast to ADM1, the coupled models deliver predictions of up to 1,000s of intracellular metabolic fluxes per species, describing intracellular metabolic pathway activity in much higher detail. Furthermore, yield coefficients which need to be specified in ADM1 are no longer required as they are implicitly encoded in the topology of the species' metabolic network. We show the feasibility of augmenting ADM1, an ordinary differential equation-based model for simulating biogas production, by FBA models implementing individual steps of anaerobic digestion. While cellular maintenance is introduced as a new parameter, the total number of parameters is reduced as yield coefficients no longer need to be specified. The coupled models provide detailed predictions on intracellular activity of microbial species which are compatible with experimental data on enzyme synthesis activity or abundance as obtained by metatranscriptomics or metaproteomics. By providing predictions of intracellular fluxes of individual community members, the presented approach advances the simulation of microbial community driven processes and provides a direct link to validation by state-of-the-art experimental techniques.

Determination of Microbial Maintenance in Acetogenesis and Methanogenesis by Experimental and Modeling Techniques.

For biogas-producing continuous stirred tank reactors, an increase in dilution rate increases the methane production rate as long as substrate input can be converted fully. However, higher dilution rates necessitate higher specific microbial growth rates, which are assumed to have a strong impact on the apparent microbial biomass yield due to cellular maintenance. To test this, we operated two reactors at 37°C in parallel at dilution rates of 0.18 and 0.07 days-1 (hydraulic retention times of 5.5 and 14 days, doubling times of 3.9 and 9.9 days in steady state) with identical inoculum and a mixture of volatile fatty acids as sole carbon sources. We evaluated the performance of the Anaerobic Digestion Model No. 1 (ADM1), a thermodynamic black box approach (TBA), and dynamic flux balance analysis (dFBA), to describe the experimental observations. All models overestimated the impact of dilution rate on the apparent microbial biomass yield when using default parameter values. Based on our analysis, a maintenance coefficient value below 0.2 kJ per carbon mole of microbial biomass per hour should be used for the TBA, corresponding to 0.12 mmol ATP per gram dry weight per hour for dFBA, which strongly deviates from the value of 9.8 kJ Cmol h-1 that has been suggested to apply to all anaerobic microorganisms at 37°C. We hypothesized that a decrease in dilution rate might select taxa with minimized maintenance expenditure. However, no major differences in the dominating taxa between the reactors were observed based on amplicon sequencing of 16S rRNA genes and terminal restriction fragment length polymorphism analysis of mcrA genes. Surprisingly, Methanosaeta dominated over Methanosarcina even at a dilution rate of 0.18 days-1, which contradicts previous model expectations. Furthermore, only 23-49% of the bacterial reads could be assigned to known syntrophic fatty acid oxidizers, indicating that unknown members of this functional group remain to be discovered. In conclusion, microbial maintenance was found to be much lower for acetogenesis and methanogenesis than previously assumed, likely due to the exceptionally low growth rates in anaerobic digestion. This finding might also be relevant for other microbial systems operating at similarly low growth rates.

Ensiling fermentation reveals pre-treatment effects for anaerobic digestion of sugarcane biomass: An assessment of ensiling additives on methane potential.

Ensiling of sugarcane trash (SCT) and sugarcane stalks (SCS) was studied to assess the effects of molasses (MOL) and lactic acid bacteria (LAB) inoculant on methane potential. The experiment was run for 70 days and monitoring parameters were analyzed at days 0, 5, 15 and 70. Biochemical methane potential (BMP) tests performed with fresh and ensiled material at day 70 showed an increase in methane potential by 24.0%, 23.4%, 1.7% and 71.1% for SCSctr, SCTctr, SCTmol and SCTmol + lab, respectively. Such improved performance is explained by the formation of organic acids (mostly acetate and lactate) which were able to decrease the pH of the silages from 5.7-5.9 to 3.8-4.2 for all SCT treatments and from 5.9 up to 3.4 for SCS treatment. Thus, the ensiling process provided similar effects to a pre-treatment at low acid concentrations, which in turn improved the digestibility of the cellulosic biomass for methane production.

The Influence of Pressure-Swing Conditioning Pre-Treatment of Cattle Manure on Methane Production.

Cattle manure is an agricultural residue, which could be used as source to produce methane in order to substitute fossil fuels. Nevertheless, in practice the handling of this slowly degradable substrate during anaerobic digestion is challenging. In this study, the influence of the pre-treatment of cattle manure with pressure-swing conditioning (PSC) on the methane production was investigated. Six variants of PSC (combinations of duration 5 min, 30 min, 60 min and temperature 160 °C, 190 °C) were examined with regards to methane yield in batch tests. PSC of cattle manure showed a significant increase up to 109% in the methane yield compared to the untreated sample. Kinetic calculations proved also an enhancement of the degradation speed. One PSC-variant (190 °C/30 min) and untreated cattle manure were chosen for comparative fermentation tests in continuously stirred tank reactors (CSTR) in lab-scale with duplicates. In the continuous test a biogas production of 428 mL/g volatile solids (VS) (54.2% methane) for untreated manure was observed and of 456 mL/g VS (53.7% methane) for PSC-cattle-manure (190 °C/30 min). Significant tests were conducted for methane yields of all fermentation tests. Furthermore, other parameters such as furfural were investigated and discussed.

Ammonia Inhibition of Anaerobic Volatile Fatty Acid Degrading Microbial Communities.

Ammonia inhibition is an important reason for reactor failures and economic losses in anaerobic digestion. Its impact on acetic acid degradation is well-studied, while its effect on propionic and butyric acid degradation has received little attention and is consequently not considered in the Anaerobic Digestion Model No. 1 (ADM1). To compare ammonia inhibition of the degradation of these three volatile fatty acids (VFAs), we fed a mixture of them as sole carbon source to three continuous stirred tank reactors (CSTRs) and increased ammonium bicarbonate concentrations in the influent from 52 to 277 mM. The use of this synthetic substrate allowed for the determination of degradation efficiencies for the individual acids. While butyric acid degradation was hardly affected by the increase of ammonia concentration, propionic acid degradation turned out to be even more inhibited than acetic acid degradation with degradation efficiencies dropping to 31 and 65% for propionic and acetic acid, respectively. The inhibited reactors acclimatized and approximated pre-disturbance degradation efficiencies toward the end of the experiment, which was accompanied by strong microbial community shifts, as observed by amplicon sequencing of 16S rRNA genes and terminal restriction fragment length polymorphism (T-RFLP) of mcrA genes. The acetoclastic methanogen Methanosaeta was completely replaced by Methanosarcina. The propionic acid degrading genus Syntrophobacter was replaced by yet unknown propionic acid degraders. The butyric acid degrading genus Syntrophomonas and hydrogenotrophic Methanomicrobiaceae were hardly affected. We hypothesized that the ammonia sensitivity of the initially dominating taxa Methanosaeta and Syntrophobacter led to a stronger inhibition of the acetic and propionic acid degradation compared to butyric acid degradation and hydrogenotrophic methanogenesis, which were facilitated by the ammonia tolerant taxa Syntrophomonas and Methanomicrobiaceae. We implemented this hypothesis into a multi-taxa extension of ADM1, which was able to simulate the dynamics of both microbial community composition and VFA concentration in the experiment. It is thus plausible that the effect of ammonia on VFA degradation strongly depends on the ammonia sensitivity of the dominating taxa, for syntrophic propionate degraders as much as for acetoclastic methanogens.

Intermittent fasting for microbes: how discontinuous feeding increases functional stability in anaerobic digestion.

BACKGROUND: Demand-driven biogas production could play an important role for future sustainable energy supply. However, feeding a biogas reactor according to energy demand may lead to organic overloading and, thus, to process failures. To minimize this risk, digesters need to be actively steered towards containing more robust microbial communities. This study focuses on acetogenesis and methanogenesis as crucial process steps for avoiding acidification. We fed lab-scale anaerobic digesters with volatile fatty acids under various feeding regimes and disturbances. The resulting microbial communities were analyzed on DNA and RNA level by terminal restriction fragment length polymorphism of the mcrA gene, 16S rRNA gene amplicon sequencing, and a [2-13C]-acetate assay. A modified Anaerobic Digestion Model 1 (ADM1) that distinguishes between the acetoclastic methanogens Methanosaeta and Methanosarcina was developed and fitted using experimental abiotic and biotic process parameters. RESULTS: Discontinuous feeding led to more functional resilience than continuous feeding, without loss in process efficiency. This was attributed to a different microbial community composition. Methanosaeta dominated the continuously fed reactors, while its competitor Methanosarcina was washed out. With discontinuous feeding, however, the fluctuating acetic acid concentrations provided niches to grow and co-exist for both organisms as shown by transcription analysis of the mcrA gene. Our model confirmed the higher functional resilience due to the higher abundance of Methanosarcina based on its higher substrate uptake rate and higher resistance to low pH values. Finally, we applied our model to maize silage as a more complex and practically relevant substrate and showed that our model is likely transferable to the complete AD process. CONCLUSIONS: The composition of the microbial community determined the AD functional resilience against organic overloading in our experiments. In particular, communities with higher share of Methanosarcina showed higher process stability. The share of these microorganisms can be purposefully increased by discontinuous feeding. A model was developed that enables derivation of the necessary feeding regime for a more robust community with higher share of Methanosarcina.

Optimization of semi-continuous anaerobic digestion of sugarcane straw co-digested with filter cake: Effects of macronutrients supplementation on conversion kinetics.

Anaerobic digestion of sugarcane straw co-digested with sugarcane filter cake was investigated with a special focus on macronutrients supplementation for an optimized conversion process. Experimental data from batch tests and a semi-continuous experiment operated in different supplementation phases were used for modeling the conversion kinetics based on continuous stirred-tank reactors. The semi-continuous experiment showed an overall decrease in the performance along the inoculum washout from the reactors. By supplementing nitrogen alone or in combination to phosphorus and sulfur the specific methane production significantly increased (P<0.05) by 17% and 44%, respectively. Although the two-pool one-step model has fitted well to the batch experimental data (R2>0.99), the use of the depicted kinetics did not provide a good estimation for process simulation of the semi-continuous process (in any supplementation phase), possibly due to the different feeding modes and inoculum source, activity and adaptation.

Demand-driven biogas production by flexible feeding in full-scale - Process stability and flexibility potentials.

For future energy supply systems with high proportions from renewable energy sources, biogas plants are a promising option to supply demand-driven electricity to compensate the divergence between energy demand and energy supply by uncontrolled sources like wind and solar. Apart expanding gas storage capacity a demand-oriented feeding with the aim of flexible gas production can be an effective alternative. The presented study demonstrated a high degree of intraday flexibility (up to 50% compared to the average) and a potential for an electricity shutdown of up to 3 days (decreasing gas production by more than 60%) by flexible feeding in full-scale. Furthermore, the long-term process stability was not affected negatively due to the flexible feeding. The flexible feeding resulted in a variable rate of gas production and a dynamic progression of individual acids and the respective pH-value. In consequence, a demand-driven biogas production may enable significant savings in terms of the required gas storage volume (up to 65%) and permit far greater plant flexibility compared to constant gas production.

Optimization of hydrolysis and volatile fatty acids production from sugarcane filter cake: Effects of urea supplementation and sodium hydroxide pretreatment.

Different methods for optimization the anaerobic digestion (AD) of sugarcane filter cake (FC) with a special focus on volatile fatty acids (VFA) production were studied. Sodium hydroxide (NaOH) pretreatment at different concentrations was investigated in batch experiments and the cumulative methane yields fitted to a dual-pool two-step model to provide an initial assessment on AD. The effects of nitrogen supplementation in form of urea and NaOH pretreatment for improved VFA production were evaluated in a semi-continuously operated reactor as well. The results indicated that higher NaOH concentrations during pretreatment accelerated the AD process and increased methane production in batch experiments. Nitrogen supplementation resulted in a VFA loss due to methane formation by buffering the pH value at nearly neutral conditions (∼ 6.7). However, the alkaline pretreatment with 6g NaOH/100g FCFM improved both the COD solubilization and the VFA yield by 37%, mainly consisted by n-butyric and acetic acids.

Critical comparison of different model structures for the applied simulation of the anaerobic digestion of agricultural energy crops.

Different model structures were compared to simulate the characteristic process variables of the anaerobic digestion of maize, sugar beet and grain silage. Depending on the type and number of the required components, it can be shown that in comparison to the complex Anaerobic Digestion Model No. 1 (ADM1) different simplified model structures can describe the gas production rate, ammonia nitrogen and acetate concentration or pH value equally well. Since the reduction of the predominantly fast kinetics of the methanogenesis, acetogenesis or acidogenesis will only have little effect on the simulation of the specific gas production, it can be proven that the hydrolysis is the rate-limiting step during the uninhibited anaerobic digestion of complex particulate substrates. However, the stoichiometric comparison reveals that the model protein gelatine is not suitable for a representative characterization of agricultural energy crops.