**1. Introduction**

The further rollback of globalization will ultimately reshape the current supply chain block, especially as more and more countries have realized how pivotal it is to have selfsufficient industries to produce strategic products such as medicine, energy, and even toilet paper rolls [1]. Aside from the public health emergency, energy security is another draconian challenge that countries across the world are reluctantly facing, although the price of crude oil did once plunge to USD 25 per barrel (158.98 L) in the middle of 2020 during the COVID-19 pandemic [2]. Whether to take bolder steps in the energy reliance transition from fossil fuel to renewable energy will make a great difference in the world that our children will be able to inherit in the future [3]. Consequently, by 2021, several

**Citation:** Liu, Y.; Liu, J.; He, H.; Yang, S.; Wang, Y.; Hu, J.; Jin, H.; Cui, T.; Yang, G.; Sun, Y. A Review of Enhancement of Biohydrogen Productions by Chemical Addition Using a Supervised Machine Learning Method. *Energies* **2021**, *14*, 5916. https://doi.org/10.3390/ en14185916

Academic Editor: Bahman Shabani

Received: 25 August 2021 Accepted: 14 September 2021 Published: 17 September 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

developed countries already started to restrict the use of fossil fuels in order to eventually achieve a shift in fuel type [4,5].

Among all sources of energy, hydrogen (H2) is one of the most favorable candidates due to its inherent appealing features: (1) high energy yield (122 kJ kg<sup>−</sup>1), (2) generation of water as a result of combustion, and (3) electricity generation through the fuel cell [6,7]. However, the current predominant H2 generation still comes from fossil-based materials via existing mature industrial chemical processes such as natural gas steam reforming (NGSR), nature gas thermal cracking (NGTC), auto-thermal reforming (ATR), coal gasification, and partial oxidation of heavier-than-naphtha hydrocarbons [8]. Consequently, the paradox of sustainability of H2 utilization and the non-renewability of H2 generation will be encountered, although the development of carbon capture storage and utilization (CCSU) such as via a mature catalytic process like Fischer–Tropsch synthesis might alleviate environmental impacts from H2 generation [9–12].

Apart from the thermal process, the biological hydrogen (BioH2) generation process also plays a supplementary role in H2 generation due to features such as versatile feedstock (lignocellulose, wet kitchen organic waste, and wastewater) and no green-house gas emissions (GHE). Despite the appealing advantages that are mentioned above, BioH2 production is hampered by its relatively lower process performance [13]. To implement BioH2 in different applications either on a decentralized or centralized basis or both, different process intensification approaches have been proposed, such as hydrolysate detoxification, mixed continuous and batch operations, co-fermentation, process optimization, and chemical addition. Among these approaches, chemical addition is considered to be one of the most attractive and practical ones because of its operational simplicity (without any additional modifications) and relatively low energy consumption [14]. However, current reports are limited to focusing on the facilitation of BioH2 production by all types of chemical additives. In contrast, the nanoparticles (NPs) as a potential type of chemical additive still lack research on their addition and the corresponding quantitative relationships, such as hydrogen yield (HY) and hydrogen evolution rate (HER) with detailed incubation conditions, especially the concentration of different metal elements.

In this paper, instead of making a simple BioH2 production enhancement comparison using the addition of NPs across literature reports, the collected data (such as HY, HER, and the substrate concentrations from literature works) were used to construct the data matrix for supervised machine learning algorithm using the developed artificial neural networks (ANNs) coupled with statistical analysis using response surface methodology (RSM) for more insightful and quantitative correlations and analysis. The review of assessing the impact of NPs additions on BioH2 production in form of HY and HER using a developed ANNs-RSM algorithm, to the best of our knowledge, has not been reported before.

#### **2. Materials and Methods**

The literature used in this review was mainly collected from the scientific databases from Web of Science, Google Scholar and Science Direct via keyword search. Various keyword groups were comprised of several words, including "dark fermentation," "biohydrogen," and "nanoparticles." With regard to the possible missing relevant literature, by using the abovementioned searching strategy, an extensive additional search process was conducted with more detailed keywords, including "trace metal," "transitional metal," "iron," "nickel," "gold," "copper," and "metal oxide." During the additional search, these mentioned keywords were also combined with the keyword "biohydrogen."

The ANNs (based on Python 2.7 platform) was deployed for data analysis. The detailed schematic diagram of the construction of the ANNs and data collection is shown in Figure S1. In this work, the widely used feed-forward three-layer networks were used. The simplified cross-out method was used for cross-validation during the data training step. The detailed descriptions of the standard procedures for this methodology can be

found in our previous works [15]. During the data training, the mean square error (*MSE*) and mean average relative residual (*MARR*) were computed as follows:

$$MSE\% = \frac{1}{N\_{sam}} \sum\_{j=1}^{N\_{sam}} \left( r\_j^{sam} - r\_j^{cal} \right)^2 \times 100\% \tag{1}$$

$$MARR\% = \frac{1}{N\_{sam}} \sum\_{j=1}^{N\_{exp}} \left( \frac{\left| r\_i^{sam} - r\_i^{cal} \right|}{r\_i^{sam}} \right) \times 100\% \tag{2}$$

where *Nsam* is the number of data, and *rsam <sup>i</sup>* and *<sup>r</sup>cal <sup>i</sup>* are actual and calculated values, respectively. The setting for allowable accuracy was 95%. For the ANNs prediction data matrix, the widely used Box–Behnken design (BBD) and the central composite design (CCD) were used to predict the data matrix generation [16]. Once the supervised data learning was complete, the analysis of variation (ANOVA) based on commercial Design Expert® Version 11 software package (Stat-Ease, Inc., Minneapolis, MN, USA) was used for statistical analysis.

#### **3. Literature Survey Comparisons**

In this paper, for the convenience of discussion, four different types of NPs (Fe-based, Au-based, Cu-based, and Ni-based) were surveyed across different studies and the results are shown in Figure 1.

**Figure 1.** Statistics of publications from Scopus and Google Scholar in regard to BioH2 production by chemical nanoparticle additions.

For each type of NPs, taking Ni-based NPs, for instance, all nickel-related species were included, such as nanoparticles such as zero-valent particles, metal oxide NiO2, etc. The number of reports on the topic of BioH2 enhancement by NPs additions has been increasing steadily since 2015. Among different NPs, the number of reports using iron-based NPs has presented a discernible trend in recent years. The impetus underlining this trend is possibly associated with its inherent appealing cost-effective feature compared to other NPs such as gold or nickel. Apart from Fe-based NPs, Ni-based NPs have experienced an appreciable increase in recent years, with an exception in 2015 [17,18]. The research interests that focus on Ni-based NPs might be pertinent to the metal cluster of hydrogenase [19]. According to recent classifications, there are three different hydrogenases, namely, [Fe], [NiFe], and [FeFe] [20,21]. During biological chemical reactions, these enzyme active centers play a pivotal role in the metabolism of proton ion-associated redox reactions. Studies have shown that [FeFe] hydrogenase catalyzes H2 generation, whereas [NiFe] hydrogenase catalyzes the consumption of H2. [NiFe] hydrogenase presents a relatively higher tolerance to the existence of oxygen and it widely exists in various types of microbial strains, whereas [FeFe] hydrogenase is relatively strict to the presence of oxygen and only exists in some algae and bacteria [22,23]. Regarding [Fe] hydrogenase, it only strictly exists in some methanogen strains [24–26].

### **4. Underlying Mechanisms of Metal Ions and Metal-Based Nanoparticles**

Many extensively studied metal ions and metal-based nanoparticles are regarded as effective additives in culture medium to facilitate BioH2 production in the dark fermentation process, including Na+, K+, NH4 +, Mg2+, Ca2+, Co2+, Zn2+, Cu2+, Fe2+/Fe3+, and Ni2+/Ni3+, among others [27–29]. Extensive studies have found that even small changes in the latter may have a significant impact on BioH2 production; hence, many strategies have been proposed based on them, such as concentration regulation, including concentration manipulation [30], size regulation [17,31], composites fabrication [23,32], and heteroatom doping [33]. In general, the enhancement of NPs addition lies in a few important facts: (i) the controllable release of mental ions that facilitates the passive transport across the membrane [34]; (ii) nanodots that facilitate the electron transport chain during metabolism, such as glycolysis [35]; and (iii) the appropriate level of NPs favorable to the hydrogenase activities (co-enzymes often contain the metal ions in the catalysis center, which ultimately enhances the rate of hydrogen generation [36]. The potential mechanisms of BioH2 enhancement are summarized in Figure 2. Therefore, in this part, this review will focus on the impact of the latter on BioH2 production and its mechanisms.

**Figure 2.** Potential mechanism of BioH2 enhancement by NPs addition.

## *4.1. Fe-Based Ions and Nanoparticles*

Iron is an important trace element in the formation of hydrogenases and other enzymes. The pre-addition of Fe in the culture medium is a widely used strategy to enhance BioH2 production in dark fermentation [37]. As illustrated in Figure 2, first, Fe is the essential element to form the metal content at the active sites of hydrogenase ([FeFe], [FeNi], and [Fe]), thus catalyzing the reduction reaction of H<sup>+</sup> to H2 [38]. Second, the presence of Fe-based NPs improves the activity of ferredoxin oxidoreductase by reducing the dissolved oxygen (DO) level and enhancing electron transfer due to the surface and quantum size effects [39,40]. In addition, Fe-based components could participate in enriching the microbial community and enhancing the growth of H2-producing bacteria [41]. The oxidative stress increases when there is a higher Fe concentration, which results in the formation of abundant oxidative radicals, thus leading to the deactivation or decomposition of enzymes [17,30].
