Modelling of biohydrogen generation in microbial electrolysis cells (MECs) using a committee of artificial neural networks (ANNs)

Figures & data

Figure 1. Topology of neural network committee with one input layer (six neurons), one hidden layer (six to ten neurons) and one output layer (one neuron).

Table 1. Database used for the development of ANN committee model.

Carbon source	No. of data points	pH	T (°C)	S (g/L)	Eapp (V)	MEC configuration	Source
Glycerol (5)	2	7	30	1	0.9	1	[11]
Acetate (1)	41	4.9–9	4–30	0.118–90.7	0.2–1.2	1–2	[5,10–14,26–34]
Glucose (8)	2	7	30	1	0.6–0.9	1	[11,5]
Cellulose (7)	1	7	30	1	0.6	2	[5]
Butyric acid (3)	1	7	30	1	0.6	1	[5]
Lactic acid (4)	1	7	30	1	0.6	1	[5]
Propionic acid (2)	1	7	30	1	0.6	1	[5]
Valeric acid (6)	1	7	30	1	0.6	1	[5]

Note: Temperature (T); substrate concentration (S); applied voltage (E_app); MEC configuration was designated as (1) for single chamber and (2) for two chambers. Numbers next to substrates were allocated to differentiate between the various types. This was done based on the molecular weights of each substrate thus acetate (59.04 g/mol), propionic acid (74.08 g/mol), butyric acid (88.11 g/mol), lactic acid (90.08 g/mol), glycerol (92.09 g/mol), valeric acid (102.13 g/mol), cellulose (176.15 g/mol) and glucose (180.16 g/mol) were ranked as numerical values 1, 2, 3, 4, 5, 6, 7 and 8, respectively.

Table 2. Ranges for input and output parameters used in the committee model development.

Parameter	Minimum	Maximum	Unit
Substrate type	1	8	–
Substrate concentration	0.1	91	g/L
pH	4.5	9	–
Temperature	4	30	°C
Applied voltage	0.2	1.2	V
*MEC configuration	1	2	–
H2	0	9	mol H2/mol substrate

Note: *MEC configuration: single chamber (1); two chamber (2); hydrogen yield (H₂). This was done based on the molecular weights, as described in Table 1.

Figure 2. The back propagation training flowchart for artificial neural network. Note: Mean square error (MSE).

Figure 3. Committee member model 1–5 (a–e) and the average (f) predicted versus observed biohydrogen yields (mol H₂/mol substrate) values for 41 experimental data sets. Note: The diagonal line illustrates expectations under a one-to-one relationship between predicted and observed values. R² = 0.90 (a); R² = 0.81 (b); R² = 0.85 (c); R² = 0.70 (d); R² = 0.80 (e); R² = 0.85 (f).

Figure 4. Observed H₂ yield (mol H₂/mol substrate) compared to the average predicted (of the five models) H₂ yield (mol H₂/mol substrate) on the nine data points used for testing of the model.

Selembo PA, Perez JM, Lloyd WA, et al. High hydrogen production from glycerol or glucose by electrohydrogenesis using microbial electrolysis cells. Int J Hydrogen Energy. 2009;34:5373–5381.

 Web of Science ®Google Scholar

Cheng SA, Logan BE. Sustainable and efficient biohydrogen production via electrohydrogenesis. Proc Natl Acad Sci USA. 2007;104(47):18871–18873.

 PubMed Web of Science ®Google Scholar