Improving total nitrogen removal using a neural network ammonia-based aeration control in activated sludge process

M. H. Husin; M. F. Rahmat; N. A. Wahab; M. F. M. Sabri

Uneingeschränkter Zugang

Improving total nitrogen removal using a neural network ammonia-based aeration control in activated sludge process

und

| 23. Sept. 2021

International Journal on Smart Sensing and Intelligent Systems

Band 14 (2021): Heft 1 (January 2021)

Über diesen Artikel

Vorheriger Artikel

Nächster Artikel

Zitieren

Online veröffentlicht: 23. Sept. 2021

Seitenbereich: 1 - 16

Eingereicht: 25. Mai 2021

Akzeptiert: 26. Aug. 2021

DOI: https://doi.org/10.21307/ijssis-2021-016

Schlüsselwörter
ABAC, Effluent, Energy consumption, Neural network, Wastewater treatment

© 2021 M. H. Husin et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Flowchart of the proposed NN ABAC methodology.

The implementation of NN ABAC control architecture in BSM1.

Performances of the last 7 days of simulation using dry weather with the PI (black line), PI-ABAC (red dotted line), and NN-ABAC (blue line).

Performances of the last 7 days of simulation using rain weather with the PI (black line), PI-ABAC (red dotted line), and NN-ABAC (blue line).

Performances of the last 7 days of simulation using storm weather with the PI (black line), PI-ABAC (red dotted line), and NN-ABAC (blue line).

Summary of recent research trend using ABAC control.

	Author	Methods	Results
PI ABAC	Uprety et al. (2015)	Feedback PID controller for ABAC to adjust DO in all aeration basins and zones	Decrease in supplemental carbon used for denitrification by 53% and overall decrease in energy consumption by 10%
	Várhelyi et al. (2018)	DO cascade, ABAC and combination of ABAC with the control of nitrate and return activated sludge recycles	ABAC combination is the most cost-saving methods (reduction of about 43%)
MPC ABAC	Santín et al. (2015a)	Fuzzy control and MPC (Feedforward ABAC)	Total Nitrogen (N_tot) violations reduced by 11.04% and 100% elimination of S_NH violations
	Santín et al. (2016)	Risk detection of effluent violation using artificial NN, fuzzy controller to improve denitrification/nitrification and MPC to improve DO tracking	N_tot violations reduced up to 97.63% and S_NH violations reduced up to 68.29% (N_tot violation strategy) N_tot violations reduced up to 78.81% and 100% elimination of S_NH violations (S_NH violation strategy)

The effluent violations under dry, rain, and storm influent.

				% of reduction
	PI	PI-ABAC	NN-ABAC	vs. PI	vs. PI-ABAC
Dry
N_tot violations (% of operating time)	17.86	11.90	11.61	‒34.99%	‒2.44%
N_tot violations (Occasions)	7	5	5	‒28.57%	0.00%
S_NH violations (% of operating time)	16.82	16.52	16.67	‒0.89%	+0.91%
S_NH violations (Occasions)	5	5	5	0.00%	0.00%
Rain
N_tot violations (% of operating time)	11.01	6.10	5.65	‒48.68%	‒7.38%
N_tot violations (Occasions)	5	3	3	‒40.00%	0.00%
S_NH violations (% of operating time)	25.60	22.92	21.58	‒15.70%	‒5.85%
S_NH violations (Occasions)	8	8	8	0.00%	0.00%
Storm
N_tot violations (% of operating time)	15.48	10.86	10.71	‒30.81%	‒1.38%
N_tot violations (Occasions)	7	5	5	‒28.57%	0.00%
S_NH violations (% of operating time)	26.34	25.15	25.15	‒4.52%	0.00%
S_NH violations (Occasions)	7	7	7	0.00%	0.00%
TSS violation (% of operating time)	0.30	0.30	0.30	0.00%	0.00%
TSS violations (Occasions)	2	2	2	0.00%	0.00%

Parameter used for LM training algorithm.

Maximum number of Epochs to train	1,000
Performance goal	0
Maximum validation failures	6
Minimum performance gradient	1e–7
Initial µu	0.001
µu decrease factor	0.1
µu increase factor	10
Maximum µu	1e10

Number of neurons suggested by the researcher and the corresponding MSE value

Researcher	Method	Number of hidden neurons	Mean square error (MSE)
Huang (2003)	$N_{h} = \sqrt{(m + 2) N} + 2 \sqrt{N / (m + 2)}$	75	0.0113080
Jinchuan and Xinzhe (2008)	$N_{h} = (N_{i n} + \sqrt{N_{p}}) / L$	28	0.0052734
Shibata and Ikeda (2009)	$N_{h} = \sqrt{N_{i} N_{o}}$	1	0.0089480

Comparison of five backpropagation algorithms.

BP algorithm	Function	MSE	Epoch	R
Levenberg–Marquardt	trainlm	0.0057795	23	0.99019
Scaled conjugate gradient	trainscg	0.0073901	27	0.98264
BFGS quasi-Newton	trainbfg	0.0074205	58	0.98849
Batch gradient descent	traingd	0.0543580	1000	0.92262
Batch gradient descent with momentum	traingdm	0.1869000	8	0.71436

The comparison of EQ, AECI, and Total OCI in dry/rain/storm weather.

				% of reduction
	PI	PI ABAC	NN ABAC	vs. PI	vs. PI ABAC
Dry
EQI (kg poll.unit s/d)	6,096.71	5,938.3021	5,978.3177	‒1.94%	+0.67%
AECI (kWh/day)	3,697.57	3,769.517	2,835.2703	‒23.32%	‒24.78%
Total OCI	16,366.30	16,500.995	15,689.4197	‒4.14%	‒4.92%
Rain
EQI (kg poll.unit s/d)	8,146.75	8,005.5647	8,029.1791	‒1.44%	+0.29%
AECI (kWh/day)	3,671.70	3,786.5543	2,832.47	‒22.86%	‒25.20%
Total OCI	15,969.35	16,133.8675	15,302.504	‒4.18%	‒5.15%
Storm
EQI (kg poll.unit s/d)	7,187.89	7,044.115	7,079.7043	‒1.51%	+0.51%
AECI (kWh/day)	3,720.76	3,830.8403	2,833.1054	‒23.86%	‒26.04%
Total OCI	17,328.67	17,403.9539	16,530.1204	‒4.61%	‒5.02%

The comparison of AECI, EQI, OCI, and SNH and Ntot violations in similar studies.

Similar studies	Proposed NN ABAC	Husin et al. (2020b)	Husin et al. (2021b)
AECI (kWh/day)	2,835.2703	3,641.69	3,749.24
EQI (kg poll.unit s/d)	5,978.3177	6,081.46	5,975.75
Total OCI	15,689.4197	16,366.30	16,435.9
N_tot violations (% of operating time)	11.61	15.77	13.8
S_NH violations (% of operating time)	16.67	16.82	16.07

Concentration thresholds of pollutants in the effluent.

Variables	N_tot [g N/m³]	COD_t [g COD/m³]	S_NH [g N/m³]	TSS [g SS/m³]	BOD₅ [g BOD/m³]
Max. values	18	100	4	30	10

The effluent quality limit.

Effluent average	S_NH (<4 g N.m^‒3)	TSS (<30 g SS.m^‒3)	N_tot (<18 g N.m^‒3)	CODt (<100 g COD.m^‒3)	BOD5 (<10 g BOD.m^‒3)
Dry
PI	2.4783	13.0248	16.8908	48.2470	2.7587
PI ABAC	2.5481	13.0244	15.8626	48.2736	2.7654
NN ABAC	2.9118	13.0233	15.3519	48.2888	2.7689
Rain
PI	3.1575	16.1970	14.7159	45.4587	3.4569
PI ABAC	3.1299	16.197	14.1804	45.4702	3.459
NN ABAC	3.2918	16.1958	13.9606	45.47	3.4581
Storm
PI	2.9953	15.2935	15.8340	47.6875	3.2065
PI ABAC	2.9965	15.2935	15.1311	47.7043	3.2103
NN ABAC	3.2386	15.2923	14.8198	47.7119	3.2115

eISSN:: 1178-5608
Sprache:: Englisch

Zeitrahmen der Veröffentlichung:: Volume Open
Fachgebiete der Zeitschrift:: Technik, Einführungen und Gesamtdarstellungen, andere

Zeitschrift RSS Feed

Improving total nitrogen removal using a neural network ammonia-based aeration control in activated sludge process

Online veröffentlicht: 23. Sept. 2021

Seitenbereich: 1 - 16

Eingereicht: 25. Mai 2021

Akzeptiert: 26. Aug. 2021

DOI: https://doi.org/10.21307/ijssis-2021-016

Schlüsselwörter
ABAC, Effluent, Energy consumption, Neural network, Wastewater treatment

© 2021 M. H. Husin et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1:

Figure 2:

Figure 3:

Figure 4:

Figure 5:

Figure 6:

Figure 7:

Figure 8:

Figure 9:

Figure 10:

Figure 11:

Summary of recent research trend using ABAC control.

The effluent violations under dry, rain, and storm influent.

Parameter used for LM training algorithm.

Number of neurons suggested by the researcher and the corresponding MSE value

Comparison of five backpropagation algorithms.

The comparison of EQ, AECI, and Total OCI in dry/rain/storm weather.

The comparison of AECI, EQI, OCI, and SNH and Ntot violations in similar studies.

Concentration thresholds of pollutants in the effluent.

The effluent quality limit.

Improving total nitrogen removal using a neural network ammonia-based aeration control in activated sludge process

Online veröffentlicht: 23. Sept. 2021

Seitenbereich: 1 - 16

Eingereicht: 25. Mai 2021

Akzeptiert: 26. Aug. 2021

DOI: https://doi.org/10.21307/ijssis-2021-016

SchlüsselwörterABAC, Effluent, Energy consumption, Neural network, Wastewater treatment

© 2021 M. H. Husin et al., published by Sciendo

This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Figure 1:

Figure 2:

Figure 3:

Figure 4:

Figure 5:

Figure 6:

Figure 7:

Figure 8:

Figure 9:

Figure 10:

Figure 11:

Summary of recent research trend using ABAC control.

The effluent violations under dry, rain, and storm influent.

Parameter used for LM training algorithm.

Number of neurons suggested by the researcher and the corresponding MSE value

Comparison of five backpropagation algorithms.

The comparison of EQ, AECI, and Total OCI in dry/rain/storm weather.

The comparison of AECI, EQI, OCI, and SNH and Ntot violations in similar studies.

Concentration thresholds of pollutants in the effluent.

The effluent quality limit.

Schlüsselwörter
ABAC, Effluent, Energy consumption, Neural network, Wastewater treatment