KR20230086849A - Dual­objective optimization for energy­saving and fouling mitigation in MBR plants using AI­based influent prediction and an integrated biological­physical model - Google Patents

Abstract

The present invention relates to a system and a method for optimizing a membrane bioreactor based on an influent artificial intelligence prediction model. More specifically, the present invention relates to a system for optimizing a membrane bioreactor based on an influent artificial intelligence prediction model, which comprises: an influent prediction model generating unit for predicting characteristics of influent after a specific time; and a harmony search for controlling a sewage treatment system based on the characteristics of the influent predicted by the influent prediction model generating unit, so that concentration of pollutants in the influent is predicted using an artificial intelligence bidirectional gate circulation unit for the influent that is rapidly changed, and a dual-purpose optimization system is configured using the harmony search so as to optimize an aeration intensity of an aerobic tank and a filtration-washing period of a membrane separation tank, thereby reducing fouling and energy consumption and improving environmental friendliness and economic feasibility.

Description

Influent artificial intelligence prediction model-based membrane bioreactor optimization system and optimization method {Dualobjective optimization for energy saving and fouling mitigation in MBR plants using AIbased influent prediction and an integrated biologicalphysical model}

The present invention relates to an influent artificial intelligence prediction model-based membrane bioreactor optimization system and optimization method. More specifically, it is about a dual-purpose optimization system for energy saving and fouling reduction in a membrane bioreactor (MBR) plant based on a two-way gate circulation unit and a harmony search algorithm among artificial intelligence technologies.

Optimal operating strategies for wastewater treatment plants have been widely used to improve economic and environmental operations. However, inadequately optimized systems have been applied in membrane bioreactor (MBR) sewage treatment plants, resulting in high energy consumption due to uncertain effects associated with complex biological and physical interactions associated with membrane fouling.

Membrane bioreactors (MBRs) combine biological treatment with physical membrane separation. MBR produces superior water quality compared to conventional wastewater treatment plants (WWTPs) and does not require additional disinfection. Despite the advantages of MBR, fouling caused by particulate adhesion to the membrane reduces MBR performance. Therefore, this has been a major obstacle to the widespread application of MBR.

Operational strategies for fouling mitigation offer the potential to make MBR economical and practical, and optimization of MBR systems could help ensure process efficiency.

Process optimization can reduce operating costs while balancing productivity and energy requirements. Optimization satisfies the purpose of the WWTP. The first is to meet the effluent discharge limit, the second is to avoid abnormal failure (safety), and the third is to minimize operating costs. Despite the importance of improving process performance, few studies have been conducted on process optimization of MBR systems.

A conventional activated sludge model 2d-based MBR system (ASM2d-MBR) has been proposed to optimize energy demand without affecting nutrient removal. An optimization technique based on a benchmark simulation model (BSM-MBR) is proposed to evaluate energy requirements and runoff quality. ASM2d model-based optimization suggested optimal set points for MBR to improve nitrogen removal and energy efficiency. A computational fluid dynamics approach was adopted for the submerged MBR to determine the optimal design and operating conditions.

However, the biological platform was derived from activated sludge systems and suffered from simplistic modeling approaches. Therefore, important information on the interaction between fouling and biological and physical processes was excluded.

In addition, a physical filtration protocol was developed to investigate the development of unremovable membrane fouling (fouling) and sludge formation in conventional membrane reactors. Three-dimensional computational fluid dynamics was used to propose hydrodynamic optimization for energy and fouling reduction. An experimental air-refining control system for membranes is proposed to reduce the air-refining flow rate and energy consumption. However, it is also difficult to consider the interaction between physical and biological processes.

Fouling, a major drawback of many MBRs, is the physical attachment of sludge to the membrane surface. Membrane fouling usually occurs in the form of foulant adhesion to form a gel/cake layer or thermodynamic filtration resistance of a gel/cake layer. In addition, the nature of the biomass, soluble microorganisms (SMPs), particulate products and anionic strength can also affect fouling.

Therefore, biological and physical processes must be integrated when attempting to optimize operations. Dynamic impact fluctuations can also affect the complex interactions of fouling mechanisms. This increases the growth demand for maintenance of nutrient removal efficiency while hindering microbial activity and increasing dissolved oxygen (DO) concentration.

The DO concentration in the MBR is very important because the diffused air sweeps across the membrane surface to clean the condensed sludge on the membrane surface and maintain aerobic conditions for nitrification. Suboptimal DO levels can affect more than 50% of operating costs.

The high uncertainty about the level of dynamic influence of the MBR can be an obstacle to DO setpoint optimization. Process control must take into account the changes that affect it to maintain efficient contaminant removal, and must anticipate the conditions that will affect successful optimization of the MBR.

Korean registered patent 10-1629240 Republic of Korea Patent Publication 10-2021-0109160 Republic of Korea Patent Publication 10-2021-0058231 Republic of Korea Patent Publication 10-2009-0078501

Therefore, the present invention has been made to solve the above conventional problems, and according to an embodiment of the present invention, the concentration of pollutants in influent is predicted by predicting rapidly changing influent using an artificial intelligence bi-directional gate circulation unit. In addition, by constructing a dual-purpose optimization system using harmony search, influent artificial The purpose is to provide an intelligent predictive model-based membrane bioreactor optimization system and optimization method.

According to an embodiment of the present invention, as a result of verification using domestic MBR plant data, energy efficiency can be improved by 12% and fouling can be reduced by 26%. It can reduce energy consumption and energy consumption, and is expected to be applied to various sewage treatment plant operation fields. For commercialization, influent artificial intelligence, which is expected to be used for energy and environmental improvement by using this technology in domestic and foreign sewage treatment plants and environmental system companies. The purpose is to provide a predictive model-based membrane bioreactor optimization system and optimization method.

On the other hand, the technical problems to be achieved in the present invention are not limited to the above-mentioned technical problems, and other technical problems that are not mentioned will become clear to those skilled in the art from the description below. You will be able to understand.

A first object of the present invention is an influent prediction model generating unit for predicting the characteristics of influent after a specific time; and a control unit controlling the sewage treatment system based on the influent characteristics predicted by the influent prediction model generation unit.

In addition, the influent prediction model generating unit utilizes a bidirectional gate circulation unit, and the bidirectional gate unit generates an influent prediction model per hour and predicts influent flow rate, COD, TSS, and TN value characteristics after a specific time. .

In addition, the hourly influent prediction model may be characterized in that influent characteristics are continuously predicted after a specific time using previously input data.

And the control unit may be characterized in that it controls the membrane bioreactor plant.

In addition, the control unit may be characterized in optimizing the control factor of the plant through harmony search.

The control unit may optimize the aeration intensity of the aerobic tank and the filtration and washing periods of the membrane separation tank.

In addition, it further includes an integrated MBR modeling unit that models the membrane bioreactor process by integrating the biological model and the physical model, wherein the biological model uses ASM-SMP and the physical model uses the RIS model to model the MBR process. can be characterized.

The control unit may optimize aeration intensity and effluent quality index based on the biological model, and optimize membrane fouling and membrane pumping energy while maintaining water productivity based on the biological model and the physical model.

A second object of the present invention is a method for optimizing a membrane bioreactor, comprising: predicting characteristics of influent after a specific time by an influent prediction model generating unit; inputting predicted influent characteristics into an integrated MBR modeling unit that models a membrane bioreactor process by integrating a biological model and a physical model; and optimizing and controlling, by a control unit, the membrane bioreactor based on the biological and physical models of the integrated MBR modeling unit.

The influent prediction model generation unit uses a bidirectional gate circulation unit, and the bidirectional gate unit generates an hourly influent prediction model, predicts the influent flow rate, COD, TSS, and TN value characteristics after a specific time, and the hourly influent prediction model may be characterized by continuously predicting influent characteristics after a specific time using previously input data.

In addition, the control unit optimizes the control factor of the membrane bioreactor plant through harmony search, and the control factor may be characterized in that the aeration intensity of the aerobic tank and the filtration and washing period of the membrane separation tank are optimized.

In addition, the biological model may use ASM-SMP, and the physical model may model the MBR process using a RIS model.

In addition, the control unit may optimize aeration intensity and effluent quality index based on the biological model, and optimize membrane fouling and membrane pumping energy while maintaining water productivity based on the biological model and the physical model.

According to the influent artificial intelligence prediction model-based membrane bioreactor optimization system and optimization method according to an embodiment of the present invention, the concentration of pollutants in the influent is predicted by prediction using an artificial intelligence bi-directional gate circulation unit for rapidly changing influent, In addition, by constructing a dual-purpose optimization system using harmony search, the aeration intensity of the aerobic tank and the filtration-cleaning period of the membrane separation tank are optimized to reduce fouling and energy consumption, thereby improving environmental performance and economic feasibility.

According to the influent artificial intelligence prediction model-based membrane bioreactor optimization system and optimization method according to an embodiment of the present invention, as a result of verification using domestic MBR plant data, energy efficiency can be improved by 12% and fouling can be reduced by 26%. It can reduce fouling and energy consumption by applying it to the operation system of domestic and foreign MBR plants, and is expected to be applied to various sewage treatment plant operation fields. It is expected to be used for energy and environmental improvement.

On the other hand, the effects obtainable in the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below. You will be able to.

The following drawings attached to this specification illustrate preferred embodiments of the present invention, and together with the detailed description of the invention serve to further understand the technical idea of the present invention, the present invention is limited only to those described in the drawings. and should not be interpreted.
1 and 2 are block diagrams of a Dual Objective Optimization (DOO) system using biological and physical processes integrated with a predictive model according to an embodiment of the present invention ((a) to (d) represent the sub-steps of the framework). (SMP is the soluble microbial product and TSS is the total suspended solids)),
3 is a conceptual block diagram of an hourly prediction model for influent conditions of an MBR plant (EEMD is an ensemble empirical mode decomposition, IMF is an eigenmode function, and DL is deep learning);
Figure 4 is a schematic description of the integrated bio-physical model (consisting of ASM-SMP and series resistance model);
5 is a diagram of the relationship between biological fouling formation and physical fouling separation;
6 shows influent TN concentration prediction results: (a) comparison of measurement data using MLR, ES, ARIMA, MLP, GRU, and BiGRU during the test period and (b) comparison of prediction errors during the test period;
7 is a box plot of optimal DO setpoints by a dual target optimization system: (a) high, (b) low influent load conditions, (c) hourly DO setpoint variation, red boxes represent times of day with high pollutant loads. indicating,
Figure 8 shows the optimization results of the membrane bioreactor: (a) SMP and TSS concentrations in the reactor, (b) the trend of the optimized relaxation duration over time and (c) the optimized filtration cycle on November 5 (membrane start-up) ( d) December 5 (TSS and SMP increase);
Figure 9: Improved economic and environmental operation by the DOO system compared to the passive and ABAC systems: (a) AE and EQI and (b) emissions nitrogen compounds (nitrite and nitrate and Kjeldahl nitrogen) from 15 to 19 November. ,
Figure 10 is (a) passive system, (b) DOO system, (c) 4 types of fouling resistance evaluation through TMP reduction and operating time increase by optimized system,
11 shows contamination monitoring of passive and DOO systems using contamination mechanism models of full blockage, medium blockage and cake filtration;
12 shows biological improvement by the optimization system in a full-scale MBR plant: (a) DO set point reduction and (b) AE reduction and EQI maintenance;
Figure 13. Assessment of fouling progression by DOO systems in full-scale MBR plants compared to manual systems: (a) trends in optimal backwash duration, (b) fouling resistance reduction, (c) TMP reduction values and increased operating time

The above objects, other objects, features and advantages of the present invention will be easily understood through the following preferred embodiments in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments described herein and may be embodied in other forms. Rather, the embodiments introduced herein are provided so that the disclosed content will be thorough and complete and the spirit of the present invention will be sufficiently conveyed to those skilled in the art.

In this specification, when an element is referred to as being on another element, it means that it may be directly formed on the other element or a third element may be interposed therebetween. Also, in the drawings, the thickness of components is exaggerated for effective description of technical content.

Embodiments described in this specification will be described with reference to cross-sectional views and/or plan views, which are ideal exemplary views of the present invention. In the drawings, the thicknesses of films and regions are exaggerated for effective explanation of technical content. Accordingly, the shape of the illustrated drawings may be modified due to manufacturing techniques and/or tolerances. Therefore, embodiments of the present invention are not limited to the specific shape shown, but also include changes in the shape generated according to the manufacturing process. For example, a region shown at right angles may be rounded or have a predetermined curvature. Accordingly, the regions illustrated in the drawings have attributes, and the shapes of the regions illustrated in the drawings are intended to illustrate a specific shape of a region of a device and are not intended to limit the scope of the invention. Although terms such as first and second are used to describe various elements in various embodiments of the present specification, these elements should not be limited by these terms. These terms are only used to distinguish one component from another. Embodiments described and illustrated herein also include complementary embodiments thereof.

Terminology used herein is for describing the embodiments and is not intended to limit the present invention. In this specification, singular forms also include plural forms unless specifically stated otherwise in a phrase. The terms 'comprises' and/or 'comprising' used in the specification do not exclude the presence or addition of one or more other elements.

In describing the specific embodiments below, several specific contents are prepared to more specifically describe the invention and aid understanding. However, readers who have knowledge in this field to the extent that they can understand the present invention can recognize that it can be used without these various specific details. In some cases, it is mentioned in advance that parts that are commonly known in describing the invention and are not greatly related to the invention are not described in order to prevent confusion for no particular reason in explaining the present invention.

Hereinafter, an hourly prediction model with a 24-hour advance time range for influent conditions based on a deep learning technique and a prediction optimization system for an MBR plant according to an embodiment of the present invention will be described.

In addition, deep learning models and conventional statistical models were compared and evaluated. The models include multi-layer perceptron (MLP), gated iteration unit (GRU), bidirectional GRU (biGRU (BiGRU)), and multiple linear regression (MLR). , overcoming the limitations of statistical models such as exponential smoothing (ES) and autoregressive integrated moving average (ARIMA).

Harmony Search (HS)-based Dual Purpose Optimization (DO) can integrate biological and physical operating set points to provide robust functionality and low energy requirements with stable effluent quality, productivity, and fouling mitigation.

In addition, the optimized system according to an embodiment of the present invention was evaluated in comparison with a passive and ammonia-based aeration control (AAC) system in consideration of economic and environmental aspects. Optimized systems according to embodiments of the present invention may help operators operate MBR plants in an efficient and environmentally responsible manner, while reducing uncertainty about the future.

A framework according to an embodiment of the present invention, including optimization of biological and physical processes integrated with prediction of influent conditions, properties, for MBR, is graphically represented in FIGS. 1 and 2 .

The first step, shown in Fig. 1(a), is the implementation of a deep learning model that predicts changes in impact conditions over the next 24 hours, including flow rates and pollutant concentrations. Contaminants that have an impact include chemical oxygen demand (COD), total nitrogen (TN), and total dietary solids (TSS). The details of the algorithm will be described in detail later.

The predicted influent is used as an input to the integrated MBR model as shown in FIG. 1(b). The integrated model simultaneously accounts for biological and physical processes and the interconnections between them through aeration and biological particulate matter. Figure 1(c) shows that the harmony search-based predictive DO of the integrated process can predict optimal operating conditions for the next 24 hours. At this stage, DOU proposes DO setpoints for both aerobic and membrane reactors, taking into account the effluent quality and operating costs of these aerobic systems. At the same time, optimal permeability can reduce pumping energy (PE) and mitigate contamination while maintaining process productivity. Objective functions of physical processes are proposed experimentally. The last step, shown in Fig. 1d, is the overall evaluation of the proposed system for MBR in terms of energy, effluent quality and fouling strength. Energy savings are evaluated by energy intensity.

Hereinafter, an hourly prediction model for influent conditions in an MBR plant will be described.

Through the case study, the influent of the plant was predicted by a model that combines data decomposition techniques with statistics and deep learning prediction techniques as shown in FIG. 2 . Impact data from the pilot and actual MBRs were collected for 1 year and 9 months, respectively.

UFC-5030f (KROHNE, Germany), CODA-500 (Horiba, Japan), HU-200TB-IM (Horiba, Japan), and TPNA-500 (Horiba, Japan) were connected to measure flow rate, COD, TSS, and TN. We then used the previous 5 days of decomposition data by using a sliding window to predict data points 24 hours ahead of impact. The sliding window method sliding the decomposed data of the previous 5 days according to the time of pre-dictation of the data that will affect the future. If the values of the internal parameters of the predictive model were calculated in the training procedure, the predictive model kept the internal parameters when using the sliding window method. The measured impact data was divided into training data (75%) and test data (25%) according to operating time. For the pilot-scale MBR sewage treatment plant, the hourly forecast model predicted data points for 2016 flow, COD, TSS and TN, respectively. The 2016 data points represented the 25% train set (2160 hour data points) excluding the first input of the prediction model and the prediction length (120 + 24 data points).

Hourly prediction models were compared to a set of measured data in terms of Mean Absolute Error (MAE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE), which measure the difference between the predicted data and the measured impact data. Evaluation of the three models can be defined by Equations 1 to 3 below.

[Equation 1]

[Equation 2]

[Equation 3]

t,

t는 각각 t 시간에 영향을 미치는 측정 및 예측 데이터이며, T는 데이터 포인트의 수이다.here,

t,

t are the measured and predicted data each affecting time t, and T is the number of data points.

Ensemble empirical mode decomposition (EEMD) is an improved method of empirical mode decomposition (EMD). EEMD overcomes a major drawback of EMD: mode mixing, in which the decomposed signal is composed of other signals. EEMD finds the exact time frequency distribution of the data even though the data is non-stationary. As an effective self-adaptive decomposition method, EEMD has been utilized in several research fields including frequency analysis, fault diagnosis, monitoring and prediction.

It uses the decomposed impact data as input to the predictive model, and the sliding window method updates the input data and prediction point data over time. Therefore, the hourly influent forecasting model continuously predicts impact conditions 24 hours later using the previous 5 days of data. In the embodiment of the present invention, existing models (MLR, ES, and ARIMA) and artificial intelligence (deep learning) models MLP, GRU, and BiGRU (bidirectional gated cyclic unit) are used. Deep learning models are a recently introduced artificial intelligence technology that has only begun to be adopted in water quality management environments. In an embodiment of the present invention, a deep learning model is implemented using an EEMD and a sliding window of influential prediction to overcome the poor prediction performance of existing prediction models. This study utilizes and compares MLP as a basic deep learning tool, and utilizes GRU and BiGRU as special deep learning options for time series data prediction.

MBR processes have been modeled by biological or physical processes in previous techniques. The biological model used a series of activated sludge models (ASM), including biological components such as SMP, and the models can be used to optimize and control aeration and effluent quality. In addition, a physical model considering the physical mechanisms of cake formation and fouling tendency can be utilized to mitigate fouling through the identified fouling mechanisms.

Recently, an integrated MBR model that considers both biological and physical processes has been identified as the best tool to simulate the MBR process. This is because the integrated model can reflect the complexity of MBR and the interrelationships between processes. Thus, an integrated model can be used to optimize and control both biological and physical processes. In an embodiment of the present invention, the MBR process is modeled using an activated sludge model soluble mi-crobial product (ASM-SMP) and a resistance-in-series (RIS) model.

ASM-SMP describes the growth and decay of biomass, carbon and nitrogen removal, and the formation and degradation of SMP. RIS describes the physical mechanism of cake layer accumulation on and inside membrane pores. Four resistances (clean membrane (Rm), pore fouling (Rp), dynamic sludge cake (Rdc) and stable sludge cake resistance (Rsc)) are included in the total resistance (Rt) to calculate the TMP value. Several interactions between the ASM-SMP and RIS models shown in Fig. 1(b) are considered. The detailed structure of the integrated ASM-SMP and RIS model is shown in FIG. The main features of the model are summarized as follows:

1. Ammonia nitrogen is converted to nitrate nitrogen by a single-step autotrophic process that requires oxygen.

2. UAP is originally produced by substrate metabolism, BAP is derived from the decay of active biomass, and SMP includes UAP and BAP.

3. TSS and SMP are transferred to a physical resistance-in-series model and cause membrane fouling.

5 shows the relationship between biological and physical mechanisms. The relationship was confirmed through experiments on laboratory and pilot-scale membranes. Biological substances such as SMP and TSS accumulate on the membrane surface and pores during the filtration process to form a foul cake layer, and the accumulated contaminants are separated during the washing process such as relaxation and backwashing. Therefore, during operation, hydraulic physical cleaning is used to reduce the fouling effect and increase the membrane capacity. Thus, the interface between the two models uses biological SMP and TSS concentrations and diffused air. The strong interactions indicate the importance of considering both biological and physical processes when formulating an optimal strategy for operating an MBR with minimal cost and environmental impact.

Optimization must take into account the complex bio-physical interactions involved in operating an MBR plant efficiently and economically. As shown in FIG. 1(c), the expected influent information for the next 24 hours was used as an input for the Harmony Search (HS)-based DOO algorithm. The HS algorithm optimized the biological set-point (aerated DO concentration) and the physical set-point (permeability) using a dual objective function that took into account complex interactions. Aeration and permeability are important factors for membrane fouling because they affect the sludge fraction of activated sludge floating and fouling, respectively.

First, DOO optimized the biological aeration process to reduce aeration energy (AE) while maintaining wastewater quality. The integrative model then interfaces the ASM-SMP and RIS models and transfers the optimized biological operating conditions to the physical process. Finally, permeability was manipulated through optimization to reduce fouling and pumping energy (PE) while maintaining water productivity.

The DOO system optimized the biological and physical processes of the pilot and full-scale MBR plants over 65 days (28 October to 31 December) and 43 days (24 August to 5 October), respectively. This is because chemical cleaning took place before and after the optimization period, during which bio-physical contamination progression was observed. Optimization results were implemented for both biological and physical processes using a proportional integral (PI) controller tuned by the integral of a time-weighted absolute error adjustment rule. We then evaluated the performance of the optimized system compared to a passive system and an ammonia-based aeration control (ABAC) system.

Referring to the biological objective function, the biological optimization shown in Fig. 1(c) suggests the following 24-h DO concentrations for aerobic and membrane reactors. The objective function for biological optimization is expressed by Equations 4 to 6 below. This minimizes AE (aeration energy) and effluent quality index (EQI).

[Equation 4]

[Equation 5]

[Equation 6]

where J _biological is the biological objective function, and w ₁ and w ₂ are |AEt| and |EQIt|, |AEt| is the minimum normalized aeration energy, |EQIt|, using the measured maximum and minimum AEt (in equation 7 below). is the minimum-maximum normalized wastewater quality index (Equation (8)) taking into account the past maximum and minimum EQIt, and DO _aero,t and DO _mem,t are the manipulated DO concentrations for the aerobic and membrane reactors at time t, respectively.

Equation 5 is the boundary condition for the DO setpoint for the two reactors. A minimum DO value of 0.5 mg/L was chosen to maintain aerobic conditions for nitrification and operate the blower in the operating range. This is because if the DO setting value is low, fan switching may occur frequently and the DO concentration may fluctuate. A minimum DO value can prevent aeration energy dissipation and aid in effective nitridation. The maximum value was selected to reduce the AE of the blower considering the passive system. Equation 6 is the effluent standard set by the Korean Ministry of Environment and is a constraint function of the objective function that determines the optimal DO setting value. AEt and EQIt are calculated by Equations 7 and 8, respectively.

[Equation 7]

[Equation 8]

where 1.8 is the transfer efficiency of 1.8 kg oxygen per kWh used, KLa is the saturated oxygen transfer coefficient, S is the oxygen saturation concentration (8 mg/L), V is the reactor volume, Qeff is the effluent flow rate, and w _pollutant is the pollutant in the pollutant unit is a weighting factor for conversion to effluent TSS (TSS _eff,t ), COD (COD _eff,t ), Kjeldahl nitrogen (Nkj _eff,t ), nitrite and nitrate (NO _eff,t ) or biological oxygen demand (BOD _eff,t ) has a value of 2, 1, 30, 10 or 2, respectively.

Next, describing the physical objective function through the membrane experiment, the physical optimization shown in FIG. 1(c) is an interface between the biological ASM-SMP and the physical RIS model. Compared to optimization of biological processes, studies on dynamic optimization of physical processes are rare. In an embodiment of the present invention, the new objective function is expressed as Equation 9 below.

For physical optimization, laboratory-scale membrane experiments are suggested. The experiment identified a relationship between fouling, PE (fouling energy) and productivity to change permeability. Reduced permeate improved pollution mitigation and pumping energy savings, but reduced water productivity.

This is consistent with the results of previous techniques evaluating physical operations using pilot and full-scale MBRs. Increased permeation accelerates the fouling process because the probability of contaminants coming into contact with the membrane increases with permeation. Through experimental results and previous research results, the physical objective function includes pollution, PE, and productivity. Optimization provides permeability of the membrane reactor over the next 24 hours. Permeance represents the permeation duration in one filtration cycle time including permeation and relaxation times. Relaxation removes contaminants from the membrane surface with diffused air.

[Equation 9]

[Equation 10]

[Equation 11]

where J _physical is the physical objective function and ratio is the manipulated transmission ratio. |PE _t | min-max normalized PE, |TMP _t | is the minimum normalized transmembrane pressure (TMP) calculated by RIS, |WP _t ^-1 | is the minimum-maximum normalized The water production (WP) amounts, w ₃ , w ₄ , and w ₅ , are |TMP _t |, |PE _t | and a weighting factor for maintaining the balance of |WP _t ^-1 |. Equation 10 is the boundary condition of transmittance; 0.6 was chosen to satisfy the amount of water supplied and 1 represents permeation only without membrane washing. PE _t was estimated using the permeate extraction of the membrane module as the effluent and using the 0.6 pump efficiency η in Equation 11. In our examples, TMP was used as the main indicator of contamination progression because it provides unambiguous information about the contamination mechanism. The reciprocal of WP (WP _t ^{- 1} ) maximizes permeate production in the objective function while minimizing pollution and operating energy consumption.

Next, Harmony Search (HS) optimization is described. HS is an optimization algorithm that finds a solution with fewer mathematical calculations and solves the optimization problem. The main parameters of HS are Harmony Memory (HM), Harmony Memory with Ratio (HMCR), Pitch Adjustment Ratio (PAR), Bandwidth (Bw) and Repetition.

All attempted solutions are stored in the algorithm's memory space, making HS a memory-based algorithm. It has advantages such as real-number coding, ease of application to various problems, and balanced exploration and utilization. To apply HM to optimize MBR, values of HM, HMCR, PAR, Bw, and iteration must be determined. The recommendation is as shown in Equation 12 below.

[Equation 12]

The choice of parameter values is sensitive and depends on the optimization problem. Therefore, various parameters can be selected depending on the target system. In addition, five parameters were determined through random search, which showed stronger performance than grid search.

In our example, we showed the best optimization strategy by selecting 100, 0.9, 0.1, 1 and 200 for HM, HMCR, PAR, Bw and iterations, respectively. The HS algorithm iteratively developed an optimal solution for the MBR and iterated until the stopping criterion (iteration) was met according to the optimization procedure.

In the embodiment of the present invention, the purpose is to propose a predictive optimization system for pilot and full-scale MBR plants through process simulation. The main problem with the MBR plant was fixing a high setpoint without taking into account the influent and the rapid fouling development of the membrane. Both of these require excessive aeration and frequent membrane replacement and can directly increase operating costs. DOO was required for both biological and physical processes to save energy considering aeration, maintain wastewater quality and reduce fouling intensity by controlling permeability.

Hereinafter, performance evaluation results of the optimization system according to an embodiment of the present invention will be described.

The total number of decomposed intrinsic mode functions (IMF) was 12, 12, 11 and 11 for influent flow, COD, TSS and TN, respectively. This indicates an irregular trend in influent COD and flow, with a relatively higher standard deviation compared to TSS and TN.

Decomposition creates a hidden sub-layer containing distinct patterns, and the generated IMF is used as an input to a predictive model to predict future influent information. This decomposition method can adaptively extract unique patterns in each abnormal influent condition by reflecting the local characteristics of influent data. Therefore, the EMD method produces different numbers of IMFs depending on the influent flow rate and pollutant concentration.

Table 1 shows the structure of the prediction model for influent data after the next 24 hours. The number of coefficients in the MLR model is the number of decomposed IMFs. The smoothing coefficient of the ES model was 0.6, which gave the best prediction between 0 and 1 values. 5, 1, and 5 were used as p, d, and q of the ARIMA model, respectively. The structure and hyperparameters of the DL predictive model were determined according to a rule of thumb.

[Table 1]

Structure of Statistical and Deep Learning Models for Predicting Hourly Influent Pollutant and Flow Rates

The number of neurons in the hidden layer must be less than the number of input data and greater than the number of output data. Thus, MLP selected 256-64-16 neurons in each layer, while GRU and BiGRU used 64 memory cells. A DL model is trained to minimize the prediction error based on the mean squared error. Adam optimizer was used to optimize the weights of neurons and memory cells. In addition, in the embodiment of the present invention, dropout for MLP and iterative dropout for GRU and BiGRU are used. Dropout is a tool to avoid overfitting problems, which increase prediction errors by over-calculating parameters.

[Table 2]

Table 2 compares the prediction performance of the statistical model and the deep learning model for predicting influent after 24 hours. RMSE, MAE and MAPE were used to evaluate predictive performance to prevent biased evaluation.

Prediction errors such as RMSE and MAE gradually decreased in the order of flow rate, COD, TSS, and TN. The magnitude of the measured flow velocity was the highest, and the magnitude of TN was relatively small. Because it was calculated, it had a single digit (see Equation 3). Because ES and ARIMA reflected the abnormal trend of the time series influent data, the RMSE, MAE, and MAPE values were lower than those of the MLR model.

Compared to statistical models, deep learning models such as MLP, GRU, and BiGRU showed better predictive performance as shown in Table 2. This is because deep learning models have hidden high-level invariant structures and the ability to interpret inherent trends in data. Also, BiGRU with low RMSE, MAE, and MAPE was more suitable for influent prediction than other prediction models. This indicates that BiGRU is superior in predicting the influent based on the update reset gate and the bi-directional data input structure, even though the influent data is abnormal.

Figure 7(a) provides representative results of predicted influent TN over sequential time in the test data. A wide variety of incoming pollutants affect the microbial activity of wastewater treatment processes. Therefore, it is important to accurately predict the dynamic influent to determine the appropriate operating conditions while accommodating the various trends.

The prediction performance of BiGRU compared to other prediction models is shown in Fig. 7(a). BiGRU predicted fluctuations in influent TN and captured unsteady trends. In contrast, ES and ARIMA outperformed MLR, but predicted influent TN with a time delay, as shown in Table 2. The MLP followed the trend of the influent TN, but as can be seen from December 10 to 14, a gap occurred between the predicted and measured TN. GRU produced predictive results similar to those produced by BiGRU. However, the GRU had a gap between predicted and measured TN when influent TN was high (November 10 to 14).

Residual errors are shown in Fig. 6(b) by calculating the residual error between the predicted TN and the measured TN for each different prediction model. BiGRU showed a smaller residual error when contrasted with other prediction results. Data decomposition reduced irrelevant information in the data and improved predictive performance. BiGRU then showed acceptable predictive performance with low RMSE, MAE and MAPE. It is inferred that this reflects an important anomaly tendency for BiGRU to outperform other predictive models while establishing the operating strategy of MBR. Therefore, we proposed an optimal operating strategy for aerobic and membrane reactors using BiGRU-based predictions for flow rate, COD, TSS and TN.

7(a) and 7(b) show optimized DO settings for aerobic and membrane reactors according to expected influent. The optimized DO settings were compared with those of the manual system (manual) and ABAC systems. Optimized DO settings for high load influent are shown in Fig. 7(a). The overall DO setpoint was reduced to less than 2mg/L, the setpoint of the manual system. In addition, the optimized DO setpoint is lower than that of the ABAC system. This is because DOO has proposed DO setpoints to improve wastewater quality and reduce energy consumption. The predicted optimal trajectories exhibited distinct divergent trends consistent with those of the influent contaminants. Housework increased the influent load from 10:00 AM to 1:00 AM. Therefore, high DO setpoints greater than 2 mg/L were mainly observed during the day. This means that the system according to the embodiment of the present invention reflects the change in pollutant load and appropriately manipulates the DO setting value for microorganisms corresponding to the predicted influent information.

The performance of biological optimization at low load is shown in Fig. 7(b). Compared to the high inlet nitrogen load, the low loading conditions required lower DO concentrations for nitrification. For the aerobic reactor, where nitrification is the main process, the average values of the DO setpoints were 1.34 and 1.26 for the high and low load conditions over 5 days, respectively. Also, the DO setpoint was generally lower than the manual setpoint and showed a similar trend to the influent load. On the other hand, ABAC regulated a relatively high DO setting value because it presented a DO setting value according to the wastewater ammonia concentration without considering other wastewater pollutants and energy consumption. Although ABAC reflects various trends in the influent and reduces the DO setpoint during lower loading times, it may be less effective in reducing energy consumption. 7(c) shows a bar graph of optimized DO setpoints for aerobic and membrane reactors in units of time. Hourly DO setpoints were mainly optimized for different inflow trends. Therefore, the DOO system reduced the operating set point based on influent conditions.

Optimal DO setpoints on a daily basis for membrane reactors, including biological operating information, and corresponding SMP and TSS concentrations were linked to physical optimization. The scouring aeration rate for the membrane was estimated by the DO set point. Biological SMP and TSS concentrations according to operating time are shown in FIG. 8 (a). FIG. 8(b) shows the relaxation duration of the transmission ratio conversion into the transmittance and the relaxation duration. The relaxation time for the entire 65 days of operation continued to increase to 3 minutes at the end of the operation time because biological contaminants were physically attached to the membrane surface by filtration.

The initiation of DOO increased permeation duration while decreasing relaxation as shown in Fig. 8(c). Due to the low accumulation of contaminants in the membrane, a long relaxation time was not required at the beginning of membrane operation. Fig. 8(d) shows that when the TSS and SMP concentrations increased on Dec. 5, the relaxation time increased, impeding the adhesion of contaminants to the membrane surface (Fig. 8(a)). The attached biological contaminants proceeded to foul and increase the TMP value. Therefore, in order to prevent fouling intensity, the relaxation time was increased according to the fouling progress.

Figures 9(a) to (d) show the operating results of the DOO system for biological and physical integration processes compared to the manual and ABAC systems over a period of 5 days. Figure 9(a) shows the consumption AE and EQI corresponding to the adjusted DO concentration in the two reactors. Results at high influent load conditions are depicted representatively as it is important to evaluate the proposed optimization system in the trap of AE reduction without considering effluent quality.

The passive system, which maintained the DO concentration at 2 mg/L, consumed 7718 kWh of power and produced 6156 kg of EQI, while the ABAC system reduced AE and EQI to 7069 kWh and 5776 kg, respectively. The optimization system according to an embodiment of the present invention reduced energy and EQI to 6947 kWh and 5715 kg, respectively, under high influent load conditions. Therefore, the system according to the embodiment of the present invention improved power consumption and effluent quality by 9.99% and 7.16%, respectively.

The improvement in power consumption was greater than the improvement in effluent quality. This is because the case study plant complied with runoff limits. However, the fixed DO setpoint of the manual system increased the AE and operating costs of the plant. Therefore, AE minimization through reduction of the DO set point was more important than EQI minimization. Figure 9(b) shows the performance of the DOO system in terms of exhaust nitrogen compounds. Although the system according to an embodiment of the present invention reduced the DO set point to reduce AE, effluents containing nitrogenous compounds were also considered. Total NO effluent is less than manual and ABAC systems. However, the Kjeldahl nitrogen (Nkj) concentration increased somewhat. This was due to the weighting for each pollutant associated with conversion to pollutant units, which were 30 for Nkj and 10 for NO. Therefore, DOO effectively reduced effluent NO rather than Nkj concentration by reducing the DO setpoint and excessive nitrification process.

Figures 10(a)-(c) compare the pollution mitigation performance of the DOO and passive operating systems, and the individual resistances during MBR execution using each system are shown in Figures 10(a) and (b), respectively. The stable sludge cake was the main resistant component causing membrane fouling in the MBR process. Stomatal fouling resistance, contributing between 5% and 52% of fouling, increased continuously over time. Therefore, it is necessary to use operating strategies to mitigate fouling by reducing the formation of stable sludge cakes and to increase the operating time of membranes. The DOO system increased the relaxation period in the filtration cycle. The stable sludge cake resistance gradually decreased and the overall resistance decreased. As the relaxation time increased, the stable sludge cake detached slightly from the membrane surface.

The overall resistance reduction of the DOO system contributed to the reduction of contamination intensity compared to the passive system. The passive system, which maintained permeation for 9 min and relaxation for 1 min, reached 85 kPa, whereas the DOO system reached 61 kPa at the end of plant operation, as shown in Fig. 10(c). A rule-based operating strategy that continuously increases the relaxation time may be one option to reduce contamination progression. However, since this strategy is built using limited causes, it is difficult to find a dynamic optimal solution in nonlinear and unsteady conditions. In this study, it was assumed that chemical film cleaning was required when the TMP reached 85 kPa, which was determined by the process operator.

The membrane of the DOO system can operate 20 days longer than the manually operated system, as shown in FIG. 10(c). Since permeate production is proportional to membrane filtration time, 5699 m3 of permeate could be produced through the optimized system. Results indicate that DOO updating the optimal permeance rate outperforms manual operation in terms of reducing membrane fouling and increasing permeation productivity.

In addition, in the present invention, the progress of fouling was evaluated using three models of complete blocking, intermediate blocking, and cake filtration. Figure 11 shows the change of the contamination mechanism model according to the operating time of the manual and DOO system. Compared to the passive system, the intermediate blocking mechanism was initiated earlier. Because DOO reduced relaxation and increased permeation time at the beginning of the operation, the accumulation of contaminants on the membrane surface and deposited particles was accelerated. Therefore, the transition from complete blocking to medium blocking was carried out from November 6th to November 3rd. In terms of cake filtration, which is the last fouling mechanism and increases fouling by forming a layer of cake, the DOO system delayed the change from intermediate interception to cake filtration to December 6th. The increased relaxation time separated contaminants from the membrane surface and maintained the intermediate blockade longer. These results show that DOO can effectively mitigate the fouling progression by controlling the transmittance.

The comprehensive performance of the DOO system is summarized in Table 3. The passive system achieved a total AE over 65 operating days of 74,759 kWh. Considering the expected influent, the DOO system reduced AE by 10.38% and improved effluent water quality by 6.47%. These results demonstrate the benefits of DOO in the trade-off of environmental and economic improvements. The optimized system also mitigated membrane fouling by 39.34% and reduced PE by 25.54%. Although the WP decreased slightly due to the short permeation time, the amount of WP still met the system requirements. In addition, energy efficiency was evaluated by calculating energy consumption (EC) per WP, including AE and PE. The optimized system showed improved energy efficiency by reducing EC/WP by 4.16%. It can also be seen that the DOO system extends the operating time of the membrane by 20 days, consuming less energy while increasing the amount of WP during membrane operation. These results show that the utilization of the DOO system can be an effective solution for economically and environmentally responsible operation of MBR plants. In the experimental example of the present invention, the DOO system was additionally evaluated by applying a full-scale MBR plant to evaluate the validity and reliability of the present invention.

[Table 3]

Hereinafter, the evaluation of the application of a dual purpose optimization (DOO) system to a full-scale MBR plant will be described.

The DOO system was used in a full-scale MBR sewage treatment plant to evaluate its performance. The BiGRU-based hourly influent prediction model used to predict inflow after the next 24 hours showed superior prediction performance compared to other prediction models. Influent predictions were applied to the DOO system for both aerobic and membrane reactors.

A major drawback of the MBR plant was the high operating cost due to the stringent 4 mg/L and 7 mg/L DO standards for the aerobic and membrane reactors, respectively. AE accounted for about 25% of the high operating cost of the MBR plant because it had to maintain water treatment stability and reduce membrane fouling with high scouring aeration intensity. Therefore, it was necessary to optimize AE while maintaining wastewater quality in order to efficiently reduce operating costs. DOO has different boundary conditions, from 3 mg DO/L to 5 mg DO/L for aerobic reactors and from 6 mg DO/L to 8 mg DO/L for membrane reactors, to ensure the stability of both biological and physical processes. used

12(a) and (b) show improved biological processes using the DOO system. The optimized DO concentration was lower than the two DO settings for the manual and ABAC systems, as shown in Fig. 12(a). This indicates that the optimization system adjusted the aeration intensity according to the influent load. ABAC showed weak performance in the DO setpoint proposal and failed to reflect the dynamic characteristics of the influent (Fig. 12(a)). ABAC was unable to find an optimal solution for the controller settings due to the very complex and non-linear features of the MBR. Therefore, it was necessary to implement a predictive optimization system to improve the economic and environmental operation of the MBR plant. The reduced DO setpoint of the optimized system is directly related to the AE consumed. Compared to the passive and ABAC systems, DOO effectively reduced AE while maintaining EQI at a similar value, as shown in Fig. 12(b). The operating energy by aeration decreased by 15.92%, while the EQI slightly decreased from 1.854 × 10 ⁶ kg to 1.843 × 10 ⁶ kg over the entire operating period. This is because optimized aeration reduced effluent NO while mitigating excessive nitrification and increasing effluent Nkj. Although the effluent NO decreased, the lower weight of the effluent NO affected the slightly reduced EQI. Biological aeration affects the physical operation in terms of membrane fouling, and a decrease in aeration intensity can weaken the scouring effect of the membrane surface, so careful optimization of the physical process, including permeation and backwashing of the membrane, was required.

The optimization system adjusted the permeability of the membrane operation from 0.87 to 1 based on the adjusted DO setpoint. A value of 0.87 represents 13 minutes permeate and 2 minutes backwash, and 1 represents a full permeate operation without backwash. The performance of the optimized physical process is shown in Figures 13(a) to (c). The optimized backwashing time was relatively short at the beginning of operation, but gradually increased to more than 1 minute as shown in FIG. 13(a). Longer backwashing effectively reduced sludge cake formation in the membrane and reduced stable sludge cake resistance. The manual system with 0.5 min of backwash showed higher values for stable sludge cake resistance as shown in Fig. 13(b). Increased backwashing reduced overall resistance and mitigated fouling. TMP decreased from 43 kPa to 31 kPa on October 5, when the operator stopped working on the membrane and started cleaning the chemicals. As a result, the optimized system was able to increase the operating time of the membrane up to 20 days until the TMP reached 43 kPa, as shown in FIG. 13(c). Additionally, the MBR plant was able to produce 38% more water compared to the manual system (18,117 m ³ to 25,097 m ³ ) due to increased uptime. This confirms that the DOO system can reduce fouling and increase permeate production to determine the optimal backwash duration.

Table 4 summarizes MBR plant performance in manual and optimized systems. The energy consumption values for aeration and pumping obtained using the DOO system are 15.92% and 16.51% lower, respectively. Reduction in aeration intensity and membrane fouling reduced the operating energy requirements. Effluent water quality remained at a similar level, but aeration decreased rapidly. It can be seen that the volume of permeate obtained using DOO is less than that obtained using the manual system. However, the optimal backwash period mitigated membrane fouling and increased membrane life. A full-scale MBR plant with a DOO system can operate longer than a manual system. Also, the energy efficiency of the optimized system, EC per permeate, improved by 12.02%. As a result, the optimization system according to the embodiment of the present invention reduced EC including AE and PE from 748,137 kWh to 628,180 kWh for 43 days. Thus, the DOO system can save 2789 kWh per day and reduce energy consumption by 1.0182 × 10 ⁶ kWh compared to the year of operation. At a price of $0.075/kWh provided by KEPCO, the proposed system could save $76,361 per year. These results indicate that the DOO system can contribute to energy savings, maintain water quality and permeate production, and reduce membrane fouling to effectively balance economic and environmental demands.

[Table 4]

A DOO system based on the integrated biological and physical model according to the present invention was developed to improve AE efficiency, maintain wastewater quality and reduce pollution. A deep learning prediction model was used to address the limited performance of existing prediction models and improve the DOO system with respect to influent parameter uncertainties. BiGRU (bidirectional gated circulating unit)-based hourly influent prediction model showed the best performance among statistical and deep learning models. In the experimental example of the present invention, the DOO system, which integrates biological and physical processes based on predicted data, was applied to pilot and MBR plants through precise simulation, and its performance was compared with manual and ABAC operating systems. The optimized system achieved improved operating trajectories in both aerobic and membrane reactors.

The optimization system according to an embodiment of the present invention improved energy efficiency by 4% and 12%, respectively, in the pilot and MBR plants while maintaining effluent quality. The optimized system also adjusted the duration of physical cleaning to account for contamination, laboratory-scale experiments to exchange PE and WP. The optimized physical cleaning time reduced PE by up to 17%, reduced fouling progression by 27%, and extended membrane life by more than 17 days. In addition, when the optimization system according to the embodiment of the present invention is applied to a full-scale MBR plant, an annual operating cost of 76,361 dollars can be saved. A major contribution of the present invention is the development of optimized biological and physical processes for MBR plant operators. DOO systems can also take into account the complex interrelationships between dynamic influent properties and biological and physical processes. According to an embodiment of the present invention, it can be extended to an optimization system to improve the economic and environmental operation of an actual MBR plant while controlling several operating variables such as activated sludge flow rate and nitrate internal cycle waste.

In addition, the device and method described above are not limited to the configuration and method of the above-described embodiments, but all or part of each embodiment is selectively combined so that various modifications can be made. may be configured.

Claims (13)

An influent prediction model generating unit that predicts the characteristics of influent after a specific time; and
A control unit for controlling the sewage treatment system based on the influent characteristics predicted by the influent prediction model generation unit; a membrane bioreactor optimization system based on an influent artificial intelligence prediction model comprising a.
According to claim 1,
The influent prediction model generator utilizes a bi-directional gate circulation unit,
The bidirectional gate unit generates an influent prediction model per hour and predicts the influent flow rate, COD, TSS, and TN value characteristics after a specific time. An influent artificial intelligence prediction model-based membrane bioreactor optimization system.
According to claim 2,
The hourly influent prediction model is an influent artificial intelligence prediction model-based membrane bioreactor optimization system, characterized in that for continuously predicting the influent characteristics after a specific time using previously input data.
According to claim 3,
The control unit is an influent artificial intelligence prediction model-based membrane bioreactor optimization system, characterized in that for controlling the membrane bioreactor plant.
According to claim 4,
The control unit is an influent artificial intelligence prediction model-based membrane bioreactor optimization system, characterized in that for optimizing the control factor of the plant through harmony search.
According to claim 5,
The control unit optimizes the influent artificial intelligence prediction model-based membrane bioreactor optimization system, characterized in that for optimizing the aeration intensity of the aerobic tank and the filtration and washing period of the membrane separation tank.
According to claim 6,
Further comprising an integrated MBR modeling unit for modeling the membrane bioreactor process by integrating the biological model and the physical model,
The biological model uses ASM-SMP, and the physical model uses a RIS model to model the MBR process.
According to claim 7,
The control unit optimizes the aeration intensity and the effluent quality index based on the biological model, and optimizes membrane fouling and membrane pumping energy while maintaining water productivity based on the biological model and the physical model. A model-based membrane bioreactor optimization system.
As a membrane bioreactor optimization method,
Predicting the characteristics of influent water after a specific time by the influent prediction model generating unit;
inputting predicted influent characteristics into an integrated MBR modeling unit that models a membrane bioreactor process by integrating a biological model and a physical model; and
A control unit optimizing and controlling the membrane bioreactor based on the biological model and the physical model of the integrated MBR modeling unit; a membrane bioreactor optimization method based on an influent artificial intelligence prediction model, comprising:
According to claim 9,
The influent prediction model generator utilizes a bi-directional gate circulation unit,
The bi-directional gate unit generates an hourly influent prediction model, predicts the influent flow rate, COD, TSS, and TN value characteristics after a specific time,
The hourly influent prediction model is an influent artificial intelligence prediction model-based membrane bioreactor optimization method, characterized in that for continuously predicting the influent characteristics after a specific time using previously input data.
According to claim 10,
The control unit optimizes the control factor of the membrane bioreactor plant through harmony search, and the control factor optimizes the aeration intensity of the aerobic tank and the filtration and washing period of the membrane separation tank. Bioreactor Optimization Method.
According to claim 11,
The biological model uses ASM-SMP, and the physical model uses a RIS model to model the membrane bioreactor process.
According to claim 12,
The control unit optimizes the aeration intensity and the effluent quality index based on the biological model, and optimizes membrane fouling and membrane pumping energy while maintaining water productivity based on the biological model and the physical model. Model-based membrane bioreactor optimization method.

Images