Abstract

The objective of this research study was the development of an intelligent system based on artificial neural networks for water distribution networks that operate with parallel pumps. The purpose of the system is to automate the process and to define the operating state of the electric motors (on, off or with partial rotation speed). The intelligent system developed is generic, which allows the application of its control structure in similar processes, and it was applied in an experimental setup that simulates a real water supply system. The performance of the network was tested experimentally under different operating conditions, including in the presence of disturbances. The settling time was, in all experiments, less than 30 seconds, the tests did not show overshoot and the maximum error was 2.9%. Results showed excellent performance in terms of pressure regulation, and it is hoped that the controller can be successfully implemented in real water distribution systems, in order to reduce water and electricity consumption, decrease maintenance costs and increase the reliability of operating procedures.

INTRODUCTION

The importance of the interdependence between water and energy is widely recognized. The energy security of the world is highly dependent on water availability, as almost all energy production technologies such as nuclear, thermoelectric and hydropower plants demand vast amounts of water (Nair et al. 2014). On the other hand, water supply services consume large amounts of energy in the treatment and transport of water from the sources to the consumers.

Sanitation companies consume 2 to 10% of the total electricity used in one country (Pelli & Hitz 2000). In the world, companies spend about $14 billion a year to pump water (Sensus 2012). James et al. (2002) estimated that Latin American countries spend $1 to $1.5 billion a year just to pump water. In Brazil, according to the National Energy Conservation Program for the Sanitation Sector of Eletrobras, the basic sanitation sector consumes between 2 and 3% of the country's electricity, 90% of which is consumed by pumping systems.

Most of the time, energy costs of pumping systems exceed the investment costs of the installations over the lifespan of projects, representing one of the main expenditures of water companies (Vieira & Ramos 2008; Costa et al. 2016). In developing countries, electricity is usually the highest cost associated with water supply (Barry 2007).

Pumping systems represent almost 20% of the energy consumed by electric motors and 25 to 50% of the total electric power needed in some industries (Europump & The Hydraulic Institute 2004). Even with the progress in conservation practices, concerns about energy use stimulates a constant search for more ‘intelligent’ technological solutions that result in more efficient and sustainable water pumping systems (Lewis & Hendrix 2012). Several studies in Europe and the United States have indicated that the industrial sector can potentially reduce energy consumption by 30 to 50% in pumping systems (Hovstadius 2007).

The optimization of pumping system operations results in significant economic impacts, and can achieve savings of millions of dollars per year in large systems. According to James et al. (2002), energy consumption in most water distribution networks (WDNs) around the world could be reduced by at least 25%, through the implementation of measures aimed at increasing energy and water efficiency; this would be equivalent to the energy consumed in all of Thailand.

In recent years, Brazilian sanitation companies and the scientific community have been trying to follow the technological development of other industrial sectors. The optimization problems associated with the operational control of WDN are complex because of their large-scale, multiple-input and multiple-output nature, as well as the various sources of additive and possibly parametric uncertainty found in the WDN (Sun et al. 2016).

There are a large number of scientific studies that deal with the optimization of pumping systems. Normally, these research studies focus on the definition of the operational routines of pumps, such as the studies by Brion & Mays (1991) and Martinez et al. (2007).

Modeling approaches for pumping systems have been based predominantly on physics and mathematical programming. The assumptions that have been made limit the applicability of these models in industry. These assumptions have neglected the dynamic characteristics of the pumping process and impede the applicability of such models in practice (Kusiak et al. 2013). Therefore, it is important to develop methods and tools that are applicable to the control of real systems, which allow – in addition to real-time monitoring – control of the flows and pressures of the networks.

In general, the pressure in the WDN can be adjusted by automatic control of pressure control valves (CVs) and variable speed pumps, in response to real-time pressure measurements at various points of the system (Page et al. 2017a, 2017b). According to Ghorbanian et al. (2017), systems with water storage and uniform pressures tend to have higher leakage rates, greater energy usage, and higher greenhouse gas (GHG) emissions than systems relying on direct pumping. The generalization of this statement is perhaps predictable but has profound implications: the higher the delivery pressure, the greater will be the amount of water loss and dissipated energy. Celi et al. (2017) presented an insight into the selection of the most suitable configuration for closed distribution systems (i.e., systems with no storage capacity). The analysis was developed considering wide ranges of operational scenarios. Results have shown that the costs are lower when the pumps are operating closer to their best efficiency point. Additional benefits in terms of costs may be obtained as a result of the introduction of variable speed drive in at least one of the station pumps.

Several different studies reported in the literature show the relation between pressure control and the decrease of energy consumption and water (Halkijevic et al. 2013; Stokes et al. 2013; Covelli et al. 2015; Rezaei et al. 2015; Covelli et al. 2016; Dai & Li 2016; Ghorbanian et al. 2017). However, due to a lack of systematic analysis of the results obtained in practical cases, progress has not always been reflected in practical actions (Vicente et al. 2015). Vicente et al. (2015) provide a comprehensive analysis of the most innovative issues related to pressure management. The methodology is based on a case-study comparison of qualitative concepts that involves published work from 140 sources.

Control can be defined as a closed feedback loop whereby the difference between a measured process variable and the desired set point is calculated. This difference is then minimized over time by the adjustment of the process (Page et al. 2017b). Page et al. (2017a) developed a proportional controller that does not require the flow in the pump to be known in order to improve the performance of the system. The parameter-less controller developed performs either better or worse than other controllers shown by the researchers.

Madoński et al. (2014) have analyzed the use of an active disturbance rejection control structure in a pressure control system. The adaptability and robustness of the proposed control scheme enhance the performance of a conventional PID-based output feedback control loop.

The paper investigates the approach with a set of experiments conducted on a laboratory-scale model of a water supply system. The proposed approach is also compared with results obtained for the sole conventional industrial PID controller. PID controllers are easily implemented in real systems, but their controlling ability is expected to be worse than that of the smart controllers. The conventional controllers are tuned for specific WDN conditions (e.g., water demand), so that the controller normally does not offers a satisfactory performance for different conditions.

The objective of this research study was to develop an intelligent control system based on artificial neural networks (ANNs), for pressure control of WDNs supplied by pumps working in parallel with variable frequency drive motors. ANNs are sets of computational methods inspired by the human brain (Sutariya et al. 2013). They are being widely used to solve problems in all areas of society, such as agriculture (Sharifzadeh et al. 2014); chemistry (Dil et al. 2016); energy generation (Salam et al. 2016); environment (Aravind et al. 2016); hydrology (Astray et al. 2016); information technology (Dahiya & Gill 2016); sanitation (Picos-Benítez et al. 2017); and water supply (Gagliardi et al. 2017).

EXPERIMENTAL SETUP

The experimental setup of water distribution (Figure 1) consists of a reservoir, two pumps (PUMP), two variable-frequency drives, a CV, five piezoresistive pressure transducers, two electromagnetic flowmeters (FT), a power source in addition to electrical and electronic installations, and a hydraulic network.

Figure 1

Experimental setup.

Figure 1

Experimental setup.

The adopted motor–pump sets are composed of a three phase 3 HP induction motor and a pump of model CAM-W10, which is manufactured by the Dancor company. The pumps are centrifuged with a maximum suction head of 6 m, maximum head of 45 m, maximum flow of 12 m3/h (nominal characteristics). Figure 2 shows the pump flow-head-power-NPSH curves.

Figure 2

Characteristic curves of the CAM-W10 pump.

Figure 2

Characteristic curves of the CAM-W10 pump.

For the control of the described experimental setup, supervisory software was developed at LabVIEW®. The management program adopts a concept of graphic programming language developed by National Instruments, called Language G. Applications were developed to group tasks together, such as the acquisition of data, analysis and logical operations, as well as online data monitoring. The language has a data stream whose nodes (operations or functions) operate on the data as soon as they are available, which is considered more efficient for process control than the traditional programming through code lines. The graphic distribution of data ‘flow’ is carried out through the links, which connect the output of one node to the input of another node. The supervisory software allowed the acquisition and saving of test data, the alteration of the operating conditions of the experimental setup (e.g., activation or not of actuators) and the adjustment of the neural parameters of the controller.

The supervisory system contains eight inputs and three outputs (analog). The analog inputs receive the signal from five pressure transducers, two flow meters and the feedback of the valve position. The analog outputs are used to vary the valve position and motor's frequency/voltage. The valve CV is responsible for the alterations on the experimental setup operation conditions. The ANN controller has an objective to keep the pressure on the pumps' output (measured by the transducer indicated in the Figure 1 as PT) near to the set point in all experiments. The pressure is controlled by the variable-frequency drives, which modify the pumps' rotational speed. The flowchart of the control is shown in Figure 3.

Figure 3

Structure of the control system.

Figure 3

Structure of the control system.

In order to evaluate the energy efficiency of the experimental setup, with and without the controller, power was measured on line, and the specific consumption of electricity (SCE) was calculated. This indicator is widely used in the literature, since it allows considering the impact of efficiency actions on the quality of the systems (Bezerra et al. 2015). It is defined as the ratio between the energy consumption (kWh) of the pumping system and the volume effectively pumped (m3) at a given time.

ANN MODELING

Due to the complexity of the operation of water supply systems and the extensive research in the area of control systems using ANNs, this was the type of controller adopted to automate the experimental setup in this study. This allowed the development of the controller without prior knowledge of the mathematical model of the plant. The controller acts on the definition and control of the actions related to the voltage supply frequency (speed of rotation) of the pumping system and the determination of how many pumps must operate simultaneously throughout the experiments.

The ANN adopted has multi-layer feedforward architecture – Multi-layer Perceptron (MLP). The learning method adopted in the research is based on the Error backpropagation methodology, which includes dynamic learning.

The recurring configuration allows ANN to retrieve past responses from feedback of the signals produced in previous instants. The architecture adopted in this research is known as Elman network, where the outputs are compared to the inputs and backpropagation of error is used to incrementally adjust connection strengths. This type of ANN was selected as an estimator of plant behavior because it has good applicability in the area of control systems.

The procedure for developing the MLP network is summarized as follows:

  • 1.

    Definition of the hidden layer number.

  • 2.

    Definition of the number of neurons in the hidden layer.

  • 3.

    Generation of random data for the neural network weights.

  • 4.

    Execution of the operation algorithm with real-time learning.

  • 5.

    Verification of performance index. If this is greater than zero, a calculation methodology for updating weight matrices, called error backpropagation, is applied.

The proposed model is an ANN composed of the input layer, a hidden layer and the output layer. As the objective of MLP is to control pressure in WDNs, the input signals used were the control variable itself – pressure (PT) – and the variables that directly influence the controlled variable: pumped flow (FT), angle of the proportional valve (CV) which controls the demand, and the delay of the power supply frequency (FP1 and FP2). Thus, the MLP neural network architecture of this investigation is composed of five inputs and two outputs.

The bias Θ of ANN was 1 (one) and the activation function chosen was the hyperbolic tangent for the neurons of the hidden layer and linear function for the neurons of the input and output layers. These activation functions were attributed based on several research studies in the literature, which obtained a low learning time and good performance indexes.

The ANN proposal was trained from input and output data measured directly on the experimental setup. The training was carried out by activating ANN and operating the plant, that is, both the pumps and the CV underwent pre-defined alterations so that this data was assimilated by the learning algorithm. Basically, the output values of the neural network were compared, in real time, with the desired values. The synaptic weights were randomly assigned and adjusted by ANN using the learning algorithm.

The number of neurons in the hidden layer and the learning coefficient, α, were 10 and 0.1, respectively. Both values were experimentally determined by trial and error. The first phase of the error backpropagation algorithm is called Forward and is responsible for defining the output from the neural network. This phase involves the following steps.

Each neuron of the input layer has a linear activation function described by Equation (1). The input variable of the neurons of the hidden layer is a result of Equation (2), while the output value of each neuron of this layer is calculated by means of Equation (3).  
formula
(1)
 
formula
(2)
 
formula
(3)
where Wab is the matrix of weights between the input layer and the hidden layer; f(.) is the activation function of the hyperbolic tangent type defined by Equation (4), and θ is the bias, whose value was defined as 1.  
formula
(4)
The input of the output layer is a result of Equation (5), while the output signal of ANN is given by Equation (6).  
formula
(5)
 
formula
(6)
where Wbc is the matrix of synaptic weights that connect the neurons of the hidden layer to the neurons of the output layer.
At the end of this phase, there is a comparison between the output signal calculated by the neural network yc and the desired value yd, a result which generates the error value, E, given by:  
formula
(7)
The second step of the algorithm is the backpropagation. It is at this stage that the weight matrices Wbc and Wab are updated. The calculation method used in this step was the Descending Gradient. After the comparison between the response calculated by the neural network and the desired response, the weight matrix Wbc is updated using Equation (8).  
formula
(8)
where (Wbc)t+1 is the weight matrix to be calculated; (Wbc)t is the current weight matrix and α is the learning rate equal to 0.1.
The adjustment of the bias value is given by Equation (9). The weight matrix between the hidden layer and the input layer is adjusted through Equation (10).  
formula
(9)
 
formula
(10)
After the adjustment of the matrix of synaptic weights, the first iteration (period) ends. If, after adjustment, the output value diverges from the desired value, it will be entered as input in the next iteration. This will cause the weights to be readjusted, and this difference will be reduced in each epoch.

As previously mentioned, at the end of each iteration of the network, the power supply frequency information is stored and becomes the input in the next iteration. Thus, the learning method minimized the flux of errors of all processing elements. This global reduction of errors continuously modified the weights until ANN met the stopping criterion, that is, the training was considered complete when the neural network reached a certain level of performance.

RESULTS AND DISCUSSION

This section presents the results of the experiments carried out to validate the proposed control system. Because of the nature of the experimental data, the results were not submitted to statistical treatment. Therefore, the values presented correspond to those collected by the equipment (meters). The experiments were as follows:

  • ■ Experiment 1 – Closed-loop test with an input reference step equal to 15, 25 and 35 m for the experimental setup operating with the minimum flow.

  • ■ Experiment 2 – Closed-loop test with an input reference step equal to 10 m for the experimental setup operating with maximum flow.

  • ■ Experiment 3 – Open-loop test with the proportional valve varying its aperture in order to simulate the demand variation of water distribution systems.

  • ■ Experiment 4 – Closed-loop test with the proportional valve varying its aperture in order to simulate the demand variation of water distribution systems.

The tests were carried out with the pumping system starting with a low frequency and the CV valve changing the experimental setup operation conditions. With this procedure, the evaluation criteria of the control system – rising time, settling time, overshoot and steady-state error – were obtained. The rising time was considered as the time necessary to reach 100% of the reference value. Meanwhile, the settling time is the time elapsed from the application of an input reference step to the time at which the amplifier output has entered and remained within a specified error band. In this research, the settling time considered an accommodation range of ±3% of the reference signal amplitude.

Experiment 1

Tests of Experiment 1 demonstrate the performance of the ANN when the flow rate adduced by the experimental setup is minimal. In order to verify the efficiency of the developed controller, three tests were carried out with different values of set point and of disturbances. The reference values for the control variable (pressure) were 15, 25 and 35 m.

The ANN defined the frequency of the engines and determined how many engines should operate simultaneously throughout the experiments. The decision-making process of the algorithm was based firstly on the control of the pressures and secondly on the energy efficiency of the system (specific consumption of electricity – SCE).

The tests were carried out with the pumping system starting at the resting position (frequency equal to zero) and with a proportional valve aperture of 30°, which corresponds to the minimum demand. With this procedure, the evaluation criteria of the control system – rising time, settling time, overshoot and steady-state error – were obtained, and these are summarized in Table 1. The results demonstrate the robustness of the system, which presented satisfactory results for the various set-points.

Table 1

Performance of the control system – Experiment 1

Reference valuesRising time (s)Settling time (s)Overshoot (m)Steady-state error (%)
15 4.6 4.6 – 2.0 
25 4.2 4.2 – 2.0 
35 3.0 3.0 – 1.7 
Reference valuesRising time (s)Settling time (s)Overshoot (m)Steady-state error (%)
15 4.6 4.6 – 2.0 
25 4.2 4.2 – 2.0 
35 3.0 3.0 – 1.7 

Figure 4 presents the behavior of the control variable – pressure – of the Experiment 1. The control system presented satisfactory results. The graphs in the figure show that the responses of the control system did not show any overshoot and that the rising time from 0 to 100% of the reference signal amplitude was less than 9 seconds in all tests. It is worth underlining that by controlling the point with the lowest pressure (with respect to the set-point value), with the objective of bringing its pressure close to the set-point value itself, it is possible to ensure that pressures in all the system respect the minimum pressure requested.

Figure 4

Behavior of the pressure control variable – Experiment 1; comparison with set-point value.

Figure 4

Behavior of the pressure control variable – Experiment 1; comparison with set-point value.

In the first two tests of Experiment 1, only PUMP-01 was in operation, since the reference pressure imposed on the controller is relatively low. In the last test, the controller determined the activation of PUMP-02, since a single pump is not sufficient to reach the predetermined pressure of 35 m. The tests demonstrated that the higher the required pressure, the higher the activation frequency of the motors – as expected. In the steady-state system, the two output variables – frequencies – remained constant, considering the absence of disturbances in the system.

The maximum error of the tests was 2.9% (0.45 m), with no transients.

A control system is considered robust when it is able to maintain the stability of the system even when subjected to disturbances. To promote a disturbance in the experimental setup, the valve located downstream of the proportional valve was abruptly closed. Figure 5 shows the influence of the disturbance in the system and the fact that the controller maintained the same previous operating conditions, quickly restoring the pressure to the desired value.

Figure 5

ANN controller response to the disturbance – Experiment 1; comparison with set-point value.

Figure 5

ANN controller response to the disturbance – Experiment 1; comparison with set-point value.

Experiment 2

Tests of Experiment 2 were performed to verify the efficiency of the ANN in operation when the flow rate of the experimental setup was at its maximum. The set-point for the control variable (pressure) was 10 m. It was not possible to carry out tests with higher reference values due to pump limitations.

The test was carried out with the pumping system starting from the resting position (frequency equal to zero) and with the opening angle of the proportional valve equal to 90°, which corresponds to the maximum demand. Figure 6 shows the behavior of the control variable (pressure). The evaluation criteria of the control system were obtained, and these are summarized in Table 2.

Figure 6

Behavior of the pressure control variable – Experiment 2; comparison with set-point value.

Figure 6

Behavior of the pressure control variable – Experiment 2; comparison with set-point value.

Table 2

Performance of the control system – Experiment 2

Reference valuesRising time (s)Settling time (s)Overshoot (m)Steady-state error (%)
15 19 37 1.45 2.9 
Reference valuesRising time (s)Settling time (s)Overshoot (m)Steady-state error (%)
15 19 37 1.45 2.9 

In Experiment 4, PUMP-02 acts with a delay in relation to PUMP-01, because it is only activated when the ANN ‘notices’ that a single pump was not enough to meet the imposed condition. From instant 45 seconds, the pump rotation speed (frequency – output variable) remains constant.

Similarly to the tests of Experiment 1, in order to promote disturbances in the experimental setup, the valve localized downstream from the proportional valve was abruptly closed. Figure 7 shows the influence of the disturbance in the system and the fact that the controller maintained the same previous operating conditions, quickly restoring the pressure to the desired value.

Figure 7

ANN controller response to the disturbance – Experiment 2; comparison with set-point value.

Figure 7

ANN controller response to the disturbance – Experiment 2; comparison with set-point value.

Experiment 3

Experiment 3 was performed with the open loop system. The experiment simulated the operation of an urban water supply system. A virtual instrument was developed in LabVIEW® to provoke the demand variation through the remote operation of the proportional valve. The results of this experiment were compared with the closed loop system (Experiment 4). Initially, the opening angle of the valve was 30°, representing the minimum demand, and at every 30 seconds a variation of 2° in the angle was imposed until the maximum demand (90°) was reached. Then, the opening position of the valve returned to the position of minimum demand following the same methodology.

The hydraulic and electric variables of Experiment 3 are shown in Figure 8. As expected, the behavior of the pressure follows an inverse pattern to that of the flow, that is, the higher the flow, the lower the pressure. The average flow rate was 11.65 m3/h.

Figure 8

Behavior of the hydraulic and electric variables – Experiment 3.

Figure 8

Behavior of the hydraulic and electric variables – Experiment 3.

Experiment 4

Experiment 4 was performed in a closed loop with the same operating conditions as Experiment 3. An input reference step of 10 m was established to analyze the electrical energy consumption and the ANN response to the alteration in the operating conditions of the experimental setup.

The ANN defined the power supply frequency and determined how many pumps should operate simultaneously throughout the tests. The time of the test was 30 minutes and followed the same operating conditions (opening of the CV) as Experiment 3. The behavior of the hydraulic and electric variables is shown in Figure 9. As expected, the system's flow follows the same pattern as the opening curve of the CV. In order to evaluate the energy efficiency of the experimental setup, with and without the ANN controller (Experiments 3 and 4), power consumption was also measured and the specific consumption of energy (SCE) of Experiment 4 calculated.

Figure 9

Behavior of the hydraulic and electric variables – Experiment 4.

Figure 9

Behavior of the hydraulic and electric variables – Experiment 4.

The control system presented satisfactory results because, despite the variation of the flow, the controller acted to adjust the pressure to the reference value. The mean pressure in the experiment was 9.76 m, which indicates that the main objective of the controller was reached.

The tests showed the same pattern of behavior, with the second pump activated only when the demand reached the value of 13 m³/h. Similarly to the previous tests, the power supply frequency of the motors remained the same. This fact corroborates with the literature, which indicates that parallel pumps must function under the same operating conditions. PUMP-01 worked during the entire experiment period, oscillating rotation between 30 Hz (lower flow) and 58 Hz at the moment of peak flow. PUMP-02 was activated at instant 642 s and switched off at instant 1,358 s, with rotation variation between 70% and 97% of its nominal rotation. With the reduction of the demanded flow, the control system decided to disconnect one pump, letting the system operate from that moment on with a single pump.

Table 3 shows the values of the energy assessment parameters. Although the variable-frequency drive consumed about 5% of the total energy and caused a decrease in the efficiency of the pumping systems (Bezerra et al. 2015), the rotation control provided a reduction in the energy consumption of 0.52 kWh/m³ to 0.14 kWh/m³, which represents a reduction of 73.66%. Therefore, the results indicated a great improvement in the operational efficiency of the experimental setup.

Table 3

Performance measures of Experiments 3 and 4

 Experiment 3Experiment 4Difference (%)
Total average flow (m3/h) 11.65 8.19 29.7 
Electric power consumption (kWh) 2.07 0.70 66.2 
Specific average energy consumption (kWh/m30.5167 0.1361 73.7 
 Experiment 3Experiment 4Difference (%)
Total average flow (m3/h) 11.65 8.19 29.7 
Electric power consumption (kWh) 2.07 0.70 66.2 
Specific average energy consumption (kWh/m30.5167 0.1361 73.7 

CONCLUSIONS

In this paper, a new algorithm for pressure control in water distribution systems was proposed. This research study developed an intelligent control system based on an ANN with multi-layer feedforward architecture – a Multi-layer Perceptron (MLP) network. The ANN was proposed for the operation of a pumping system with parallel pumps with variable frequency drive motors. MLPs have already been used successfully in several applications in engineering; however, their application in the specific case of this study is unknown.

This intelligent system was applied to an experimental setup that simulates a real water supply system. The performance of ANN was tested experimentally under different operating conditions, including in the presence of disturbances. It provided a smooth start and satisfactorily maintained service pressure within previously established limits. The settling time in all experiments was less than 30 seconds and the maximum steady-state error was 2.9%. Therefore, the main conclusion of the research study is that the proposed ANN architecture is suitable for the control of pumps working in parallel, and that it has a robust performance for the range of operations of the experimental arrangement.

Scientific research investigations, such as the present study, have greater probability of success and viability in real systems, since they are flexible and allow for the necessary adaptations. It is expected that the proposed controller can be applied to other similar water distribution systems without the need for modeling the system to be controlled.

The application of the monitoring system allows a significant increase in the effectiveness, efficiency and reliability of the WDNs, providing lower operating costs. The operation is defined on the basis of the real demand, which makes the most efficient use of the available water while supporting conservation and environmental efforts. Besides that, the application of the proposed system enables an increase in hydraulic and energy efficiency of pumping system operations, energy saving, increased reliability, reduction in the volume of water consumed and water losses, increase of equipment useful life, minimization of hydraulic transients, increase in reliability of operating procedures, minimization of the need for system stops and elimination of jumps of production), elimination of the need for valves to start and stop the engine and modernization of plants.

ACKNOWLEDGEMENTS

This research has been partially supported by Centrais Elétricas Brasileiras SA (Eletrobras) and Brazilian National Council of Technological and Scientific Development (Conselho Nacional de Desenvolvimento Científico e Tecnológico, CNPq, Process 458994/2014-6).

REFERENCES

REFERENCES
Aravind
,
T. P. A.
,
Haneef
,
M. M.
,
Sulthan
,
M. M. T.
&
Solomon
,
M. I.
2016
Monitoring, analysis and modelling of ambient air quality status at Indoshell Mould Ltd, Sidco, Coimbatore using artificial neural network
.
Themed Section: Engineering and Technology
2
(
2
),
1172
1117
.
Astray
,
G.
,
Soto
,
B.
,
Lopez
,
D.
,
Iglesias
,
M. A.
&
Mejuto
,
J. C.
2016
Application of transit data analysis and artificial neural network in the prediction of discharge of Lor River, NW Spain
.
Water Science and Technology
73
(
7
),
1756
1767
.
Barry
,
J. A.
2007
Watergy: Energy and Water Efficiency in Municipal Water Supply and Wastewater Treatment
.
Cost-Effective Savings of Water and Energy. Alliance to Save Energy
,
Washington, DC
,
USA
.
Bezerra
,
S.
,
Silva
,
S.
,
Gomes
,
H.
&
Salvino
,
M.
2015
Energy savings in pumping systems: application of fuzzy system
.
Science & Engineering Journal
24
(
1
),
71
78
.
Brion
,
L. M.
&
Mays
,
L. W.
1991
Methodology for optimal operation of pumping stations in water distribution systems
.
Journal of Hydraulic Engineering
117
(
11
),
1551
1569
.
Celi
,
C. L.
,
Iglesias-Rey
,
P. L.
&
Solano
,
F. M.
2017
Energy optimization of supplied flows from multiple pumping stations in water distributions networks
.
Procedia Engineering
186
,
93
100
.
Costa
,
L. H. M. K.
,
Prata
,
B. A.
,
Ramos
,
H. M.
&
Castro
,
M. A. H.
2016
A branch-and-bound algorithm for optimal pump scheduling in water distribution networks
.
Water Resources Management
30
(
3
),
1037
1052
.
Covelli
,
C.
,
Cozzolino
,
L.
,
Cimorelli
,
L.
,
Della Morte
,
R.
&
Pianese
,
D.
2015
A model to simulate leakage through joints in water distribution systems
.
Water Science and Technology: Water Supply
15
(
4
),
852
863
.
Covelli
,
C.
,
Cozzolino
,
L.
,
Cimorelli
,
L.
,
Della Morte
,
R.
&
Pianese
,
D.
2016
Optimal location and setting of PRVs in WDS for leakage minimization
.
Water Resources Management
30
(
5
),
1803
1817
.
Dahiya
,
M.
&
Gill
,
S.
2016
Secured bluetooth authentication using artificial neural networks
.
International Journal of Research in Computer and Communication Technology
5
,
244
248
.
Dil
,
E. A.
,
Ghaedi
,
M.
,
Ghaedi
,
A. M.
,
Asfaram
,
A.
,
Goudarzi
,
A.
,
Hajati
,
S.
&
Gupta
,
V. K.
2016
Modeling of quaternary dyes adsorption onto ZnO-NR-AC artificial neural network: analysis by derivative spectrophotometry
.
Journal of Industrial and Engineering Chemistry
34
(
25
),
186
197
.
Europump & Hydraulic Institute
2004
Variable Speed Pumping: A Guide to Successful Applications
.
Cornwall
,
UK
.
Gagliardi
,
F.
,
Alvisi
,
S.
,
Franchini
,
M.
&
Guidorzi
,
M.
2017
A comparison between pattern based and neural network short-term water demand forecasting models
.
Water Science and Technology: Water Supply
17
(
5
),
1426
1435
.
Halkijevic
,
I.
,
Vukovic
,
Z.
&
Vouk
,
D.
2013
Frequency pressure regulation in water supply systems
.
Water Science and Technology: Water Supply
13
(
4
),
896
905
.
Hovstadius
,
G.
2007
Tackling soaring energy costs in North America
.
World Pumps
2007
(
492
),
38
40
.
James
,
K.
,
Campbell
,
S. L.
&
Godlove
,
C. E.
2002
Watergy – Taking Advantage of Untapped Energy and Water Efficiency Opportunities in Municipal Water Systems
.
Alliance to Save Energy
,
Washington, DC
,
USA
Kusiak
,
A.
,
Zeng
,
Y.
&
Zhang
,
Z.
2013
Modeling and analysis of pumps in a wastewater treatment plant: a data-mining approach
.
Engineering Applications of Artificial Intelligence
26
(
7
),
1643
1651
.
Lewis
,
C.
&
Hendrix
,
M.
2012
Smart grid and AMI for water utilities
.
Journal American Water Works Association
104
(
9
),
58
61
.
Madoński
,
R.
,
Nowicki
,
M.
&
Herman
,
P.
2014
Application of active disturbance rejection controller to water supply system
. In:
33rd Chinese Control Conference (CCC)
,
4401
4405
.
Martinez
,
F.
,
Hernandez
,
V.
,
Alonso
,
J. M.
,
Rao
,
Z.
&
Alvisi
,
S.
2007
Optimizing of the operation of the Valencia water distribution network
.
Journal of Hydroinformatics
9
(
1
),
65
78
.
Nair
,
S.
,
George
,
B.
,
Malano
,
H. M.
,
Arora
,
M.
&
Nawarathna
,
B.
2014
Water–energy–greenhouse gas nexus of urban water systems: review of concepts, state-of-art and methods
.
Resources, Conservation and Recycling
89
,
1
10
.
Page
,
P. R.
,
Abu-Mahfouz
,
A. M.
&
Mothetha
,
M. L.
2017a
Pressure management of water distribution systems via the remote real-time control of variable speed pumps
.
Journal of Water Resources Planning and Management
143
(
8
),
04017045
.
Page
,
P. R.
,
Abu-Mahfouz
,
A. M.
&
Yoyo
,
S.
2017b
Parameter-less remote real-time control for the adjustment of pressure in water distribution systems
.
Journal of Water Resources Planning and Management
143
(
9
),
04017050
.
Pelli
,
T.
&
Hitz
,
H. U.
2000
Energy indicators and savings in water supply
.
Journal American Water Works Association
92
(
6
),
55
62
.
Picos-Benítez
,
A. R.
,
López-Hincapié
,
J. D.
,
Chávez-Ramírez
,
A. U.
&
Rodríguez-García
,
A.
2017
Artificial intelligence based model for optimization of COD removal efficiency of an up-flow anaerobic sludge blanket reactor in the saline wastewater treatment
.
Water Science and Technology
75
(
6
),
1351
1361
.
Rezaei
,
H.
,
Ryan
,
B.
&
Stoianov
,
I.
2015
Pipe failure analysis and impact of dynamic hydraulic conditions in water supply networks
.
Procedia Engineering
119
,
253
262
.
Salam
,
M. A.
,
Ang
,
S. P.
,
Rahman
,
Q. M.
&
Malik
,
O.
2016
An estimation of magnetic field strength near substation using artificial neural network
.
International Journal of Electronics and Electrical Engineering
4
(
2
),
166
171
.
Sensus
2012
Water 20/20: Bringing Smart Water Networks into Focus
.
Sharifzadeh
,
S.
,
Clemmensen
,
L. H.
,
Borggaard
,
C.
,
Støier
,
S.
&
Ersbøll
,
B. K.
2014
Supervised feature selection for linear and non-linear regression of L*a*b* color from multispectral images of meat
.
Engineering Applications of Artificial Intelligence
27
,
211
227
.
Stokes
,
J. R.
,
Horvath
,
A.
&
Sturm
,
R.
2013
Water loss control using pressure management: life-cycle energy and air emission effects
.
Environmental Science and Technology
47
,
10771
10780
.
Sun
,
C. C.
,
Puig
,
V.
&
Cembrano
,
G.
2016
Combining CSP and MPC for the operational control of water networks
.
Engineering Applications of Artificial Intelligence
49
,
126
140
.
Sutariya
,
V.
,
Groshev
,
A.
,
Sadana
,
P.
,
Bhatia
,
D.
&
Pathak
,
Y.
2013
Artificial neural network in drug delivery and pharmaceutical research
.
Open Bioinformatics Journal
7
(
1
),
49
62
.
Vicente
,
D. J.
,
Garrote
,
L.
,
Sánchez
,
R.
&
Santillán
,
D.
2015
Pressure management in water distribution systems: current status, proposals, and future trends
.
Journal of Water Resources Planning and Management
142
(
2
),
04015061
.