Irrigation and Drainage Systems 17: 179–193, 2003. © 2003 Kluwer Academic Publishers. Printed in the Netherlands.
Local automatic control modes in an experimental irrigation canal MANUEL RIJO Department of Agricultural Engineering, Universidade de Évora, Apartado 94, Colégio da Mitra, 7002-554 Évora, Portugal (E-mail:
[email protected])
Abstract. The paper presents two local PI automatic control modes developed, tuned and verified in an experimental automatic canal. The canal was used as a physical model and it will support further studies on canal automation domain. The control algorithms are installed in local PLC’s with the objective of controlling the water depths immediately upstream and downstream of the corresponding check gates. The strategy adopted to tune the PI controllers was to compare the frequency response of the state space model, which relies on the linearized Saint-Venant equations, with a simple model considering each canal pool as a reservoir (reservoir model). The obtained results for the control gains were validated by field tests. A few simulation results are also presented as well as the experimental facility and the main configuration of the SCADA developed for the canal supervision and monitoring. Key words: canal automation, downstream control, irrigation canal, local automatic control, PI controller, upstream control
Introduction It is nowadays clear and well accepted that water is becoming a scarce natural resource. In the near future, this scarcity will present one of the biggest problems that modern societies have to face. Irrigation is the main water user and so intelligent management of openchannel conveyance and delivery systems are necessary to achieve higher water savings within a short period of time. The main purpose of an automatic canal control is to optimize the water supply in order to match the expected or aleatory demands at the offtake level. In real situations and with traditional management tools, an open-channel water conveyance and delivery system is very difficult to manage, especially if there is a demand-oriented operation (Clemmens 1987). Remote monitoring and control systems are becoming more and more cost-effective water management tools due to the permanent cost reduction and higher accuracies of the dedicated equipment such as computers,
180 controllers, remote terminal units, communication equipment, sensors and software. Nowadays, most of the programmable controllers used for canal control permit the simultaneous installation of several control modes that can be activated according to different water delivery scenarios; this is impossible with more traditional controllers (hydro mechanical, electromechanical and analogical controllers). Automatic canal control in connection with SCADA systems (supervisory control and data acquisition systems) can improve irrigation canal management, significantly increasing water use efficiency and the quality of the deliveries and, at the same time, saving labor and reducing building construction (canal dimensions) (Rijo 1999). Canal automation has become a significant research area. However, most of the research teams only use numerical simulators, without having the possibility to test and verify their mathematical approaches with prototypes or physical models. This is the route taken in this study and it is, probably, its main strength. The paper discusses two local PI (Proportional-Integral) control modes – upstream and downstream control – developed, tuned, installed and verified in an experimental physical automatic canal, designed to support control model studies. The paper also presents the strategy adopted in tuning the two PI controllers, some simulation results, and describes the experimental facility and the SCADA main configuration, developed for the supervision and monitoring of the canal.
General description of the experimental facility The Hydraulics and Canal Control Centre (NuHCC) of the University of Évora (Portugal) is the experimental facility where the automatic canal used to run simulations presented in the paper is located. NuHCC is composed by a central post, an automatic canal, a traditional canal and two reservoirs. Central post Located just in front of the two canals. The installation consists of a control office, where the master controller and the server PC are installed, and a multipurpose room (research, exposition and demonstration). In the PC, a SCADA application is installed to ensure the automatic canal control and supervision (Almeida et al. 2002). The PC is simultaneously a SCADA and internet server, the SCADA application being a web application. Accordingly, any process may be controlled by the central post, taking
181 into account that all the considered variables are available through a modbus connection protocol (Almeida et al. 2002). Automatic canal A lined trapezoidal canal with 0.15 m cross section bottom width, sides slope of 1:0.15 (V:H) and 0.900 m deep. With a length of 141 m 1.5×10−3 longitudinal bottom slope, it was designed for delivering water at a rate of 0.090 m3 s−1 . The adopted cross section answers to the digital control and water level sensor needs. The design maximizes water level variations for a certain flow perturbation, in order to detect and record the smallest flow perturbations. With uniform flow, the cross section causes 1 mm variation of water depth for a perturbation of 0.228×10−3 m3 s−1 and, with the help of the water level sensor installed, it is possible to record variations of around 0.700 mm. The canal is equipped with four rectangular sluice gates (C2, C3, C4 and C5, Figure 1), operated by electric actuators. These gates divide the canal in four pools with lengths of respectively 35 m (1st, 2nd and 3rd pools) and 36 m (4th pool – downstream). The flow at the first three gates is of the orifice type (under the gate) and the last one is an overshot gate (flow over the gate). The last gate was installed at the downstream section of the canal to discharge the remaining flow to the return canal and, at the same time, to control the flow dynamics within the downstream pool. There are also four offtakes, one for each gate. Each of them was designed as an orifice in the canal wall, equipped with an additional steel pipe (offtake box, Figure 1) where an electromagnetic flowmeter is installed (MC2, MC3, MC4, MC5, Figure 1) as well as a motorized butterfly valve (V2, V3, V4 and V5, Figure 1). The outflows go directly to the traditional canal, located at a lower level (Figure 1). In order to make possible the digital control in real time, water level sensors were installed within the offline stilling wells (Figure 1), three for each pool, located respectively at the beginning, middle and end of the pool (SNJ, SNMT, SNM, Figure 1). The canal inlet is equipped with a Monovar type flow control valve (Alsthom 2000) and also with an electromagnetic flowmeter (respectively V1 and MC1, Figure 1). Each offtake-gate group is controlled and monitored (offtake valve and gate positions) by a local programmable logic controller (PLC), after appropriate programming. Each PLC receives water depths information from the two closest upstream sensors and from the one immediately downstream of the gate. Figure 1 shows the operational area for each controller (AREA
Figure 1. Local downstream and upstream control modes.
182
183 2. . . .5) and the matching box where the controller and the electric boards (box 2. . . 5) are installed. For the AREA 1, the controller controls canal inlet water pumps and the Monovar valve and receives water depth information from the SNJ1 sensor, within the canal, and from the MNU ultrasonic sensor, within the control reservoir. These five PLC’s are linked to the master PLC at the central post with a modbus network. Each PLC has enough memory and speed capacities for hosting several control programs. Its selection can be done by an HMI (Human Machine Interface) terminal or in the SCADA application. Traditional canal The lined return canal to the storage reservoir, located parallel to the automatic, is upstream-controlled to convey the outflows from the automatic canal without any problems. It is equipped with an AMIL gate, a duckbill weir and two Neyrpic type offtakes (Kraatz & Mahajan 1975). The corresponding outflows return to the storage reservoir through a buried pipe (Figure 1), providing installation with no water losses, except for insignificant evaporation effects. Reservoirs Located at the head of the automatic canal (Figure 1). They guarantee the automatic canal inflows and also control the operation of two submersible pumps that deliver water to the higher reservoir, taken from the storage reservoir located at a lower level. At the downstream end of the automatic canal begins the traditional canal that ends in the water storage reservoir. This canal closes the described hydraulic system (Figure 1). Inside the reservoirs, level floats B1. . . B6 are installed and the ultrasonic level transmitter MNU (Figure 1) to control the pumps operation. The web site of the NuHCC (http://canais.nuhcc.uevora.pt) provides detailed information and photos about all the equipment, electric boards, communications, reservoir capacities and canal dimensions.
Local automatic feedback PI controllers As already mentioned, each local PLC controls a group offtake-check gate. For each device, the controller can control the position or the flow or permit its direct control (open/close). These algorithms (Rijo et al. 2001) are not shown; the paper only presents the local PI upstream and downstream water depths control modes for the check gates.
184
Figure 2. Local downstream and upstream control modes.
Figure 3. Feedback PI controller action.
Local automatic control is accomplished with control equipment located at the site of the control gate, and by using water level information from adjacent pools (Baume et al. 1999). Considering the variable location given in reference to the check gate, if the controlled variable (target variable controlled by the control algorithm) is the water depth immediately downstream or upstream from the gate, the control mode is designated, accordingly, by local downstream (gate G1, Figure 2a) or local upstream control (gate G2, Figure 2). To control these two water depths, a reset anti-windup PI controller was designed (the PI loop is only active if the estimated control gate position is between pre-defined minimum and maximum values). With a feedback control algorithm, the controlled variable is measured and any deviation from the corresponding setpoint value – error – is fed into the control algorithm to produce a corrective action. External disturbances (expected or unexpected inflows and/or outflows) are indirectly taken into account through their effects on the output of the system (Figure 3). Using the error as input, PI loop calculates gate openings or flow rates in real time to maintain the target values for the controlled water depth. The two local control modes here presented are, in fact, just one. But, as a positive error determines a gate closing or a negative error determines a gate opening for the upstream control, the behavior is the opposite for the downstream control (signals, Figure 3).
185
Figure 4. Upstream/downstream full control algorithm.
Figure 4 presents the full algorithm for the upstream and downstream control, including the PI loop and the gate position controller, where (Rijo et al. 2001): – href erence , is the water depth setpoint, upstream or downstream from the check gate; – hmeasured , is the measured water depth, at the same cross section of href erence ; – e, is the error, difference between href erence and hmeasured ; – h , within the dead band filter, is the dead band width for the error (5 mm); – Tf , within the second order filter, is the filter time constant (10 s); – Kp and KI are, respectively, the proportional and integral gains for the PI algorithm; – Qest , is the estimated flow (PI output) for the corresponding check gate; – Xest , is the estimated check gate position, considering Qest ; – Kx , is the gain used for the Xest calculation (roughly considered as constant); – Xref , is the gate position setpoint for the position controller; – Xmin and Xmax , are, respectively, the minimum and maximum positions for the check gate; – Xmed , is the measured check gate position; – x and T , are, respectively, the hysteresis and the time interval considered between consecutive actuator starts, for the installed position controller of the type “Bang-Off-Bang”. The calculated error e is first passed through the dead band filter and when the error is smaller than the pre-defined band width h , this block doesn’t load to any change in the output value of the entire algorithm (dead band filter output is zero).
186 To eliminate measurement noise and to increase the controller stability (Ogata 1997) a second-order filter, with Tf as time constant, was inserted between the dead filter and the PI block. The PI is the main component of the implemented algorithm. For the two cases here presented, the PI control output is the estimated flow Qest (t) for the check gate at the current time t, obtained according to the equations: Qest (t) = Q1 (t) + Q2 (t) Q1 (t) = Kp e(t) Q2 (t) = KI t0 e(t)dt
(1)
The proportional term Q1 (t), which is the current value of the error e(t) multiplied by the parameter Kp , constitutes the basic control action, which responds directly to the magnitude of the error signal. The integral term Q2 (t) provides the necessary control action to reduce the steady-state error (Molina & Miles 1996). The parameters Kp and KI are, respectively, the proportional and the integral gains and are the calibration parameters of the control model. In irrigation canals the PID algorithm is very often reduced to a PI controller, as in this case. The differential or derivative action is used when, in slow process, action must be taken as soon as possible after an upset or else the time to recover will be too long. Derivative action is not typically used in irrigation canal control algorithms, being very difficult to tune properly (Burt et al. 1998). Using finite differences approximation for digital implementation, the PI algorithm installed in each local controller has the form: Qest (t) = Q1 (t) + Q2 (t − 1) + KI t e(t)
(2)
Once the PI control action is calculated, the corresponding gate opening Xest is estimated multiplying the PI output by the parameter Kx (Figure 4), roughly obtained using the standard submerged flow gate discharge equation for the first three gates and the weir discharge equation for the fourth one. In order to ensure that a position out of the hardware range (xmin − xmax , respectively, zero and 800 mm), is not transmitted to the position actuator, a saturation block filter (5th block, Figure 4) calculates the new gate position Xref . The position controller ensures that the estimated position will be applied to the gate by the actuator. Rijo et al. (2001) present all the control parameter values for the two local modes presented. The time interval t for the discrete-time digital PI control (eq. 2) is 125×10−3 s, greater than the specific program cycle of the controller (the needed time interval to read inputs, to perform calculations, to send outputs . . . ).
187 The process of selecting the controller parameters to meet a given performance specification is known as controller tuning. Because most PID (or PI) controllers are adjusted on site, different types of tuning rules have been proposed in the literature. Also, automatic tuning methods have been developed and some of the PID controllers may possess on-line automatic tuning capabilities (Ogata 1997). Most of these tuning techniques are very useful for SISO systems (Ogata 1997). But, when the system consists of many interconnected non-linear subsystems such as in irrigation canals with multiple pools and gates, tuning becomes challenging due to the existing interactions between the various canal pools. In such cases, control parameters are usually tuned by trial and error or by optimization techniques with the help of a simulation model (Baume et al. 1999). The strategy adopted in the present study to tune the PI controllers was to compare the frequency response of the state space model, which relies on the linearized Saint-Venant equations, with a simple model that considers each canal pool as a reservoir (reservoir model) (Ratinho et al. 2002). The Saint-Venant equations were linearized assuming near steady state conditions (Schuurmans 1995; cited by Hof 1996)). This fact ensures that, for small deviations from the considered setpoint, the resulting equations would still describe the hydraulic system behavior. The assumed values to determine the steady flow profile within each canal pool were the project nominal features, i.e., 0.090 m3 s−1 and 0.700 m (as water depth setpoint). The Bode plots of the reservoir model and the finite difference model of the Saint-Venant equations gave approximate answers up to a certain frequency value (Ratinho et al. 2002). The tested controllers were designed taking this fact into consideration. The three pairs of gains (Kp ; KI ) obtained as best results were submitted to field validation by Ratinho et al. (2002). Three different scenarios for the field tests were defined by allowing the flow to be modified between 0.020 m3 s−1 and 0.080 m3 s−1 with return to the first value, considering steps of 0.010 m3 s−1 , 0.020 m3 s−1 and 0.030 m3 s−1 (respectively 11%, 22% and 33% of the canal design flow) on the inflow hydrograph (upstream control) or on the outflows (downstream control). As an example of the controller field validation, in Figure 5, a few results in terms of water depth variations around the used target 0.700 m are compared at the first gate (C2, Figure 1) level, considering the local upstream control mode. The figure only presents the field results for the two best controllers tested – controllers A (Kp = −0.6; KI = −0.006) and B (Kp = −0.8; KI = −0.01). As seen in the upper part of the figure, due to noise sensor measurements the differences between the two controllers are not very visible. However, it is noticeable that controller A can quickly
188
Figure 5. Field validation of the two best upstream water depth controllers for the gate C2, considering sensor-recorded water depth variations around the setpoint (700 mm) and inflow canal variations of 0.010 m3 s−1 (upper part) and 0.02 m3 s−1 (lower part) (adapted from (Ratinho et al. 2002).
bring the water depth back to the setpoint. The differences between the two controllers can be better observed in the lower part of the Figure 5, where it becomes clear that controller A has a smaller overshot. So, considering the field test results, controller A was selected. The proportional and integral control gains installed are the same for all gates, obviously with positive signals for the downstream control, as the hydrodynamics of the four canal pools are nearly the same (similar geometry and, for the considered frequencies, having a reservoir behavior).
Field hydraulic simulations results Figures 6, 7 and 8 present a few results (sensor records) of the field hydraulic simulations, with the analyzed two control modes installed in each local controller. Figure 6 presents the main menu of the developed SCADA application. This synoptic is a complete representation of the automatic canal, reflecting all inputs and outputs and includes the initial condition for the upstream local control simulation. The opening of all valves (%) and the respective flows
Figure 6. Main menu for the entire canal of the SCADA application, showing the schematic operation for the initial condition of the upstream local control mode.
189
190
Figure 7. Upstream local control mode field simulation results at the gate C3 level, for an inflow decrement of 15 ls−1 (upper part).
191
Figure 8. Downstream local control mode field simulation results at the gate C2 level, for an outflow increment of 15 ls−1 at the offtake V4 (upper part).
192 (l/s) (offtakes and Monovar valves), the gate openings and the water depths (in mm) are visible. For the field hydraulic simulations, a set point of 700 mm for all gates was defined (the observed values of the sensors are within the considered dead band, 5 mm). For the initial condition (Figure 6), an inflow of 50 ls−1 (Monovar) and outflows of 5 ls−1 , 10 ls−1 and 15 ls−1 , respectively, were used for the offtakes 1, 2 and 3 (or V2, V3 and V4, Figure 1). Figure 7 presents the response of the gate C3 (Figure 1) to an inflow decrement of 15 ls−1 (canal head). As it is visible on the lower part of the figure, the controller brings the water depth back to the setpoint in a fast way (about 5 min) for stabilization on the setpoint). Figure 8 presents the response of the gate C2 (Figure 1), for the installed downstream local control mode, to a sudden demand of 15 ls−1 for the offtake V4 (Figure 1). Initially, the inflow was 49 ls−1 to answer to the outflows demands (10 ls−1 , 10 ls−1 and 29 ls−1 , respectively for the offtakes V2 and V5 and downstream canal section). As can be seen at the upper part of the figure, the new demand is instantaneously satisfied using the water volume within the pool (local downstream control logics, Figure 2a). At the time 10:20, the inflow (56 ls−1 ) is still less than the total outflow (64 ls−1 ). The setpoint is quickly obtained, as can be seen on the lower part of the figure.
Conclusions and perspectives The gain parameter values installed for the two control modes proved their robustness and accuracy for several canal operation scenarios. The tuning technique used also proved capable of properly tuning the desired controller. The main goal of the described automatic canal is to support studies on canal automation, namely through comparative studies of the several mathematical approaches found in the literature. For these reasons, it is necessary to continue to develop and install other control modes, such as downstream distant control modes, to achieve a good panoply of possibilities for testing and demonstration to irrigation system managers and engineers.
References Alsthom, Gec. (Ed) 2000. Monovar Control Valve. A65-61 A, Armentières, France. Almeida M., Figueiredo J. & Rijo M. 2002. SCADA Configuration and Control Modes Implementation on an Experimental Water Supply Canal. Proc. of 10th IEEE Mediterranean Conference on Control and Automation (MED2002), Lisbon , CD-ROM, Book of Abstracts, pag. 53.
193 Baume J.P., Malaterre P.O. & Sau J. 1999. Tuning of PI controllers for an irrigation canal using optimization tools. In: A.J. Clemmens & S.S. Anderson (Eds) Modernization of Irrigation Water Delivery Systems (pp 483–500). (Proc. of the USCID Workshop, Phoenix), USCID. Burt C.M. et al. 1998. Improved Proportional-Integral (PI) Logic for canal automation. Journal of Irrigation and Drainage Engineering 124(1): 53–57. Clemmens A.J. 1987. Delivery system schedules and required capacities. In: Zimbelman, D.D. (Ed) Planning, Operation, Rehabilitation and Automation of Irrigation Water Delivery Systems (pp 18–34). ASCE, New-York. Holf, A. 1996. Control Design for Water Management Systems, TU Delft. Kraatz, D.B. & Mahajan, I.K. 1975. Small hydraulic structures. Irrigation and Drainage Paper, no. 26, FAO, Rome. Molina L.S. & Miles, J.P. 1996. Control of an irrigation canal. Journal of Hydraulic Engineering 122(7): 403–410. Ogata K. 1997. Modern Control Engineering. Prentice Hall International Inc., International edition. Ratinho T., Figueiredo J. & Rijo M. 2002. Modeling, Control and Field Tests on an Experimental Irrigation Canal. Proc. of 10th IEEE Mediterranean Conference on Control and Automation (MED2002), Lisbon , CD-ROM, Book of Abstracts, pag. 53. Rijo M. 1999. SCADA of an upstream controlled irrigation canal system. In: A.J. Clemmens & S.S. Anderson (Eds) Modernization of Irrigation Water Delivery Systems (pp 123–136). Proc. of the USCID Workshop, Phoenix), USCID. Rijo M. et al. 2001. Experimental canal for irrigation water delivery and the study of control and operation strategies for the water saving (in Portuguese), Final Technical Report of the Research Project INTERREG II-C, SIDReg-98.01.02.00002. Schuurmans, J. 1995. Control of Water Levels in Open Channels. Ph.D thesis, TU Delft.
