Fuzzy model identification and control system design for coagulation chemical dosing of potable water

Transcription

1 Fuzzy model dentfcaton and control system desgn for coagulaton chemcal dosng of potable water C.-L. Chen* and P.-L. Hou** *Dept of Chemcal Engneerng, Natonal Tawan Unversty, Tape 106, Chnese Tawan (E-mal: **Tape Water Department, Tape 10617, Chnese Tawan (E-mal: Abstract A practcal feedforward control system wth fuzzy feedback trm s presented for controllng the dosng strategy of Changhsng Water Purfcaton Plant of Tape Water Department (TWD). The feedforward move s based on a regresson model from successful dosng data between 2001 and The expert operators adjustng actons n meetng the settled water turbdty trend and ph values are adopted to establsh the fuzzy feedback control rules. A host of feld tests have demonstrated the effectveness of the proposed control strategy. Keywords Chemcal dosng; coagulaton; feedforward control; fuzzy control Introducton The Changhsng purfcaton plant of Tape Water Department s responsble for supplyng hgh qualty drnkng water to more than a mllon ctzens around Tape cty. In order to promote the qualty of potable water, many scholars and practtoners have been tryng to mprove effectve management and protecton of water resources, purfcaton processng technques, qualty analyss and mantenance of dstrbuton networks (Jang, 1997; Kang et al., 1999). Less prorty s addressed on the systematc study of qualty montorng nstrument and automaton engneerng although these progresses are more n demand by the feld operators of the purfcaton plant. When the qualty and dosng data are collected and ntegrated, n addton to on-lne automatc dosng control system for predcton and adjustment, then t s expected to come up wth n-tme most optmum dosng n the changes of coagulant, the hgh turbdty raw water brought by typhoon and storm water, as well as the fant nucle water whch s dffcult to handle. It wll help to reduce the probablty of operatonal error and the cost of water treatment chemcals. Ths paper ams at modellng the human experts experence and then developng a feedforward and fuzzy feedback control system for the Changhsng plant for supplyng hgh qualty drnkng water. Water Scence & Technology: Water Supply Vol 6 No 3 pp Q IWA Publshng 2006 Purfcaton processng procedure and water qualty montorng system The fundamental procedure of a typcal purfcaton process ncludes: (a) coagulaton chemcal dosng; (b) sedmentaton; (c) fltraton; and (d) dsnfecton, whch s manly used to elmnate turbdty and to sterlze, as shown n Fgure 1. Flowchart of coagulaton chemcal dosng of potable water A flowchart of coagulatng chemcal dosng of potable water s shown n Fgure 2. We can then characterze the feedforward and feedback control strateges accordng to control subjects. do: /ws

2 C.-L. Chen and P.-L. Hou Fgure 1 Process flow sheet for Changhsng purfcaton plant The feedforward control strategy Chemcal coagulaton procedure ncludes chemcal dosng, rapd mxng (enablng the coagulant to dsperse rapdly and to collde wth turbdty partcles) and slow mxng (formng coarse floc). Durng the ste operaton, the most mportant ssue s to determne a correct dosng formula. However, the dosng patterns prevalng n purfcaton plants are stll based on personal experence. Jar tests should be added to obtan the optmal Fgure 2 Flowchart of coagulaton chemcal dosng of potable water 98

3 dosng data as the ste dosng nstructon durng the perod of abnormal or hgh turbdty water qualty. Partcularly, jar test s usually too late to meet practcal needs durng the raw water qualty rapd change perod. More often than not to add hgh level of coagulant would be taken to meet wth the sudden stuaton, and t s easy to cause mproper purfcaton qualty and extra chemcal costs. Furthermore, water treatment s stll n step-dosng manpulaton to deal wth the contnuous changng raw water qualty, but lacks an effectve fne-tunng mechansm. If a contnuous adjustment s made n dosng parallel to the changes n turbdty, the precson of dosng control should be able to be mproved. Thus, some successful dosng coagulaton emprcal data are appled to look for the correlaton between raw water turbdty and conductvty. Such an emprcal dosng quantty wll be used as the bass for automatc dosng to arrve at the optmzed operaton. Feedforward control wth PAC/alumnum sulfate and NaOH coagulant The successful dosng data of poly alumnum chlorde (PAC, contanng 10% Al 2 O 3 effectve component) and the alumnum sulfate (contanng 8% Al 2 O 3 effectve component) of Changhsng purfcaton plant are collected, whch ncludes turbdty and conductvty of raw water, dosng concentraton of PAC, alumnum sulfate and NaOH and ph values of water after mxng. Multple regresson of the correlatve parameters was conducted usng least squares (Penny and Lndfeld, 2000). The followng polynomals correlate the PAC/alumnum sulfate dosng levels and the turbdty of raw water (Turb) wthn two nephelometrc turbdty unt (NTU) ranges. C.-L. Chen and P.-L. Hou PAC ¼ 5:4656 þ 0:3529 Turb 2 0:00156 Turb 2 þ 0: Turb 3 for 3 NTU, Turb, 400 NTU ðr 2 ¼ 0:9771Þ ð1þ PAC ¼ 27:3535 þ 0:0411 Turb for 400 NTU, Turb, 1; 500 NTU ðr 2 ¼ 0:9973Þ ð2þ Alum ¼ 10:0778 þ 0:4454 Turb 2 0:00165 Turb 2 þ 0: Turb 3 for 3 NTU, Turb, 400 NTU ðr 2 ¼ 0:9563Þ ð3þ Alum ¼ 33:0788 þ 0:0795 Turb for 400 NTU, Turb, 1; 500 NTU ðr 2 ¼ 0:9709Þ ð4þ It s also found that both the conductvty and the ph value of raw water decreased wth ncreasng turbdty and the acd coagulant addton also causes the water qualty to be at acdty level. The 45% NaOH s used to adjust the ph value of the mxed water to mantan the optmum coagulaton condton. Hence, f PAC s used as coagulant, NaOH nonlnear equaton can be found, where Cond represents conductvty of raw water: NaOH ¼ 6: :0866 Cond þ 0:0179 Turb þ 0:00028 Cond 2 2 0: Turb 2 ðr 2 ¼ 0:9973Þ ð5þ Another NaOH non-lnear equaton can also be found when alumnum sulfate s used as coagulant as shown below, where Alum represents the dosage of alumnum sulfate: NaOH ¼ 21:4855 þ 0:1413 Alum þ 0:00321 Alum 2 2 0: Alum 3 ðr 2 ¼ 0:9973Þ ð6þ Control dosng tmes n conjuncton wth varatons n turbdty of raw water The man dsturbances of the feedforward control acton nclude dstant Chngtan raw water turbdty and conductvty. When detectng the raw water turbdty and conductvty 99

4 C.-L. Chen and P.-L. Hou of Chngtan, the dosng adjustment n Changhsng dverson well wll occur 90 mnutes later. Nevertheless, the dosng tmng should be adjusted accordng to the trend of raw water turbdty due to the nfluence of flud flowng and mass transfer dffuson n the process of water transportaton. The adjustng tme should be advanced to 80 mnutes later n Changhsng dverson well when the turbdty of raw water n Chngtan wer s on an upward trend. The adjustment needs to be delayed to be 100 mnutes later f the raw water turbdty s on the downward trend. Ths experence can be expressed verbally as: R1 : IF S 1. 0 THEN Dos: tme s ¼ 80 mn s R2 : IF S 1, 0 THEN Dos: tme s 90 þ 10 ¼ 100 mn s S 1 ¼ raw water turbdtyðtþ 2 raw water turbdtyðt 2 1Þ Dt Feedback control strategy Apply settled water turbdty to perform feedback control (I) Under dfferent turbdty ranges of raw water, the average turbdty of varous settlng pond water of Changhsng purfcaton plant s used as the controlled varable to perform the automatc adjustment of dosage regresson equaton multplers. The results of jar tests are also appled for manual adjustment of the multpler of dosage regresson equaton. The feedback control adjustment s performed every 100 mnutes (90 mnutes after dosng and 10 mnutes for averagng the turbdty of varous settlng pond water). Based on the average of turbdty of settlng pond water calculated from dfferent raw water turbdty ranges applyng PAC and alumnum sulfate coagulants, respectvely, some condtonal descrptve statements can be nferred to express fuzzy control rules: R3 : IF S 1 0 and X j. A j THEN change Dos: Tme ¼ 100 mn and DðtÞ ¼B 1 Dðt 2 100Þ S 1 ¼ raw water turbdtyðtþ 2 raw water turbdtyðt 2 1Þ Dt where S 1 s the slope of raw water turbdty; denotes the raw water turbdty of dfferent range; j represents settlng pond water of dfferent unts; X j s turbdty of settlng pond water n actual raw water turbdty range (NTU); A j s turbdty of settlng pond water n dfferent turbdty range of raw water (NTU); DðtÞ and Dðt 2 100Þ are tmely dosage and dosage 100 mnutes ago (ppm), and B 1 s a multpler of 1.1 tmes of dosage regresson value. Cope wth surgng turbdty of settled water and automatcally reduce dosage to perform feedback control (II) Affected by season, temperature, algae, mcrobes and outsde pollutants, raw water qualty s caprcous all the tme. Under the same turbdty the chemcal and physcal characterstcs of collod are not always the same. It s unavodable to deal wth a lot of emergency n the water treatment. In statstcally processng the operaton and control strateges of dosng related to the surge of turbdty of settled water, a fuzzy control mode can be establshed. 100 Input verbal varable scaled rate of change X ¼ {SS, S, M, L}. Where SS, S, M, and L represent slghtly, lght, medum and large, respectvely, upward trends of turbdty of settled water. The membershp functons of scales rate of change are shown n Fgure 3.

5 Fgure 3 The membershp functon of scaled rate of change C.-L. Chen and P.-L. Hou Output verbal varable ncreased rato of dosage Y ¼ {Z, S, M}. Where: Z represents nl adjustment of dosage; S represents slght adjustment of dosage; and M represents large adjustment of dosage, respectvely. The specfc sngletons are used as output operaton ntenstes. The membershp functons of ncreased rato of dosage are shown n Fgure 4. Establshng fuzzy control rule base wth the operaton and control experence of turbdty surge of settled water. Rule 1 : IF X s SS THEN Y ¼ Y 1 ¼ Y Z ¼ 0 Rule 2 : IF X s S THEN Y ¼ Y 2 ¼ Y S ¼ 0:2 Rule 3 : IF X s M THEN Y ¼ Y 3 ¼ Y S ¼ 0:2 Rule 4 : IF X s L THEN Y ¼ Y 4 ¼ Y M ¼ 0:4 The ncreased rato of dosage can be obtaned from the weghted average (Sun and Yang, 1994). 8 P f SS ðxþy Z þ f S ðxþy S ; f 0 # X, 0:1 f ðxþy Y ¼ X >< f S ðxþy S þ f M ðxþy S ; f 0:1 # X, 0:25 P ¼ f ðxþy ¼ f ðxþ f M ðxþy S þ f L ðxþy M ; f 0:25 # X, 0:4 >: f L ðxþy M ; f 0:4 # X Fgure 4 The membershp functon of ncreased rato of dosage 101

6 C.-L. Chen and P.-L. Hou Meanwhle, durng the perod of low turbdty and stable raw water qualty, operators tend to keep the dosage unchanged when the water output s just as the well condton. We expermented durng a low turbdty of raw water (about 10NTU), to decrease dosage from 7 ppm of PAC to 6 ppm, and then to descend to 5 ppm. The water turbdty held steady at 1.2 NTU. In ths research we are tryng to establsh searchng for optmum dosng mechansm by gradually reducng the dosage under automatc control manner whle the output water qualty s mantaned n a good condton, so as to attan the goal of reducng the alumnum salt content n output water and the cost of water treatment. Ths research adopted a tral and error approach to perform the optmum dosng chan rule, but the output of dosage s not supposed to exceed the upper and lower lmt of dosage. The fuzzy control rules are shown n the followng: R4 : IF S 1 < 0 and S 2 # 0 THEN change Dos: Tme ¼ 100 mn s and D 2 ðtþ ¼D 2 ðt 2 100Þ 2 1 and DðtÞ $ D mn R5 : IF S 1 < 0 and S 2. 0 THEN change Dos: Tme ¼ 100 mn and D 2 ðtþ ¼ð1þYÞD 2 ðt 2 100Þ and DðtÞ # D max where raw water turbdtyðtþ 2 raw water turbdtyðt 2 1Þ S 1 ¼ ; turbdty slope of raw water Dt settled water turbdtyðtþ 2 settled water turbdtyðt 2 1Þ S 2 ¼ ; turbdty slope of settled water Dt D 2 ðtþ s the n-tme dosage (ppm) at Changhsng dverson well; DðtÞ s total dosage output; D max and D mn are upper and lower lmts of dosage. Lmtaton of maxmum and mnmum dosages In order to avod the dosng devaton of dosng automatc control due to system and detector abnormalty, an upper and lower dosage lmt control s mposed to gude the feedback control operaton adjusted wthn a safe range, where D R ðtþ represents regresson concentraton of dosage (ppm), B 2 and B 3 are multplers of 0.8 and 1.4, respectvely. R6 : D mn ¼ B 2 D R ðtþ R7 : D max ¼ B 3 D R ðtþ Mxed water ph feedback control (III) Coagulant dosng must be controlled at the most sutable ph value to flocculate. Except for the approxmate dosng of NaOH by the feedforward system, t also needs to measure the ph value of mxed water to adjust the dosage of NaOH. 102 Establshng a fuzzy control rule base for target ph value by applyng PAC/alumnum sulfate coagulant. The ph control experence of mxed water wth PAC coagulant s analyzed, and the relatonshp of turbdty of raw water to ph dstrbuton of mx water can be nduced. 1. Turbdty of raw water at 300 and 1,500 NTU, ph value should be mantaned at Turbdty of raw water at between 120 and 299 NTU, ph should be Turbdty of raw water at under 119 NTU, ph should stay above 7.1 Use raw water turbdty as the measurng varable and target ph value as operatng varable, the control knowledge base can be desgned as below. L, M and H represent low, medum and hgh raw water turbdty phases, respectvely, for nput verbal

7 varable X, the raw water turbdty; H, M and L represent settngs at hgh, medum and low ph values, respectvely, for output verbal varable Y, the target ph value. Here, some specfc sngletons are adopted as the output operatng ntensty: Rule 1 : IF X s L THEN Y ¼ Y 1 ¼ Y H ¼ 7:1 Rule 2 : IF X s M THEN Y ¼ Y 2 ¼ Y M ¼ 7:0 Rule 3 : IF X s H THEN Y ¼ Y 3 ¼ Y L ¼ 6:9 Target ph value can be derved from the followng weghted average approach: P 8 f ðxþy f H ðxþy L þ f M ðxþy M ; f 300 # X # 1; 500 Y ¼ X >< P ¼ f ðxþy ¼ f M ðxþy M þ f L ðxþy H ; f 120 # X, 299 f ðxþ >: f L ðxþy H ; f 0 # X, 119 C.-L. Chen and P.-L. Hou Establshng fuzzy condtons and rules for NaOH dosng. As the detected ph s very nmble, the ph feedback control system wll establsh NaOH dosng fuzzy control condtonal rules through every 10 mnutes adjustment of NaOH dosng, R8 : IF ph. Y THEN change Dos: Tme ¼ 10 mn s and NaOH DOS:ðtÞ ¼NaOH DOS:ðt 2 10Þ 2 1 R9 : IF ph, Y THEN change Dos: Tme ¼ 10 mn s and NaOH DOS:ðtÞ ¼NaOH DOS:ðt 2 10Þþ1 Feld operaton verfcaton The feld dosng operaton at Changhsng purfcaton plant durng 2003 and 2004 has appled the regresson values of coagulants and NaOH as the bass for dosng operaton and the results of treatment are substantally conformed to expectaton. The followng are two cases of feld operatons. Case 1. On 10 January 2005, when the raw water turbdty s mantaned at 5 NTU, PAC dosage reduced from 9 ppm to 7.5 ppm and decreased further to 6.5 ppm. The dosng range s not devatng from the upper/lower lmt of dosng at PAC regresson value 7.19 whle raw water turbdty s at 5 NTU. The turbdty of settled water was mantaned at 1.5 NTU, and the turbdtes of settlng pond water were also lower than the control value. ph of mxed water was mantaned at , whch conformed to the target value set n ph control rule base. It s verfed that when the turbdty of raw water s stable, the automatc dosng reducton strategy between upper and lower lmts of dosage s ndeed feasble. Case 2. On 7 February 2005 at 4 pm the turbdty of raw water rose from 5 to 7.7 NTU and PAC dosage mantaned at 6.3 ppm. The upper and lower lmts of dosng of raw water turbdty 5 NTU PAC dosng regresson value 7.2 ppm and raw water turbdty 7.7 NTU PAC dosng regresson value 8.1 ppm, and settled water turbdty rose from 1.0 NTU to 1.7 NTU. Whle the turbdty of settled water s lower than the control values, f dosng s adjusted upward followng the ncrease trend of raw water turbdty, t wll not occur wth turbdty peak of settled water. If the regresson equaton s appled to adjust dosage to cope wth the change n turbdty, t wll help to stablze the qualty of water output. Meantme, durng the perod of 07:30 11:30 pm, raw water turbdty was mantaned at 6 NTU. From the rse of turbdty of settled water, t s derved that scaled rate of change ¼ 0.24 at 8:30 pm, PAC dosage ncreases from 6.3 ppm to 7.5 ppm wth an ncrease of dosng rato about 0.2. It conformed to the control rule of feedback control (II). After process reacton, the turbdty of settled water dropped from 1.3 NTU to 1.0 NTU. Ths also proved that at 103

8 the surge of settled water and fltrate, applyng operaton and control strateges to reduce the turbdty of settled water and fltrate can mprove treatment of abnormal water qualty. As well as mxed water ph at s not devatng from the target ph value set. C.-L. Chen and P.-L. Hou Concluson The water purfcaton s a complcated tme-varyng system whch gves rse to uncertanty, urgency and safety consderatons. In ths research, the statstcal result of emprcal operaton was appled to form the feedforward control strategy wth feedback trm. By completely realzng the dosng control actons of human doman experts, the proposed fuzzy control method can help mnmzng feld operators errors n water purfcaton operaton. A host of feld tests have demonstrated the effectveness of the proposed control strategy. References Jang, C.L. (1997). Research n the Automatc Montorng System of Purfcaton Treatment Coagulant Dosng. Master Thess, Insttute for Envronment Engneerng Research, Natonal Tawan Unversty. Kang, S.F., Lao, S.L. and Yu, R.F. (1999). Study n the Optmzaton n Operaton of Potable Water Treatment System. Water Works Assocaton of the Republc of Chna. Penny, J. and Lndfeld, G. (2000). Numercal Methods Usng Matlab. Prentce Hall, NJ, USA. Sun, T.Y. and Yang, Y.K. (1994). Fuzzy Control - Theory, Practces and Applcaton. Chuan Hwa Technology Lbrary Corp., Tape. 104