Title of Invention  AN APPARATUS FOR PROVIDING A CONTROL VECTOR 

Abstract  An adaptive predictive expert control system for controlling singleinput singleoutput, or multivariable timevariant processes with known or unknown parameters and with or without time delay, is disclosed. The adaptive predictive expert control system of the present invention adds an expert block (14) into the operation of previously known adaptive predictive control systems. This expert block based on rules and in the evolution of the process variables, determines and/or modifies the operation of the driver block (9), control block (8) and adaptive feedback mechanism (6) of the previous art, in order to improve the performance, robustness and stability of the overall control system. 
Full Text  FORM 2 THE PATENTS ACT, 1970 (39 of 1970) COMPLETE SPECIFICATION (See Section 10, rule 13) APPARTUS FOR PROVIDING A. CONTROL. VECTOR SANCHEZ, JUAN, MARTIN of PLAZA VALLE DE LA JAROSA, 77, E2 8035, MADRID, SPAIN , SPANISH national The following specification particularly describes the nature of the invention and the manner in which it is to be performed :  ADAPTIVE PREDICTIVE EXPERT CONTROL SYSTEM BACKGROUND OF THE INVENTION Field of the Invention The present invention is directed to an adaptive predictive expert control system for controlling singleinput singleoutput, or multivariable timevariant processes, with known or unknown parameters and with or without time delay. More particularly, the present invention is directed to a system with an expert block controlling previously known adaptivepredictive control systems. This expert block operates using rules that can determine and/or modify the operation of a driver block, a control block and an adaptive mechanism. The expert block can accommodate evolution of the process input/output (I/O) variables used with the above blocks and mechanisms. Application of the expert block to control systems defined in previous adaptivepredictive methodologies improves performance, robustness and stability of the overall system. Description of the Related Prior Art The application of Knowledge based rules applied to selection of PID tuning parameters is discussed by Coelho AAR et al. In the disclosure entitled "MULTI LOOP ADAPTIVE CONTROLLERS IN NONLINEAR PROCESS CONTROL", proceedings of the 30 IEEE Conference on Decision and Control, Vol. 3, Brighton, England, December 1991 (1991 1211), pages 2909291 2 XP000279392. Use of PIP controllers in process control appiications is well known, while the use of knowledge based rules to modify the parameters ot"a PID controller is a more recent development. The application of adaptivepredictive control systems using adaptivepredictive controllers is well known. Such adaptivepredictive control systems are described in European Patent No. 0037579, issued August 20, 1986, entitled "AdaptivePredictive Control Method and AdaptivePredictive Control System", U.S. Patent No. 4,358,822 issued November 9, 1982, entitled "AdaptivePredictive Control System" and United Kingdom Patent No. 1583545, dated July 1,1977 and entitled "Improvements In an Relating to Control Systems", all issued to the present applicant. The adaptivepredictive controllers are operable to predict, by means of an adaptivepredictive (AP) model included in the control block, the value of a set of dynamic process output variables. The set of dynamic process output variables form a dynamic process output vector at a future sampling instant. The set of dynamic process output variables also generate at each sampling instant, using the AP model, a predicted control vector that causes the predicted dynamic process output vector to equal a desired dynamic process output vector at the future sampling instant. The desired dynamic process output vector is generated by a driver block according to desired performance criterion. In addition, adaptivepredictive controllers include an adaptive mechanism that periodically updates the parameters of the AP model within the control block. The updates occur in such a way that the difference between the actual value of the dynamic process output vector at the future sampling instant and the value of the dynamic process output vector predicted by the control block is reduced towards zero. Adaptivepredictive control systems have proven reliability and excellent performance when applied to industrial processes. However, their performance, robustness and stability become less reliable when the controlled process is very non linear, is time varying and/or evolves in the presence of strong noises and perturbations. In these situations, it must be determined when the adaptive predictive control can be applied successfully, and when it may be advantageous to use the available real time process information to model the input/output relationship of the process. Thus, a new control solution is desired wherein: a) The experience in the application of adaptivepredictive control could be used (i) to develop rules to determine in real time when adaptivepredictive control is advisable; and (ii) when adaptivepredictive control is advisable, to develop additional rules to determine how it must be applied and when adaptation of the AP model parameters must be performed. b) When adaptivepredictive control is not advisable, the experimental knowledge of the human operator should be taken in to account by a further Set of rules that will apply an "intelligent" control vector to the process. , The present invention is an improvement over the previously disclosed adaptivepredictive control systems, as indicated in the above mentioned U.S. Patent Nos, 4,358,822 and 4,197,576, and United Kingdom Patent No. 1583545. SUMMARY OF THE INVENTION The adaptivepredictive expert control system of the present invention adds an expert block into the operation of previously known adaptivepredictive control systems. The expert block determines and/or modifies the operation of the driver block, the control block and the adaptive mechanism of the previous art. The expert block operates with a set of rules, for example: a) A first set of rules which can determine whether or not the control block can use the AP model to generate a control vector by applying adaptivepredictive control as defined by the previous art. b) When the AP model can be used to generate the control vector, a second set of rules can determine whether or not the parameters of the AP model can be updated from the real time process I/O variables measurements. c) When the AP model can be lined to generate the control vector, a third set of rules can determine whether or not control limits applied to the predicted control vector must be reduced appropriately. d) When the AP model should not be used to generate the control vector, the control block will use a fourth set of rules, based on the human operator control experience, to generate the control vector to be applied to the process. e) When the AP model can be used to generate the control vector, a iifth set of rules can determine whether or not the performance criterion of the driver block must be redefined. f) When the AP model can be used to generate the control vector, a sixth set of rules can determine whether or not the parameters of the AP model must be reinitialized to some predefined values The above considered sets of rules, within the expert block, imitate in different ways human intelligence. For instance, these rules can take into account specific domains in which the process I/O variables may reside and the length of "time of residence", understanding by "time of residence" the number of consecutive control periods that the process I/O variables remain in a specific domain. The relation between the sets of rules and the specific domains and times of residence may be defined, for instance, as follows: 1) The first set of rules may examine a first domain and a first time of residence for a dynamic process output vector, containing at least one process output variable. The first set of rules can then determine that adaptivepredictive control will be applied when the dynamic process output vector resides in the first domain for a period of time in excess of the predetermined first time of residence. 2) The second set of rules may examine a second domain and a second time of residence for the dynamic process output vector. The second set of rules can determine that adaptation by updating the AP model parameters should be stopped while adaptivepredictive control is applied. The updates to the AP model can be 5 c) When the AP model can be used to generate the control vector, a third set of rules can determine whether or not control limits applied lo the predicted control vector must be reduced appropriately. d) When the AP model should not be used to generate the control vector, the control block will use a fourth set of rules, based on the human operator control experience, to generate the control vector to be applied to the process. e) When the AP model can be used to generate the control vector, a fifth set of rules can determine whether or not the performance criterion of the driver block must be redefined. f) When the AP model can be used to generate the control vector, a sixth set of rules can determine whether or not the parameters of the AP model must be reinitialized to some predefined values. The above considered sets of rules, within the expert block, imitate in different ways human intelligence. For instance, these rules can take into account specific domains in which the process I/O variables may reside and the length of "time of residence", understanding by "time of residence" the number of consecutive control periods that the process I/O variables remain in a specific domain. The relation between the sets of rules and the specific domains and times of residence may be defined, for instance, as follows: 1) The first set of rules may examine a first domain and a first time of residence for a dynamic process output vector, containing at least one process output variable. The first set of rules can then determine that adaptivepredictive control will be applied when the dynamic process output vector resides in the first domain for a period of time in excess of the predetermined first time of residence. 2) The second set of rules may examine a second domain and a second time of residence for the dynamic process output vector. The second set of rules can determine that adaptation by updating the AP model parameters should be stopped while adaptivepredictive control is applied. The updates to the AP model can be halted when the dynamic process output vector resides in the second domain for a period of time in excess of the second time of residence. 3) The third set of rules may examine a third domain and a third time of residence for the dynamic process output vector. The third set of rules can carry out an appropriate tightening of control limits while adaptivepredictive control is applied. The tightening of control limits is applied when the dynamic process output vector resides in the third domain for a period of time in excess of the third time of residence. In addition, while the dynamic process output vector is in the domain for adaptivepredictive control, the expert block will always be able to modify the control block parameters and/or to redefine the driver block performance criterion taking into account the particular operating conditions of the process and the desired performance of the control system. Thus: , 4) The fifth set of rules may examine the dynamic process output vector evolution to redefine the driver block performance criterion according to a desired control system performance. A first set of subdomains related to the domain for application of adaptivepredictive control is defined according to a desired control system performance criteria. The redefinition of the driver block performance criterion can occur when the dynamic output process vector enters and resides in a subdomain of the first set of subdomains for a predetermined time of residence. 5) The sixth set of rules may examine the dynamic process output vector evolution to determine when the parameters of the AP model should be partially or totally reinitialized. The reinitialization can use the experimental knowledge available for the process dynamics in a subdomain of a second set of subdomains. Again, the second set of subdomains is related to the domain for application of the adaptivepredictive control. The reinitialization can be applied when the dynamic process output vector resides in a subdomain of the second set of subdomains for a predetermined time of residence. 7 The distinctive feature of the present invention allows the adaptivepredictive expert control system to control processes submitted to the influence of noises and perturbations and/or with a highly nonlinear dynamic and time varying nature in such a way that the perfonnance, robustness and stability of the new system is significantly better than that of the previous art, as explained in the following. As previously discussed, the functioning of the expert block allows to combine adaptivepredictive control with rule based control. The first one will be used in the domain of operation where it is possible to describe satisfactorily the process dynamic behavior by means of an AP model and, thus, achieve precise, high performance control, as described in the previous art, and the second one in the domain of operation where it is advisable to use the experience of the human operator and emulate his behavior by means of rules. As a matter of fact, inmost industrial applications of adaptivepredictive controllers, they are on automatic mode only in a certain domain of operation and, when the process goes away from this domain, the operator takes manual control over the plant. Thus, this distinctive feature of this invention allows to integrate the operators knowledge within the automatic control system, increasing in this way its robustness, autonomy and stability in the overall operation of the plant. Also another distinctive feature of the present invention allows to stop the operation of the adaptive mechanism when the process should not apply adaptivepredictive control or when the process I/O variables attain a certain domain, in which the effect of noises and perturbations may cause an undesirable drift on the AP model parameters within the control block. For instance, this may happen when the process approaches stability and, in this situation, the level of noises and perturbations is relatively large in relation with the variation on the control vector. Then, it is advisable to disregard the process I/O variables information for estimation purposes and, therefore, to stop the updating of the AP 8 model parameters. Otherwise the above mentioned drift in the AP model parameters originates unstability in the operation of the control system. In a similar way an additional feature of the present invention allows to avoid the problem of large erratic predictive control actions that may for instance be originated when adaptivepredictive control tries to compensate for stochastic noises and perturbations, added on the dynamic process output vector, and the level of said stochastic noises and perturbations is relatively high in relation with the actual overall tendency on the evolution of the dynamic process output vector. The definition of a domain for limited control solves this problem providing a smooth and efficient control action, in spite of the stochastic noises that may act on the process, and avoiding high frequency large excursions on the control vector. A further additional feature of the present invention allows to reinitialize totally or partially the parameters of the AP model within the control block when the dynamic process output vector under adaptivepredictive control evolves attaining subdomains of operation where important, even discontinuous, changes in the process dynamics may occur, as happens in a PH process or in a high performance aircraft. This further additional feature of the present invention uses the knowledge available about the process dynamics on said subdomains of operation to avoid the significant deterioration that said changes in the process dynamics may cause in the performance of the control system. Another feature of the present invention allows modification of the driver block performance criterion to accommodate the desired performance of the control system to the operating conditions when required, thus improving the overall performance of the system. In the following detailed description of the invention, the adaptivepredictive expert control system is presented as an extension of the previous adaptivepredictive control system disclosed in the U.S. Patent No. 4,197,576, in order to allow a better and simple understanding of the invention, 9 but also includes.features of oilier disclosures of the previous ail, such as (hose of the European Patent No. 0037579 and the U.S. Patent No. 4,358,822. For instance, in the U.S. Patent No. 4,197,576 the prediction horizon used is equal to 1, but the invention may be directly applied just as easily to an embodiment with a larger prediction horizon, as that considered in European Patent No. 0037578 and U.S. Patent No. 4,358,822. It is important to emphasize that the invention is an extension of the previous art on adaptivepredictive control and therefore may be applied as an extension of any of the embodiments of said previous art. The above, and other objects, features and advantages of the present invention will become apparent from the following description read in conjunction with the accompanying drawings, in which like reference numerals designate the same elements. BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a schematic representation showing the general and conceptual structure of adaptivepredictive control systems, according to prior art embodiments; Fig. 2 is a schematic representation showing the general and conceptual structure of the adaptivepredictive expert control system according to the present invention; Fig. 3(a) is a graph showing the output of an adaptivepredictive expert controlled process; and Fig. 3(b) is a graph showing the control signal for the process of Fig. 3(a). DETAILED DESCRIPTION OF THE INVENTION Referring now to the drawings, wherein like numerals indicate like elements, there is shown in Fig. 2 a block diagram which provides a pictorial representation of both the apparatus carrying out the process being controlled and the adaptivepredictive expert controf system which controls the operation of the 10 apparatus. It should be recognized, however, that various elements of this diagram do not necessarily represent physically separate entities but rather indicate the different functions which are carried out by a single digital computer. Of course, since different elements of the cqmputer combine to carry out the different functions of the process, each element in Fig. 2 may properly be considered a separate means for carrying out the specified function. From the comparison between Figs. 1 and 2, it can be seen that basically the present invention adds and integrates the operation of the expert block 14 modifying the operation of the prior art system. The operation of expert block 14 is described as a part of the overall description of the adaptive predictive expert control system as follows. At any sampling instant k, a human or an automatic operator 2 may decide to directly apply a control vector u(k) to the apparatus 10 carrying out the process being controlled, as shown by path 1. Alternatively, operator 2 may decide on automatic operation of the adaptivepredictive expert control system. When the adaptivepredictive expert control system is under automatic operation, two different actions may consecutively be performed at each sampling instant k, which are described in the following: (A) Identification action: In the identification action, the control vector u(k) is applied to both the apparatus 10 carrying out the process being controlled and the identification block 4 as shown in Fig. 2. The identification block 4 uses and adaptivepredictive (AP) model stored in computation block 5 to compute an estimated incremental process output vector d(k). The parameters of the AP model can be initialized or reinitialized by expert block 14, taking into account the information received from sensor 12 and the sixth set of rules described above. The sixth set of rules considers the second set of subdomains related to the domain for application of adaptivepredictive control and their respective times of residence. An error vector e(k), which represents the difference between the actual and the estimated incremental process output vectors y(k) and d(k), 11 respeetively, is used to update the parameters of the previously mentioned adaptivepredictive model through an adaptive feedback mechanism 6. However, this updating will take place only if permitted by expert block 14. Expert block 14 makes an update decision based on Information received from sensor 12, input from an operator 2 (input vePtor v(k), described below) and the second set of rules described above. The second set of rules considers the second domain and time of residence for the dynamic process output vector, to decide whether to stop adaptation, as set forth in some detail in operation (C) below. The control vector u(k) is delayed by r + 1 sampling periods in delay block 11 before being acted upon by computation block 5. (B) Control action: The above identification action may take place always before the control action is executed. As shown by path 7 the control vector u{k) to be applied to apparatus 10 carrying out the process to be controlled is computed by,control block 8. This computation is carried out under the control of expert block 14, based on the information received from sensor 12 and the first set of rules described above, which considers the first domain and time of residence for an adaptivepredictive control. The computation can be carried out in the following two ways: 1. Predictive control computation: this computation will be performed by control block 8 using the same updated AP model as the identification block 4, in such a manner that a predicted control vector u(k) is first computed. Said predicted control vector u(k) makes the desired incremental output vector of the process d 1 (k + r + 1), at the sampling instant k + r +1, equal to the corresponding predicted incremental process output vector d1" (k + r + 1), at the same instant k + r + 1, where r is the number of sampling time delays observed or conveniently considered in the process. The desired incremental process output vector d1 (k + r + 1) is explicitly or implicitly computed at the instant k by the driver block 9 in response to input vector v(k) from operator 2. The input of driver block 9 is 12 direetly or indirectly set by the operator of the control system and represents the desired setpoint value of the output of the process. Driver block 9 generates, according to a certain performance criterion, the desired process output trajectory which is the desired dynamic trajectory over which the process output should reach the desired steady output, i.e. the said setpoint value. The performance criterion of driver block 9 can be initially defined and later on redefined by expert block 14, taking into account the information received from sensor 12 and the fifth set of rules described above. The fifth set of rules considers the first set of subdomains of the domain for adaptivepredictive control. The value of the desired process output trajectory is dp (k I r I 1) at sampling instant k + r + l. The vector dp (k + r + 1) permits the incremental process output vector d1 (k + r + 1) to be computed, as set forth in some detail in the description of operation (G). As u(k) would not act in the process output vector until the instant k + r+l, the desired incremental output vector of the process d1 (k + r + 1) must be explicitly or implicitly known at instant k in order to compute u(k). The number r represents the number of sampling time delays considered in the process. Secondly, the predicted control vector u(k) will be limit checked, as part of the predictive control computation in control block 8, in order to obtain the control vector u(k) to be applied to apparatus 10 carrying out the process to be controlled. Expert block 14, based on the information received from sensor 12 and the third set of rules, which considers the third domain and time of residence for limited control, will determine when the control limits to be applied on said predicted control vector will have to be reduced in control block 8, as set forth in some detail in operation (I). 2. Intelligent control computation: In this case, the control vector to be applied to apparatus 10 will be computed by control block 8 based on said fourth set of rules which imitates the human operator behavior and that may be 13 malci ialized using any of the well known intelligent, ex perl or fuzzy control techniques. To properly control the process carried out by apparatus 10 the adaptivepredictive expert control system uses absolute and incremental values of the output, input and measurable disturbance vectors of the process. The sequence of specific operations that the adaptive predictive expert control system will cany out at every sampling instant k during its automatic operation is described as follows: (A) Measurement (by sensor 12), and, if it is considered convenient, filtering of the output variables of the process carried out by apparatus 10 to obtain the dynamic process output vector yp(k), the dimension of which is considered to be n. (B) Computation of the incremental output vector y(k) (in computation block 13) by: y(k) = yp(k)yp(kY) (1) where y is an integer that can be conveniently chosen and which represents the number of sampling periods upon which the incremental process output vector y(k) is computed. (C) Computation in expert block 14 of the variables that determine whether the dynamic process output vector yp(k) is in the domain for adaptivepredictive control and/or in the domain for stopping adaptation and/or in the domain for limited control, respectively, and their corresponding times of residence. A way of calculating these variables is described in the following points: 1. The variables pcd(k) and pct(k) indicate if yp(k) is in the domain for adaptivepredictive control and its time of residence, respectively, and their values arc obtained as follows: (i) if yp(k) PCD, then pcd(k) = 1, otherwise pcd(k) = 0, 14 being PCI) the domain for adapliveprediclivc control, i.e. the set of values of yp (k) that may determine the application of adaptivepredictive control. This set may be defined by the operator, for instance, by means of an upper limit, yPCU, and a lower limit, ypC2 on the value of yp(k). (ii) if pcd(k) = 1, then pct(k) = pct(k  1) + 1, otherwise pct(k) = 0. 2. The variables sad(k) and sat(k) indicate if yn(k) is in the domain for stopping adaptation and its time of residence, respectively, and their values are obtained as follows: (i) if yp(k) e SAD, then sad(k) = 1, otherwise sad(k) = 0, being SAD the domain for stopping adaptation, i.e. the set of values of yp(k) inside PCD that may determine that the operation of the adaptive feedback mechanism 6 is stopped. This set may be dcllned by the operator, for instance, by means of an upper limit, ysau, and a lower limit, ysau, on the value of yp(k) around the setpoint value v(k). (ii) If sad(k) = 1, then sat(k) = sat(k 1) + 1, otherwise sat(k) = 0. 3. The variables lcd(k) and lct(k) indicate if yp(k) is in the domain for limited control and its time of residence, respectively, and their values are obtained as follows: (i) if yp(k) € LCD, then lcd(k) = 1, otherwise lcd(k) = 0, being LCD the domain for limited control, i.e. the set of values of yp(k) inside PCD that may determine the reduction of the control limits on the control vector to be applied on apparatus 10. This set may be defined by the operator, for instance, by means of an upper limit, yleu, and a lower limit ylel, on the value of yp(k) around the setpoint value v(k). (ii) If lcd(k) = 1, then lct(k) = Ict(k  1) + 1, otherwise lct(k) = 0. From the values of the above defined variables expert block 14 will compute the variables pc(k), sa(k) and lc(k) as described in the following: 15 • If pecl(k) I and pct(k) • petl • 0 then pc(k) l, otherwise pc(k) 0. • If sad(k) = 1 and sat(k) > satl > 0 then sa(k) = l, otherwise sa(k)= 0. • If lcd(k) = 1 and lct(k) > lctl > 0 then lc(k) = 1, otherwise Ic(k) = 0. The values of pctl, satl and lctl may be appropriately selected by the operator. From the values of the previously defined variables pc(k), pa(k) and lc(k), expert block 14 will determine the sequence of specific operations here considered, according to the following rules: (a) Operations (G), (H) and (1) below, which allow the computation of adaptivepredictive control, will be executed only if the value of the variable pc(k) is equal to 1, and particularly the control limits on the control signal to be applied to apparatus 10 will be reduced in (I) only if the variable lc(k) is equal to 1. (b) The following operations (D), (E) and (F), which allow the updating of the AP model parameters, will be executed only if the value of pc(k) is equal to 1 and the value of sa(k) is equal to 0. (c) Finally, operation (J), the last of the sequence, will be executed only if the variable pc(k) is equal to 0. Also in operation (C) the computation in expert block 14 of the variables, that determine if the dynamic process output vector is in anyone of the subdomains of said first and second sets of subdomains of said domain for adaptivepredictive control and their corresponding times of residence, may be carried out in a similar way to that described above for the variables previously considered in this operation. (D) Computation in identification block 4 ( if pc(k) = 1 and sa(k) = 0 ) of the estimated incremental process output vector d(k) by the AP model, which can be defined by: 16 (2) Where the vector u (k  r  1) and w (k  i r2) are obtained by: (3) (4) where Up(k  i  r) and Wp(k  1  r2) are the control and the measurable disturbances vectors, respectively, of dimensions n1 and m, at the sampling instants k  i  r and k  i  r2, respectively. In equation 2, the integers h, f and g can be conveniently chosen, and likewise the integers r1 and r2 can also be conveniently chosen taking into account the available or forecasted measurements of the output and disturbances vectors, respectively. The matrices Aj (k  1), Bi (k 1) and Ci (k  1) of the AP model have appropriate dimensions and their values, that correspond to past values before being updated at time k, can be initialized and reinitialized totally or partially by expert block 14 taking into account the information received from sensor 12 and the sixth set of rules described above, which considers said first set of subdomains of said domain for adaptivepredictive control and their respective times of residence. If the dimension of the control vector is bigger than the dimension of the dynamic process output vector then, in most cases, supplementary conditions should be added to obtain a unique solution, or simply some of the control vector components can be included in the disturbance vector; as a particular case it will be considered that n1 =n. 17 (E) Computation (if pc(k) = 1 and sa(k) = 0) of the incremental error vector by: (5) (F) Computation in adaptive feedback mechanism 6 (if pc(k) = 1 and sa(k) = 0) of the updated values al inslant k of the parameters ajjq(k) and bijq(k) and cijq(k), that are the elements in the jth row and qth column of the matrices Aj(k), Bi(k) and Ci(k), respectively, by means of any of the algorithms defined in the previous art, for instance by means of the following algorithms: (6) (7) (8) where ej(k), yq(k  i  r1 ), uq(k  i  r) and Wq(k  i  r2) are the corresponding components of the vectors e(k), y(k  i  q), u(k  i  r) and w(k  i  r2), respectively. The coefficients βaijq, βbijq and βcijq can be conveniently tuned and aj(k) are variable gains, which may be chosen, for instance, as follows: (9) (G) Computation in driver block 9 (if pc(k) =1) of the desired process output vector and the desired incremental process output vector at sampling instant k + r + 1, which can be carried out explicitly or implicitly using any of the designs 18 of the previous art on adaptivepredictive control and particularly the general one defined in U.S.A. patent No. 4,358,822, considered as follows: At each sampling instant k, the driver block 9 selects a desired dynamic output trajectory between sampling instants k + r + 1 and k + r + 1 + X with A, > 0, said desired trajectory being equal to a specific process output trajectory, between sampling instants k + r + 1 and k + r + 1 + X, that the adaptivepredictive model predicts would be caused by a specific sequence of future control vectors between sampling instants k and k + X, and such that the specific process output trajectory and the specific future control vector sequence optimize a chosen performance criterion in which a future reference trajectory of process output vectors can be explicitly considered, said reference trajectory can be periodically redefined as a function of the previously measured process output and evolve according to the desired dynamics to the setpoint. The optimization of the chosen performance criterion can be obtained through the minimization of a certain index, I, which characterizes the selected performance criterion. Said index may vary as a function of at least one of a reference trajectory of process output vectors, a set point of said process, previously measured and predicted process output and control vectors, constraints on the values of said control vectors and any other parameter or variables that influence the control performance of said process. Several indexes were considered by the way of examples in the description of the previous art presented in European Patent No. 0037579 and U.S.A. patent 4,358,822. As an illustrative example we will here consider, for reasons of simplicity, the following one: (10) Note that this index is a particular case of said general design previously considered in which X is equal to zero, y"p(k + r + 1), yn(k + r + 1  r1  i) and 19 v (k I I  i) me the predicted process outpul vector, the process output vector and the driver block input.vector at ihe sampling instants k + r + 1, k + r + 1  r\  i and k + 1  i, respectively, v(k + 1  i) is a vector of dimension n, that is generated directly or indirectly by the operator; and the matrices F; (i = 1, t) and Hi (i =1, s), as well as the integers t and s, can be chosen freely, to take into account the desired dynamics (i.e. to define the desired dynamics). The minimization of index (10) implies: (10a) where dp(k + r + 1) of dimension (n x 1) is the desired process output vector at sampling instant k + r + 1, i.e. the specific value of the predicted process output y"p(k + r + 1) at the sampling time instant k + r + 1 that minimizes index (10). .From the value of the desired process output vector dp(d + r +1), computed by equation (10a), the desired incremental output vector d1 (k + r + 1) can be easily computed in various manners; a particular one, usually convenient when y > r, is given by the following equation: .01) If found necessary the value of d1 (k + r + 1) can be limit checked. The choice of index (10) was illustrated in the experimental examples of the previous art. The computation of the predicted control vector, up(k), and the predicted incremental control vector at sampling instant k, u(k), at sampling instant k, which minimizes index (10) is considered in operation (H) below. According to one of the distinctive features of the present invention, the initial index selected, to determine the operation of the driver block, can also be redefined on real time by expert block 14 taking into account the information 20 received from sensor 12 and said fifth set of rules, which considers said second set of subdomains of said domain for adaptivepredictive control and their corresponding times of residence. (I I) The computation of the predicted control vector and the predicted incremental control vector in control block 8 (if pc(k) = l) is intimately related to the above considered operation (G). This compulation was considered and illustrated for several indexes, by the way of examples, in the description of the previous art presented in European Patent No. 0037579 and U.S.A. Patent No. 4,358,822. In fact, operation (G) determines at least implicitly the value of the predicted control vector at instant k, since the predicted control vector corresponds to the value at instant k of said specific future control vector sequence that minimizes the index considered in operation (G). This is equivalent to saying that the predicted control vector at instant k is the one that makes the predicted process output at instant k + r + 1 equal to the value of the desired process output trajectory at the same future sampling instant k + r + 1. Therefore, in general, the explicit computation of the predicted control vector at instant k can be made, using the AP model, from the value of the desired process output trajectory at instant k + r + 1. Particularly, in the case in which the performance index is defined by (10), and integers y, r1 and r 2 are chosen such that y > r, r1 > r and r2 > r, the considered explicit computation may be carried out according to the following: 1. From the updated AP model (updated by the output of adaptive feedback mechanism 6), the predicted incremental process output vector d1 "(k + r + l) at the sampling instant k + r + 1, will depend upon the predicted incremental control vector u(k) and is given by the equation: h f dt"(k + r+l)= ΣA,(k)y(k + r+lr1i)+ Σ Bi(k) u(k + 1  i) (12) i=l i=2 g + Σ C,k) w(k + r + 1 r2  1) + B1(k) u(k) 21 The predicted incremental control vector u(k) is computed by making the corresponding predicted incremental process output vector d j "(k + r + !) equal to the desired incremental output vector d1 (k + r + 1), and is given by: 2. From u(k), the predicted process control vector up(k) will be computed by: (14) where up(k  y) is the limited process control vector applied to the apparatus 10 at instant k  y. The computation of this limited process control vector at instant k is considered in the next operation. (I) Computation in control block 8 (if pc(k) = 1) of the limited process control vector, up(k), to be applied to the apparatus 10 carrying out the process being controlled, is made by applying absolute and incremental limits to the previously computed predicted process control vector up(k). Absolute upper upU and lower up1 limits conveniently chosen will be first applied to the predicted process control vector Up(k) to obtain a first value of up(k). Therefore, Up(k) will be made equal to up(k), unless up(k) is beyond the chosen absolute limits, in which case, Up(k) will be made equal either to u pu or to upl. Secondly, in order to compute the final value for Up(k), an incremental limit ujI (k) > 0 will be applied to said first value of up(k) as follows: (i) 22 above llic equations 1, 3, 4, 1 1, 14 and 15 need to be respectively modified as follows: Likewise, when it is considered convenient to give specific constant values to some of the AP model parameters (for instance, because of a certain knowledge of the process), these values can be given to the respective parameters, and the corresponding coefficients will be set to zero. Also, it is possible to stop the updating operations of the adaptivepredictive model parameters as long as it is considered convenient. If desired the identification action may be performed at any time, even when the control vector is not computed by the adaptive predictive expert control system by carrying out operations (A), (B), (D), (E) and (F), and this identification action can be performed in real time or offline and even in between the sampling intervals. It will be observed that in the operation (II) to compute u(k), the matrix Bl(k) must be inverted. The risk of singularity of matrix B [(k) can almost always be avoided by adding time delays to the components of the input and output process vectors, and controlling the resultant process. An illustrative experimental example of this procedure was presented in the previous art. 24 i.X£l..vR. IM RN"f! AI , J"XAMPl ,P. The adaptive predictive expert control system, previously described, has been implemented for the control of a simulated pH process, as described in the following. A pH process typically mixes a flow of an acid solution with a flow of a base splution and tries to control the resultant pi I of the mix by manipulating the flow rate of the base solution. This pH control has been for many years an example of control difficulties in nonlinear chemical processes. The main difficulty comes from the fact that the gain of the process changes drastically (increases) when the pH value of the mix enters in a region around the value of 7. The pH set point is usually in this region. The simulated pi I process considers that the dynamic relationship between the incremental pH of the mix at instant k, Apl I(k), and the incremental percentage of flow rate of the base solution, u(k), is described, when the absolute value of the pH of the mix is lower than 4 4 or bigger than 9.6, by the following equation: pH(k)= 1.0089 pH(kl) 0.0879 pH(k2) (22) + 0.01622 u(k 1)  0.00833 u(k  2) However, when the absolute value of the pH of the mix is between 4.4 and 9.6, the above equation (22) becomes: pH(k)= 1.0089 pH(kl) 0.0879 pH(k2) + 0.1622 u(k 1) 0.0833 u(k 2) (23) 25 The gain in cqunlion (23) is ten (imes grenter than in cquation (22.). Thus, when (he absolute value of the pll enters the region between 4.4 and 9.6, the gain of the simulated process increases by ten times, which represents the main difficulty to control this kind of processes. Additionally, the simulation adds a measurement noise of zero mean and 0.1 standard deviation on the pH measured value. The adaptive predictive expert system of this invention, solves easily this problem by defining an adaptive predictive control domain and a lower and an upper expert control domains. Said adaptive predictive control domain will be defined by a lower limit, ypcl, and an upper limit, ypCU, whose values in this case are 4.4 and 9.6, respectively. Said lower expert control domain, will correspond to the values of pl1 under 4.4, and said upper expert control domain to the values of pH over 9.6. Additionally, a domain for stopping adaptation and a domain for limited control may be defined assigning the following values to their lower and upper limits, respectively: ysal = 6, Ysau = 8, ylcu = 6.5 y ylc =7.5. Thus the adaptive predictive expert control system of this invention can now be applied to the previously considered simulated pH process, according to the sequence of specific operations (A) to (J), by: (i) assigning values to the limits on the time of residence for the adaptive predictive control domain, pctL the stopping adaptation domain, satl and limited conuol domain. letl. ln this , . (ii) assigning values to the parameters that determine the execution of adaptive predictive control, according to operations (A) to (I). In this implementation the assigned values for the main of said parameters were: y = 1, A. 26  Additionally, the driver block parameters correspond to a desired second order dynamics with a damping ratio equal to I and a time constant equal to 1.5 control periods, and the initial values for the adaptive predictive model parameters are: al(0)  1, a2(0) = 0.2, bl(0) = 0.1 and b2(0) = 0.1. As it may be observed, these initial values differ from those of equation (23) of the simulated process in the adaptive predictive domain. (iii) determining the expert control to be applied in those expert domains previously defined. In this implementation, the expert control applied in said lower expert domain was Up(k) = 48, and in said upper expert domain was Up(k) = 51. These last two expert control signals were those that an expert operator would apply to drive the pH value towards the adaptive predictive control domain. Figs. 3(a) and 3(b) show, from the beginning of the control action, the results"of a 100 control instants experiment, in which the simulated pH process was controlled from an initial pH value of 2, corresponding to a control signal of 22, to the set point of 7. In FIG. 3(a), the evolution of the pH value is shown, within the first expert domain and the adaptive predictive control domain, until it reaches the set point value. Fig. 3(b) represents the evolution of the control signal over the same period of time. It should be noted that for the control problem of a pll 1 process, the adaptive predictive expert control system has provided a highly desirable solution. This solution is notable in light of the fact that the problem of controlling a pH process is a long and oftencited example of nonlinear chemical processes, for which it is difficult to provide adequate control. In summary, the adaptive predictive expert control system described adds an expert block into the operation of previously known adaptive predictive control systems for controlling singleinput singleoutput, or multivariable timevariant processes with known or unknown parameters and with or without time delay. 27 This export block, bused on rules and observable evolution of the process variables, determines and/or modi fies the opuralion of the drivur block, control block and adaptive .feedback mechanism of the previous art. in order lo improve (he performance, robustness and stability of the overall control system While examples of the application ol"different adaptivepredictive control schemes have been cited in the above discussion, it should be evident that any adaptivepredictive control scheme can be used to implement the present invention. That is. the success of the invention does not depend with particularity on the adaptivepredictive scheme used. 28 CLAIM: 1. An apparatus to provide a control vector to a device carrying out a dynamic process, comprising:an input vector containing at least one input variable of said process; an output vector containing at least one output variable of said process; a process model effective to provide predicted values of said output vector at a plurality of future sampling instants; a predicted control vector determined at a sampling instant k effective to produce a predicted output vector at a future sampling instant k+r+1 equal to a desired output vector at said future sampling instant k+r+1; applying control limits on said predicted control vector to produce a limited control vector; an intelligent control vector derived from empirical operator experience; a first set of rules effective to select one of said limited control vector and said intelligent control vector as said control vector; and said device varies said input vector in accordance with said control vector. 2. An apparatus to provide a control vector according to claim 16, further including:a second set of rules effective to determine if parameters of said process model should be updated effective to reduce a difference between an actual process output vector at a sampling instant k+r+1 and a predicted process output vector at said sampling instant k+r+1; a third set of rules effective to determine if said control limits applied on said predicted control vector should be reduced; a fourth set of rules effective to generate said intelligent control vector further based on an evolution of at least one of said at least one input variable and said at least one output variable; an index of performance that considers process output vector trajectories that said process model predicts would be caused by future control vector sequences over a prediction horizon [lambda]+l, being [lambda]+l a positive integer; said index varying as a function of at least one of a reference trajectory of process output vectors, a set point of said process, previously measured and 29 predicted process output and control vectors, constraints on values of said control vectors and any other parameter or variables that influence control performance of said process; said desired output vector and said predicted control vectors are obtained by minimizing said index of performance; a fifth set of rules effective to redefine in real time said index of performance based on an evolution of at least one of said at least one input variable and said at least one output variable; and a sixth set of rules effective to reinitialized said parameters of said process model based on an evolution of at least one of said at least one input variable and said at least one output variable. 3. An apparatus to provide a control vector according to claim 17, wherein:said first set of rules is determined taking into account at least one domain and a corresponding time of residence for at least one of said at least one input variable and said at least one output variable related to application of adaptive predictive control; said second set of rules is determined taking into account at least one domain and a corresponding time of residence for at least one of said at least one input variable and said at least one output variable related to stopping adaptation of said process model; said third set of rules is determined taking into account at least one domain and a corresponding time of residence for at least one of said at least one input variable and said at least one output variable related to reducing control limits or said predicted control vector; said fourth set of rules can be selected from at least one of a known intelligent, expert and fuzzy control technique; said fifth set of rules is determined taking into account at least one subdomain and a corresponding time of residence for at least one of said at least one input variable and said at least one output variable of a domain related to application of adaptive predictive control; and said sixth set of rules is determined taking into account at least one subdomain and a corresponding time of residence for at least one of said at least one input 

inpct200200065mumabstract(15072004).doc
inpct200200065mumabstract(15072004).pdf
inpct200200065mumcancelled page(15072004).pdf
inpct200200065mumclaims(granted)(15072004).doc
inpct200200065mumclaims(granted)(15072004).pdf
inpct200200065mumcorrespondence ipo(12102006).pdf
inpct200200065mumcorrespondence(25012005).pdf
inpct200200065mumdrawing(15072004).pdf
inpct200200065mumform 19(03102003).pdf
inpct200200065mumform 1a(15072004).pdf
inpct200200065mumform 2(granted)(15072004).doc
inpct200200065mumform 2(granted)(15072004).pdf
inpct200200065mumform 3(15072004).pdf
inpct200200065mumform 5(15072004).pdf
inpct200200065mumpower of attoreny(18012002).pdf
Patent Number  211160  

Indian Patent Application Number  IN/PCT/2002/00065/MUM  
PG Journal Number  45/2007  
Publication Date  09Nov2007  
Grant Date  17Oct2007  
Date of Filing  18Jan2002  
Name of Patentee  ADAPTIVE PREDICTIVE EXPERT CONTROL ADEX SL  
Applicant Address  PLAZA VALLE DE LA JAROSA, 77,E28035,MADRID, SPAIN  
Inventors:


PCT International Classification Number  G05B 1/00  
PCT International Application Number  PCT/IB00/01368  
PCT International Filing date  20000628  
PCT Conventions:
