초록 ▼

A novel 3-Input-3-Output (3×3) Model-Free Adaptive (MFA) controller with a set of artificial neural networks as part of the controller is introduced. A 3×3 MFA control system using the inventive 3×3 MFA controller is described to control key process variables including Power, Steam Throttle Pressure

A novel 3-Input-3-Output (3×3) Model-Free Adaptive (MFA) controller with a set of artificial neural networks as part of the controller is introduced. A 3×3 MFA control system using the inventive 3×3 MFA controller is described to control key process variables including Power, Steam Throttle Pressure, and Steam Temperature of boiler-turbine-generator (BTG) units in conventional and advanced power plants. Those advanced power plants may comprise Once-Through Supercritical (OTSC) Boilers, Circulating Fluidized-Bed (CFB) Boilers, and Once-Through Supercritical Circulating Fluidized-Bed (OTSC CFB) Boilers.

대표청구항 ▼

1. A method of controlling a 3-Input-3-Output (3×3) process having three main-processes and six sub-processes, each of said main-processes having an input and an output responsive to the said input, each of said sub-processes having an input and an output, each of said main-processes having a measur

1. A method of controlling a 3-Input-3-Output (3×3) process having three main-processes and six sub-processes, each of said main-processes having an input and an output responsive to the said input, each of said sub-processes having an input and an output, each of said main-processes having a measured process variable, which is the summation of the output of the main-process and outputs from two corresponding sub-processes; the 3-Input-3-Output (3×3) process having unknown relationships between their inputs and outputs, without approximating or modeling said relationship, comprising: a) selecting a setpoint representing a desired value for the measured process variable for each of the main-processes;b) obtaining an error value which is a function of the difference between said setpoint and said measured process variable for each of the main-processes;c) providing a hardware 3-Input-3-Output (3×3) controller that has three main controllers, each corresponding to one of the main-processes, and six compensators, each corresponding to one of the sub-processes;d) for each of the main processes, applying said error value as the sole input to its corresponding main controller whose inputs are time-delayed functions of said error value, and whose output is a control value combined with the outputs from two adjacent compensators to produce the total control output of the main controller;e) configuring each said main controller by entering a time constant based on the time constant of the corresponding main-process being controlled;f) for each of the sub-processes, providing a compensator which uses the control value from the corresponding main controller as its input, and uses the time constant of the main controller as its time constant;g) applying the total control output of each main controller to vary the input of its corresponding main-process and two sub-processes; andh) continuously iteratively varying the parameters of each of the main controllers to help minimize said error value for each of the main-processes. 2. The method of claim 1, further comprising adding said error value to the control value for each of the main-processes. 3. The method of claim 1, for each of the main-processes, further comprising adding the error value to said control value, in which the error value can be increased or decreased by a manually adjustable gain parameter. 4. The method of claim 1, wherein the 3-Input-3-Output (3×3) process is a Power-Pressure-Temperature (PPT) process of a Boiler-Turbine-Generator (BTG) unit of a conventional coal-fired power boiler, or a Once-Through Supercritical (OTSC) Boiler, or a Circulating Fluidized-Bed (CFB) Boiler, or a Once-Through Supercritical Circulating Fluidized-Bed (OTSC CFB) Boiler. 5. A 3-input-3-output (3×3) model-free adaptive (MFA) control system comprising: a) a 3-Input-3-Output (3×3) process to be controlled, further comprising: i) three main-processes G11, G22, G33 having their corresponding process outputs x11(t), x22(t), x33(t), responsive to their corresponding control signals u1(t), u2(t), u3(t) respectively, andii) six sub-processes G21, G31, G12, G32, G13, G23 having their corresponding process outputs x21(t), x31(t), x12(t), x32(t), x13(t), x23(t), wherein the process produces three measured process variables y1(t), y2(t), y3(t) for the corresponding main-processes, each of the process variables being affected by the output of its corresponding main-process, the outputs of two corresponding sub-processes signals as follows: y1(t)=x11(t)+x12(t)+x13(t)+d1(t),y2(t)=x21(t)+x22(t)+x23(t)+d2(t),y3(t)=x31(t)+x32(t)+x33(t)+d3(t);wherein d1(t), d2(t) and d3(t) are disturbance signals; andb) a hardware 3-Input-3-Output (3×3) model-free adaptive (MFA) controller further comprising: i) three main-controllers C11, C22, C33 responsive to error signals e1(t), e2(t) and e3(t) respectively and having their corresponding control outputs y11(t), v22(t) v33(t); andii) six compensators C21, C31, C12, C32, C13, C23 having their corresponding outputs v21(t), v31(t), v12(t), v32(t), v13(t), v23(t) wherein:iii) the error signals e1(t), e2(t), e3(t) are iteratively calculated as follows:e1(t)=r1(t)−y1(t), if G11 is direct actinge1(t)=[r1(t)−y1(t)], if G11 is reverse actinge2(t)=r2(t)−y2(t), if G22 is direct actinge2(t)=−[r2(t)]−y2(t)], if G22 is reverse actinge3(t)=r3(t)−y3(t), if G33 is direct actinge3(t)=−[r3(t)]−y3(t)], if G33 is reverse actingwhere a process is direct acting if the process output increases as the process input increases, and a process is reverse acting, where a process is direct acting if the process output increases as the process input increases, and a process is reverse acting if the process output decreases as the process input increases, and r1(t), r2(t) and r3(t) are set points; iv) each of the compensators uses a control output from a corresponding main-controller as its input, and uses the time constant of the corresponding main controller as its time constant; andv) each of the main-controllers has a combined control output being iteratively calculated based on its corresponding control output and the outputs from two corresponding compensators as follows:u1(t)=v11(t)+v12(t)+v13(t),u2(t)=v21(t)+v22(t)+v23(t),u3(t)=v31(t)+v32(t)+v33(t),wherein, the combined control output of each main-controller is the input to its corresponding main-process and two sub-processes in order to control the measured process variable of its corresponding main-process. 6. w) The control system of claim 5, wherein: a) the three main-controllers C11, C22, and C33, each have a corresponding normalized error value E111(n), E122(n), and E133(n) being computed with a normalization function N(.) as follows: E111⁡(n)=Kc11Tc11⁢N⁡(e1⁡(t)),⁢E122⁡(n)=Kc22Tc22⁢N⁡(e2⁡(t)),⁢E133⁡(n)=Kc33Tc33⁢N⁡(e3⁡(t)), in which, n denotes the nth iteration; e1(t), e2(t), and e3(t) are the error signals as the controller inputs, Kc1l>0, Kc22>0, and Kc33>0 are the controller gains, and Tc11>0, Tc22>0, and Tc33>0 are the time constants for C11, C22, and C33, respectively; b) the main-controller C11, C22, and C33 each includes a neural network with an input layer including a plurality of input neurons arranged to receive normalized, scaled and delayed forms of said error signal, a hidden layer including a plurality of hidden neurons each arranged to sum the signals received by each of said input neurons weighted by an individual first weighting factor, and an output neuron which sums the individually weighted outputs of hidden neurons, the weighting factors for said hidden neuron outputs being iteratively adjusted, and an activation function f(x) substantially of the form f⁡(x)=0,if⁢⁢xba where a is an arbitrary constant and b=½; c) the main-controllers are arranged to produce their control outputs v11(t), v22(t), v33(t) substantially based on the following difference equations and online learning algorithms of the neural network: For Controller C11: pj11⁡(n)=∑i=1N⁢wij11⁡(n)⁢Ei11⁡(n),qj11⁡(n)=apj11⁡(n)+b,v11⁡(n)=Kc11⁢e1⁡(n)+100⁡[a⁢∑j=1N⁢hj11⁡(n)⁢qj11⁡(n)+b],Δ⁢⁢wij11⁡(n)=a2⁢η11⁢e1⁡(n)⁢Ei11⁡(n)⁢hj11⁡(n),Δ⁢⁢hj11⁡(n)=a⁢⁢η11⁢e1⁡(n)⁢qj11⁡(n); For Controller C22 pj22⁡(n)=∑i=1N⁢wij22⁡(n)⁢Ei22⁡(n),qj22⁡(n)=apj22⁡(n)+b,v22⁡(n)=Kc22⁢e2⁡(n)+100⁡[a⁢∑j=1N⁢hj22⁡(n)⁢qj22⁡(n)+b],Δ⁢⁢wij22⁡(n)=a2⁢η22⁢e2⁡(n)⁢Ei22⁡(n)⁢hj22⁡(n),Δ⁢⁢hj22⁡(n)=a⁢⁢η22⁢e2⁡(n)⁢qj22⁡(n); For Controller C33 pj33⁡(n)=∑i=1N⁢wij33⁡(n)⁢Ei33⁡(n),qj33⁡(n)=apj33⁡(n)+b,v33⁡(n)=Kc33⁢e2⁡(n)+100⁡[a⁢∑j=1N⁢hj33⁡(n)⁢qj33⁡(n)+b],Δ⁢⁢wij33⁡(n)=a2⁢η33⁢e2⁡(n)⁢Ei33⁡(n)⁢hj33⁡(n),Δ⁢⁢hj33⁡(n)=a⁢⁢η33⁢e2⁡(n)⁢qj33⁡(n); in which, n denotes the nth iteration; a is an arbitrary constant and b=½ which are bounded by the activation function f(x); η11>0, η22>0, and η33>0 are the learning rates for C11, C22, and C33, respectively; pj11, pj22, pj33, qj11, qj22, and qj33 are variables of the neural network for C11, C22, and C33, respectively; wij11, wij22, wij33, hj11, hj22, and hj33 are weighting factors of the neural network for C11, C22, and C33, respectively; Kc11>0, Kc22>0, and Kc33>0 are the controller gains for C11, C22, and C33, respectively; and Ei11(n), Ei22(n), Ei33(n) are the delayed signals of E111(n), E122(n), and E133(n), respectively; and d) the compensators have their corresponding outputs C21(S), C31(S), C12(S), C32(S), C13(S), C23(S) substantially in the form of the following Laplace transfer functions: C21⁡(S)=V21⁡(S)V11⁡(S)=Ks21⁢Kc21Tc21⁢S+1.C31⁡(S)=V31⁡(S)V11⁡(S)=Ks31⁢Kc31Tc31⁢S+1.C12⁡(S)=V12⁡(S)V22⁡(S)=Ks12⁢Kc12Tc12⁢S+1.C32⁡(S)=V32⁡(S)V22⁡(S)=Ks32⁢Kc32Tc32⁢S+1.C13⁡(S)=V13⁡(S)V33⁡(S)=Ks13⁢Kc13Tc13⁢S+1.C23⁡(S)=V23⁡(S)V33⁡(S)=Ks23⁢Kc23Tc23⁢S+1. in which, S is the Laplace transform operator; V11(S), V22(S), V33(S), V21(S), V31(S), V12(S), V32(S), V13(S), and V23(S) are the Laplace transform of signals v11(t), v22(t), v33(t), v21(t), v31(t), v12(t), v32(t), v13(t), and v23(t), respectively; Kc21>0, Kc31>0, Kc12>0, Kc32>0, Kc13>0, and Kc23>0 are the compensator gains; Ks21, Ks31, Ks12, Ks32, Ks13, Ks23 are the compensator sign factors, and Tc11>0, Tc22>0, and Tc33>0 are the time constants for the main controllers C11, C22, and C33, respectively. 7. The control system of claim 6, in which the compensators have said compensator sign factors being selected based on the acting types of the main-processes Gll and sub-processes Glm as follows: Kslm=1, if Gll and Glm have different acting typesKslm=−1,if Gll and Glm, have the same acting type where l=1, 2, 3; m=1, 2, 3; and l≠m. 8. The control system of claim 5, in which the 3-Input-3-Output (3×3) process is a Power-Pressure-Temperature (PPT) process of a Boiler-Turbine-Generator (BTG) unit of an electrical power or energy generation plant, in which the 3×3 process has the following main-processes and sub-processes: for G11, the input is Firing Rate and output is Power,for G22, the input is Throttle Valve Position and output is Steam Throttle Pressure,for G33, the input is Water Feed and output is Master Steam Temperature,for G21, the input is Firing Rate and output is Steam Throttle Pressure,for G31, the input is Firing Rate and output is Master Steam Temperature,for G12, the input is Throttle Valve Position and output is Power,for G32, the input is Throttle Valve Position and output is Master Steam Temperature,for G13, the input is Water Feed and output is Power,for G23, the input is Water Feed and output is Steam Throttle Pressure. 9. The control system of claim 8, in which the 3-Input-3-Output (3×3) process is a Power-Pressure-Temperature (PPT) process of a Boiler-Turbine-Generator (BTG) unit of a conventional coal-fired power boiler, or a Once-Through Supercritical (OTSC) Boiler, or a Circulating Fluidized-Bed (CFB) Boiler, or a Once-Through Supercritical Circulating Fluidized-Bed (OTSC CFB) Boiler.