A design method of a speed model predictive controller of a traction elevator based on a CARIMA model

Abstract

This invention discloses a method for designing a speed model predictive controller for a traction elevator based on the CARIMA model, comprising the following steps: Step 1: Setting the prediction field of view p; Step 2: Rearranging the traction equations of the elevator traction machine into a generalized model prediction general form and calculating the H and Q matrices; Step 3: Calculating the torque current change value in the current step based on the control law predicted by the generalized model; Step 4: Updating the torque current value i in the current step based on the torque current value in the previous step. q (k)=i q (k-1)+Δi q (k), where k represents the current period; Step 5: The torque current value obtained in the previous step is used as the input of the current controller and enters the current controller; Step 6: Adjust the input weighting value λ of the cost function J according to the actual control effect; Step 7: The current value becomes the value of the previous period and enters the next prediction period. In the next period, repeat the above steps 3 to 5 to perform rolling optimization control.

Description

Technical Field

[0001] This invention relates to the field of elevator control technology, and in particular to a design method for a speed model predictive controller for traction elevators based on the CARIMA model. Background Technology

[0002] The CARIMA model stands for Controlled Auto-regressive Integrated Moving Average. CARIMA is used because it includes predicted values ​​of disturbances, thus providing unbiased predictions in steady state without needing to consider uncertainties introduced by certain parameters.

[0003] The CARIMA model has the following characteristics: (1) It can describe a class of non-stationary disturbances; (2) It can guarantee that the steady-state error of the system output is 0; (3) The CARIMA model can naturally incorporate integral action into the control law, so the deviation caused by step load disturbance will be naturally eliminated.

[0004] Existing traction elevator speed controllers generally use PI controllers, which utilize their inherent low-pass characteristics for disturbance rejection. However, disturbance rejection is not a primary design consideration, and there is no specific design for disturbance rejection. This type of control method has limited ability to comprehensively control these uncertainties, and its parameter adjustments are more based on on-site experience for targeted adjustments (ad hoc). There is also no specific way to deal with disturbances from the load cells. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a design method for a traction elevator speed model predictive controller based on the CARIMA model, which addresses the aforementioned technical problems existing in the PI controller generally used for traction elevator speed controllers.

[0006] To achieve the above-mentioned objectives, the present invention provides a method for designing a traction elevator speed model predictive controller based on the CARIMA model, comprising the following steps:

[0007] Step 1: Set the prediction horizon (p);

[0008] Step 2: Rearrange the elevator traction machine's drive equations into the generalized model prediction (GPC) form.

[0009] Calculate the H and Q matrices; where: H is a p×1 dimensional matrix, and H is a p×p dimensional matrix. Let P be a p×1 dimensional matrix, and let P be a p×(m-1) dimensional matrix. Let Q be a (m-1)×1 dimensional matrix, and let P be a p×n dimensional matrix. Let be an n×1 dimensional matrix, where ← refers to past measurements or calculated values, and → refers to future predictions, Δ = 1 - z. -1 Let P be the difference factor term, and Z denote the difference operator. Since m = 1, then P and Z are different factors. The matrix is ​​0;

[0010] Step 3: Calculate the torque current change value in the current step based on the control law predicted by the generalized model. Where E = [1 0 … 0] 1×p The coefficient matrix represents the torque current change in the current cycle that is only predicted. As a reference velocity curve matrix, H is the feedback velocity matrix measured by the speed sensor. T H is a real symmetric matrix, λ is the input weighting value of the cost function J, and λI is also a real symmetric matrix;

[0011] Step 4: Update the torque current value i in the current step based on the torque current value from the previous step. q (k)=i q (k-1)+Δi q (k), where k represents the current period; i q (k-1) represents the torque current of the previous cycle;

[0012] Step 5: The torque current value obtained in the previous step is used as the input to the current controller and enters the current controller;

[0013] Step 6: Adjust the input weighting value λ of the cost function J in Step 3 based on the actual control effect;

[0014] Step 7: The current value becomes the value of the previous period and enters the next prediction period. In the next period, steps 3 to 5 above are repeated to perform rolling optimization control (Receding Horizon Control).

[0015] In a preferred embodiment of the present invention, the H and Q matrices are:

[0016] in,

[0017]

[0018]

[0019] A1 = a1-1

[0020] A2 = -a1

[0021] For permanent magnet synchronous traction machines:

[0022]

[0023] in, It is the permanent magnet flux linkage, J is the moment of inertia on the shaft side of the traction system, B is the damping coefficient of the traction system, and p p For extreme logarithms, T s Sampling time;

[0024] For induction asynchronous traction machines

[0025]

[0026] Among them, L r L is the equivalent leakage inductance on the rotor side. m It's mutual induction, i sd The current is the direct-axis current equivalent to the stator side, and n is the reduction ratio of the worm gear reducer.

[0027] Using the properties of the Toeplitz matrix, for Solve the problem.

[0028]

[0029] in,

[0030]

[0031] If the predicted field of view p is odd:

[0032]

[0033] If the predicted field of view p is even:

[0034]

[0035] By adopting the above technical solution, this invention establishes a controlled autoregressive integral moving average model of permanent magnet synchronous motor and asynchronous induction motor, and uses a generalized model predictive control algorithm to achieve speed control, thereby improving the robustness of the elevator control system under external disturbances and its own parameter disturbances. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the equivalent model of the induction asynchronous motor of the present invention.

[0037] Figure 2 For speed regulation during field weakening, i sd and L mWith magnetic flux A diagram illustrating the adjustments made between them.

[0038] Figure 3 This is a flowchart illustrating the design method of the traction elevator speed model prediction controller based on the CARIMA model according to the present invention. Detailed Implementation

[0039] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0040] The background technology of this invention can be found in the following documents:

[0041] [1] "Multivariable Generalized Predictive Control of Diagonal CARIMA Model" by Li Qi'an and Chu Jian, Journal of Zhejiang University (Engineering Science), No. 4, 2006;

[0042] [2] "A Novel GPC Algorithm and Its Implementation in PMSM Direct Torque Control" Lin Ruiquan, Huang Tao, Wang Chunying, Journal of Harbin University of Science and Technology, October 2011;

[0043] [3] Design of Network Predictive Controller for Permanent Magnet Synchronous Motor, Hu Jibao, Zhang Yan, China Science and Technology Papers Online, Vol. 9, No. 16, August 2016;

[0044] [4] Solving Symmetric Linear Equations by LDLT Decomposition Method Zhu Songsheng, Nanjing Normal University

[0045] Based on the CARIMA model, this invention proposes a design method for an optimal controller applicable to traction elevators using generalized model predictive control. By setting the prediction field and input signal weighting coefficients, the optimal torque current with the cost function is obtained for current controller control.

[0046] 1. The general form of generalized model prediction is:

[0047]

[0048] in, H is a p×1 dimensional matrix, and H is a p×p dimensional matrix. Let P be a p×1 dimensional matrix, and let P be a p×(m-1) dimensional matrix. Let Q be a (m-1)×1 dimensional matrix, and let P be a p×n dimensional matrix. This is an n×1 dimensional matrix. Where ← refers to past measurements or calculated values, and → refers to future predictions. The dimension p is the prediction horizon, and Δ = 1 - z. -1For the difference factor, z represents the difference operator.

[0049]

[0050] The general form of a CARIMA model is:

[0051]

[0052] in

[0053] a(z) = 1 + a1z -1 +a2z -2 …+a n z -n

[0054] b(z) = b1z -1 +b2z -2 ...+b m z -m

[0055] ζ k For random disturbance variables with a mean of 0, the entire The term is obtained by multiplying both sides of the disturbance prediction term by Δ:

[0056] a(z)Δy k =b(z)Δu k +T(z)ζ k

[0057] This result is called the Incremental Form, which uses an incrementing input term to obtain the output.

[0058] Assuming the disturbance is constant over time—a key assumption of this method—once an increasing form is adopted, the difference factor Δ = 1 - z -1 Because of this, the disturbance value is the same in the current and previous timeframes, so the disturbance term becomes 0 after multiplying by Δ. Therefore, in the assumptions of this method, it is believed that the disturbance term is 0 when using an increasing form.

[0059] make

[0060] A(z)=a(z)Δ=(1+a1z -1 +a2z -2 …+a n z -n (1-z) -1 )=1+A1z -1 +A2z -2 …+A n z -n

[0061] Where A1 = a1 - 1, A2 = -a1 + a2, A3 = a2 - a3...

[0062] In fact, A i The coefficients are obtained by convolution of a(z) and the Δ coefficients, that is, by solving for the overlapping area determined by the points where the Δ coefficients slide over the a(z) coefficients, satisfying:

[0063]

[0064] Where coef refers to the coefficient of Δ, and satisfies the following constraints:

[0065] i,j>0

[0066] j≤i

[0067] (i-j+1)∈[0,1]

[0068] Based on the above description, the model can be written as:

[0069] A(z)y k =b(z)Δu k

[0070] Using one-step prediction, we get:

[0071] y k+1 +A1y k +A2y k-1 +…+A n y k-n+1 =b1Δu k +b2Δu k-1 +…+b m Δu k-m+1

[0072] Iteratively predicting in one step, predicting in p steps, yields:

[0073]

[0074] Among them, y k+1 This represents the predicted output value at the next time step, y. k y represents the measurement value at the current moment. k-1 This represents the measurement value at the previous moment, and so on.

[0075] The next goal is to reorganize the CARIMA model into a standard model for generalized model predictive control, as described above. The above equation can be reorganized as follows:

[0076]

[0077] The corresponding matrix yields:

[0078]

[0079] in,

[0080]

[0081] By rearranging and eliminating terms, we can obtain the output prediction matrix. The solution formula

[0082]

[0083] Combining the above equation with the GPC standard equation but

[0084]

[0085] The dimensions are p×p, p×(m-1), and p×n, respectively.

[0086] This involves C. A The problem of finding the inverse matrix, C A It is a lower left triangular matrix, and also a Toeplitz matrix. To find the inverse of a Toeplitz matrix, we can use the following property of the Toeplitz matrix.

[0087]

[0088] This transforms the inversion operation into matrix multiplication, resulting in...

[0089]

[0090] Design the cost function:

[0091]

[0092] The purpose of this design is to ensure that J is a scalar, where, This is called the error weighted sum. Let λ be the input weighted sum, and λ be the input weight value.

[0093] Expanding the above equation, we get...

[0094]

[0095] Since J is a scalar, according to the linear property, each of its expansion terms is also a scalar.

[0096]

[0097] It can be used to combine like terms to obtain...

[0098]

[0099] Substitution

[0100]

[0101] Organized

[0102]

[0103] Since J is a scalar, each term is equal to its transpose.

[0104]

[0105] so,

[0106]

[0107] To find the minimum value of the cost function, J must satisfy the condition... The partial derivative is 0. This uses two formulas for matrix differentiation, one of which is...

[0108]

[0109] Where A is a real symmetric matrix,

[0110]

[0111] Another one is

[0112]

[0113] Because (A) T A) T =A T A, then in the above solution of the cost function J, H T H is a real symmetric matrix, and λI is also a real symmetric matrix. The matrix obtained by adding the two is H. T H+λI is also a real symmetric matrix, thus...

[0114]

[0115] Solving for the given information

[0116]

[0117] In actual prediction implementation, only the control value of the current step is controlled, so it needs to be multiplied by a coefficient matrix E = [1 0 … 0]. 1×p

[0118]

[0119] in:

[0120]

[0121] This is the reference signal curve matrix.

[0122]

[0123] This is the matrix of feedback signals measured by the sensor.

[0124] In this way, speed controller control can be achieved simply by updating the control value for the current cycle. The following section will illustrate this with a practical application in traction motors.

[0125] 1. When using a permanent magnet synchronous motor

[0126] For a permanent magnet synchronous motor, its electromagnetic torque equation is:

[0127]

[0128] Where, p p For extreme logarithms, It is a direct-axis magnetic flux linkage. For cross-axis flux linkage, i d For direct-axis current, i q For quadrature axis current,

[0129]

[0130] in, It is a permanent magnet flux linkage, L d It is a direct-axis inductor, L q It is a quadrature axis inductor

[0131] Substituting into the above equation, we can obtain

[0132]

[0133] For surface-mounted permanent magnet synchronous motors

[0134] L d =L q

[0135] but

[0136]

[0137] The mechanical equations of the traction system are:

[0138]

[0139] Among them, T L For the load torque of the traction system, ω m ω is the angular velocity of the traction system, which can be measured by a velocity sensor; J is the moment of inertia on the shaft side of the traction system; and B is the damping coefficient of the traction system, which can be obtained through two uniform motions at different speeds, specifically, two motions at the same position with the same load.

[0140] When the two uniform velocities are ω1 and ω2 respectively, assuming the direction of the driving force is opposite to the direction of the load, then

[0141] T e1 -(T L +Bω1)=T e2 -(T L +Bω2)

[0142] Among them, T e1 and T e2 It is the average of the driving forces at two different speeds, which gives us

[0143]

[0144] It is assumed that within a sampling period dt, the load torque belongs to a large inertial system, and the inertial lag constant is much larger than dt, T L Remain unchanged, that is

[0145] Backward differential discretization yields

[0146]

[0147] Where z is the current value, z-1 is the previous state value, and T s The sampling period.

[0148] rearranged items

[0149]

[0150] Following the general form of the CARIMA model described above, get

[0151] a(z) = 1 + a1z -1

[0152] in,

[0153]

[0154] The load torque ripple function is:

[0155] Lin Ruiquan et al., in "A Novel GPC Algorithm and Its Implementation in PMSM Direct Torque Control," and Hu Jibao et al., in "Design of Network Predictive Controller for Permanent Magnet Synchronous Motors," introduce that actual engineering experimental data shows that the load torque of permanent magnet synchronous motors in machine tools, traction elevators, and electric vehicles exhibits smooth ergodic characteristics. Especially in elevators, during a single operation, the car operates in a closed loop with a fixed passenger capacity. At this time, the elevator's load torque is the imbalance between the car side and the counterweight side. As mentioned above, it is assumed that within a sampling period dt, the load torque belongs to a large inertial system, and the inertial lag constant is much larger than dt. L Remain unchanged, that is This aligns with the CARIMA model's requirement of slowly time-varying disturbances, treating load torque as a system disturbance term.

[0156] Multiplying both sides by the difference factor Δ, we get

[0157]

[0158] Because the disturbance changes slowly, it is multiplied by the difference factor Δ = 1 - z -1 Post-perturbation term It is 0.

[0159] Organize into standard form A(z)y k =b(z)Δu k have to

[0160] A(z)ω m (z)=b(z)Δi q (z)

[0161] in,

[0162] A1 = a1-1

[0163] A2 = -a1

[0164]

[0165] The goal is to organize this into a generalized form of generalized model prediction. Since the inverse matrix is ​​required when calculating matrices such as H, P, and Q, the prediction field of view p should generally not be set too far to account for computational costs.

[0166] In actual calculations because The dimensions of each matrix can be calculated, namely p×p, 0, and p×2. The main task is to find C. A The inverse matrix p×p

[0167] in,

[0168] Using the properties of the Toeplitz matrix described above

[0169] Because A(k) = 1 + A1k -1 +A2k -2 =a(k)Δ=(1+a1k) -1 (1-k) -1 )

[0170] but

[0171]

[0172] in,

[0173]

[0174] Using the expansion of power series

[0175]

[0176] We can obtain,

[0177]

[0178] in,

[0179]

[0180] If the predicted field of view p is odd

[0181]

[0182] If the predicted visual field p is even

[0183]

[0184] Thus obtain

[0185]

[0186] Thus, by transforming the Topelitz matrix into a multiplication operation, the resulting matrix is ​​substituted into the control law.

[0187]

[0188] in,

[0189] E = [1 0 … 0] 1×p

[0190]

[0191] This is the reference velocity curve matrix.

[0192]

[0193] This is the feedback velocity matrix measured by the speed sensor.

[0194] Obtaining the torque current change value in the current step is equivalent to an integrator, thus eliminating steady-state error. Therefore, the torque current value in the current step is...

[0195] i q (z)=i q (z-1)+Δi q (z)

[0196] This torque current value is substituted into the current controller as the optimal input for the current controller.

[0197] 1. When using an induction asynchronous motor, see [link to relevant documentation]. Figure 1 Its equivalent model adopts a six-element model, which consists of stator resistance, rotor resistance, mutual inductance, stator inductance, rotor inductance, and slip resistance. Figure 2 Chinese R s R is the stator resistance. r L is the rotor resistance equivalent to the stator side. s L is the equivalent leakage inductance on the stator side. r L is the equivalent leakage inductance on the rotor side. m It's mutual induction. It is the equivalent slip resistance;

[0198] Its electromagnetic torque equation is:

[0199]

[0200] Where i sd To represent the direct-axis current equivalent to the stator side, when regulating speed below the base frequency, the current at the no-load rated speed is generally used as the direct-axis current. See [link / reference]. Figure 2 During field weakening speed regulation, L m and i sd It varies with the operating point. As shown in the figure, to maintain constant power output when the output voltage reaches the limit value, magnetic weakening and speed boosting are required. sd and L m It will be based on the magnet chain The decrease is adjusted according to the curve.

[0201] i sq The equivalent quadrature-axis current on the stator side varies with torque. L m It is equivalent mutual inductance, L r This is the equivalent inductance on the rotor side, and n is the reduction ratio of the worm gear reducer. The worm gear reducer is used to improve the load-carrying capacity of the asynchronous motor. According to the law of conservation of energy, the input power equals the output power, and the torque at both ends of the reducer has the following relationship:

[0202] T e 'Ω'=T e Ω

[0203]

[0204] Among them, T e Ω' is the electromagnetic torque on the shaft side of the asynchronous motor, Ω' is the angular velocity on the shaft side of the asynchronous motor, and T is the electromagnetic torque on the shaft side of the asynchronous motor. e Ω is the torque at the traction end, n is the angular velocity at the traction end, and n is the reduction ratio of the reducer.

[0205] Based on the mechanical equations of the traction system described above, Here ω m This represents the equivalent angular velocity at the traction end, i.e., Ω and T mentioned above. L It is the load torque at the traction end of the asynchronous motor.

[0206] Similarly, assuming that within a sampling period dt, the load torque belongs to a large inertial system, the inertial lag constant is much larger than dt, T L Remain unchanged, that is

[0207] It can be written in state-space equation form.

[0208]

[0209] Backward differential discretization yields

[0210]

[0211] Where k is the current value, k-1 is the previous state value, and T s The sampling period.

[0212] Organized

[0213]

[0214] As described in the section on permanent magnet synchronous motors.

[0215] According to the general form of the CARIMA model, get

[0216] a(k) = 1 + a1k -1

[0217] in,

[0218]

[0219] Multiplying both sides of the load torque fluctuation function by the difference factor Δ, we get

[0220]

[0221] Because the disturbance changes slowly, it is multiplied by the difference factor Δ = 1 - k. -1 Post-perturbation term It is 0.

[0222] Arranged into standard form A(k)y k =b(k)Δu k have to

[0223] A(k)ω m (k)=b(k)Δi q (k)

[0224] in,

[0225] A1 = a1-1

[0226] A2 = -a1

[0227]

[0228] The goal is to organize this into a generalized form of generalized model prediction.

[0229] In actual calculations because The dimensions of each matrix can be calculated, namely p×p, 0, and p×2. The main task is to find C. A The inverse matrix p×p

[0230] in,

[0231]

[0232] Using the properties of the Toeplitz matrix described above

[0233] Because A(k) = 1 + A1k -1 +A2k -2 =a(k)Δ=(1+a1k) -1 (1-k)-1 ),but

[0234]

[0235] in,

[0236]

[0237] Using the expansion of power series

[0238]

[0239] We can obtain,

[0240]

[0241] in,

[0242]

[0243] If the predicted field of view p is odd

[0244]

[0245] If the predicted visual field p is even

[0246]

[0247] get

[0248]

[0249] Thus, the matrix obtained by inverse transformation of the Toeplitz matrix and multiplication is substituted into the control law.

[0250]

[0251] in,

[0252] E = [1 0 … 0] 1×p

[0253]

[0254] This is the reference velocity curve matrix.

[0255]

[0256] The feedback speed is measured by the speed sensor.

[0257] Obtaining the torque current change value in the current step is equivalent to an integrator; therefore, the torque current value in the current step is i. q (k)=i q (k-1)+Δiq (k)

[0258] This torque current value is substituted into the current controller as the optimal input for the current controller.

[0259] Regarding the optimal input of the current controller (H) T H+λI) -1 Two methods are provided for solving the inverse matrix.

[0260] The first approach is the traditional Gauss-Jordan elimination method, which involves constructing a linear equation (H... T When solving for H+λI)x=I, in the original matrix H T Based on H+λI, add an identity matrix of the same order to construct an augmented matrix [H]. T H+λI|I], then [H T H in H+λI|I] T H+λI is transformed into an identity matrix using elementary row operations, and subsequent identity matrices are also transformed accordingly. The preceding matrix H... T The matrix corresponding to the identity matrix after H+λI is transformed is the invertible matrix obtained.

[0261] The second approach is based on the LDLT decomposition method, which is described below:

[0262] As mentioned above, H T H+λI is a real symmetric matrix, when H T When all the principal minors of H+λI are not zero, the LDLT decomposition method can be used to solve the problem.

[0263] When matrix H T When the principal minors of H+λI are not zero, there is a unique Doolittle decomposition H. T H+λI=LU, where

[0264]

[0265] It is a lower triangular matrix, with L being its abbreviation.

[0266]

[0267] It is an upper triangular matrix, with U being its abbreviation.

[0268] Extract each row of matrix U sequentially. ii ,get

[0269]

[0270] in,

[0271]

[0272] It is a diagonal matrix, and D is its abbreviation.

[0273]

[0274] so

[0275]

[0276] Because H T H+λI is a real symmetric matrix, therefore we have

[0277] (H T H+λI) T =H T H+λI

[0278] It can be obtained

[0279]

[0280] Based on the uniqueness of the decomposition, we obtain

[0281]

[0282] Right now

[0283]

[0284] When H T When all the principal minors of H+λI are not zero, H T H+λI can be uniquely decomposed into

[0285] H T H+λI=LDL T

[0286] in,

[0287]

[0288] When H T H+λI satisfies LDL T When decomposing the conditions, the solution formula (1) can be obtained.

[0289]

[0290] in,

[0291] aik For H T Elements of the H+λI matrix

[0292] k = 1, 2, ..., n

[0293] i = k+1, k+2, ..., n

[0294] When solving H T When H+λI is the inverse matrix, we have (H T H+λI)x=b, where x=(H T H+λI) -1 b is the identity matrix I, and we obtain

[0295] LDL T x = b

[0296] The above equation can be decomposed into a system of three equations.

[0297] Ly=b

[0298] Dz=L

[0299] L T x = z

[0300] Formula (2) can be obtained.

[0301]

[0302] The x-matrix is ​​the desired (H) matrix. T H+λI) -1

[0303] It can be proven that the computational cost of solving a system of symmetric equations using the LDLT decomposition method is approximately It requires less computation than solving using Gaussian elimination.

[0304] See Figure 3 The general steps of the design method for a speed model predictive controller for a traction elevator based on the CARIMA model are as follows:

[0305] 1. Set the predicted field of view p;

[0306] 2. Rearrange the elevator traction machine's drive equations into the generalized model prediction (GPC) form. Calculate the H and Q matrices. in,

[0307]

[0308]

[0309] A1 = a1-1

[0310] A2 = -a1

[0311] For permanent magnet synchronous traction machines

[0312]

[0313] in, It is the permanent magnet flux linkage, J is the moment of inertia on the shaft side of the traction system, B is the damping coefficient of the traction system, and p p For extreme logarithms, T s Sampling time.

[0314] For induction asynchronous traction machines

[0315]

[0316] Among them, L r L is the equivalent leakage inductance on the rotor side. m It's mutual induction, i sd The current is the direct-axis current equivalent to the stator side, and n is the reduction ratio of the worm gear reducer.

[0317] Using the properties of the Toeplitz matrix, for Solve the problem.

[0318]

[0319] in,

[0320]

[0321] If the predicted field of view p is odd

[0322]

[0323] If the predicted visual field p is even

[0324]

[0325] 3. Based on the control law of GPC, calculate the torque current change value in the current step.

[0326]

[0327] Where E = [1 0 … 0] 1×p The coefficient matrix represents the torque current change in the current cycle that is only predicted.

[0328]

[0329] This is the reference velocity curve matrix.

[0330]

[0331] This is the feedback velocity matrix measured by the speed sensor.

[0332] 4. Update the torque current value i in the current step based on the torque current value from the previous step. q (k)=i q (k-1)+Δi q (k), where k represents the current period;

[0333] 5. The torque current value obtained in the previous step is used as the input to the current controller and enters the current controller;

[0334] 6. Adjust the input weighting value λ of the cost function J based on the actual control effect;

[0335] 7. The current value becomes the value of the previous period and enters the next prediction period. In the next period, the three steps (3-5) mentioned above are repeated to perform rolling optimization control (Receding Horizon Control).

Claims

1. A method for designing a speed model predictive controller for a traction elevator based on the CARIMA model, characterized in that, Includes the following steps: Step 1: Set the predicted field of view p; Step 2: Rearrange the elevator traction machine's drive equations into the generalized model prediction (GPC) form. Calculate the H and Q matrices; where: H is a p×1 dimensional matrix, and H is a p×p dimensional matrix. Let P be a p×1 dimensional matrix, and let P be a p×(m-1) dimensional matrix. Let Q be a (m-1)×1 dimensional matrix, and let P be a p×n dimensional matrix. Let p be an n×1 dimensional matrix, where the dimension p is the predicted field of view p described in step 1, ← refers to past measurements or calculated values, and → refers to predicted values ​​for the future, Δ=1-z -1 Let z be the difference factor term, and z denote the difference operator. Since m = 1, therefore P and The matrix is ​​0; Step 3: Calculate the torque current change value in the current step based on the control law predicted by the generalized model. Where E = [1 0 … 0] 1×p The coefficient matrix represents the torque current change in the current cycle that is only predicted. As a reference velocity curve matrix, H is the feedback velocity matrix measured by the speed sensor. T H is a real symmetric matrix, λ is the input weighting value of the cost function J, and λI is also a real symmetric matrix; Step 4: Update the torque current value i in the current step based on the torque current value from the previous step. q (k)=i q (k-1)+Δi q (k), where k represents the current period; i q (k-1) Torque current of the previous cycle; Step 5: The torque current value obtained in the previous step is used as the input to the current controller and enters the current controller; Step 6: Adjust the input weighting value λ of the cost function J based on the actual control effect; Step 7: The current value becomes the value of the previous period and enters the next prediction period. In the next period, steps 3 to 5 above are repeated to perform rolling optimization control.

2. The design method for a traction elevator speed model predictive controller based on the CARIMA model as described in claim 1, characterized in that, The H and Q matrices are: in, A1 = a1-1 A2 = -a1 For permanent magnet synchronous traction machines: in, It is the permanent magnet flux linkage, J is the moment of inertia on the shaft side of the traction system, B is the damping coefficient of the traction system, and p p For extreme logarithms, T s Sampling time; For induction asynchronous traction machines Among them, L r L is the equivalent leakage inductance on the rotor side. m It's mutual induction, i sd The current is the direct-axis current equivalent to the stator side, and n is the reduction ratio of the worm gear reducer. Using the properties of the Toeplitz matrix, for Solve the problem. in, If the predicted field of view p is odd: If the predicted field of view p is even:

Images