A kalman filter-based speed and load torque observer for traction motor and a method of observing

Abstract

The application discloses a traction motor speed and load torque observer based on Kalman filtering, which is represented by the following space equation for a permanent magnet synchronous motor: and is represented by the following space equation for an induction asynchronous motor: The application further discloses a traction motor speed and load torque observation method based on Kalman filtering, which comprises the following steps: an initialization step, a prediction step, a correction step and a convergence step.

Description

Technical Field

[0001] This invention relates to the field of traction motor speed observation and load torque observation technology, and particularly to a traction motor speed and load torque observer and observation method based on Kalman filtering. Background Technology

[0002] The control system is subject to uncertainty, which manifests in the following three aspects: First, there is no perfect mathematical model; second, the disturbances of the system are uncontrollable and difficult to model; and third, the sensors used for measurement also have errors.

[0003] The purpose of the Kalman filter is to find the optimal estimate of the system state vector at time k, given the state observation vector at time k and the basic mathematical model of the system. This minimizes the posterior covariance matrix (i.e., minimizes the trace of the matrix, thus minimizing the variance and consequently the posterior estimated error), which in turn minimizes the variance. In other words, it seeks a suitable Kalman gain so that the state estimate approaches the true value.

[0004] In Kalman filtering, the covariance matrices Q and R are used to describe the variance information of the Gaussian distribution. The Gaussian distribution is characterized by being describable using only its mean and variance, and it remains Gaussian even after passing through a linear system. This ensures that the filter can perform iterative state estimation. Nearly half a century of research and practice have shown that Kalman filtering is the optimal linearization filter, yielding the best estimates when used in linear systems.

[0005] When an elevator system starts, if a low-precision speed sensor, such as an ABZ incremental speed sensor, is used, compared to an analog signal input SinCos sensor, it cannot obtain an AB pulse signal per unit speed measurement cycle at very low speeds. The general processing is to use the speed information from the previous state, which is equivalent to a delay filter. The speed measurement is inaccurate, which leads to a deviation between the speed control value given by the controller and the actual control value required, resulting in shaking during startup.

[0006] like Figure 1The diagram shows the AB phase of the ABZ incremental speed sensor and the SIN and COS phases of the SINCOS sensor. The AB phase is a square wave signal, with phase A leading phase B by 90°, and SINCOS is an analog signal. In practical applications, the ABZ incremental speed sensor uses the upper and lower edges of the AB phases for counting, while the SINCOS sensor uses analog signal sampling for phase angle demodulation. Therefore, at very slow speeds, the ABZ incremental speed sensor may not capture an edge within two speed sampling intervals, even though a speed change is actually occurring. The SINCOS sensor, using analog signal sampling, can obtain the angle change in real time, thus providing the real-time speed change value. Therefore, when using the ABZ incremental speed sensor, there will be inaccurate speed measurement intervals during elevator startup.

[0007] Additionally, during startup, the controller provides a feedforward torque based on the weighing value measured by the weighing switch installed at the bottom of the car. This reduces the burden on feedback control during startup, resulting in smaller initial feedback deviations and allowing the system to follow the reference control value more quickly. However, if the weighing value deviates significantly from the actual value due to signal drift from the weighing switch (usually a magnetic induction switch), it will affect the feedforward preset torque, thus impacting the startup process. Summary of the Invention

[0008] One of the technical problems to be solved by this invention is to provide a traction motor speed and load torque observer based on Kalman filtering to address the shortcomings of existing traction motor speed and load torque observation methods. The Kalman filter is applied to the traction motor system as an equivalent linear system, which can achieve optimal filtering.

[0009] The second technical problem to be solved by this invention is to provide a method for observing the speed and load torque of a traction motor based on Kalman filtering, which addresses the shortcomings of existing methods for observing the speed and load torque of a traction motor. The Kalman filtering method is applied to the traction motor system as an equivalent linear system, which can achieve optimal filtering.

[0010] To achieve the aforementioned objectives, this invention provides a Kalman filter-based speed and load torque observer for a traction motor. The traction motor can be either a permanent magnet synchronous motor or an induction asynchronous motor, wherein:

[0011] 1) For permanent magnet synchronous motors, the speed and load torque observers of traction motors based on Kalman filtering are represented by the following space equations:

[0012] x k =Ax k-1 +Bu k-1 +wk-1

[0013] z k =Hx k +v k

[0014]

[0015] Where, x k Let x represent the current state matrix. k-1 Let ω represent the state matrix of the previous period. m (k) represents the current angular velocity of the traction system, ω m (k-1) represents the angular velocity of the traction system in the previous cycle, T L (k) represents the current load torque of the traction system, T L (k-1) represents the load torque of the traction system in the previous cycle, A represents the state transition matrix, and B is the input matrix. μ T is the damping coefficient of the traction system. s Where J is the sampling period, J is the moment of inertia on the shaft side of the traction system, and p is the number of pole pairs. It is a permanent magnet flux linkage, z k H represents the measurement matrix, and H represents the observation matrix;

[0016] 2) For induction asynchronous motors, the speed and load torque observers of traction motors based on Kalman filtering are represented by the following space equations:

[0017] x k =Ax k-1 +Bu k-1 +w k-1

[0018] z k =Hx k +v k

[0019]

[0020] Where, x k Let x represent the current state matrix. k-1 Let ω represent the state matrix of the previous period. m (k) represents the current angular velocity of the traction system, ω m (k-1) represents the angular velocity of the traction system in the previous cycle, T L (k) represents the current load torque of the traction system, T L (k-1) represents the load torque of the traction system in the previous cycle, A represents the state transition matrix, and B is the input matrix. μ T is the damping coefficient of the traction system. sWhere J is the sampling period, p is the moment of inertia on the shaft side of the traction system, n is the reduction ratio, and L is the sampling period. m It is equivalent mutual inductance, L r It is the equivalent inductance on the rotor side, i d The no-load current at rated speed when speed is adjusted below the base frequency is used as the direct-axis command current; z k H represents the measurement matrix, and H represents the observation matrix.

[0021] A method for observing the speed and load torque of a traction motor based on Kalman filtering includes the following steps:

[0022] 1) Initialization steps

[0023] Initialize according to the following equation: ω m (0)=0, i q (0) = 0, T L (0) = k, P0 = I, where: ω m (0) represents the angular velocity of the traction system at initialization, i q (0) is the quadrature-axis current at initialization, T L (0) is the load torque of the traction system at initialization, P0 is the covariance of the 2×2 initial prior estimation error, k is the initial weighing value, and I is the 2×2 identity matrix.

[0024] The covariance matrices Q and R for the estimated noise and the measured noise are both 2×2 matrices, determined based on the actual system model and sensor characteristics;

[0025] When k=1, the first iteration is performed to calculate the prior estimate. in This represents the prior estimate of the first iteration, which is a 2×1 matrix;

[0026] Calculate the initial prior error covariance matrix

[0027] in, Let A denote the covariance matrix of the prior estimation error in the first iteration, and let A denote the state transition matrix. T Let A represent the transpose of the state transition matrix, Q represent the covariance matrix of the process noise, and A and Q represent the state transition matrix. Both are 2×2 matrices;

[0028] Calculate the initial Kalman gain.

[0029] Where K1 represents the initial Kalman gain, H TR represents the transpose of the observation matrix, and R represents the covariance matrix of the measurement noise.

[0030] K1 is a 2×2 matrix;

[0031] 2) Prediction Steps

[0032] Calculate the priori estimated values.

[0033] in: This represents the current prior estimate. u represents the posterior estimate of the previous period. k-1 Let A represent the input value of the previous cycle, and B represent the state transition matrix. Both B and B are 2×1 matrices;

[0034] When using a permanent magnet synchronous motor

[0035] Among them, B μ T represents the damping coefficient of the traction system. s Where J is the sampling period, J is the moment of inertia on the shaft side of the traction system, and p is the number of pole pairs. It is a permanent magnet flux chain;

[0036] When using an induction asynchronous motor

[0037] Among them, B μ T represents the damping coefficient of the traction system. s Where J is the sampling period, p is the moment of inertia on the shaft side of the traction system, n is the reduction ratio, and L is the sampling period. m It is equivalent mutual inductance, L r It is the equivalent inductance on the rotor side, i d It generally represents the no-load current at rated speed, and is used as the direct shaft command current.

[0038] Calculate the prior error covariance matrix

[0039] in, Let P represent the current prior estimate error covariance matrix. k-1 This represents the covariance matrix of the posterior estimation error of the previous period. and P k-1 Both are 2×2 matrices;

[0040] 3) Calibration steps

[0041] Calculate Kalman gain

[0042] Among them, K k K represents the Kalman gain. k It is a 2×2 matrix, and the operation of inverting it is not complicated;

[0043] Calculate the posterior estimate

[0044] in, Let Z represent the current posterior estimation matrix. k Represents the current measurement matrix. It is a 2×1 matrix. The residual matrix;

[0045] Update the posterior estimation error covariance matrix

[0046] Among them, P k Let P be the covariance matrix representing the current posterior estimation error. k It is a 2×2 matrix;

[0047] When entering the next iteration cycle, the matrix with index k in the current cycle will be changed to k-1 in the next cycle.

[0048] 4) Convergence Steps

[0049] Set a convergence time t, and obtain the result after reaching that convergence time t. The members in the value are the observed values ​​of motor speed and load torque, which are used for speed control and feedforward compensation preset torque current during elevator system startup, respectively.

[0050] Because of the above technical solution, during elevator system startup, after the traction mechanism brake opens, there is a period of zero-speed maintenance. Due to the use of low-precision speed sensors, such as ABZ incremental speed sensors, compared to analog signal input SINCOS sensors, at very low speeds, the AB pulse signal is not obtained per unit speed measurement cycle. The typical approach is to use the speed information from the previous state, which is essentially a delayed filter, resulting in inaccurate speed measurement. This leads to a deviation between the control input given by the controller and the actual required control input, causing swaying during startup. The speed observer essentially compensates for the accuracy of the speed sensor at very low speeds, correcting the speed control input.

[0051] The correction of load torque is also reflected during startup. The controller provides a feedforward torque based on the weighing value fed back from the car's weighing switch. This reduces the burden on feedback control during startup, allowing the system to follow the reference control value more quickly. When the weighing switch (usually a magnetic induction switch) experiences signal drift, causing a significant deviation between the weighing value and the actual value, it affects the feedforward preset torque, impacting the startup process. The load torque observer essentially compensates for this torque in such situations, correcting the feedforward preset torque. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the AB phase of the ABZ incremental speed sensor and the SIN and COS phases of the SINCOS sensor.

[0053] Figure 2 This is a schematic diagram of the equivalent model of an induction asynchronous motor.

[0054] Figure 3 For i sd L m and A diagram illustrating the relationship between the changes.

[0055] Figure 4 This is a flowchart illustrating the method for observing the speed and load torque of a traction motor based on Kalman filtering. Detailed Implementation

[0056] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0057] 1. A brief description of the construction of the Kalman filter model

[0058] The purpose of the Kalman filter is to find the optimal estimate of the system state vector at time k, given the basic mathematical model of the system and the state observation vector at time k, such that the posterior covariance matrix is ​​minimized (i.e., the trace of the matrix is ​​minimized, thus minimizing the variance, and consequently minimizing the posterior estimated error). In other words, it seeks a suitable Kalman gain so that the state estimate approximates the true value. This will be explained in detail below.

[0059] The standard form of the state-space equation of a Kalman filter is:

[0060] x k =Ax k-1 +Bu k-1 +w k-1

[0061] z k =Hx k +v k

[0062] Where: A is the state transition matrix, x is the state variable, and xi is the state variable. k Let x be the current state variable matrix. k-1 Let B be the state variable matrix of the previous period, B be the input matrix, and u be the input signal matrix. k-1 Let w be the previous input signal matrix, and w be the process noise matrix introduced by the control system itself, which follows a Gaussian distribution with expectation of 0 and covariance of Q, i.e., w ~ P(0, Q). k-1 Let z be the noise matrix of the previous cycle, and z be the actual sensor measurement value. k For this sensor measurement, H is the observation matrix, and v is the measurement noise, which follows a Gaussian distribution with expectation of 0 and covariance of R, i.e., v ~ P(0, R). k This is the measurement noise matrix for the previous period;

[0063] Let the state variable matrix x be n×1 dimensional and the sensor measurement matrix z be m×1 dimensional. Then A is n×n dimensional, B is n×n dimensional, w is n×1 dimensional, H is m×n dimensional and v is m×1 dimensional.

[0064] Taking the model described in this invention as a two-dimensional state-space system as an example, then

[0065]

[0066] Define the following estimates, error matrix, and covariance:

[0067] Prior Estimate Based on the ideal system model, there is no noise influence, and the estimated value is generated from the previous cycle's iteration;

[0068] Posteriori Estimate Estimates based on current measurement results;

[0069] Priority Error Estimate in It follows the expected value of 0 and the covariance is The Gaussian distribution, i.e.

[0070] Posterior error estimation in It follows a pattern with expected value of 0 and covariance of P. k The Gaussian distribution, i.e.:

[0071] Covariance of prior estimation error

[0072] Covariance P of prior estimation error k :

[0073] As stated above,

[0074]

[0075] because It follows a pattern with expected value of 0 and covariance of P. k Gaussian distribution, variance because but That is, want To minimize it, we need to minimize its variance, i.e., P. k The trace is the smallest, that is...

[0076] In the description of the uncertainty relationship above,

[0077] Q represents the modeling error, which is related to the linearity of the computer's error model and the error introduced by discretization. Input errors are also taken into account here.

[0078] The selection of R depends on the characteristics of the sensor. If the initial values ​​of P0, Q, and R cannot be obtained precisely, and only the possible ranges are known, then the larger possible values ​​should be conservatively adopted. If the accurate prior information of Q, R, and P0 is unknown, the value of Q should be appropriately increased to increase the weight of utilizing the real-time measurement values.

[0079] The calculation of prior and posterior estimates requires following these steps:

[0080] A. Based on the previous posterior estimate Calculate the prior estimate at the current time, i.e.

[0081]

[0082] B. Prior estimates of the measurement results

[0083] C. Calculate the posterior estimate at the current time step. Let G = K k H, then And Kk ∈[0, H -1 ], Let G be the residual matrix. When G = 0, i.e., K... k When = 0, This means that the measurement noise is high at this point, so we choose to trust the prior estimate more; when G=1, i.e., K k =H -1 hour, If the calculated error is large at this point, we choose to trust the measured value. Therefore, the objective is to find the Kalman gain K. k To minimize the error, i.e. (Actual value). Therefore, as mentioned above, K k The value of is related to the error, from the covariance matrix P of the error. k Seeking K k .

[0084]

[0085] z k =Hx k +v k Substitution

[0086]

[0087] Expand

[0088]

[0089] Will Substitution

[0090]

[0091] because and v k They are mutually independent and their expected value is 0, that is... E[v k T ] = E[v k If ] = 0, then according to the properties of expectation, the expectation of a real matrix is ​​itself, and we have

[0092]

[0093] Then P k The equation is transformed into

[0094]

[0095] Similarly, perform the expansion of mutually independent multiplicative matrix variables, transforming the real matrix IK... k H and K k And its transpose is proposed. and v k v k T These are covariance matrices, which are not independent and cannot be separated.

[0096]

[0097] Expand

[0098]

[0099] Open by transposing.

[0100]

[0101]

[0102] As mentioned earlier, to minimize the error, i.e., to minimize the variance of the error, P... k The trace is the smallest.

[0103] The objective is then min tr(P) k )

[0104]

[0105] For tr(P) k Please provide information about K. k The derivative,

[0106]

[0107] because R and R are real symmetric matrices, therefore R = R T ,but

[0108]

[0109] To find the extreme value, let get

[0110]

[0111] To determine concavity / convexity, tr(P) k Find the second derivative, and we get

[0112] ( (It is a real symmetric matrix)

[0113] From the above,

[0114] In this method, v1 represents the measurement noise of the speed sensor, and v2 represents the measurement noise of the weighing sensor.

[0115] The above description is explained in detail in the following text, please refer to the following text.

[0116] variance Since E[v1] = 0, then Similarly, Generally, v1 and v2 belong to two independent measurement systems with low correlation. Therefore, the autocorrelation is much greater than the cross-correlation, and this method is based on this assumption. That is, the sequential principal minor of R is greater than 0. In this method, H is the identity matrix, a real symmetric square matrix. Both are real symmetric square matrices

[0117] Following the same analysis method as above

[0118]

[0119] In this method, v1 represents the prior estimation error of speed, and v2 represents the prior estimation error of weighing.

[0120] The above description is explained in detail in the following text, please refer to the following text.

[0121] variance because but Similarly, Generally speaking, and These are two independent estimation systems with low correlation, so the autocorrelation will be much greater than the cross-correlation. This method is based on this assumption. Right now The order of principal minors is greater than 0. This is a linearity property of linear algebra. The principal minors of the matrix formed by adding R and R must also be greater than 0. These conditions satisfy the necessary and sufficient conditions for a positive definite matrix. Based on the relationship between positive definiteness, positive semi-definiteness, and the concavity / convexity of functions, if... As tr(P) k If the Hessian matrix of is positive definite, then tr(P) k ) is a strongly convex function, that is, for tr(P) k Please provide information about K. k The derivative of K is 0, thus K is obtained. k tr(P) time kThe value is the minimum, which means that the variance of the error is minimized.

[0122] In K k In this formula, only It is unknown, so it needs to be based on The definition is Please find out.

[0123]

[0124] x k and Substituting the definition into x, we get x k =Ax k-1 +Bu k-1 +w k-1 and get

[0125]

[0126] Open the transpose, and you will get

[0127]

[0128] Perform an expansion of the independent multiplicative matrix variables to extract the real matrix A. and w k-1 If they are mutually independent and their expected values ​​are all 0, then The above formula can be rearranged to obtain

[0129]

[0130] Thus, the iterative relationship between the prior estimated covariance matrix and the previous period's posterior estimated covariance matrix is ​​obtained and substituted into... In this process, the relationship between the posterior estimated covariance matrix and the prior estimated covariance matrix is ​​obtained.

[0131]

[0132] The above steps can guarantee finding the optimal Kalman gain K. k Used to estimate state variables And iterative calculations are performed.

[0133] The following section will introduce its practical application in traction motors.

[0134] When using a permanent magnet synchronous motor

[0135] For a permanent magnet synchronous motor, its electromagnetic torque equation is:

[0136]

[0137] Where p is the pole logarithm, i d For direct-axis flux linkage, i is quadrature-axis flux linkage. d For direct-axis current, i q Quadrature axis current

[0138] Will

[0139]

[0140] in, It is a permanent magnet flux linkage, L d It is a direct-axis inductor, L q It is a quadrature-axis inductance; substituting into the above equation, we get

[0141]

[0142] For surface-mounted permanent magnet synchronous motors

[0143] L d =L q

[0144] but

[0145]

[0146] The mechanical equation of the traction system is

[0147]

[0148] Among them, T L For the load torque of the traction system, ω m Let J be the angular velocity of the traction system, J be the moment of inertia on the shaft side of the traction system, and B be the moment of inertia. μ Let be the damping coefficient of the traction system. This coefficient can be obtained through two uniform motions at different speeds. Specifically, at the same position and with the same load, at two uniform speeds of ω1 and ω2 respectively, assuming the driving force direction is opposite to the load direction, then...

[0149] T e1 -(T L +B μ ω1)=T e2 -(T L +B μ ω2)

[0150] Among them, T e1 and T e2 It is the average of the driving forces at two different speeds, which gives us

[0151]

[0152] 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

[0153] The above mechanical equations can then be written in state-space equation form.

[0154]

[0155] Backward differential discretization yields

[0156]

[0157] Where k is the current value, k-1 is the previous state value, and T s The sampling period.

[0158] Organized

[0159]

[0160]

[0161] The standard form of the state-space equations of the corresponding Kalman filter

[0162] x k =Ax k-1 +Bu k-1

[0163] z k =Hx k

[0164] in,

[0165]

[0166] When using an induction asynchronous motor, see [link to relevant documentation]. Figure 2 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 In the middle, R s R is the stator resistance. r For equivalence

[0167] Rotor resistance to the stator side, L 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.

[0168] Its electromagnetic torque equation is:

[0169]

[0170] Where i sd To represent the direct-axis current equivalent to the stator side, when regulating speed below the fundamental frequency, the current at the no-load rated speed is generally used as the direct-axis current. During field weakening speed regulation, L... m and i sd It changes with the operating point. For example... Figure 3 As shown, when the output voltage reaches the limit value to maintain constant power output, magnetic weakening and speed boosting are required. sd and L m It will be based on the magnet chain Adjust according to the curve as the value decreases. 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:

[0171] T e 'Ω'=T e Ω

[0172]

[0173] 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.

[0174] 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 This refers to the load torque at the traction end of the asynchronous motor. Similarly, it is assumed that within one sampling period dt, the load torque belongs to a large inertial system, with an inertial lag constant much greater than dt, T. L Remain unchanged, that is

[0175] It can be written in state-space equation form.

[0176]

[0177] Backward differential discretization yields

[0178]

[0179] Where k is the current value, k-1 is the previous state value, and T s The sampling period.

[0180] Organized

[0181]

[0182] The standard form of the state-space equations of the corresponding Kalman filter

[0183] x k =Ax k-1 +Bu k-1

[0184] z k =Hx k

[0185] in,

[0186]

[0187] See Figure 4 The present invention provides a method for observing the speed and load torque of a traction motor based on Kalman filtering, which consists of four main steps: initialization, prediction, correction, and convergence.

[0188] 4.1 Initialization Steps

[0189] 4.1.1 Initialization

[0190] ω m (0)=0, i q (0)=0,T L (0) = k, P0 = I

[0191] Where P0 is a 2×2 identity matrix and k is the initial weighing value.

[0192] The covariance matrices Q and R for estimated noise and measured noise are both 2×2 matrices, determined based on the actual system model and sensor characteristics.

[0193] 4.1.2 When k=1, perform the first iteration to calculate the prior estimate. in A 2×1 matrix

[0194] 4.1.3 Calculate the initial prior error covariance matrix

[0195]

[0196] Among them, A and Both are 2×2 matrices

[0197] 4.1.4 Calculate the initial Kalman gain.

[0198]

[0199] in, K1 is a 2×2 matrix

[0200] 4.2 Prediction Steps

[0201] 4.2.1 Calculate Priori Estimated Values

[0202]

[0203] in Both B and B are 2×1 matrices

[0204] When a permanent magnet synchronous motor is used,

[0205]

[0206] When using an induction asynchronous motor

[0207]

[0208] 4.2.2 Calculate the prior error covariance matrix

[0209]

[0210] in, and P k-1 Both are 2×2 matrices

[0211] 4.3 Calibration Steps

[0212] 4.3.1 Calculate Kalman gain

[0213]

[0214] Among them, K k It is a 2×2 matrix, and the operation of inverting it is not complicated.

[0215] 4.3.2 Calculate the posterior estimate

[0216]

[0217] in, It is a 2×1 matrix. For residuals

[0218] 4.3.3 Update the posterior estimation error covariance matrix

[0219]

[0220] Among them, P k It is a 2×1 matrix

[0221] 4.3.4 Enter the next cycle (step 4.2) iteration. The matrix with index k in the current cycle is changed to k-1 after entering the next cycle;

[0222] 4.4 Convergence Steps

[0223] Set a convergence time t, and obtain the result after reaching that convergence time t. The members in the value are the observed values ​​of motor speed and load torque, which are used for speed control and feedforward compensation preset torque current during elevator system startup, respectively.

[0224] In this invention, during elevator system startup, after the traction mechanism brake is engaged, there is a period of zero-speed maintenance. Due to the use of low-precision speed sensors, such as ABZ incremental speed sensors, compared to analog signal input SINCOS sensors, AB pulse signals are not obtained per unit speed measurement cycle at very low speeds. The typical approach is to reuse speed information from the previous state, which is essentially a delayed filter, resulting in inaccurate speed measurement. This leads to a deviation between the speed control input from the controller and the actual required control input, causing swaying during startup. The speed observer essentially compensates for the accuracy of the speed sensor at very low speeds, correcting the speed control input.

[0225] The correction of load torque is also reflected during startup. The controller provides a feedforward torque based on the weighing value fed back from the car's weighing switch. This reduces the burden on feedback control during startup, allowing the system to follow the reference control value more quickly. When the weighing switch (usually a magnetic induction switch) experiences signal drift, causing a significant deviation between the weighing value and the actual value, it affects the feedforward preset torque, impacting the startup process. The load torque observer essentially compensates for this torque in such situations, correcting the feedforward preset torque.

Claims

1. A method for observing the speed and load torque of a traction motor based on Kalman filtering, characterized in that, Includes the following steps: 1) Initialization steps Initialize according to the following equation: ,in: The angular velocity of the traction system at initialization. The quadrature-axis current at initialization. The load torque of the traction system at initialization. for The covariance of the initial prior estimation error, To initialize the weighing value, for The identity matrix; For the covariance matrices Q and R of the estimated noise and the measured noise, respectively, both are... The values ​​of the members of the matrix are determined based on the actual system model and sensor characteristics; When k=1, the first iteration is performed to calculate the prior estimate. ,in Let represent the prior estimate of the first iteration, . Matrix; Calculate the initial prior error covariance matrix in, The covariance matrix represents the prior estimation error of the first iteration. Represents the state transition matrix. This represents the transpose of the state transition matrix. The covariance matrix A and denoted by A represent the process noise. Both are 2×2 matrices; Calculate the initial Kalman gain. ,in, Indicates the initial Kalman gain. This represents the transpose of the observation matrix. The covariance matrix represents the measurement noise. ； 2) Prediction Steps Calculate the priori estimated values. ,in This represents the current prior estimate. This represents the posterior estimate from the previous period. Let A represent the input value of the previous cycle, and B represent the state transition matrix. , and All Matrix; When using a permanent magnet synchronous motor ,in, This represents the damping coefficient of the traction system. The sampling period is The moment of inertia is the rotational inertia on the shaft side of the traction system. For extreme logarithms, It is a permanent magnet flux chain; When using an induction asynchronous motor ,in, This represents the damping coefficient of the traction system. The sampling period is The moment of inertia is the rotational inertia on the shaft side of the traction system. For extreme logarithms, The reduction ratio, It is equivalent mutual inductance. It is the equivalent inductance on the rotor side. The no-load current at rated speed is used as the direct-axis command current. Calculate the prior error covariance matrix ,in, This represents the current prior estimate error covariance matrix. This represents the covariance matrix of the posterior estimation error of the previous period. and ; 3) Calibration steps Calculate Kalman gain ,in, Indicates Kalman gain, ; Calculate the posterior estimate ,in, Denotes the current posterior estimation matrix. Represents the current measurement matrix. It is a 2×1 matrix. The residual matrix; Update the posterior estimation error covariance matrix ,in, The covariance matrix represents the current posterior estimation error. It is a 2×2 matrix; When entering the next iteration cycle, the matrix with index k in the current cycle will be changed to k-1 in the next cycle. 4) Convergence Steps Set a convergence time The convergence time is reached. The result obtained later The members in the value are the observed values ​​of motor speed and load torque, which are used for speed control and feedforward compensation preset torque current during elevator system startup, respectively.

Images