Thermal state space time-varying parameter estimation method based on mechanism data hybrid drive

Through the time-varying parameter estimation method based on the hybrid drive of mechanism data, the forgetting factor least squares method of discrete transfer function is improved, and the known and unknown coefficients are separated, the problems of complexity and computational complexity of thermal equipment are solved, and the state space is realized Accurate time-varying parameter estimation and improvement of computational efficiency of the model.

CN119026331APending Publication Date: 2024-11-26LIYANG RES INST OF SOUTHEAST UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411039055.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

In modern industry, the thermal equipment model is complex and nonlinear, resulting in a large deviation from the state space model and the actual output. The existing technology has shortcomings in reducing the computational complexity, especially in the determination of the system state space model.

Method used

Using the time-varying parameter estimation method based on mechanical data hybrid drive, the forgetting factor least squares estimation calculation method of discrete transfer function is improved, known and unknown coefficient vectors are separated, the estimation calculation method is simplified, and the highest order requirement of excitation signals is reduced, and the state space is realized. Time-varying parameter estimates of the model.

Benefits of technology

It effectively reduces the errors of the state space model and device output, simplifies the complexity of the estimation algorithm, reduces the highest order requirements of the excitation signal, and improves the accuracy and calculation efficiency of parameter estimation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119026331A_ABST
    Figure CN119026331A_ABST
Patent Text Reader

Abstract

The invention discloses a thermotechnical state space time-varying parameter estimation method based on mechanism data hybrid drive, which comprises the following steps: analyzing the working characteristics and working principle of a model, selecting proper state space identification model input and output variables, and obtaining a state space model based on a mechanism model; obtaining a discrete state space model through state space system discretization; solving a coefficient of the discrete transfer function model by adopting a recursive least square method with a forgetting factor; and solving a time-varying parameter in the state space model through the estimated coefficient vector. According to the method, a forgetting factor least square estimation algorithm of a discrete transfer function is improved, and time-varying parameter estimation of a state space model is realized; meanwhile, by separating known and unknown coefficient vectors, the number of to-be-estimated parameters is reduced, the complexity of an estimation algorithm is simplified, and the highest order requirement of an excitation signal is reduced.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of industrial control, and in particular relates to a method for estimating time-varying parameters of thermal state space based on mechanism data hybrid drive. Background Art

[0002] In modern industrial systems, thermal equipment models have become the basis for system optimization and control. Generally, equipment models can be modularized, but they have the characteristics of multivariable, strong coupling and nonlinearity. The model building process is relatively complicated, and there is a large difference between the modeling output and the actual output. Therefore, in order to facilitate subsequent control, establishing a mechanism-data dual-driven identification model has become a key task.

[0003] However, as the operating conditions change, the deviation between the actual nonlinear mechanism model and the identified state space model increases, and the output value predicted by the constant state space model also deviates. Therefore, this patent takes typical thermal equipment and heat exchangers as examples, and proposes a state space model time-varying parameter estimation method based on the forgetting factor recursive least squares method (RLS, Recursive Least Squares) to update the state space parameters and reduce the error between the state space model and the equipment output.

[0004] Invention patent CN202310665747.8 uses the least squares support vector machine method based on the support vector machine (SVM) for parameter estimation of the battery life distribution model, which reduces the difficulty of calculation and improves the operation speed. It is widely used in parameter estimation. Invention patent CN202111554956.2 introduces a moment of inertia identification method based on the recursive least squares method with a forgetting factor, which improves the system control performance of the motor under conditions such as changes in the load moment of inertia. However, the above patents are aimed at specific physical models, and none of them considers the determined quantities in the state space model of the system from the mechanism, so they are all insufficient in reducing the computational complexity. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a method for estimating time-varying parameters in thermal state space based on hybrid drive of mechanism data. The forgetting factor least squares estimation algorithm of discrete transfer function is improved to realize the time-varying parameter estimation of state space model; at the same time, by separating known and unknown coefficient vectors, the number of parameters to be estimated is reduced, the complexity of the estimation algorithm is simplified, and the highest order requirement of the excitation signal is reduced. Technical solution: A method for estimating time-varying parameters in thermal state space based on hybrid drive of mechanism data of the present invention comprises the following steps:

[0006] Step 1: Analyze the working characteristics and working principle of the model, select appropriate input and output variables of the state space identification model, and obtain the state space model based on the mechanism model;

[0007] Step 2: discretize the state space system to obtain a discrete state space model;

[0008] Step 3, using the recursive least square method with forgetting factor to solve the coefficients of the discrete transfer function model;

[0009] Step 4: Solve for the time-varying parameters in the state-space model using the estimated coefficient vector.

[0010] Furthermore, step 1 is specifically as follows: analyzing the working characteristics and working principle of the thermal standby heat exchanger, the temperature equations of the hot fluid wall and the cold fluid are:

[0011]

[0012] Where T is the temperature of the hot and cold streams; V is the volume of the hot and cold streams of the heat exchanger; m is the mass flow rate of the hot and cold streams; c p is the specific heat capacity of the stream; ρ is the density of the hot and cold streams; A is the heat transfer area of ​​the subsegment; K is the heat transfer coefficient of the subsegment, subscripts h, w and c represent the hot fluid, wall and cold fluid respectively, i and i-1 represent the current subsegment and the previous subsegment of the fluid or wall respectively, then the state space equations of the fluid and the wall are constructed as follows:

[0013] in

[0014] According to the heat transfer equation, the state space coefficient term is obtained as:

[0015]

[0016] Taking the temperature of the i-1 section of the hot and cold fluids in the heat exchanger as the input, the temperature of the hot fluid, wall and cold fluid in the i-th section as the state quantity, and the temperature of the hot fluid and cold fluid in the i-th section as the output, the mechanism model is converted into a state space model. After bringing in the structural parameters and physical property parameters, the following two-input and two-output system state space model is obtained:

[0017]

[0018] Change the input variables within the constraints, observe the derived state space coefficient matrix, and select the parameters with more obvious changes as the parameters to be estimated according to actual needs; other constants or coefficients with little change are fixed as known parameters to form a state space gray box model;

[0019] Taking the change of the physical properties of cold fluid as an example, A(3,2) and A(3,3) are selected as typical variables to be estimated, and time-varying parameter estimation is performed to obtain the continuous-time state-space model in the following form:

[0020]

[0021] y(t)=C(t)x(t)+D(t)u(t)

[0022] Furthermore, step 2 is specifically as follows: by discretizing the state space system, a discrete state space model of the following form is obtained:

[0023] x((k+1)T)=G(kT)x(kT)+H(kT)u(kT)

[0024] y(kT)=C(kT)x(kT)+D(kT)u(kT).

[0025] Assume that the input u is obtained by sampling the zero-order holder, take the differential equation as the general solution of the first-order ordinary differential equation, take (k+1)T and kT as the upper limit of integration, and solve to obtain:

[0026]

[0027] The above formula is processed by difference to become:

[0028] x((k+1)T)=e AT x(kT)+A -1 (e AT -I)Bu(kT)

[0029] The coefficient term of the discrete state space is defined as:

[0030] G=e AT

[0031] H=A -1 (e AT -I)B

[0032] The discretized state space format is obtained.

[0033] x((k+1)T)=G(kT)x(kT)+H(kT)u(kT)

[0034] y(kT)=C(kT)x(kT)

[0035] Perform z-transform on the above equation,

[0036] X(z)=(zI-G) -1 HU(z)

[0037] Y(z)=C(zI-G) -1 HU(z)

[0038] For the exponential function e of the matrix A ,

[0039]

[0040] Take the first two terms of the discrete form of the exponential function, e A ≈I+A, substituting into the coefficients G and H in the z-transformation formula, we get:

[0041] X(z)=((z-1)I-AT) -1 BTU(z)

[0042] Y(z)=C((z-1)I-AT) -1 BTU(z)

[0043] According to the state space, the discrete transfer function is obtained according to the above steps:

[0044]

[0045] In the formula, u1 and u2 are input variables, y is the output variable, is the time-varying parameter of the state space.

[0046] Furthermore, step 3 is specifically as follows: using a recursive least square method with a forgetting factor to solve the coefficients to be estimated of the discrete transfer function model;

[0047] For the discrete transfer function, taking the first item of the output as an example, it is written as follows:

[0048]

[0049] The basic idea of ​​the least squares method is to transform the system model and convert the parameters in the system state matrix and input matrix into state quantities.

[0050] y(k)=h T (k)θ

[0051] For the equation in the above form, the recursive formula of the least squares method for the forgetting factor is:

[0052]

[0053] in,

[0054]

[0055] When starting the recursive formula, the initial value The value method is

[0056] P(0)=αI

[0057]

[0058] Among them, α is a sufficiently large positive real number, ε is a zero vector or a sufficiently small positive real vector;

[0059] By subtracting the known terms, the order of the excitation signal is reduced, thus accelerating the convergence of the estimated parameters.

[0060] Consider the following ARX model,

[0061] A(z -1 )y(k)=B(z -1 )u(k)

[0062] A(z -1 )=1+a1z -1 +...+a M z -M +a M+1 z -(M+1) +...+a m z -m

[0063] B(z -1 )=b0+b1z -1 +...+b N z -N +b N+1 z -(N+1) +...+b m z -n

[0064] M,N,m,n∈Z, and M≤m,N≤n

[0065] The parameter vector θ of the ARX model is defined as:

[0066] θ T =[a1,a2,...,a M ,a M+1 ,...,a m ,b0,b1,b2...,a N ,a N+1 ,...,b n ]

[0067] Let the estimated parameter vector be The difference between the actual output and the estimated output of the object, that is, the residual ε, is:

[0068]

[0069] Assume that from a1 to a M are the known parameters derived from the mechanism model, from a M+1 to a m is a time-varying parameter to be estimated; suppose it is known that from b0 to b N is a known coefficient, from b N+1 to b nThe unknown and known quantities of the transfer function coefficients are usually not from a1 to a M It is continuous, but the known numbers can be commutatively transferred to the front of all the known terms so that from a1 to a M are all known, and the sequence of the input signal is adjusted;

[0070] The estimated output for the kth observation is:

[0071]

[0072] y0(k) is a known coefficient,

[0073]

[0074] h0 T (k)=[-y(k-1),...,-y(kM),u(k-1),...,u(kN)]

[0075] θ0(k)=[a1(k),a2(k),...,a M (k),b0(k),b1(k),b2(k),...,b N (k)]

[0076] Similarly, the actual output value y1(k) of the model after subtracting the known terms is defined as:

[0077] y1(k)=y(k)-y0(k)=y(k)-h0 T (k)θ0(k)=h1 T (k)θ1(k)

[0078] h1 T (k)=[-y(kM-1),...,-y(km); u(kN-1),...,u(kn)]

[0079] θ1=[a M+1 (k),a M+2 (k),...,a m (k); b N+1 (k),b N+2 (k),...,b n (k)]

[0080] Therefore, we only need to use y1(k) as the new output value to replace y(k) to obtain the optimal estimate of the original transfer function, and the RLS recursive formula of the time-varying parameter is obtained as follows:

[0081]

[0082] Furthermore, step 4 is as follows: solve the time-varying parameters, bring them into the RLS case, and obtain the estimated coefficient vector The coefficient vector It can also be expressed by the time-varying parameters of the state space The overdetermined system of equations is expressed by choosing Equations with larger coefficients form a positive definite system of equations to be solved.

[0083]

[0084] The order of the excitation signal u needs to be greater than or equal to (mM)+(nN)+1 to ensure that the open-loop information identification experiment is sufficient. The full parameter estimation requires at least m+n+1 order excitation.

[0085] The present invention also discloses a computer device, comprising a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method of the present invention.

[0086] The present invention also discloses a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the steps of the method of the present invention are implemented.

[0087] The present invention also discloses a computer program product, comprising a computer program / instruction, which implements the steps of the method of the present invention when executed by a processor.

[0088] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages:

[0089] The present invention improves the forgetting factor least squares estimation algorithm of discrete transfer function and realizes time-varying parameter estimation of state space model; at the same time, by separating known and unknown coefficient vectors, the number of parameters to be estimated is reduced, the complexity of the estimation algorithm is simplified, and the highest order requirement of the excitation signal is reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 It is a flow chart of the method for estimating time-varying parameters of the state space model of the present invention.

[0091] Figure 2 is the parameter estimation result 1 of the case of the present invention;

[0092] Figure 3 This is the parameter estimation result 2 of the case of the present invention; the dotted line represents the true value of each parameter, the solid line represents the estimated value of each parameter, the dotted line with the solid line overlaps with it represents the time-varying parameter, and the dotted line without overlapping with it represents the known parameter subtracted in the RLS method. DETAILED DESCRIPTION

[0093] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0094] like Figure 1 As shown, step 1 of the present invention takes a typical thermal hot standby heat exchanger as an example to analyze the working characteristics and working principle of the thermal equipment.

[0095] The temperature equations for the hot fluid wall and the cold fluid are:

[0096]

[0097] Where T is the temperature of the hot and cold streams; V is the volume of the hot and cold streams of the heat exchanger; m is the mass flow rate of the hot and cold streams; c p is the specific heat capacity of the stream; ρ is the density of the hot and cold streams; A is the heat transfer area of ​​the sub-segment; K is the heat transfer coefficient of the sub-segment. Subscripts h, w and c represent hot fluid, wall and cold fluid respectively, i and i-1 represent the current sub-segment and the previous sub-segment of the fluid or wall respectively, then the state space equations of the fluid and the wall can be constructed as:

[0098] in

[0099] According to the heat transfer equation, the state space coefficient term is obtained as:

[0100]

[0101] Select appropriate model input and output variables. In this case, the temperature of the i-1 section of the hot and cold fluids in the heat exchanger is used as the input, the temperature of the hot fluid, wall and cold fluid in the i-th section is used as the state quantity, and the temperature of the hot fluid and cold fluid in the i-th section is used as the output. The mechanism model is converted into a state space model. After the structural parameters and physical property parameters are brought in, the following two-input and two-output system state space model is obtained.

[0102]

[0103] Change the input variables within the constraints and observe the derived state space coefficient matrix. According to actual needs, take the parameters with more obvious changes as the parameters to be estimated; other constants or coefficients with little change are fixed as known parameters. Construct the state space gray box model.

[0104] This case takes the change of the physical properties of cold fluid as an example, selects A(3,2) and A(3,3) as typical variables to be estimated, and performs time-varying parameter estimation. The continuous-time state-space model in the following form is obtained:

[0105]

[0106] y(t)=C(t)x(t)+D(t)u(t)

[0107] Step 2, by discretizing the state space system, we get a discrete state space model in the following form:

[0108] x((k+1)T)=G(kT)x(kT)+H(kT)u(kT)

[0109] y(kT)=C(kT)x(kT)+D(kT)u(kT).

[0110] Assuming that the input u is obtained by sampling the zero-order holder, the differential equation is taken as the general solution of the first-order ordinary differential equation, and (k+1)T and kT are taken as the upper limit of integration respectively, and the solution is obtained:

[0111]

[0112] The above formula can be processed by difference to become:

[0113] x((k+1)T)=e AT x(kT)+A -1 (e AT -I)Bu(kT)

[0114] The coefficient term of the discrete state space is defined as:

[0115] G=e AT

[0116] H=A -1 (e AT -I)B

[0117] The discretized state space format is obtained.

[0118] x((k+1)T)=G(kT)x(kT)+H(kT)u(kT)

[0119] y(kT)=C(kT)x(kT)

[0120] Perform z-transform on the above equation,

[0121] X(z)=(zI-G) -1 HU(z)

[0122] Y(z)=C(zI-G) -1 HU(z)

[0123] For the exponential function e of the matrix A ,

[0124]

[0125] Take the first two terms of the discrete form of the exponential function, e A≈I+A, substituting into the coefficients G and H in the z-transformation formula, we get:

[0126] X(z)=((z-1)I-AT) -1 BTU(z)

[0127] Y(z)=C((z-1)I-AT) -1 BTU(z)

[0128] According to the case state space, the discrete transfer function is obtained according to the above steps,

[0129]

[0130] In the formula, u1 and u2 are input variables, y is the output variable, is the time-varying parameter of the state space.

[0131] Step 3: Use the recursive least square method with forgetting factor to solve the coefficients to be estimated of the discrete transfer function model.

[0132] For the discrete transfer function, taking the first term of the output as an example, it can be written as follows:

[0133]

[0134] The basic idea of ​​the least squares method is to transform the system model and convert the parameters in the system state matrix and input matrix into state quantities.

[0135] y(k)=h T (k)θ

[0136] For the equation in the above form, the recursive formula of the least squares method for the forgetting factor is:

[0137]

[0138] in,

[0139]

[0140] When starting the recursive formula, the initial value The value method is

[0141] P(0)=αI

[0142]

[0143] Among them, α is a sufficiently large positive real number, and ε is a zero vector or a sufficiently small positive real vector.

[0144] Considering that some coefficients are known, in order to reduce the computational complexity, the order of the excitation signal can be reduced by subtracting the known terms, thereby accelerating the convergence of the estimated parameters.

[0145] Consider the following ARX model,

[0146] A(z -1 )y(k)=B(z -1 )u(k)

[0147] A(z -1 )=1+a1z -1 +...+a M z -M +a M+1 z -(M+1) +...+a m z -m

[0148] B(z -1 )=b0+b1z -1 +...+b N z -N +b N+1 z -(N+1) +...+b m z -n

[0149] M,N,m,n∈Z, and M≤m,N≤n

[0150] The parameter vector θ of the ARX model is defined as:

[0151] θ T =[a1,a2,...,a M ,a M+1 ,...,a m ,b0,b1,b2...,a N ,a N+1 ,...,b n ]

[0152] Let the estimated parameter vector be The difference between the actual output and the estimated output of the object, that is, the residual ε, is:

[0153]

[0154] Assume that we know from a1 to a M are the known parameters derived from the mechanism model, from a M+1 to a m is a time-varying parameter to be estimated; assuming that N is a known coefficient, from b N+1 to b n is the coefficient to be estimated. In the actual processing process, the unknown and known quantities of the general transfer function coefficients are not from a1 to a MIt is continuous, but the known numbers can be commutatively transferred to the front of all the known terms so that from a1 to a M are all known, and the sequence of the input signal is adjusted without loss of generality.

[0155] The estimated output for the kth observation is:

[0156]

[0157] y0(k) is a known coefficient,

[0158]

[0159] h0 T (k)=[-y(k-1),...,-y(kM),u(k-1),...,u(kN)]

[0160] θ0(k)=[a1(k),a2(k),...,a M (k),b0(k),b1(k),b2(k),...,b N (k)]

[0161] Similarly, the actual output value y1(k) of the model after subtracting the known terms is defined as:

[0162] y1(k)=y(k)-y0(k)=y(k)-h0 T (k)θ0(k)=h1 T (k)θ1(k)

[0163] h1 T (k)=[-y(kM-1),...,-y(km); u(kN-1),...,u(kn)]

[0164] θ1=[a M+1 (k),a M+2 (k),...,a m (k); b N+1 (k),b N+2 (k),...,b n (k)]

[0165] Therefore, we only need to use y1(k) as the new output value to replace y(k) to obtain the optimal estimate of the original transfer function, and the RLS recursive formula of the time-varying parameter is obtained as follows:

[0166]

[0167] Step 4: Solve the time-varying parameters and bring them into the RLS case to obtain the estimated coefficient vector The coefficient vector It can also be expressed by the time-varying parameters of the state space The overdetermined system of equations is expressed by choosing Equations with larger coefficients are solved by forming a system of positive equations to avoid the phenomenon of large numbers eating up small numbers.

[0168]

[0169] The order of the excitation signal u needs to be greater than or equal to (mM)+(nN)+1 to ensure that the open-loop information identification experiment is sufficient. Full parameter estimation requires at least m+n+1 order excitation. The time-varying parameter estimation method proposed in this patent can reduce the order of the excitation signal by M+N.

[0170] In order to test the parameter estimation effect under variable working conditions, the heat transfer coefficient of the cold fluid was changed at 400s to obtain a new set of transfer functions. The estimated results of the coefficients are as follows: Figure 2-3 As shown in the figure. The dotted lines represent the true values ​​of each parameter, the solid lines represent the estimated values ​​of each parameter, the dotted lines that overlap with the solid lines represent the time-varying parameters, and the dotted lines that do not overlap with the solid lines represent the known parameters that are subtracted in the RLS method. It can be observed that these known parameters do not change during the process of changing working conditions. Under the effect of the forgetting factor, the parameter estimation algorithm will reach the next relatively stable result after the next 100 observations.

[0171] Taking the previous 400 estimates as an example, after the 400th estimate is completed, the parameter estimate is

[0172]

[0173] For the case of two unknown parameters, five The overdetermined system of equations, where The two items with smaller coefficient ratios are used for parameter estimation.

[0174]

[0175] Available Original model settings It can be seen that this method has a smaller estimation error.

[0176] Similarly, any parameter in the state space can be estimated by using the state space band variable The transfer function coefficients are obtained by estimating the time-varying parameters of the state-space model, and the The solution of can complete the gray box model estimation of the state space of any unknown coefficients.

Claims

1. A method for estimating time-varying parameters of thermal state space based on mechanism data hybrid drive, characterized in that: The steps include: Step 1: Analyze the working characteristics and working principle of the model, select appropriate input and output variables of the state space identification model, and obtain the state space model based on the mechanism model; Step 2: discretize the state space system to obtain a discrete state space model; Step 3, using the recursive least square method with forgetting factor to solve the coefficients of the discrete transfer function model; Step 4: Solve for the time-varying parameters in the state-space model using the estimated coefficient vector.

2. According to claim 1, a method for estimating time-varying parameters of thermal state space based on mechanism data hybrid drive is characterized in that: Step 1 is as follows: Analyze the working characteristics and working principle of the thermal standby heat exchanger. The temperature equations of the hot fluid wall and the cold fluid are: Where T is the temperature of the hot and cold streams; V is the volume of the hot and cold streams of the heat exchanger; m is the mass flow rate of the hot and cold streams; c p is the specific heat capacity of the stream; ρ is the density of the hot and cold streams; A is the heat transfer area of ​​the subsegment; K is the heat transfer coefficient of the subsegment, subscripts h, w and c represent the hot fluid, wall and cold fluid respectively, i and i-1 represent the current subsegment and the previous subsegment of the fluid or wall respectively, then the state space equations of the fluid and the wall are constructed as follows: in According to the heat transfer equation, the state space coefficient term is obtained as: Taking the temperature of the i-1 section of the hot and cold fluids in the heat exchanger as the input, the temperature of the hot fluid, wall and cold fluid in the i-th section as the state quantity, and the temperature of the hot fluid and cold fluid in the i-th section as the output, the mechanism model is converted into a state space model. After bringing in the structural parameters and physical property parameters, the following two-input and two-output system state space model is obtained: Change the input variables within the constraints, observe the derived state space coefficient matrix, and select the parameters with more obvious changes as the parameters to be estimated according to actual needs; other constants or coefficients with little change are fixed as known parameters to form a state space gray box model; Taking the change of the physical properties of cold fluid as an example, A(3,2) and A(3,3) are selected as typical variables to be estimated, and time-varying parameter estimation is performed to obtain the continuous-time state-space model in the following form: y(t)=C(t)x(t)+D(t)u(t).

3. According to claim 1, a method for estimating time-varying parameters of thermal state space based on mechanism data hybrid drive is characterized in that: Step 2 is as follows: by discretizing the state space system, a discrete state space model is obtained as follows: x((k+1)T)=G(kT)x(kT)+H(kT)u(kT) y(kT)=C(kT)x(kT)+D(kT)u(kT). Assume that the input u is obtained by sampling the zero-order holder, take the differential equation as the general solution of the first-order ordinary differential equation, take (k+1)T and kT as the upper limit of integration, and solve to obtain: The above formula is processed by difference to become: x((k+1)T)=e AT x(kT)+A -1 (this AT -I)Bu(kT) The coefficient term of the discrete state space is defined as: G=e AT H=A -1 (e AT -I)B The discretized state space format is obtained. x((k+1)T)=G(kT)x(kT)+H(kT)u(kT) y(kT)=C(kT)x(kT) Perform z-transform on the above equation, X(z)=(zI-G) -1 HU(z) Y(z)=C(zI-G) -1 HU(z) For the exponential function e of the matrix A , Take the first two terms of the discrete form of the exponential function, e A ≈I+A, substituting the coefficients G and H in the z-transformation formula, we get X(z)=((z-1)I-AT) -1 BTU(z) Y(z)=C((z-1)I-AT) -1 BTU(z) According to the state space, the discrete transfer function is obtained according to the above steps: In the formula, u1 and u2 are input variables, y is the output variable, is the time-varying parameter of the state space.

4. According to claim 1, a method for estimating time-varying parameters of thermal state space based on mechanism data hybrid drive is characterized in that: Step 3 is specifically as follows: using the recursive least square method with a forgetting factor to solve the coefficients to be estimated of the discrete transfer function model; For the discrete transfer function, taking the first item of the output as an example, it is written as follows: The basic idea of ​​the least squares method is to transform the system model and convert the parameters in the system state matrix and input matrix into state quantities. y(k)=h T (k)θ For the equation in the above form, the recursive formula of the least squares method for the forgetting factor is: in, When starting the recursive formula, the initial value K(0), The value method is P(0)=αI Among them, α is a sufficiently large positive real number, ε is a zero vector or a sufficiently small positive real vector; By subtracting the known terms, the order of the excitation signal is reduced, thus accelerating the convergence of the estimated parameters. Consider the following ARX model, A(z -1 )y(k)=B(z -1 )u(k) A(from -1 )=1+a1z -1 +...+a M from -M +a M+1 from -(M+1) +...+a m from -m B(from -1 )=b0+b1z -1 +...+b N from -N +b N+1 from -(N+1) +...+b m from -n M,N,m,n∈Z, and M≤m,N≤n The parameter vector θ of the ARX model is defined as: θ T =[a1,a2,...,a M ,a M+1 ,...,a m ,b0,b1,b2...,a N ,a N+1 ,...,b n ] Let the estimated parameter vector be The difference between the actual output and the estimated output of the object, that is, the residual ε, is: Assume that from a1 to a M are the known parameters derived from the mechanism model, from a M+1 to a m is a time-varying parameter to be estimated; suppose it is known that from b0 to b N is a known coefficient, from b N+1 to b n The unknown and known quantities of the transfer function coefficients are usually not from a1 to a M It is continuous, but the known numbers can be commutatively transferred to the front of all the known terms so that from a1 to a M are all known, and the sequence of the input signal is adjusted; The estimated output for the kth observation is: y0(k) is a known coefficient, h0 T (k)=[-y(k-1),...,-y(kM),u(k-1),...,u(kN)] θ0(k)=[a1(k),a2(k),...,a M (k),b0(k),b1(k),b2(k),...,b N (k)] Similarly, the actual output value y1(k) of the model after subtracting the known terms is defined as: y1(k)=y(k)-y0(k)=y(k)-h0 T (k)θ0(k)=h1 T (k)θ1(k) h1 T (k)=[-y(kM-1),...,-y(km);u(kN-1),...,u(kn)] θ1=[a M+1 (k),the M+2 (k),...,the m (k);b N+1 (k),b N+2 (k),...,b n (k)] Therefore, we only need to use y1(k) as the new output value to replace y(k) to obtain the optimal estimate of the original transfer function, and the RLS recursive formula of the time-varying parameter is obtained as follows:

5. According to claim 1, a method for estimating time-varying parameters of thermal state space based on mechanism data hybrid drive is characterized in that: Step 4 is as follows: solve the time-varying parameters, bring them into the RLS case, and get the estimated coefficient vector The coefficient vector It can also be expressed by the time-varying parameters of the state space The overdetermined system of equations is expressed by choosing Equations with larger coefficients form a positive definite system of equations to be solved. The order of the excitation signal u needs to be greater than or equal to (mM)+(nN)+1 to ensure that the open-loop information identification experiment is sufficient. The full parameter estimation requires at least m+n+1 order excitation.

6. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method of claim 1.

7. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.

8. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to claim 1 are implemented.