Power electronic equipment mechanism data fusion modeling method and device, medium and product

Abstract

The invention relates to a power electronic equipment mechanism data fusion modeling method and device, a medium and a product. Comprising the following steps: acquiring a node voltage equation set of a preset node of a component of the nonlinear system of the power electronic equipment, and solving the node voltage equation set according to a preset condition to obtain a value of a nonlinear state variable of the nonlinear system; constructing a neural network based on a Koopman operator as a first neural network; based on the value of the nonlinear state variable, performing multiple iterative training on the first neural network within a preset time period, and determining a loss value of the first neural network after each training is completed; when the loss value of the determined first neural network is smaller than a preset target loss value after the last training in the preset time period is completed, obtaining a mechanism data fusion model of the nonlinear system according to the first neural network after the multiple iterative training in the preset time period is completed; and efficient and accurate modeling of the power electronic equipment is realized.

Description

Technical Field

[0001] The present application relates to the technical field of power electronic equipment modeling, and in particular to a power electronic equipment mechanism data fusion modeling method, device, medium and product. Background Art

[0002] In the modeling of power electronic equipment, traditional technologies usually use mechanism modeling methods such as analytical formulas, equivalent circuits and finite element calculations. The calculation scale involved in these mechanism modeling methods is too large or even unable to converge. Therefore, it is urgent to propose a new power electronic equipment modeling method. Summary of the invention

[0003] Based on this, it is necessary to provide a power electronic equipment mechanism data fusion modeling method, device, medium and product that can realize efficient and accurate modeling of power electronic equipment in response to the above technical problems.

[0004] In a first aspect, the present application provides a method for fusion modeling of mechanism data of power electronic equipment, wherein the power electronic equipment at least includes a nonlinear system;

[0005] The modeling method comprises:

[0006] Obtaining a node voltage equation group of a preset node of the components of the nonlinear system, and solving the node voltage equation group according to preset conditions to obtain a value of a nonlinear state variable of the nonlinear system;

[0007] Construct a neural network based on the Koopman operator as the first neural network;

[0008] Based on the value of the nonlinear state variable, the first neural network is trained multiple times within a preset period of time, wherein the multiple iterative trainings within the preset period of time are performed sequentially, and the loss value of the first neural network is determined once after each training is completed; and

[0009] After completing the last training within a preset time period, when the determined loss value of the first neural network is less than the preset target loss value, the mechanism data fusion model of the nonlinear system is obtained based on the first neural network that has completed multiple iterative trainings within the preset time period.

[0010] In one embodiment, the first neural network is trained multiple times within a preset period of time based on the value of the nonlinear state variable, wherein the multiple iterative trainings within the preset period of time are performed sequentially, and the loss value of the first neural network is determined once after each training is completed, including:

[0011] Based on the value of the nonlinear state variable, the first neural network that has completed the (i-1)th parameter update is trained for the i-th time, where i is a positive integer and 1<i;

[0012] Determining, based on a preset loss function and according to the value of the nonlinear state variable, an i-th loss value of the first neural network after completing the i-th training;

[0013] According to the i-th loss value, controlling the first neural network that has completed the i-th training to perform the i-th parameter update;

[0014] Based on the value of the nonlinear state variable, the first neural network that has completed the i-th parameter update is trained for the (i+1)th time.

[0015] In one embodiment, the modeling method further comprises:

[0016] Based on the value of the nonlinear state variable, training the first neural network for a first time, wherein the value of the parameter of the first neural network for the first training is a preset initial value;

[0017] Determining a first loss value of the first neural network after completing the first training based on a preset loss function and according to the value of the nonlinear state variable;

[0018] According to the first loss value, the first neural network that has completed the first training is controlled to perform the first parameter update.

[0019] In one embodiment, the preset period includes the 1st to Mth step sub-periods performed sequentially, wherein M is a positive integer, 2≤M, at least two step sub-periods are the same or different, the sum of all step sub-periods is equal to the preset period, each step sub-period is configured with a preset expected loss value, the preset expected loss value of the jth step sub-period is less than the preset expected loss value of the j-1th step sub-period, wherein j is a positive integer, 1<j≤M, and the preset expected loss value of the Mth step sub-period is equal to the preset target loss value;

[0020] The modeling method also includes:

[0021] After completing the last training in the j-1th step sub-period, if the determined loss value of the first neural network is less than the preset expected loss value of the j-1th step sub-period, continue to perform multiple iterative training in the j-1th step sub-period for the first neural network that has completed multiple iterative training in the j-1th step sub-period, wherein the multiple iterative training in the step sub-period is performed sequentially;

[0022] After completing the last training in the j-1th step sub-period, if the determined loss value of the first neural network is greater than or equal to the preset expected loss value of the j-1th step sub-period, the training is exited.

[0023] In one embodiment, the step of training the first neural network after the i-1th parameter update based on the value of the nonlinear state variable comprises:

[0024] Mapping the value of the nonlinear state variable from the nonlinear space to the linear space through the encoder to obtain the value of the linear state variable;

[0025] Based on the general solution of the node voltage equations, and according to the Koopman eigenvalue and the value of the linear state variable, the predicted linear state variable value is obtained by dynamic evolution in the linear space;

[0026] The predicted linear state variable value is mapped from the linear space back to the nonlinear space through the decoder.

[0027] In one of the embodiments, the first neural network further includes an auxiliary neural network;

[0028] The modeling method also includes:

[0029] Based on the value of the linear state variable, the Koopman eigenvalue is determined by the auxiliary neural network.

[0030] In one embodiment, the first neural network includes an encoder and a decoder, wherein the encoder is an encoder based on a Koopman characteristic function, and the decoder corresponds to the encoder;

[0031] Determining, based on a preset loss function and according to the value of the nonlinear state variable, an i-th loss value of the first neural network after completing the i-th training, comprising:

[0032] Determining, based on a preset autoregressive error function and according to the value of the nonlinear state variable, an i-th autoregressive error of the first neural network after completing the i-th training;

[0033] Determining, based on a preset prediction error function and according to the value of the nonlinear state variable, an i-th prediction error of the first neural network after completing the i-th training;

[0034] Determining, based on a preset linear error function and according to the value of the nonlinear state variable, an i-th linear error of the first neural network after completing the i-th training;

[0035] According to the i-th autoregressive error and its corresponding preset first weighting coefficient, the i-th prediction error and its corresponding preset second weighting coefficient, and the i-th linear error and its corresponding preset third weighting coefficient, determine the i-th loss value of the first neural network after completing the i-th training.

[0036] In a second aspect, the present application also provides a power electronic equipment mechanism data fusion modeling device, wherein the power electronic equipment at least includes a nonlinear system;

[0037] The modeling device comprises:

[0038] A nonlinearization module, used to obtain a node voltage equation group of a preset node of the components of the nonlinear system, and solve the node voltage equation group according to preset conditions to obtain a value of a nonlinear state variable of the nonlinear system;

[0039] A neural network module is used to construct a neural network based on the Koopman operator as the first neural network;

[0040] A network training module, configured to perform multiple iterative training on the first neural network within a preset time period based on the value of the nonlinear state variable, wherein the multiple iterative training within the preset time period is performed sequentially, and a loss value of the first neural network is determined once after each training is completed;

[0041] A model building module is used to obtain a mechanism data fusion model of the nonlinear system based on the first neural network that has completed multiple iterations of training within a preset time period, when the loss value of the first neural network determined after the last training within a preset time period is less than a preset target loss value.

[0042] In a third aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0043] In a fourth aspect, the present application further provides a computer program product, comprising a computer program, which implements the steps of the method described in the first aspect when executed by a processor.

[0044] The above-mentioned power electronic equipment mechanism data fusion modeling method, device, medium and product, the power electronic equipment at least includes a nonlinear system, by obtaining the node voltage equation group of the preset nodes of the components of the nonlinear system, and solving the node voltage equation group according to preset conditions, the value of the nonlinear state variable of the nonlinear system is obtained; then, a neural network based on the Koopman operator is constructed as the first neural network; further, based on the value of the nonlinear state variable, the first neural network is subjected to multiple iterative training within a preset time period, wherein the multiple iterative training within the preset time period is performed sequentially, and the loss value of the first neural network is determined once after each training is completed; finally, when the loss value of the first neural network determined after the last training within the preset time period is less than the preset target loss value, the mechanism data fusion model of the nonlinear system is obtained according to the first neural network that has completed the multiple iterative training within the preset time period, thereby realizing the mechanism data fusion modeling of the power electronic equipment.

[0045] In this way, the embodiment of the present application integrates the mechanism relationship of the nonlinear system of the power electronic equipment into the data modeling of the nonlinear system through a deep learning method based on the Koopman operator, thereby establishing a mechanism data fusion model of the nonlinear system, that is, the pure mechanism model of the nonlinear system is replaced by the mechanism data fusion model, which not only makes full use of the efficiency and accuracy of data modeling, realizes efficient and accurate modeling of the nonlinear system, and further realizes efficient and accurate modeling of power electronic equipment, but also, compared with the black box model established by data modeling in traditional technology, since the mechanism relationship of the nonlinear system is integrated into the data modeling of the nonlinear system, the mechanism data fusion model is more interpretable, thereby making the mechanism data fusion model of the power electronic equipment more interpretable. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 A schematic diagram of a flow chart of a method for fusion modeling of power electronic equipment mechanism data in one embodiment;

[0048] Figure 2A A schematic diagram of a topological structure of a single IGBT and a topological structure of a single FWD in one embodiment;

[0049] Figure 2B for Figure 2A Grayscale image of

[0050] Figure 3 A schematic diagram of a circuit structure involved in a process of discretizing hardware into an equivalent admittance and an equivalent current source by using the reverse Euler method in an embodiment;

[0051] Figure 4A The computational principle of the Koopman operator linearization of a nonlinear system is illustrated in one embodiment;

[0052] Figure 4B for Figure 4A Grayscale image of

[0053] Figure 5 This is one of the flowchart diagrams of step S106 in one embodiment;

[0054] Figure 6 This is a second flow chart of step S106 in one embodiment;

[0055] Figure 7 A schematic diagram of a process flow of a power electronic equipment mechanism data fusion modeling method in one embodiment;

[0056] Fig. 8A is one of the training schematic diagrams of the first neural network in one embodiment;

[0057] Figure 8B for Fig. 8A Grayscale image of

[0058] Fig. 9 is a flow chart of step S502 in one embodiment;

[0059] Fig. 10A FIG2 is a second schematic diagram of training of a first neural network in one embodiment;

[0060] Fig. 10B for Fig. 10A Grayscale image of

[0061] Fig.11 is a flow chart of step S504 in one embodiment;

[0062] Fig.12 A schematic diagram of a modular multi-level converter type high voltage direct current transmission topology structure in one embodiment;

[0063] Fig.13 One of the simulation diagrams of a data fusion modeling method for a power electronic equipment mechanism in one embodiment;

[0064] Fig.14 The second simulation diagram of the data fusion modeling method for power electronic equipment mechanism in one embodiment;

[0065] Fig.15 The third simulation diagram of the data fusion modeling method for power electronic equipment mechanism in one embodiment;

[0066] Fig.16 A schematic diagram of the structure of a power electronic equipment mechanism data fusion modeling device in one embodiment;

[0067] Fig.17 FIG. 1 is a schematic diagram of the structure of a computer device in an embodiment. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0069] (I) The present application embodiment provides a power electronic equipment mechanism data fusion modeling method, which can be applied to large-scale, modular, high-voltage, and large-capacity power electronic equipment modeling, such as but not limited to the modeling of flexible DC converter valve towers, power electronic transformers, etc. The power electronic equipment includes at least a nonlinear system, such as a power electronic equipment including a nonlinear system and other systems other than the nonlinear system.

[0070] In an exemplary embodiment, referring to Figure 1 , provides a power electronic equipment mechanism data fusion modeling method, the modeling method includes the following steps S102~S108.

[0071] S102, obtaining a node voltage equation group of preset nodes of components of a nonlinear system of power electronic equipment, and solving the node voltage equation group according to preset conditions to obtain a value of a nonlinear state variable of the nonlinear system.

[0072] Among them, the nonlinear system of power electronic equipment can be understood as a system in power electronic equipment that is composed of parts whose mechanism relationships are unclear, multiple electrical parameters are difficult to determine, and / or nonlinear solutions are too complicated.

[0073] The nonlinear system of power electronic equipment usually contains a large number of components. The components can be nonlinear or linear. For example, the components are, but not limited to, nonlinear capacitors, nonlinear switches, and linear resistors; wherein, the nonlinear switches are, but not limited to, nonlinear diodes, nonlinear bipolar junction transistors (Bipolar Junction Transistor, BJT), nonlinear metal oxide semiconductor field effect transistors (Metal Oxide Semiconductor Field Effect Transistor, MOSFET), nonlinear insulated gate bipolar transistors (Insulated Gate Bipolar Transistor, IGBT), etc.

[0074] Each component may include at least two preset nodes. The preset nodes of a component may be understood as nodes that are preset in the equivalent circuit or topological structure of the component. Figure 2A , Figure 2A The topology of a single IGBT and the topology of a single FWD (Free Wheeling Diode) are exemplarily illustrated; in the topology of the IGBT, the static characteristics of the IGBT are characterized by MOSFET and BJT, and the dynamic characteristics of the IGBT are characterized by each junction capacitance Ceg, Cdg, Cgs, Cds, Cec and each diffusion capacitance Ceb, Cdif. The preset nodes of the IGBT are, for example, nodes f1, f2, f3, f4, f5, f6, and f8, that is, a single IGBT may include 7 preset nodes, and nodes f7, f6, and f2 may be preset nodes of the FWD, that is, a single FWD may include 3 preset nodes. In addition, Figure 2A middle , , All represent resistance, , , All represent inductance; Gate, Emitter, and Collector represent gate, emitter, and collector respectively.

[0075] For each preset node, at least one node voltage equation is listed according to its mechanism relationship in the equivalent circuit or topological structure, and the node voltage equation can be a nonlinear differential algebraic equation. All node voltage equations of all preset nodes of a component are connected to obtain the node voltage equation group of the component. When solving the node voltage equation group of the component, the preset conditions used may include preset initial conditions and preset boundary conditions for the node voltage equation group of the component, and the preset initial conditions include, for example, preset initial node voltage vectors. Solve the node voltage equation group of the component to obtain a time domain numerical solution, which is a specific numerical value of a nonlinear state variable, that is, the value of the nonlinear state variable; multiple different numerical solutions at a certain moment can constitute a numerical solution vector at that moment, and multiple numerical solution vectors at different moments can constitute a data set; finally, the input of the first neural network can be a data set, and the value of the nonlinear state variable is input into the first neural network in the form of a data set.

[0076] Exemplarily, when writing the node voltage equation group of a component, the electrical parameters of hardware such as MOSFET, BJT, capacitor, etc. in the equivalent circuit or topological structure of the component are related to the node voltages of the preset nodes at both ends of the component. According to the electrical parameters of the hardware, the functional relationship (iv) between the current flowing through the hardware and the node voltages at both ends of the hardware can be written. Then, according to Kirchhoff's current law, the functional relationship (iv) of all the hardware of the component can be linked to form a nonlinear differential algebraic equation group of the component, that is, the node voltage equation group of the component.

[0077] It is not possible to directly obtain the analytical solution of the node voltage equations listed, and a numerical method is required to obtain the time domain numerical solution. The numerical method includes discretizing a hardware into an equivalent admittance and an equivalent current source through the Backward Euler Method.

[0078] For example, taking capacitance as an example, Figure 3 For the capacitor shown in (a), the dynamic current-voltage (iv) relationship is:

[0079] (1);

[0080] in, and are the voltage and current at both ends of the capacitor respectively. C in formula (1) to formula (5) is the capacitance value of the capacitor. arrive The points can be obtained:

[0081] (2);

[0082] The integral part of equation (2) is approximated by the reverse Euler method, and we can get:

[0083] ,final (3);

[0084] Equation (3) can be viewed as a Figure 3 (b) The equivalent admittance shown and the equivalent current source A parallel circuit where:

[0085] (4);

[0086] time The equivalent current source With an interval of one time step The last moment The capacitor voltage is related to:

[0087] (5);

[0088] As mentioned above, each of all the hardware is discretized into an equivalent admittance and an equivalent current source through the inverse Euler method. For each hardware, the value of the equivalent admittance at time k and the value of the equivalent current source at time k are both related to the electrical parameters at time k. The electrical parameters at time k are related to the voltage at both ends of the hardware nodes at time k-1. Therefore, the value of the equivalent admittance and the value of the equivalent current source are only related to the voltage at both ends of the hardware nodes. The equivalent admittance and equivalent current source of all the hardware in the equivalent circuit or topological structure of the components are spliced ​​into an equivalent admittance matrix through the topological relationship between the hardware. and the equivalent current source vector Ohm's Law Formula , is the node voltage vector, and the node voltage vector at time k can be calculated by an iterative method based on the node voltage vector at time k-1.

[0089] For example, taking the solution of the node voltage vector of the component at time k (the node voltage vector represents the node voltage of each preset node of the component) as an example:

[0090] According to the node voltage vector at time k-1, the equivalent admittance matrix at time k is obtained for the first time and the equivalent current source vector ; The equivalent admittance matrix obtained at the first time k and the equivalent current source vector Substitute into the formula , the first inverse solution obtains the node voltage vector .

[0091] If the node voltage vector at time k-1 is equal to the node voltage vector obtained by the first inverse solution If the absolute value of the difference is less than the preset threshold, the node voltage vector is obtained by the first inverse solution. As the node voltage vector at time k.

[0092] If the node voltage vector at time k-1 is equal to the node voltage vector obtained by the first inverse solution If the absolute value of the difference is greater than or equal to the preset threshold, the node voltage vector is obtained according to the first inverse solution. , the equivalent admittance matrix at time k is obtained for the second time and the equivalent current source vector ; The equivalent admittance matrix obtained at time k for the second time and the equivalent current source vector Substitute into the formula , the second inverse solution obtains the node voltage vector .

[0093] If the node voltage vector is obtained by the first inverse solution The node voltage vector is obtained by the second inverse solution If the absolute value of the difference is less than the preset threshold, the node voltage vector is obtained by the second inverse solution. As the node voltage vector at time k.

[0094] If the node voltage vector is obtained by the first inverse solution The node voltage vector is obtained by the second inverse solution If the absolute value of the difference is greater than or equal to the preset threshold, the node voltage vector is obtained according to the second inverse solution. , and the equivalent admittance matrix at time k is obtained for the third time and the equivalent current source vector ; The equivalent admittance matrix at time k obtained for the third time and the equivalent current source vector Substitute into the formula , the third inverse solution obtains the node voltage vector .

[0095] If the node voltage vector is obtained by the second inverse solution The node voltage vector is obtained by the third inverse solution If the absolute value of the difference is less than the preset threshold, the node voltage vector is obtained by the third inverse solution. As the node voltage vector at time k.

[0096] If the node voltage vector is obtained by the second inverse solution The node voltage vector is obtained by the third inverse solution If the absolute value of the difference is greater than or equal to the preset threshold, the node voltage vector is obtained according to the third inverse solution. and formula , continue the fourth inverse solution to obtain the node voltage vector , and judge whether the node voltage vector Unew obtained by the third inverse solution is the same as the node voltage vector Unew obtained by the fourth inverse solution The absolute value of the difference between the two values ​​and the preset threshold value is calculated until the node voltage vector is obtained by the inverse solution. The node voltage vector is obtained by inverse solution from the previous step The absolute value of the difference is less than the preset threshold, so the node voltage vector is obtained by inverse solution. As the node voltage vector at time k, the node voltage vector at time k is finally obtained; in addition, when solving the node voltage vector of the component at time k=1, the equivalent admittance matrix at time k=1 can be obtained for the first time based on the preset initial node voltage vector and the equivalent current source vector , and continue with subsequent calculations.

[0097] The node voltage vector obtained at time k can be used as the vector of values ​​of the nonlinear state variables at time k, that is, the nonlinear numerical solution vector at time k.

[0098] S104, constructing a neural network based on the Koopman operator as the first neural network.

[0099] The Koopman operator is a linear, infinite-dimensional operator. The Koopman operator acts on the measurement function , which is equivalent to measuring the function Mapping functions in nonlinear systems Compound phase, namely:

[0100] ;

[0101] ;in, is the Koopman operator.

[0102] refer to Figure 4A , Figure 4A The computational principle of the Koopman operator linearized nonlinear system is illustrated in an exemplary manner, including the nonlinear numerical solution vector at time k in the nonlinear system. The numerical solution vector of the linear system at k moments The mapping relationship of and The dimensions may be the same or different; multiple moments are represented as 1…k…; is the nonlinear numerical solution vector at time k+1 in the nonlinear system, is the linear numerical solution vector of the predicted k+1 time in the linear system, is the Koopman eigenvalue matrix in the time domain, and t means the time in the time domain. As mentioned above, when t is at time k, It can be obtained according to the Koopman eigenvalue vector at time k, is the Koopman eigenvalue vector at time k, The dimensions and The dimensions are the same, for The elements in the Koopman eigenvalue vector form a diagonal matrix, and the other elements are filled with 0, and the k-time point can be obtained. ; In the linear space, the prediction of the value of the linear state variable is only the prediction from time k to time k+1, and the time between time k and time k+1 is one time step , time step The specific size of can be configured according to actual needs, and the embodiment of the present application does not limit this; that is:

[0103] ;

[0104] ;

[0105] ;

[0106] Based on this, the Koopman operator can be used to constrain the layer structure and activation function of a basic neural network in deep learning to construct a neural network based on the Koopman operator, and the neural network based on the Koopman operator is used as the first neural network.

[0107] S106, based on the value of the nonlinear state variable, performing multiple iterative training on the first neural network within a preset time period, wherein the multiple iterative training within the preset time period is performed sequentially, and a loss value of the first neural network is determined after each training is completed.

[0108] In each training, the value of the nonlinear state variable can be input into the first neural network to train the first neural network. Based on a pre-set loss function, a loss value can be determined for the first neural network after each training. The loss value can reflect the accuracy of the trained first neural network. In iterative training, the loss value determined for the first neural network after the last training is completed is used to guide the next training of the first neural network.

[0109] S108, after completing the last training within the preset time period, when the determined loss value of the first neural network is less than the preset target loss value, a mechanism data fusion model of the nonlinear system is obtained based on the first neural network that has completed multiple iterations of training within the preset time period.

[0110] Exemplarily, if the size of the preset time period is configured as 240 minutes, the first neural network will be iteratively trained multiple times in sequence based on the value of the nonlinear state variable within the 240 minutes. Wherein, when the loss value of the first neural network determined after the last training within the 240 minutes is less than the preset target loss value, it means that after the iterative training of the first neural network within the 240 minutes, a first neural network with an accuracy that meets the target value is obtained. At this time, the mechanism data fusion model of the nonlinear system of the power electronic equipment can be constructed according to the parameters of the first neural network that has completed the iterative training within the 240 minutes, wherein the parameters of the first neural network are, for example, but not limited to, weights and biases.

[0111] It can be understood that the embodiment of the present application does not limit the specific size of the preset time period, which can be configured according to actual needs; the total number of training times within the preset time period is determined by the duration of completing one iterative training, which is not limited by the embodiment of the present application.

[0112] (II) The following is an exemplary description of how to perform each training in multiple iterative trainings within a preset period of time, which is also applicable to the case where i=1:

[0113] In an exemplary embodiment, referring to Figure 5 , step S106 includes the following steps S502~S508.

[0114] S502, based on the value of the nonlinear state variable, perform an i-th training on the first neural network that has completed the i-1-th parameter update, where i is a positive integer and 1<i.

[0115] The updating of the parameters of the first neural network may include updating the weights and biases of the first neural network, that is, the parameters of the first neural network may include the weights and biases of the first neural network. The value of the nonlinear state variable may be input into the first neural network that has completed the i-1th parameter update to perform the i-th training on the first neural network that has completed the i-1th parameter update.

[0116] S504, based on a preset loss function and according to the value of the nonlinear state variable, determining the i-th loss value of the first neural network after completing the i-th training.

[0117] S506, according to the i-th loss value, controlling the first neural network that has completed the i-th training to perform the i-th parameter update.

[0118] S508, based on the value of the nonlinear state variable, perform the (i+1)th training on the first neural network that has completed the i-th parameter update.

[0119] Exemplarily, if the size of the preset time period is configured as 240 minutes, the first neural network will be iteratively trained multiple times in sequence based on the value of the nonlinear state variable within the 240 minutes. After the i-th training is completed and the i-th loss value is determined, the i-th reference gradient can be determined according to the i-th loss value, and the i-th reference gradient is input to the first neural network that has completed the i-th training, so that the first neural network that has completed the i-th training performs the i-th parameter update; then, the value of the nonlinear state variable is input to the first neural network that has completed the i-th parameter update, so as to perform the i+1th training on the first neural network that has completed the i-th parameter update.

[0120] Optionally, step S106 also includes: based on the value of the nonlinear state variable, training the first neural network for the first time, wherein the value of the parameter of the first neural network for the first training is a preset initial value, and the preset initial value may include a preset initial weight and a preset initial bias, etc.; then, based on a preset loss function and according to the value of the nonlinear state variable, determining a first loss value of the first neural network after completing the first training; then, determining a first reference gradient according to the first loss value, and inputting the first reference gradient into the first neural network that has completed the first training, so that the first neural network that has completed the first training performs a first parameter update; then, inputting the value of the nonlinear state variable into the first neural network that has completed the first parameter update, to perform a second training on the first neural network that has completed the first parameter update.

[0121] (III) The following is an exemplary description of how to perform multiple iterations of training within a preset period of time:

[0122] In an exemplary embodiment, the preset period includes the 1st to Mth step sub-periods performed sequentially, wherein M is a positive integer, 2≤M, at least two step sub-periods are the same or different, the sum of all step sub-periods is equal to the preset period, each step sub-period is configured with a preset expected loss value, the preset expected loss value of the jth step sub-period is less than the preset expected loss value of the j-1th step sub-period, wherein j is a positive integer, 1<j≤M, and the preset expected loss value of the Mth step sub-period is equal to the preset target loss value. In addition, at least two step sub-periods are continuous or spaced apart.

[0123] Exemplarily, the size of the preset time period is configured to be 240 minutes, and M is configured to be 5. The preset time period is divided into the 1st to 5th step sub-periods which are performed sequentially, namely: the 1st step sub-period [0~20 minutes], the 2nd step sub-period (20 minutes~40 minutes], the 3rd step sub-period (40 minutes~60 minutes], the 4th step sub-period (60 minutes~120 minutes], the 5th step sub-period (120 minutes~240 minutes]. Among them, when performing iterative training within the 240 minutes, the 1st to the 5th step sub-periods are also performed sequentially, and multiple iterative trainings are performed sequentially in each step sub-period.

[0124] refer to Figure 6 , step S106 includes the following steps S602~S604.

[0125] S602, after completing the last training in the j-1th step sub-period, when the determined loss value of the first neural network is less than the preset expected loss value of the j-1th step sub-period, the first neural network that has completed multiple iterative training in the j-1th step sub-period continues to perform multiple iterative training in the j-th step sub-period.

[0126] S604, after completing the last training in the j-1th step sub-period, if the determined loss value of the first neural network is greater than or equal to the preset expected loss value of the j-1th step sub-period, exit the training.

[0127] In the embodiment of the present application, the meaning of the step is that the preset expected loss value of the j-th sub-period is less than the preset expected loss value of the j-1-th sub-period, so as to perform step-by-step iterative training within the preset period to improve modeling efficiency. That is to say, after completing the last training in the j-1th step sub-period, if the loss value of the determined first neural network is greater than or equal to the preset expected loss value of the j-1th step sub-period, the training is directly exited, indicating that after completing the iterative training of the first neural network in the j-1th step sub-period, the first neural network with the expected accuracy is not obtained, and thus the last parameter update in the j-1th step sub-period and the iterative training in all step sub-periods after the j-1th step sub-period are no longer performed; only after completing the last training in the j-1th step sub-period, if the loss value of the determined first neural network is less than the preset expected loss value of the j-1th step sub-period, it is indicated that after completing the iterative training of the first neural network in the j-1th step sub-period, the first neural network with the expected accuracy is obtained, and thus the last parameter update in the j-1th step sub-period and the iterative training in all step sub-periods after the j-1th step sub-period are continued, thereby improving the modeling efficiency.

[0128] Optionally, refer to Figure 7 After exiting the training in step S604, the power electronic equipment mechanism data fusion modeling method further includes the following steps S702 to S706.

[0129] S702, constructing a neural network based on a Koopman operator as a second neural network, wherein the structure of the second neural network is different from that of the first neural network.

[0130] S704, based on the value of the nonlinear state variable, performing multiple iterative training on the second neural network within a preset time period, wherein the multiple iterative trainings are performed sequentially, and a loss value of the second neural network is determined after each training is completed.

[0131] S706, after completing the last training within the preset time period, when the determined loss value of the second neural network is less than the preset target loss value, a mechanism data fusion model of the nonlinear system is obtained based on the second neural network that has completed multiple iterations of training within the preset time period.

[0132] In this embodiment, after exiting the training in step S604, the first neural network will no longer be iteratively trained for multiple times within a preset period of time based on the value of the nonlinear state variable. Instead, the first neural network will be invalidated, and a neural network based on the Koopman operator will be reconstructed as the second neural network. The structure of the second neural network is different from that of the first neural network. Then, based on the value of the nonlinear state variable, the second neural network will be iteratively trained for multiple times within a preset period of time.

[0133] Similarly, during the process of performing multiple iterative training on the second neural network within a preset time period, after completing the last training within the j-1th step sub-period, if the loss value of the second neural network determined is greater than or equal to the preset expected loss value of the j-1th step sub-period, the training is exited, and the second neural network is no longer trained for multiple iterative training within the preset time period based on the value of the nonlinear state variable. Instead, the second neural network is invalidated, and a neural network based on the Koopman operator is reconstructed as the third neural network. The structure of the third neural network is different from the structure of the second neural network and the structure of the first neural network. Then, based on the value of the nonlinear state variable, the third neural network is trained for multiple iterative training within the preset time period... The embodiment of the present application follows this until a neural network based on the Koopman operator is found. After completing the last training within the preset time period, the loss value of the determined neural network is less than the preset target loss value, and then the mechanism data fusion model of the nonlinear system is obtained according to the neural network that has completed multiple iterative training within the preset time period.

[0134] (IV) The following is a further exemplary description of how to perform each training in multiple iterative trainings within a preset period of time, which is also applicable to the case where i=1:

[0135] In an exemplary embodiment, referring to Fig. 8A , the first neural network includes an encoder ( ) and the decoder ( ), where the encoder is an encoder based on the Koopman characteristic function, and the decoder corresponds to the encoder. The encoder, decoder and Koopman operator act on the numerical solution vector:

[0136] ;

[0137] ;

[0138] ;

[0139] = ;

[0140] = .

[0141] Optionally, the first neural network also includes an auxiliary neural network ( ), the Koopman eigenvalue can be determined by an auxiliary neural network based on the value of the linear state variable. That is, = ,in, for The vector composed of the Koopman eigenvalues ​​corresponding to each element in .

[0142] refer to Fig. 9 In step S502, based on the value of the nonlinear state variable, the first neural network that has completed the i-1th parameter update is trained for the i-th time, including the following steps S902~S906.

[0143] S902, mapping the value of the nonlinear state variable from the nonlinear space to the linear space through an encoder to obtain the value of the linear state variable.

[0144] S904, based on the general solution of the nonlinear differential algebraic equations and according to the Koopman eigenvalues ​​and the values ​​of the linear state variables, dynamically evolve in the linear space to obtain the predicted values ​​of the linear state variables.

[0145] The general solution based on the nonlinear differential algebraic equations may be in a format according to the general solution of the differential equations. The nonlinear differential algebraic equations may be node voltage equations.

[0146] S906: Map the predicted linear state variables from the linear space back to the nonlinear space through the decoder.

[0147] In this embodiment, the Koopman eigenvalue corresponds to the value of the linear state variable one by one, and the value of the linear state variable changes with the change of the value of the nonlinear state variable. Fig. 10A It can be understood that, since the first neural network includes an encoder, a decoder and an auxiliary neural network, performing multiple iterative training on the first neural network within a preset period of time is actually performing multiple iterative training on the encoder, the decoder and the auxiliary neural network simultaneously within a preset period of time. Therefore, although the numerical solution vector input to the first neural network in two adjacent trainings is the same, the Koopman eigenvalue vectors in the two adjacent trainings are different because the auxiliary neural networks in the two adjacent trainings are different; and for the same training, if the numerical solution vector input to the first neural network is different, that is, the input of the first neural network is different, the corresponding Koopman eigenvalue vector will also be different.

[0148] in, ; ;

[0149] n is The dimension of The discrete time involved in writing the node voltage equations using the trapezoidal rule is consistent with .

[0150] Optionally, refer to Fig.11 In step S504, based on a preset loss function and according to the value of the nonlinear state variable, the i-th loss value of the first neural network after completing the i-th training is determined, including the following steps S1102~S1108.

[0151] S1102, determining an i-th autoregressive error of the first neural network after completing the i-th training based on a preset autoregressive error function and according to the nonlinear state variable.

[0152] Exemplarily, the autoregressive error function is preset for:

[0153]

[0154] Here, MSE is the mean square error.

[0155] S1104: Determine, based on a preset prediction error function and in accordance with the nonlinear state variable, an i-th prediction error of the first neural network after completing the i-th training.

[0156] Exemplarily, the preset prediction error function for:

[0157]

[0158] in, is the number of prediction steps in linear space, which can be a positive integer greater than 1.

[0159] S1106: Determine, based on a preset linear error function and in accordance with the nonlinear state variable, an i-th linear error of the first neural network after completing the i-th training.

[0160] Exemplarily, a linear error function is preset for:

[0161]

[0162] Here, T is the number of time steps.

[0163] S1108, determining the i-th loss value of the first neural network after completing the i-th training based on the i-th autoregressive error and its corresponding preset first weighting coefficient, the i-th prediction error and its corresponding preset second weighting coefficient, and the i-th linear error and its corresponding preset third weighting coefficient.

[0164] In the embodiment of the present application, as important indicators for evaluating the accuracy of the mechanism data fusion model, three loss functions are introduced in the loss function part, namely, autoregressive error function, prediction error function, and linear error function, and they are linearly combined. The modeling of different nonlinear systems can be adapted by adjusting the size of the first weighting coefficient, the second weighting coefficient and / or the third weighting coefficient.

[0165] (V) The following is an exemplary simulation of the data fusion modeling method for power electronic equipment mechanism:

[0166] Fig.12 It is the basic topology of modular multilevel converter-high voltage direct current (MMC-HVDC), which includes multiple bridge arms, each of which is composed of N sub-modules SM and a bridge arm inductor L0. Usually, the topology of the sub-module SM is a half-bridge topology, which includes two upper and lower IGBTs and two upper and lower freewheeling diodes (i.e., freewheeling diode D1 and freewheeling diode D2). is the voltage of capacitor C0 of submodule SM, and are the current and output voltage of the submodule SM respectively.

[0167] The difficulty of MMC simulation lies mainly in its complex topological structure and the large number of IGBTs contained therein, which results in too many circuit nodes and too large a scale of matrix solution. Direct solution will make the simulation time too long and difficult to apply, and when the number of MMC levels is too large, it may even cause the simulation to not converge. In this regard, system-level analysis shows that the high-frequency component of the bridge arm injection current ip (ipa, ipb, ipc) decays rapidly, so the change of the bridge arm injection current ip is very small between two adjacent time steps. Therefore, in the embodiment of the present application, the voltage and current between the submodule SM and the bridge arm are decoupled for circuit division, and all submodules SM are separated from the corresponding bridge arm to form the smallest circuit to reduce the computational burden and improve numerical stability.

[0168] The following is a single submodule SM of MMC as the research object. The submodule SM consists of two upper and lower IGBTs, two upper and lower freewheeling diodes (i.e., freewheeling diode D1 and freewheeling diode D2) and a capacitor C0. As mentioned above, the submodule SM includes 14 preset nodes (f1~f8 multiplied by 2=16), among which the node f8 of the upper IGBT coincides with the node f1 of the lower IGBT, that is, they are merged into a common node; the node f8 of the lower IGBT serves as the reference node of the upper and lower IGBTs. The voltage simulation of these 14 nodes is shown in the following figure. Fig.13 As shown, the overall result error is small; Fig.13 There are two rows of diagrams, each with seven columns of diagrams. In the upper row, from left to right, they are diagrams of nodes f1, f2, f3, f4, f5, f6 and f8 of the upper IGBT; in the lower row, from left to right, they are diagrams of nodes f1, f2, f3, f4, f5, f6 and f8 of the lower IGBT. The line True represents the simulation of the numerical solution, and the line Predicate represents the simulation of the output of the mechanism data fusion model. The line Predicate is not much different from the line True, which is enough to show that the output of the mechanism data fusion model is highly accurate.

[0169] In practical applications, it is impossible to measure the voltage of the nodes inside the submodule SM. In order to meet the actual application, the input IGBT current, the control signal Vg of the input f4 node, and the output voltage of the submodule SM are selected to be measurable. The voltage of capacitor C0 of submodule SM Simulate again and the result is as follows Fig.14 ;in, Fig.14There are two rows of diagrams, each with three columns. In the upper row, from left to right, the diagrams are the current flowing through the upper IGBT, the current flowing through the lower IGBT, and the control signal Vg of the upper IGBT; in the lower row, from left to right, the diagrams are the control signal Vg of the lower IGBT, the output voltage of the submodule SM, and the output voltage Vg of the submodule SM. The voltage of capacitor C0 of submodule SM ; the line True represents the simulation of the numerical solution, and the line Predicate represents the simulation of the output of the mechanism data fusion model.

[0170] In addition to the prediction accuracy of the mechanism data fusion model and the simulation acceleration effect brought by the mechanism data fusion model, the interpretability of the mechanism data fusion model is another key point of the technical solution of the embodiment of the present application. Interpretability refers to explaining the mechanism relationship satisfied by the data in the calculation process on a theoretical basis. The examples designed in this paper are all time-dependent nonlinear differential algebraic equations. The input is the numerical solution vector of the nonlinear differential algebraic equations at time k, and the output is the numerical solution vector at time k+1.

[0171] Fig.15 The Koopman eigenvalues ​​found by the submodule SM through the auxiliary neural network are shown. Data with a time length of 128 are randomly selected from the numerical solution vectors at multiple moments (for example, from the numerical solution vector at moment k=1 to the numerical solution vector at moment k=128). Each numerical solution vector includes 10 numerical solutions selected from 14 numerical solutions. The data are calculated after being mapped by the encoder and then through the function The obtained Koopman eigenvalues ​​and the relationship between the input and output of the auxiliary neural network. In this simulation, in order to fully display the relationship between the 14 node voltages, the input Scaling to the interval [0, 1], the corresponding Remain unchanged.

[0172] Fig.15 The left figure shows: Each line represents the Koopman eigenvalue of a numerical solution vector at time 1 to 128. In the time window of 128, Almost unchanged, and distributed around the value 1. This is because during the calculation process, the length of time fed to the first neural network is 128 each time. Within this time length, due to the simulation time step Even though the Koopman eigenvalue at time k depends on the value of the nonlinear state variable at time k, the Koopman eigenvalues ​​should be equal or approximate within the small interval of time length 128, and since the numerical solution vectors at the previous and next moments are close in size, the Koopman eigenvalues ​​should be distributed around 1.

[0173] Fig.15 The right picture shows: The linear state variable values ​​of the 14 node voltages and the predicted linear state variable values ​​strictly satisfy the linear relationship. This is because the calculation process of the mechanism data fusion model is to first map the values ​​of the nonlinear state variables to the linear space through the encoder, and then calculate the Koopman eigenvalues ​​corresponding to the values ​​of the linear state variables through the auxiliary neural network. Complete the prediction, which is also the calculation principle required by the general solution format of the nonlinear differential algebraic equations. To meet this principle, the values ​​of the linear state variables and the predicted values ​​of the linear state variables strictly satisfy the linear relationship, and the simulation results are consistent with the theoretical reasoning.

[0174] The embodiment of the present application applies the Koopman operator to the data modeling process, which essentially completes the mapping of nonlinear systems to linear systems through data models, converts the iterative solution process into a simple calculation process, thereby simplifying the modeling process. The entire modeling of the embodiment of the present application includes at least: solving the node voltage equation group, loss function design, neural network architecture design, and especially introduces the Koopman operator to realize the linearization of nonlinear systems and the accelerated solution of the refined model of power electronic equipment.

[0175] In an exemplary embodiment, reference Fig.16 A power electronic equipment mechanism data fusion modeling device is also provided, the power electronic equipment includes a nonlinear system, and the modeling device includes:

[0176] The nonlinearization module 1610 is used to obtain a node voltage equation group of a preset node of a component of a nonlinear system, and solve the node voltage equation group according to preset conditions to obtain a value of a nonlinear state variable of the nonlinear system;

[0177] A neural network module 1620, used to construct a neural network based on a Koopman operator as a first neural network;

[0178] A network training module 1630 is used to perform multiple iterative training on the first neural network within a preset time period based on the value of the nonlinear state variable, wherein the multiple iterative training within the preset time period is performed sequentially, and a loss value of the first neural network is determined after each training is completed;

[0179] The model building module 1640 is used to obtain a mechanism data fusion model of the nonlinear system based on the first neural network that has completed multiple iterations of training within a preset time period when the loss value of the first neural network determined after the last training within a preset time period is less than a preset target loss value.

[0180] In an exemplary embodiment, the network training module 1630 is also used to perform an i-th training on the first neural network that has completed the i-1th parameter update based on the value of the nonlinear state variable, where i is a positive integer and 1<i; determine the i-th loss value of the first neural network after completing the i-th training based on a preset loss function and according to the value of the nonlinear state variable; control the first neural network that has completed the i-th training to perform the i-th parameter update based on the i-th loss value; and perform an i+1-th training on the first neural network that has completed the i-th parameter update based on the value of the nonlinear state variable.

[0181] In an exemplary embodiment, the network training module 1630 is also used to perform a first training on the first neural network based on the value of the nonlinear state variable, wherein the parameter values ​​of the first neural network performing the first training are preset initial values; based on a preset loss function and according to the value of the nonlinear state variable, determine the first loss value of the first neural network after completing the first training; and according to the first loss value, control the first neural network that has completed the first training to perform a first parameter update.

[0182] In an exemplary embodiment, the preset period includes the 1st to Mth step sub-periods performed in sequence, wherein M is a positive integer, 2≤M, at least two step sub-periods are the same or different, the sum of all step sub-periods is equal to the preset period, each step sub-period is configured with a preset expected loss value, the preset expected loss value of the jth step sub-period is less than the preset expected loss value of the j-1th step sub-period, wherein j is a positive integer, 1<j≤M, and the preset expected loss value of the Mth step sub-period is equal to the preset target loss value; the network training module 1630 is also used to complete the j-1th step sub-period The network training module 1630 is further configured to exit the training when, after the last training in the j-1th step sub-period, the loss value of the first neural network determined is less than the preset expected loss value of the j-1th step sub-period, for the first neural network that has completed multiple iterative trainings in the j-1th step sub-period, the multiple iterative trainings in the j-1th step sub-period are continued, wherein the multiple iterative trainings in the step sub-periods are performed sequentially, and the network training module 1630 is further configured to exit the training when, after the last training in the j-1th step sub-period is completed, the loss value of the first neural network determined is greater than or equal to the preset expected loss value of the j-1th step sub-period.

[0183] In an exemplary embodiment, the first neural network includes an encoder and a decoder, wherein the encoder is an encoder based on the Koopman characteristic function, and the decoder corresponds to the encoder; the network training module 1630 is also used to map the values ​​of the nonlinear state variables from the nonlinear space to the linear space through the encoder to obtain the values ​​of the linear state variables; based on the general solution of the node voltage equation group, and according to the Koopman eigenvalues ​​and the values ​​of the linear state variables, the predicted values ​​of the linear state variables are dynamically evolved in the linear space; the predicted values ​​of the linear state variables are mapped from the linear space back to the nonlinear space through the decoder.

[0184] In an exemplary embodiment, the first neural network further includes an auxiliary neural network; the network training module 1630 is further configured to determine the Koopman eigenvalue through the auxiliary neural network based on the value of the linear state variable.

[0185] In an exemplary embodiment, the network training module 1630 is also used to determine the i-th autoregressive error of the first neural network after completing the i-th training based on a preset autoregressive error function and according to the value of the nonlinear state variable; determine the i-th prediction error of the first neural network after completing the i-th training based on a preset prediction error function and according to the value of the nonlinear state variable; determine the i-th linear error of the first neural network after completing the i-th training based on a preset linear error function and according to the value of the nonlinear state variable; determine the i-th loss value of the first neural network after completing the i-th training based on the i-th autoregressive error and its corresponding preset first weighting coefficient, the i-th prediction error and its corresponding preset second weighting coefficient, and the i-th linear error and its corresponding preset third weighting coefficient.

[0186] Each module in the above-mentioned power electronic equipment mechanism data fusion modeling device can be implemented in whole or in part by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of the above modules.

[0187] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Fig.17As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device is used to store related data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for fusion modeling of power electronic equipment mechanism data is implemented. Those skilled in the art can understand that Fig.17 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0188] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0189] In one embodiment, a computer program product is provided, comprising a computer program, which implements the steps in any of the above method embodiments when executed by a processor.

[0190] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited thereto. The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of the present application. The above-mentioned embodiments only express several implementation methods of the present application, and the description is more specific and detailed, but it cannot be understood as a limitation on the scope of the patent of this application. It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present application, and these modifications and improvements are all within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be subject to the attached claims.

Claims

1. A power electronic equipment mechanism data fusion modeling method, characterized in that: The power electronic equipment at least includes a nonlinear system; The modeling method comprises: Obtaining a node voltage equation group of a preset node of the components of the nonlinear system, and solving the node voltage equation group according to preset conditions to obtain a value of a nonlinear state variable of the nonlinear system; Construct a neural network based on the Koopman operator as the first neural network; Based on the value of the nonlinear state variable, the first neural network is trained multiple times within a preset period of time, wherein the multiple iterative trainings within the preset period of time are performed sequentially, and the loss value of the first neural network is determined once after each training is completed; and After completing the last training within a preset time period, when the determined loss value of the first neural network is less than the preset target loss value, the mechanism data fusion model of the nonlinear system is obtained based on the first neural network that has completed multiple iterative trainings within the preset time period.

2. The modeling method according to claim 1, characterized in that: The method of performing multiple iterative training on the first neural network within a preset time period based on the value of the nonlinear state variable, wherein the multiple iterative training within the preset time period is performed sequentially, and the loss value of the first neural network is determined once after each training is completed, including: Based on the value of the nonlinear state variable, the first neural network that has completed the (i-1)th parameter update is trained for the i-th time, where i is a positive integer and 1<i; Determining, based on a preset loss function and according to the value of the nonlinear state variable, an i-th loss value of the first neural network after completing the i-th training; According to the i-th loss value, controlling the first neural network that has completed the i-th training to perform the i-th parameter update; Based on the value of the nonlinear state variable, the first neural network that has completed the i-th parameter update is trained for the (i+1)th time.

3. The modeling method according to claim 2, characterized in that: The modeling method also includes: Based on the value of the nonlinear state variable, training the first neural network for a first time, wherein the value of the parameter of the first neural network for the first training is a preset initial value; Determining a first loss value of the first neural network after completing the first training based on a preset loss function and according to the value of the nonlinear state variable; According to the first loss value, the first neural network that has completed the first training is controlled to perform the first parameter update.

4. The method according to claim 2, characterized in that: The preset period includes the 1st to Mth step sub-periods that are performed sequentially, wherein M is a positive integer, 2≤M, at least two step sub-periods are the same or different, the sum of all step sub-periods is equal to the preset period, each step sub-period is configured with a preset expected loss value, the preset expected loss value of the jth step sub-period is less than the preset expected loss value of the j-1th step sub-period, wherein j is a positive integer, 1<j≤M, and the preset expected loss value of the Mth step sub-period is equal to the preset target loss value; The modeling method also includes: After completing the last training in the j-1th step sub-period, if the determined loss value of the first neural network is less than the preset expected loss value of the j-1th step sub-period, continue to perform multiple iterative training in the j-1th step sub-period for the first neural network that has completed multiple iterative training in the j-1th step sub-period, wherein the multiple iterative training in the step sub-period is performed sequentially; After completing the last training in the j-1th step sub-period, if the determined loss value of the first neural network is greater than or equal to the preset expected loss value of the j-1th step sub-period, the training is exited.

5. The method according to claim 2 or 3, characterized in that: The first neural network includes an encoder and a decoder, wherein the encoder is an encoder based on a Koopman characteristic function, and the decoder corresponds to the encoder; The step of performing the i-th training on the first neural network after the i-1-th parameter update based on the value of the nonlinear state variable comprises: Mapping the value of the nonlinear state variable from the nonlinear space to the linear space through the encoder to obtain the value of the linear state variable; Based on the general solution of the node voltage equations, and according to the Koopman eigenvalue and the value of the linear state variable, the predicted linear state variable value is obtained by dynamic evolution in the linear space; The predicted linear state variable value is mapped from the linear space back to the nonlinear space through the decoder.

6. The method according to claim 5, characterized in that The first neural network also includes an auxiliary neural network; The modeling method also includes: Based on the value of the linear state variable, the Koopman eigenvalue is determined by the auxiliary neural network.

7. The method according to claim 2 or 3, characterized in that: The determining, based on a preset loss function and according to the value of the nonlinear state variable, the i-th loss value of the first neural network after completing the i-th training comprises: Determining, based on a preset autoregressive error function and according to the value of the nonlinear state variable, an i-th autoregressive error of the first neural network after completing the i-th training; Determining, based on a preset prediction error function and according to the value of the nonlinear state variable, an i-th prediction error of the first neural network after completing the i-th training; Determining, based on a preset linear error function and according to the value of the nonlinear state variable, an i-th linear error of the first neural network after completing the i-th training; According to the i-th autoregressive error and its corresponding preset first weighting coefficient, the i-th prediction error and its corresponding preset second weighting coefficient, and the i-th linear error and its corresponding preset third weighting coefficient, determine the i-th loss value of the first neural network after completing the i-th training.

8. A power electronic equipment mechanism data fusion modeling device, characterized in that: The power electronic equipment at least includes a nonlinear system; The modeling device comprises: A nonlinearization module, used to obtain a node voltage equation group of a preset node of the components of the nonlinear system, and solve the node voltage equation group according to preset conditions to obtain a value of a nonlinear state variable of the nonlinear system; A neural network module is used to construct a neural network based on the Koopman operator as the first neural network; A network training module, configured to perform multiple iterative training on the first neural network within a preset time period based on the value of the nonlinear state variable, wherein the multiple iterative training within the preset time period is performed sequentially, and a loss value of the first neural network is determined once after each training is completed; A model building module is used to obtain a mechanism data fusion model of the nonlinear system based on the first neural network that has completed multiple iterations of training within a preset time period, when the loss value of the first neural network determined after the last training within a preset time period is less than a preset target loss value.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images