Method for Extracting Circuit Parameters of DAB Converter Based on Physics-Informed Neural Network

Abstract

The present invention discloses a method for extracting circuit parameters of a DAB converter based on a physics-informed neural network, belonging to the technical field of circuit parameter extraction. The specific steps are as follows: construct a time-domain expression of the DAB converter; establish a physical connection between the collected power quantity information data set and the data set of parameters to be extracted; establish a time-domain recurrence relationship based on the physical connection established in step S2; construct a physics-informed neural network for DAB converter parameter identification that fuses data mechanism and physical information. The physics-informed neural network updates the circuit parameter data serving as network weights by predicting the collected power quantity information, and simultaneously achieves accurate prediction of power quantity information and precise identification of parameter data. By adopting the above method, by inputting a small amount of inductor current and output voltage data of the DAB converter, and under the scenario of considering the accuracy of the collected data affected by noise and different modulation methods, high-precision circuit parameter extraction can be realized, with strong robustness and good generalization performance.

Description

Technical Field

[0001] The present invention relates to the technical field of circuit parameter extraction, and particularly to a method for extracting circuit parameters of a DAB converter based on a physics-informed neural network. Background Art

[0002] Dual-active-bridge (DAB) converters have attracted much attention due to their characteristics such as high power density, electrical isolation, and bidirectional power transmission ability, and are widely used in fields such as electrified transportation, distributed generation, and aerospace. Further, the control optimization and status monitoring of DAB converters can greatly improve system efficiency, reduce hardware costs, and lower the risk of failures. And to achieve efficient control optimization and accurate status monitoring, the accurate extraction of DAB converter parameters is a particularly important link.

[0003] Traditional converter parameter extraction methods are mainly based on physical information, and these methods rely on additional sensors and complex mathematical calculations. When the order of the system model is too high, the extremely high model complexity will inevitably lead to a huge amount of calculation. In recent years, data-driven parameter extraction methods using artificial intelligence (AI) tools have been widely applied, but they require a large amount of data to construct the potential relationship between the input (voltage, current signals) and output (parameter values to be extracted), and have poor practicability in data-sparse fields such as power electronics.

[0004] There is a method in the prior art for extracting parameters of a Buck converter (step-down converter) based on a physics-informed neural network. It forms a physics-informed neural network by sequentially connecting a data-driven network composed of a fully connected neural network and a physics-informed network implemented by the Runge-Kutta method, and can achieve the extraction of the main parameters of the Buck circuit. The disadvantages are: it is only applicable to scenarios with a large inductor current ripple, which is contrary to the efficient operation of the converter. The topology and operating state of the Buck converter are relatively simple, and the technical difficulty is relatively low.

[0005] There is a method in the prior art for extracting the DC-side capacitor and AC-side inductor values of a three-phase inverter based on a physics-informed neural network. Disadvantages: Its hardware experiments show that the maximum error in the extraction of the AC-side inductor value of this method exceeds 14%, and the maximum error in the extraction of the DC-side capacitor value exceeds 5%, and the extraction accuracy is too low.

[0006] In the prior art, there is a method of using the genetic algorithm backpropagation (GA-BP) algorithm to construct the mapping relationship between the terminal voltage and terminal current of a DAB converter and the converter parameters. Disadvantages: It is a pure data-driven method that requires a large amount of data for training to obtain an accurate relationship, and there are disadvantages such as the risk of overfitting, large data requirements, poor generalization, and poor local deployment ability.

[0007] Currently, there is no method for parameter extraction based on a physics-informed neural network applicable to complex DC-AC-AC-DC systems (such as DAB converters, etc.). Moreover, most current parameter extraction methods based on neural networks are only applicable to specific converter types with specific control strategies or modulation methods, and there are obvious limitations in generalization, making it difficult to apply to real-world scenarios. Summary of the Invention

[0008] The object of the present invention is to provide a method for extracting circuit parameters of a DAB converter based on a physics-informed neural network to solve the above technical problems.

[0009] To achieve the above object, the present invention provides a method for extracting circuit parameters of a DAB converter based on a physics-informed neural network, and the specific steps are as follows:

[0010] Step S1: Construct the time-domain differential expression of the DAB converter;

[0011] Step S2: Determine the data set of collected power quantity information and the data set of parameters to be extracted, and establish the physical connection between the data set of collected power quantity information and the data set of parameters to be extracted;

[0012] Step S3: Establish a time-domain recurrence relationship based on the physical connection established in Step S2;

[0013] Step S4: Construct a physics-informed neural network for parameter identification of the DAB converter by fusing data mechanism and physical information and train it. This physics-informed neural network updates the circuit parameter data serving as the weights of the physics-informed neural network by predicting the collected power quantity information, and simultaneously realizes the prediction of power quantity information and the identification of parameter data.

[0014] Preferably, when the product of the primary-side voltage of the transformer and the secondary-side voltage of the transformer is greater than zero and less than zero within the same period, the time-domain differential expression of the DAB converter is established as follows:

[0015] ;

[0016] Wherein, is the inductor current, and They are respectively the primary side equivalent voltage of the output capacitor voltage and the primary side equivalent voltage of the output voltage; , is the output capacitor voltage, is the turns ratio of the primary and secondary sides of the transformer; , is the output voltage, represents time;

[0017] is the input voltage; is the inductance value; is the relative state variable of the primary and secondary side switches. When , ; when , , and are respectively the primary side voltage of the transformer and the secondary side voltage of the transformer;

[0018] , is the equivalent parameter value of the inductor and the switch resistance, is the parasitic resistance value of the inductor, is the on-resistance value of the switch;

[0019] , is the equivalent parameter value of the capacitor parasitic resistance, is the parasitic resistance value of the capacitor;

[0020] , is the equivalent parameter value of the resistor, is the load resistance value;

[0021] , is the equivalent parameter value of the capacitor, is the capacitance value.

[0022] Preferably, step S2 is specifically as follows:

[0023] The elements included in the power consumption information data set are the inductor current and the output voltage;

[0024] The data set of parameters to be extracted is ;

[0025] When the inductor current is used as the input variable, the differential linearization relationship is as follows:

[0026] ;

[0027] Among them, is the non-linear operator determined by the inductor current and its expression is as follows:

[0028] ;

[0029] When the output voltage is used as the input variable, the differential linearization relationship is as follows:

[0030] ;

[0031] Among them, is the non - linear operator determined by the equivalent output voltage , and the expression is as follows:

[0032] .

[0033] Preferably, step S3 is specifically as follows:

[0034] Step S31: Assume that the inductor current states at the starting time and the ending time are respectively and , and the output voltage states are respectively and , and satisfy ; At the same time, assume that there are unobservable intermediate states during the time period , and the intermediate states are: , , , where is the time coefficient, , , …, , and the intermediate state points and are implicit points;

[0035] Step S32: Taking half of the switching period as the unit and based on the inductor current waveform, the starting time is selected as the first turning point time of the inductor current waveform, and a random time within the set threshold is set as the intermediate time between the second turning point time and the third turning point time. The point corresponding to this time is the quasi - mid - point, and the ending time is selected as the second turning point time and the third turning point time. Then the inductor current and the equivalent output voltage at the starting time correspond to and respectively, and the endings correspond to and respectively. The points corresponding to the starting time and the ending time are explicit points;

[0036] Step S33: To couple the implicit points and the explicit points, use The implicit Runge-Kutta method of order, through a set of constant parameters uniquely determined by the order , and are both relational constants, is the time coefficient constant in step S31;

[0037] Establish the forward and backward recurrence relations of the inductor current and the equivalent output voltage respectively to couple the explicit points and the implicit points;

[0038] The expression of the backward recurrence relation is as follows:

[0039] ;

[0040] The expression of the forward recurrence relation is as follows:

[0041] .

[0042] Preferably, step S4 is specifically as follows:

[0043] Step S41: The data-driven network adopts a fully connected neural network, and the input is the electrical quantity information of the explicit point corresponding to the starting moment , the electrical quantity information of the explicit point corresponding to the starting moment , the variable S and the time interval between the starting moment and the ending moment , the bias and weight are , and the output is the electrical quantity information of the implicit point at the intermediate moment predicted by the data-driven part and ;

[0044] Step S42: Establish the physical information network part according to the backward recurrence relation expression and the forward recurrence relation expression. The input is the electrical quantity information of the implicit point at the intermediate moment predicted by the data-driven part and , the bias and weight are , and the electrical quantity information of the explicit points corresponding to the starting moment and the ending moment is respectively deduced through the forward recurrence relation and the backward recurrence relation , , and ;

[0045] Step S43: Connect the data-driven network and the physical information network in series to form a physical information neural network for parameter identification of the DAB converter fused by data mechanism and physical information;

[0046] Step S44: Based on the simulation and experimental platform, collect the power data of the DAB converter according to Step S3 to form the collected power information dataset of the physics-informed neural network. During the collection stage of the simulation data, add instances with analog-to-digital conversion noise and sampling noise, and instances using single-phase-shift modulation and dual-phase-shift modulation.

[0047] Step S45: Determine the loss function and the training stop condition, and train the physics-informed neural network for the DAB converter parameter identification in Step S43. After reaching the stop condition, output the extracted parameter data.

[0048] Preferably, in Step S45, the expression of the loss function is as follows:

[0049] ;

[0050] where is the loss function, and are the predicted values of the inductor current at the starting time and the ending time respectively, and are the predicted values of the output voltage at the starting time and the ending time respectively;

[0051] The training stop condition is as follows:

[0052] When the number of training cycles reaches the first set number, or twice in a row ;

[0053] or the number of training cycles reaches the second set number.

[0054] Therefore, the present invention adopts the above-mentioned method for extracting circuit parameters of a DAB converter based on a physics-informed neural network, and the beneficial effects are as follows:

[0055] (1) Construct a physics-informed neural network of the DAB converter by fusing data mechanism and physical information, and realize high-precision circuit parameter extraction by inputting a small amount of key data of the inductor current and output voltage of the DAB converter.

[0056] (2) In the scenario of considering the influence of analog-to-digital conversion noise and sampling noise on the accuracy of the collected data, the accuracy of the extraction result is less affected, and it has strong robustness.

[0057] (3) Considering two modulation methods of single-phase-shift modulation and dual-phase-shift modulation, the parameters can be accurately extracted, and it has considerable generalization.

[0058] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Description of the Drawings

[0059] Figure 1 It is a flowchart of the method for extracting circuit parameters of the DAB converter based on the physics-informed neural network of the present invention;

[0060] Figure 2 It is an application diagram of an embodiment of the present invention;

[0061] Figure 3 It is a circuit topology diagram of the DAB converter adopted by the present invention;

[0062] Figure 4 It is an equivalent circuit diagram of the DAB converter adopted by the present invention;

[0063] Figure 5 It is a schematic diagram of the explicit start point, end point and implicit points selected by the present invention;

[0064] Figure 6 It is a structure diagram of the physics-informed neural network constructed by the present invention;

[0065] Figure 7 It is a convergence process diagram of the physics-informed neural network of the present invention;

[0066] Figure 8 It is a comparison diagram of the experimental, simulation and prediction results of the inductor current of the solution proposed by the present invention.

[0067] Figure 9 It is a comparison diagram of the experimental, simulation and prediction results of the output voltage of the solution proposed by the present invention. Detailed Embodiment

[0068] In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "upper", "lower", "inner", "outer", etc. is based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship in which the product of the present invention is usually placed during use. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In the description of the present invention, it should also be noted that unless otherwise clearly specified and limited, the terms "set", "installed", "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the internal communication of two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0069] The following will describe in detail the embodiments of the present invention in conjunction with the accompanying drawings.

[0070] As Figures 1-2 shown, a method for extracting circuit parameters of a DAB converter based on a physics-informed neural network is as follows:

[0071] Step S1: Construct the time-domain expression of the DAB converter.

[0072] When the product of the primary-side voltage and the secondary-side voltage of the transformer is greater than zero and less than zero within the same period, for a general DAB converter, as Figure 3 shown, perform equivalence. As Figure 4 shown, its time-domain differential expression can be established as follows:

[0073] ;

[0074] Wherein, is the inductor current, and are respectively the primary-side equivalent voltage of the output capacitor voltage and the primary-side equivalent voltage of the output voltage; , is the output capacitor voltage, is the turns ratio of the primary and secondary sides of the transformer; , is the output voltage, represents time. is the input voltage; is the inductance value; is the relative state variable of the primary and secondary switches. When , ; when , , and are respectively the primary-side voltage and the secondary-side voltage of the transformer. , is the equivalent parameter value of the inductor and the switch resistance, is the parasitic resistance value of the inductor, is the on-resistance value of the switch. , is the equivalent parameter value of the capacitor parasitic resistance, is the parasitic resistance value of the capacitor. , is the equivalent parameter value of the resistance, is the load resistance value. , is the equivalent parameter value of the capacitor, is the capacitance value. In any specific modulation mode (dual-phase-shift modulation mode and single-phase-shift modulation mode), as long as > 0 and < 0 both occur simultaneously within one period, and the time-domain expression can be established.

[0075] Step S2: Determine the collected power consumption information data set and the data set of parameters to be extracted, and establish the physical connection between the collected power consumption information data set and the data set of parameters to be extracted.

[0076] The elements in the collected power consumption information data set include the inductor current and the output voltage. The data set of parameters to be extracted is .

[0077] When the inductor current is used as the input variable, the differential linearization relationship is as follows:

[0078] ;

[0079] Among them, is the non-linear operator determined by the inductor current , and the expression is as follows:

[0080] ;

[0081] When the output voltage is used as the input variable, the differential linearization relationship is as follows:

[0082] ;

[0083] Among them, is the non-linear operator determined by the equivalent output voltage , and the expression is as follows:

[0084] .

[0085] Step S3: Establish a time-domain recurrence relationship based on the physical connection established in Step S2.

[0086] Step S3 is specifically as follows:

[0087] Step S31: Assume that the inductor current states at the starting time and the ending time are respectively and , and the output voltage states are respectively and , and satisfy ; At the same time, assume that there are unobservable intermediate states within the time period, and the intermediate states are: , , , where is the time coefficient, , ,…, , the intermediate state point and are implicit points, such as Figure 5 shown.

[0088] Step S32: Taking half of the switching period as a unit and based on the inductor current waveform, the starting moment is selected as the first turning point moment of the inductor current waveform. A random moment within the set threshold is set at the intermediate moment between the second and third turning point moments, and the point corresponding to this moment is the quasi-midpoint. The ending moment is selected as the second and third turning point moments. Then, the starting moments of the inductor current and the equivalent output voltage correspond to and respectively, and the endings correspond to and respectively. The starting moment and the ending moment corresponding points are explicit points, such as Figure 5 shown.

[0089] Step S33: To couple the implicit points and explicit points, use order implicit Runge-Kutta method, through the constant parameter set uniquely determined by the order , and are all relationship constants, is the time coefficient constant in Step S31. The order of the Runge-Kutta method of the physics-informed neural network is not fixed and can be adjusted according to actual recognition requirements. In this scenario, the 10th-order Runge-Kutta method is selected.

[0090] Establish the forward and backward recurrence relationships of the inductor current and the equivalent output voltage respectively to couple the explicit points and implicit points;

[0091] The expression of the backward recurrence relationship is as follows:

[0092] ;

[0093] The expression of the forward recurrence relationship is as follows:

[0094] .

[0095] The forward and backward recurrence relationships lay the foundation for the subsequent establishment of the physics-informed network part.

[0096] Step S4: Construct a physics-informed neural network for DAB converter parameter identification by fusing data mechanism and physical information, and train it. This physics-informed neural network updates the circuit parameter data serving as the weights of the physics-informed neural network by predicting the collected power information, while achieving accurate prediction of power information and precise identification of parameter data.

[0097] Specifically, step S4 is as follows:

[0098] Step S41: The data-driven network adopts a data-driven fully connected neural network, with the input being the power information of the explicit point corresponding to the starting moment 、the power information of the explicit point corresponding to the starting moment 、the variable S, and the time interval between the starting moment and the ending moment , with the bias and weights being , and the output being the power information of the implicit point at the intermediate moment predicted by the data-driven part and .

[0099] Step S42: Establish the physics-informed network part according to the backward recurrence relation expression and the forward recurrence relation expression. The input is the power information of the implicit point at the intermediate moment predicted by the data-driven part and , with the bias and weights being , and respectively derive the power information of the explicit points corresponding to the starting moment and the ending moment through the forward recurrence relation and the backward recurrence relation 、 、 and .

[0100] The structure of the data-driven neural network is not fixed, and the number of hidden layers and the number of neurons in each layer can be adjusted according to actual identification requirements. In this scenario, the selected data-driven network is a fully connected neural network with 4 hidden layers, each containing 50 neurons. Its activation functions are the Leaky Relu function and the tanh function, and its bias and weights are .

[0101] Step S43: Connect the data-driven network and the physics-informed network in series to form a physics-informed neural network for DAB converter parameter identification by fusing data mechanism and physical information, as shown in Figure 6 .

[0102] Step S44: Based on the simulation and experimental platform, collect the power data of the DAB converter according to Step S3 to form the collected power information dataset of the physics-informed neural network. In the acquisition stage of the simulation data, add examples with analog-to-digital conversion noise and sampling noise and examples using single-phase-shift modulation and dual-phase-shift modulation. In this embodiment, in the experimental data acquisition stage, the converter is controlled using single-phase-shift modulation.

[0103] Step S45: Determine the loss function and training stop conditions, and train the physics-informed neural network for the DAB converter parameter identification in Step S43. After reaching the stop condition, output the extracted parameter data.

[0104] In Step S45, the expression of the loss function is as follows:

[0105] ;

[0106] where is the loss function, , and the physics-informed neural network reduces the value of the loss function by continuously updating its as well as are also continuously updated, which respectively affect the calculation processes of the data-driven network and the physics-information network, thereby promoting the convergence of the predicted value to the true value. As shown in Figure 7 , and are the predicted values of the inductor current at the starting time and the ending time respectively, and are the predicted values of the output voltage at the starting time and the ending time respectively;

[0107] The training stop conditions are as follows:

[0108] When the number of training cycles reaches the first set number, the first set number in this embodiment is 1000, or continuously twice ;

[0109] or the number of training cycles reaches the second set number, and the second set number is 20000.

[0110] After reaching the above stop conditions, output the final value, which is the final unknown dataset of parameters to be extracted. The comparison chart of the experimental, simulation, and prediction results is as shown in Figures 8-9As shown. In all simulation examples, the average percentage error of the extracted parameters does not exceed 1.5%. In the experiments conducted on the hardware-in-the-loop platform, the average percentage error of all extracted parameters does not exceed 6.8%, verifying the effectiveness and accuracy of the method.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that they can still modify or equivalently replace the technical solutions of the present invention, and these modifications or equivalent replacements cannot make the modified technical solutions deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for extracting circuit parameters of a DAB converter based on a physics-informed neural network, characterized in that The specific steps are as follows: Step S1: Construct the time-domain differential expression of the DAB converter; Step S2: Determine the collected power quantity information data set and the to-be-extracted parameter data set, and establish the physical connection between the collected power quantity information data set and the to-be-extracted parameter data set; Step S3: Establish the time-domain recurrence relationship based on the physical connection established in Step S2; Step S4: Construct a physical information neural network for DAB converter parameter identification by fusing data mechanism and physical information and perform training. This physical information neural network updates the circuit parameter data serving as the weights of the physical information neural network by predicting the collected power quantity information, and simultaneously realizes the prediction of power quantity information and the identification of parameter data.

2. The method for extracting circuit parameters of a DAB converter based on a physics-informed neural network according to claim 1, wherein: When the product of the primary-side voltage and the secondary-side voltage of the transformer is greater than zero and less than zero within the same period, the time-domain differential expression of the DAB converter is established as follows: ； Among them, is the inductor current, and are respectively the primary equivalent voltage of the output capacitor voltage and the primary equivalent voltage of the output voltage; , is the output capacitor voltage, is the turns ratio of the primary and secondary windings of the transformer; , is the output voltage, represents time; is the input voltage; is the inductance value; is the relative state variable of the primary and secondary switches. When , ; when , , and are the primary-side voltage and secondary-side voltage of the transformer respectively; , is the equivalent parameter value of the inductor and the switching resistance, is the parasitic resistance value of the inductor, is the on-resistance value of the switch; , is the equivalent parameter value of the parasitic resistance of the capacitor, is the parasitic resistance value of the capacitor; , is the equivalent resistance parameter value, is the load resistance value; , is the equivalent capacitance parameter value, is the capacitance value.

3. The method for extracting circuit parameters of a DAB converter based on a physics-informed neural network according to claim 2, wherein Step S2 is specifically as follows: The elements in the collected power quantity information data set include inductor current and output voltage; The parameter dataset to be extracted is ; When the inductor current is used as the input variable, the differential linearization relationship is as follows: ； Among them, is a non-linear operator determined by the inductor current The expression is as follows: ； When the output voltage is used as the input variable, the differential linearization relationship is as follows: ； Among them, is a non-linear operator determined by the equivalent output voltage, and the expression is as follows: 。 4. A method for extracting circuit parameters of a DAB converter based on a physics-informed neural network according to claim 2, characterized in that Step S3 is specifically as follows: Step S31: Assume that at the starting moment and the ending moment , the inductor current states are respectively and , and the output voltage states are respectively and , and they satisfy ; Meanwhile, assume that there are unobservable intermediate states during the time period of , and the intermediate states are: , , , where is the time coefficient, , ,…, , and the intermediate state points and are implicit points; Step S32: Taking half of a switching period as a unit and based on the inductor current waveform, the starting moment is selected as the first turning point moment of the inductor current waveform. A random moment within the set threshold is set at the middle moment between the second turning point moment and the third turning point moment. The point corresponding to this moment is the quasi-midpoint. The ending moment is selected as the second turning point moment and the third turning point moment. Then the starting moments of the inductor current and the equivalent output voltage correspond to and respectively. The endings correspond to and respectively. The starting moment and the ending moment corresponding points are explicit points; Step S33: To couple implicit points and explicit points, use the implicit Runge-Kutta method of order determined uniquely by a set of constant parameters , and both being relationship constants, where is the time coefficient constant in step S31; Establish the forward and backward recurrence relationships of the inductor current and the equivalent output voltage respectively to couple the explicit point and the implicit point; The expression of the backward recurrence relationship is as follows: ； The expression of the forward recurrence relationship is as follows: 。 5. A method for extracting circuit parameters of a DAB converter based on a physics-informed neural network according to claim 2, characterized in that Step S4 is specifically as follows: Step S41: The data-driven network adopts a fully connected neural network, and the input is the power information of the explicit point corresponding to the starting moment , the power information of the explicit point corresponding to the starting moment , variable S and the time interval between the starting moment and the ending moment , and the bias and weights are , and the output is the power information of the implicit point at the intermediate moment predicted by the data-driven part and ; Step S42: Establish a physical information network according to the backward recurrence relation expression and the forward recurrence relation expression, with the input being the power information of the implicit points at the intermediate moment predicted by the data-driven network and , with the bias and weight being . Derive the power information of the explicit points corresponding to the starting moment and the ending moment respectively through the forward recurrence relation and the backward recurrence relation , , and ; Step S43: Connect the data-driven network and the physical information network in series to form a physical information neural network for DAB converter parameter identification by fusing data mechanism and physical information; Step S44: Through the simulation and experimental platform, collect the power data of the DAB converter according to Step S3 to form the collected power quantity information data set of the physical information neural network. In the collection stage of the simulation data, add examples with analog-to-digital conversion noise and sampling noise added, and examples using single-phase-shift modulation and double-phase-shift modulation; Step S45: Determine the loss function and the training stop condition to train the physical information neural network for DAB converter parameter identification in Step S43. After reaching the stop condition, output the extracted parameter data.

6. A method for extracting circuit parameters of a DAB converter based on a physics-informed neural network according to claim 5, characterized in that In Step S45, the expression of the loss function is as follows: ； Among them, is the loss function, and are the predicted values of the inductor current at the starting time and the ending time respectively; and are the predicted values of the output voltage at the starting time and the ending time respectively. The training stop condition is as follows: When the number of training cycles reaches the first set quantity, or twice in a row ; Or the number of training cycles reaches the second set quantity.

Images