Multi-parameter identification method for dc-dc boost converter

Abstract

The application provides a DC-DC step-up converter multi-parameter identification method, comprising the following steps: constructing a continuous state space model of a DC-DC step-up converter; discretizing the continuous state space model of the DC-DC step-up converter in step S1; constructing a physical information nested neural network, wherein the physical information nested neural network comprises a data driving layer and a physical layer in sequence from input to output, and the discretized continuous state space model of the DC-DC step-up converter is taken as a training model of the physical layer; obtaining training parameters, inputting the training parameters into the physical information nested neural network, constructing pseudo labels, inputting the pseudo labels into a loss function of the physical information nested neural network for semi-supervised training; obtaining identification parameters of the DC-DC step-up converter in real time, and inputting the identification parameters into the trained physical information nested neural network to obtain parameter identification results.

Description

Technical Field

[0001] This invention relates to a method for identifying parameters of power equipment, and more particularly to a method for identifying multiple parameters of a DC-DC boost converter. Background Technology

[0002] DC-DC boost converters are a type of switching power electronic device widely used in power management, electric vehicles, photovoltaic power generation, portable devices, and other fields. During long-term operation, the system parameters (such as inductance, capacitance, equivalent series resistance (ESR), load resistance, and power device on-resistance) will dynamically change due to the combined effects of external environmental changes, operating condition fluctuations, and component aging. These system parameters not only reflect the current operating status of the converter but are also closely related to its health level, energy efficiency, and control performance. Therefore, the ability to accurately identify the system parameters of DC-DC boost converters in real time is a key prerequisite for the efficiency, stability, and reliability analysis of DC-DC boost converters.

[0003] In existing technologies, multi-parameter identification methods for DC-DC boost converters, such as parameter estimation methods combining wavelet denoising and recursive least squares, and methods integrating data-driven and physical modeling, all require additional high-frequency sampling devices. This is because boost converters experience uncertain and discontinuous output voltage transients at each switching instant. These transients are mainly caused by sudden changes in the current direction in the output capacitor branch and are influenced by multiple parameters, including the output capacitance, load resistance, and output capacitor ESR. Conventional sampling strategies used in existing converter control systems (such as those based on DSP28335 and ARM chips) typically sample the system state only once at the intersection of the carrier and modulation waves. Since the output voltage abrupt changes occur on the nanosecond to microsecond scale, and the sampling circuit has unavoidable delays, the actual sampling point often only captures the state before the abrupt change, making it difficult to obtain crucial voltage information after the abrupt change. The additional sampling devices not only increase system cost and complexity but also reduce system reliability.

[0004] Therefore, in order to solve the above-mentioned technical problems, it is urgent to propose a new technical approach. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a multi-parameter identification method for DC-DC boost converters. This method utilizes a continuous state-space model of the DC-DC boost converter and constructs a physical information nested neural network. The discretized continuous state-space model is used as the training model for the physical layer of the physical information nested neural network. Pseudo-labels are set to perform semi-supervised training on the physical information nested neural network, thereby effectively capturing the state information after output voltage transients. This significantly improves the parameter identification accuracy of the DC-DC boost converter without requiring additional data acquisition equipment, thus simplifying the overall system structure, improving system reliability, and demonstrating strong generalization ability.

[0006] This invention provides a multi-parameter identification method for a DC-DC boost converter, comprising the following steps:

[0007] S1. Construct a continuous state-space model of the DC-DC boost converter;

[0008] S2. Discretize the continuous state-space model of the DC-DC boost converter in step S1;

[0009] S3. Construct a physical information nested neural network, which includes a data-driven layer and a physical layer from input to output, and uses the continuous state-space model of the discretized DC-DC boost converter as the training model of the physical layer.

[0010] S4. Obtain the training parameters and input them into the physical information nested neural network. Construct pseudo-labels and input them into the loss function of the physical information nested neural network for semi-supervised training.

[0011] S5. Acquire the identification parameters of the DC-DC boost converter in real time, and input the identification parameters into the trained physical information nested neural network to obtain the parameter identification results.

[0012] Furthermore, constructing the continuous state-space model of the DC-DC boost converter specifically includes:

[0013]

[0014] Where: i L V represents the inductor current of the DC-DC boost converter. o V represents the output voltage of the DC-DC boost converter. in R represents the input voltage of the DC-DC boost converter, L represents the input inductance of the DC-DC boost converter, and R represents the input voltage of the DC-DC boost converter. L The input inductor's parasitic resistance is represented by C, the output capacitor of the DC-DC boost converter is represented by R. CR represents the parasitic resistance of the output capacitor, and V represents the load resistance. F R represents the diode's forward voltage drop. dson S represents the on-resistance of the switch in the DC-DC boost converter. w Indicates the switch state; when the switch is on, S... w =1, S when the switch is off w =0;

[0015] λ represents the parameters to be identified, including the input inductance L and the parasitic resistance R of the input inductance. L Output capacitor C, parasitic resistance R of output capacitor C Load resistance R, diode forward voltage drop V F and the on-resistance R of the switch dson .

[0016] Furthermore, the construction of the physical layer training model based on the continuous state-space model of the discretized DC-DC boost converter specifically includes:

[0017]

[0018] Among them, c i b j and a ij The coefficient is determined by querying the implicit Longgokuta Butcher table. i and j represent the index of the hidden state, which is an integer between 1 and q. k represents the sequence number of the switching action. Δt represents the time interval of the on / off state.

[0019] Furthermore, in step S4, constructing pseudo-tags specifically includes:

[0020]

[0021] in: The pseudo-label value represents the output voltage after the instantaneous voltage change during switching. The pseudo-label value represents the inductor current after the instantaneous change in current during switching. This represents the output voltage before the voltage transient at the moment of switching. It represents the inductor current after the instantaneous voltage change during switching.

[0022] Furthermore, in step S4, the loss function is specifically as follows:

[0023]

[0024] in: This represents the inductor current output from the physical layer. This indicates the output voltage from the physical layer.

[0025] Furthermore, the data-driven layer of the physical information nested neural network is a feedforward fully connected neural network, specifically including an input layer, an output layer, and a hidden layer; wherein: the hidden layer uses the Sigmoid activation function, and the output layer does not use an activation function;

[0026] The calculation formula for each neuron in the data-driven layer is as follows:

[0027]

[0028] Where: l is the layer index of the neural network, w ij b represents the weight between the i-th neuron in layer l-1 and the j-th neuron in layer l. j It is the bias of the j-th neuron in the l-th layer, N I σ is the total number of neurons in the (l-1)th layer, and σ(.) represents the nonlinear equation.

[0029] The beneficial effects of this invention are as follows: By using a continuous state-space model of a DC-DC boost converter and constructing a physical information nested neural network, the discretized continuous state-space model is used as the training model for the physical layer of the physical information nested neural network. Pseudo-labels are set to perform semi-supervised training on the physical information nested neural network, thereby effectively capturing the state information after the output voltage transient. This effectively improves the parameter identification accuracy of the DC-DC boost converter without the need for additional data acquisition equipment, thus simplifying the structure of the entire system, improving the reliability of the system, and the method has strong generalization ability. Attached Figure Description

[0030] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0031] Figure 1 This is a schematic diagram of the process of the present invention.

[0032] Figure 2 This is a schematic diagram of the DC-DC boost converter topology of the present invention.

[0033] Figure 3 This is a schematic diagram of the physical information nested neural network structure of the present invention.

[0034] Figure 4 This is a schematic diagram illustrating the data acquisition process of the present invention. Detailed Implementation

[0035] The present invention will be further described in detail below:

[0036] This invention provides a multi-parameter identification method for a DC-DC boost converter, comprising the following steps:

[0037] S1. Construct a continuous state-space model of the DC-DC boost converter; such as... Figure 2 As shown, the actual structure of the boost converter in this invention is a BOOST boost circuit, which uses MOSFETs as switches. In the actual structure, there is no resistor R. dson and R C To illustrate clearly, the on-resistance of the switch and the parasitic resistance of the capacitor C are considered as a single resistor and formed... Figure 2 The topological structure in the model is used to establish a continuous state-space model.

[0038] S2. Discretize the continuous state-space model of the DC-DC boost converter in step S1; during the discretization process, the implicit Rungokuta method is used for discretization.

[0039] S3. Construct a physical information nested neural network, which includes a data-driven layer and a physical layer from input to output, and uses the continuous state-space model of the discretized DC-DC boost converter as the training model of the physical layer.

[0040] S4. Obtain the training parameters and input them into the physical information nested neural network. Construct pseudo-labels and input them into the loss function of the physical information nested neural network for semi-supervised training.

[0041] S5. The identification parameters of the DC-DC boost converter are acquired in real time and input into the trained physical information nested neural network to obtain the parameter identification results. Through the above method, the continuous state space model of the DC-DC boost converter is used to construct the physical information nested neural network. The discrete continuous state space model is used as the training model of the physical layer of the physical information nested neural network. Pseudo-labels are set to perform semi-supervised training on the physical information nested neural network, which can effectively capture the state information after the transient change of the output voltage. This can effectively improve the parameter identification accuracy of the DC-DC boost converter without the need to add other data acquisition equipment, thereby simplifying the structure of the entire system, improving the reliability of the system, and the method has strong generalization ability.

[0042] The DC-DC boost converter collects system status data, including inductor current i, each time the modulated wave intersects with the carrier wave. L Output voltage v o And the switching sequence signal. Due to the hardware delay of the sampling circuit, conventional sampling can only acquire the system state of the output voltage before the transient, and cannot acquire the voltage after the transient. Assuming the state before the transient is "-", the state after the transient is "+", and the switching action is k, then the instantaneous value of i... L -and v o- The time interval Δt between each switching state and the switching state S w Observable physical quantities will be used as inputs to the neural network, i.e., as training parameters, and the data acquisition process is as follows: Figure 4 As shown.

[0043] The identification parameters mentioned above are the same as the training parameters. The training parameters are the historical operating data of the DC-DC boost converter, while the identification parameters are parameters acquired in real time. Here, the identification parameters are not the parameters to be identified, but the current real-time operating status data of the DC-DC boost converter used to identify the parameters to be identified.

[0044] In this embodiment, constructing the continuous state-space model of the DC-DC boost converter specifically includes:

[0045]

[0046] Where: i L V represents the inductor current of the DC-DC boost converter. o V represents the output voltage of the DC-DC boost converter. in R represents the input voltage of the DC-DC boost converter, L represents the input inductance of the DC-DC boost converter, and R represents the input voltage of the DC-DC boost converter. L The input inductor's parasitic resistance is represented by C, the output capacitor of the DC-DC boost converter is represented by R. C R represents the parasitic resistance of the output capacitor, and V represents the load resistance. F R represents the diode's forward voltage drop. dson S represents the on-resistance of the switch in the DC-DC boost converter. w Indicates the switch state; when the switch is on, S... w =1, S when the switch is off w =0;

[0047] λ represents the parameters to be identified, including the input inductance L and the parasitic resistance R of the input inductance. L Output capacitor C, parasitic resistance R of output capacitor C Load resistance R, diode forward voltage drop V F and the on-resistance R of the switch dson .

[0048] The training model for the physical layer, constructed based on the continuous state-space model of the discretized DC-DC boost converter, specifically includes:

[0049]

[0050]

[0051] Among them, c i bj and a ij The coefficients are determined by querying the implicit Longgokuta Butcher table. i and j represent the indices of the hidden states, which are integers between 1 and q. k represents the sequence number of the switching action, and Δt represents the time interval of the on / off state. Through the above transformation, the continuous spatial state model can be used to train the neural network by embedding physical information, thereby improving the accuracy of the final recognition result.

[0052] In this embodiment, step S4, constructing pseudo-tags specifically includes:

[0053]

[0054] in: The pseudo-label value represents the output voltage after the instantaneous voltage change during switching. The pseudo-label value represents the inductor current after the instantaneous change in current during switching. This represents the output voltage before the voltage transient at the moment of switching. This represents the inductor current after the instantaneous voltage change during switching. As described in the background section, since it is impossible to collect information about the output voltage transient in the existing system structure, this pseudo-label can enrich the training data, ensuring that the neural network does not lose the response characteristics of the entire system during parameter identification, thereby ensuring the accuracy of the final identification result.

[0055] In this embodiment, the loss function in step S4 is specifically:

[0056]

[0057] in: This represents the inductor current output from the physical layer. The output voltage of the physical layer is represented by this loss function. Pseudo-labels are added to the loss function to achieve semi-supervised training, thereby effectively ensuring the accuracy of the final recognition result.

[0058] like Figure 3 As shown: In this embodiment, the data-driven layer of the physical information nested neural network is a feedforward fully connected neural network, specifically including an input layer, an output layer, and a hidden layer; wherein: the hidden layer uses the Sigmoid activation function, and the output layer does not use an activation function;

[0059] The calculation formula for each neuron in the data-driven layer is as follows:

[0060]

[0061] Where: l is the layer index of the neural network, w ijb represents the weight between the i-th neuron in layer l-1 and the j-th neuron in layer l. j It is the bias of the j-th neuron in the l-th layer, N I σ is the total number of neurons in the (l-1)th layer, and σ(.) represents the nonlinear equation.

[0062] 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 it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for multi-parameter identification of a DC-DC boost converter, characterized in that: Includes the following steps: S1. Construct a continuous state-space model of the DC-DC boost converter; S2. Discretize the continuous state-space model of the DC-DC boost converter in step S1; S3. Construct a physical information nested neural network, which includes a data-driven layer and a physical layer from input to output, and uses the continuous state-space model of the discretized DC-DC boost converter as the training model of the physical layer. S4. Obtain the training parameters and input them into the physical information nested neural network. Construct pseudo-labels and input them into the loss function of the physical information nested neural network for semi-supervised training. S5. Acquire the identification parameters of the DC-DC boost converter in real time, and input the identification parameters into the trained physical information nested neural network to obtain the parameter identification results; Constructing a continuous state-space model of a DC-DC boost converter specifically includes: (1)； in: This represents the inductor current of the DC-DC boost converter. This indicates the output voltage of the DC-DC boost converter. This represents the input voltage of the DC-DC boost converter. This represents the input inductance of the DC-DC boost converter. The parasitic resistance of the input inductor. This represents the output capacitor of the DC-DC boost converter. This represents the parasitic resistance of the output capacitor. Indicates the load resistance. This indicates the forward voltage drop of the diode. This represents the on-resistance of the switch in the DC-DC boost converter; Indicates the switch state; when the switch is on. When the switch is turned off ; Indicates the parameters to be identified, including the input inductance. Parasitic resistance of input inductor Output capacitor Parasitic resistance of the output capacitor Load resistance Diode forward voltage drop and the on-resistance of the switch ; The training model for the physical layer, constructed based on the continuous state-space model of the discretized DC-DC boost converter, specifically includes: in, , and The coefficients are determined by querying the implicit Longgokuta Butcher table. and The index representing the hidden state is an integer between 1 and q. Indicates the sequence number of the switch action. This indicates the time interval between the on / off states.

2. The multi-parameter identification method for DC-DC boost converter according to claim 1, characterized in that: In step S4, constructing pseudo-tags specifically includes: (6)； in: The pseudo-label value represents the output voltage after the instantaneous voltage change during switching. The pseudo-label value represents the inductor current after the instantaneous change in current during switching. This represents the output voltage before the voltage transient at the moment of switching. It represents the inductor current after the instantaneous voltage change during switching.

3. The multi-parameter identification method for DC-DC boost converter according to claim 2, characterized in that: In step S4, the loss function is specifically as follows: (7) in: This represents the inductor current output from the physical layer. This indicates the output voltage from the physical layer.

4. The multi-parameter identification method for a DC-DC boost converter according to claim 1, characterized in that: The data-driven layer of the physical information nested neural network is a feedforward fully connected neural network, specifically including an input layer, an output layer, and a hidden layer; wherein: the hidden layer uses the Sigmoid activation function, and the output layer does not use an activation function; The calculation formula for each neuron in the data-driven layer is as follows: (8)； Where: l is the layer index of the neural network, w ij b represents the weight between the i-th neuron in layer l-1 and the j-th neuron in layer l. j It is the bias of the j-th neuron in the l-th layer, N I σ is the total number of neurons in the (l-1)th layer, and σ(.) represents the nonlinear equation.

Images