EIS online detection system and method for high-power fuel cell system

Abstract

The application provides an EIS online detection system and method for a high-power fuel cell system, a main loop of which adopts a two-way parallel Boost converter topology structure, and a small signal excitation injection is performed through an auxiliary DC / DC converter-2; a control system comprises a sampling system, a PWM driving circuit, a communication module and an embedded controller; a battery equivalent circuit model is established; when the high-power fuel cell system is running, a multi-frequency sinusoidal periodic signal optimized by a non-dominated sorting genetic algorithm based on Tent chaotic mapping improvement is injected into an output current of the auxiliary DC / DC converter-2; a single cell or stack voltage is sampled, current data is measured, a voltage drop corresponding to the internal resistance is calculated, a time domain signal is converted into a frequency domain signal, and battery impedance information in a modified Fouquet equivalent circuit model is obtained; characteristic parameter data is imported into a least square support vector machine (LSSVM) for training and testing, and in-situ detection and fault diagnosis are performed online; and the application can provide a new way for active diagnosis and early warning of faults.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fuel cell systems, in particular to an EIS online detection system and method for a high-power fuel cell system. BACKGROUND

[0002] The wide application of hydrogen energy in the fields of transportation and energy provides an effective way for the recycling of carbon-free energy. Among them, fuel cells play a key role as the core technology in the development and application of hydrogen energy. Fuel cells have the advantages of high energy efficiency, good environmental friendliness, high dynamic response, etc. However, under different operating conditions, fuel cells often face frequent start-stop and large discharge depth, which puts high requirements on their performance, especially in terms of rapid start-up and dynamic response capability. Frequent start-stop and severe working changes can easily lead to membrane drying, water flooding and other faults of fuel cells, affecting their stability and life. Moreover, since high-power fuel cells have large output current, small signal excitation current is easily submerged by high DC bias and difficult to detect. Therefore, it is necessary to accurately detect the operating state parameters of fuel cell stacks and individual cells to provide data support for fault diagnosis systems. However, due to the multi-physical field coupling characteristics and material individuality of fuel cells, it is difficult to directly obtain the internal state parameters. Electrochemical impedance spectroscopy (EIS) can superimpose various disturbance excitation signals on fuel cells, and extract and calculate features with electrochemical equivalent circuit models, so as to obtain the internal state information of fuel cells, which can effectively monitor and diagnose faults. SUMMARY

[0003] The present application proposes an EIS online detection system and method for a high-power fuel cell system, proposes a modified Fouquet equivalent circuit model, and adopts a two-way parallel Boost converter topology structure to monitor and calculate the internal state information of fuel cell stacks during operation, injects a multi-frequency sinusoidal periodic signal optimized by a non-dominated sorting genetic algorithm based on Tent chaotic mapping into an auxiliary DC / DC converter, and then uses EIS characteristic data to realize the characterization of partial faults, providing a new way for active diagnosis and early warning of faults.

[0004] The present application adopts the following technical solutions.

[0005] An EIS online detection system for a high-power fuel cell system, comprising a main circuit and a control system; the main circuit adopts a two-way parallel Boost converter topology structure and performs small signal excitation injection through an auxiliary DC / DC converter; the control system comprises a sampling system, a PWM drive circuit, a communication module and an embedded controller; an equivalent circuit model of the battery is established;

[0006] The main circuit comprises a fuel cell, a DC / DC converter and a direct current load connected in sequence; the control system comprises a sampling system, a PWM driving circuit, a communication module and an embedded controller; the sampling system comprises a CVM inspector, a voltage sensor and a high-precision current Hall sensor; the CVM inspector integrated in the sampling system is used for real-time processing and calculation of collected fuel cell voltage to calculate data of single cell voltage consistency state, average voltage and total voltage; the voltage sensor is used for measuring total fuel cell output voltage; the current sensor is used for measuring fuel cell output current; and the communication module is connected to the upper computer.

[0007] The main circuit adopts a two-way parallel Boost converter topology structure, one way for output power transmission to meet the load demand, and the other way for injection of a multi-frequency sinusoidal periodic excitation signal with low DC bias to avoid detection difficulty caused by high DC bias of the high-power fuel cell and improve detection precision of small signal excitation.

[0008] The detection method of the EIS online detection system for the high-power fuel cell system, when the high-power fuel cell system is running, a multi-frequency sinusoidal periodic signal optimized by a non-dominated sorting genetic algorithm based on Tent chaotic mapping is injected into an auxiliary DC / DC converter output current; single or stack voltage is sampled by a fuel cell voltage inspection CVM, and current data are measured by a Hall current sensor and other elements, and voltage drop corresponding to internal resistance is calculated according to open circuit voltage and output voltage of the fuel cell, and time domain signals are converted into frequency domain signals by fast Fourier transform; battery impedance information in a modified Fouquet equivalent circuit model is obtained according to amplitude and phase fitting at specific frequency points; characteristic parameter data are imported into a least square support vector machine LSSVM for training and testing, and in-situ detection and fault diagnosis are performed online.

[0009] The multi-frequency sinusoidal periodic excitation signal in the diagnosis system is optimized by a non-dominated sorting genetic algorithm based on Tent chaotic mapping, the phase of different frequency sinusoidal signals is optimized, the crest factor of the combined multi-frequency sinusoidal signal is reduced, and the problem of reduced EIS measurement precision caused by excessive amplitude superposition of the excitation signal is avoided.

[0010] The high-power fuel cell is improved by an equivalent circuit model, and a Fouquet model composed of four elements of ohmic internal resistance R ohm , activation internal resistance R ct , Warburg diffusion resistance Z w and constant phase Z CPE is replaced by a series Foster circuit. wTo further accurately simulate the low and high frequency characteristics of high-power fuel cells, a double-layer distributed capacitance C dl is adopted to form a modified Fouquet equivalent circuit model composed of series and parallel ohmic resistance R ohm , activation resistance R ct , concentration resistance R w , double-layer distributed capacitance C dl and concentration capacitance C w ; wherein, the ohmic resistance R ohm , activation resistance R ct indicate the degree of membrane dryness, the concentration resistance R w indicates the degree of water flooding, and the double-layer distributed capacitance C dl indicates the electrical characteristics generated by the bipolar plate and electrolyte of the stack.

[0011] The online detection method of the detection method is used to identify the normal, membrane dryness and water flooding states of the fuel cell stack;

[0012] The data for LSSVM model training and model verification are derived from experimental samples with clear conclusions and contain state labels; the training set data and test set data are randomly divided in a ratio of 3:1;

[0013] The least squares support vector machine LSSVM uses the ohmic resistance, activation resistance and impedance at 2, 4, 8, 16, 32, 64, 128, 256 and 1024 frequency points in three states as feature vectors, and the label values are normal, membrane dryness and water flooding, corresponding to numbers 1, 2 and 3 respectively.

[0014] In the detection method, the sampling system collects voltage and current information, and converts the analog signals into digital signals in real time through an A / D converter; one signal is input into the communication module and then into the embedded controller for pre-data processing and calculation, and the other signal is input into the DC / DC controller for PID operation and then through the PWM drive circuit to realize stable voltage or current control.

[0015] The communication module uses CAN field bus protocol to transmit real-time data to the host computer and information storage medium for display and storage respectively.

[0016] The fuel cell voltage inspection CVM directly interacts with the communication module without passing through the A / D converter.

[0017] The method comprises the following steps:

[0018] Step S1, the initial population is valued by the NSGA-II genetic algorithm based on the Tent chaotic mapping, the convergence ability of the algorithm is improved, the phase of the sinusoidal excitation signal under different frequencies is optimized by using the improved algorithm, and the optimization target is the crest factor, that is, the ratio of the peak value of the signal to twice the effective value;

[0019] The frequency values of each sinusoidal component of the sinusoidal excitation signal are 2, 4, 8, 16, 32, 64, 128, 256 and 1024, and the amplitude values of each sinusoidal component are all 4A, and the phase value is 0 rad by default.

[0020] Step S2, a modified Fouquet equivalent circuit model is established to describe the Nyquist characteristics of the fuel cell; wherein the internal resistance of the fuel cell is simplified as a circuit composed of ohmic resistance R ohm , activation resistance R ct , concentration resistance R w , double-layer distributed capacitance C dl and concentration capacitance C w ;

[0021] Step S3, the optimized multi-frequency sinusoidal excitation signal is superimposed on the output current of the auxiliary DC / DC converter of the fuel cell system through the PWM driving circuit;

[0022] Step S4, after the excitation signal is injected, the discrete sequences of the response voltage and the excitation current are synchronously collected by the sampling system, and then the fast Fourier transform is performed to obtain the frequency domain voltage complex sequence and the frequency domain current complex sequence; the impedance corresponding to the excitation frequency is calculated by impedance spectrum, that is, the response voltage of each frequency sinusoidal component of the excitation current signal and the internal resistance under each frequency sinusoidal excitation, the amplitude and phase information of the impedance corresponding to the frequency point are obtained, and the Nyquist curve is fitted; the single cell voltage signal collected by the CVM is directly input into the communication module and uploaded to the upper computer;

[0023] Step S5, the calculated impedance information is used as the original data set for fault diagnosis training and testing, wherein there are 20 characteristic data and 1 label column; the data set is input into the LSSVM vector machine to complete fault diagnosis.

[0024] Step S1 specifically includes the following steps:

[0025] The function of the multi-frequency sinusoidal excitation signal is:

[0026]

[0027] In the formula, f(t) is the frequency of the excitation model at time t.

[0028] The initialization population of the NSGA-II genetic algorithm is obtained by Tent chaotic mapping.

[0029] The optimization target of the NSGA-II genetic algorithm is defined as a crest factor representing the uniformity of signal amplitude distribution in the entire time domain, and the calculation formula is:

[0030]

[0031] In the formula, I ac,max is the maximum value of the multi-frequency sinusoidal excitation signal; I ac,min is the minimum value of the multi-frequency sinusoidal excitation signal; I eff is the effective value of the multi-frequency sinusoidal excitation signal. The expression of the multi-frequency sinusoidal excitation signal effective value I eff is:

[0032]

[0033] Further, step S2 specifically includes the following steps:

[0034] The impedance of the modified Fouquet equivalent circuit model can be represented as:

[0035]

[0036] According to the above formula, the real part and the imaginary part of the impedance are calculated, and the Nyquist curve is drawn in the complex plane according to the relationship between the real part and the imaginary part of the modified Fouquet equivalent circuit model at each frequency, which can represent the electrochemical impedance spectrum of the fuel cell.

[0037] Step S4 specifically includes the following steps:

[0038] The sampling system includes CVM patrolers, Hall current sensors, voltage sensors, main components, and precision resistors, analog-to-digital converters ADC, and other sampling required components; wherein the CVM patroler can monitor the internal of the battery stack, measure the current operating state of a single battery, and transmit real-time data to the main controller through the field bus protocol. The detailed voltage data measured by the CVM patroler can be used for fault diagnosis, and when voltage imbalance, unit failure and other problems occur, the fault alarm can be directly sent out to trigger and record the voltage data at the time of failure; step S5 specifically includes the following steps:

[0039] The LSSVM simplifies the standard SVM by converting the constraint conditions into a square loss function to simplify the optimization process;

[0040] According to the SRM criterion, the objective function and the constraint condition of the LSSVM are:

[0041]

[0042] In the formula, ξ iis a prediction error; c is a regularization parameter or penalty parameter for controlling the degree of sample penalty on the prediction error ξ i LSSVM handles errors by minimizing the square loss, avoiding the introduction of slack variables; by constructing a Lagrange transformation polynomial, the following transformation is performed:

[0043]

[0044] The partial derivatives of the above formula w, b, e k , α k are taken respectively and the derivatives are zero, that is, the linear equations are obtained, and the solution is the calculation formula of the LSSVM algorithm:

[0045]

[0046] In the formula, K(x, x i ) is a kernel function.

[0047] The application provides an EIS online detection system and method for a high-power fuel cell system, comprising a main loop and a control system. The main loop adopts a two-way parallel Boost converter topology structure, and a small signal excitation signal is injected through an auxiliary DC / DC converter-2. The control system comprises a sampling system, a PWM driving circuit, a communication module and an embedded controller; an equivalent circuit model of the battery is established; when the high-power fuel cell system is running, a multi-frequency sinusoidal periodic signal optimized by a non-dominated sorting genetic algorithm based on Tent chaotic mapping improvement is injected into the output current of the auxiliary DC / DC converter-2; the fuel cell voltage is patrolled through a CVM sampling single or stack voltage, and current data are measured by combining a Hall current sensor and other elements, the voltage drop corresponding to the internal resistance of the fuel cell is calculated according to the open-circuit voltage and the output voltage of the fuel cell, and the time-domain signal is converted into a frequency-domain signal through fast Fourier transform. According to the amplitude and phase fitting at a specific frequency point, the battery impedance information in the corrected Fouquet equivalent circuit model is obtained; the characteristic parameter data are imported into a least square support vector machine (LSSVM) for training and testing, and in-situ detection and fault diagnosis are performed online; compared with the prior art, the application has the following beneficial effects:

[0048] 1. The scheme can realize the functional integration of signal optimization, equivalent circuit parameter identification and fuel cell EIS fault diagnosis, provides a high-precision high-power fuel cell EIS fault diagnosis method, and opens up a new idea for in-situ detection of high-power fuel cells.

[0049] 2. The method adopts a two-way parallel Boost converter topology structure, and an excitation signal is injected through an auxiliary DC / DC converter, thereby avoiding the influence of high-power fuel cell direct current bias on the small current excitation signal.

[0050] 3、The method can optimize the excitation signal, obtain impedance information of multiple frequencies at one time, improve the speed of EIS time domain measurement, and effectively avoid the influence of excessive amplitude caused by signal superposition.

[0051] 4、The system integrates a CVM inspector, can process and calculate voltage data of the fuel cell in real time, obtains parameters such as single-cell voltage consistency state, average voltage and total voltage, and provides strong technical support for in-situ monitoring of high-power fuel cells, online rapid measurement of EIS and research and development of fault diagnosis technology.

[0052] 5、The application measures EIS of the high-power fuel cell online, monitors and calculates information of the internal state of the fuel cell stack during operation, and then realizes representation of partial faults by using impedance characteristic data, thereby providing a new way for active diagnosis and early warning of faults. BRIEF DESCRIPTION OF DRAWINGS

[0053] The application will be further described in detail below in combination with the drawings and specific embodiments:

[0054] Figure 1 is a hardware structure schematic diagram of a preferred embodiment of the application; Fig. 1 Figure 1 is a hardware structure schematic diagram of a preferred embodiment of the application;

[0055] Figure 2 is a control system circuit principle schematic diagram of the preferred embodiment of the application; Fig. 2 Figure 2 is a control system circuit principle schematic diagram of the preferred embodiment of the application;

[0056] Figure 3 is an EIS schematic diagram of a fuel cell in a healthy state simulated in the embodiment; Fig. 3 Figure 3 is an EIS schematic diagram of a fuel cell in a healthy state simulated in the embodiment;

[0057] Figure 4 is an EIS obtained by simulating change of an ohmic resistance in the embodiment, wherein the solid line part is a normal fuel cell EIS, the dashed line part is a fault fuel cell EIS, and the circle dashed line part is a fault fuel cell EIS schematic diagram estimated by the FFT algorithm in the present application; Fig. 4 Figure 4 is an EIS obtained by simulating change of an ohmic resistance in the embodiment, wherein the solid line part is a normal fuel cell EIS, the dashed line part is a fault fuel cell EIS, and the circle dashed line part is a fault fuel cell EIS schematic diagram estimated by the FFT algorithm in the present application;

[0058] Figure 5 is an EIS schematic diagram obtained by simulating change of an activation resistance in the embodiment; Fig. 5 Figure 5 is an EIS schematic diagram obtained by simulating change of an activation resistance in the embodiment;

[0059] Figure 6 is a LSSVM support vector machine fault diagnosis result schematic diagram in the embodiment (in the figure, sample class labels 1, 2 and 3 respectively represent three states of normal, membrane dry and flooding). Fig. 6 Figure 6 is a LSSVM support vector machine fault diagnosis result schematic diagram in the embodiment (in the figure, sample class labels 1, 2 and 3 respectively represent three states of normal, membrane dry and flooding). DETAILED DESCRIPTION

[0060] The application will be further described in detail below in combination with the drawings and specific embodiments:

[0061] It should be noted that the following detailed description is illustrative only, and is intended to provide further description in connection with the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the application belongs.

[0062] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application; as used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise; it will be further understood that the terms "comprises" and / or "comprising", when used in this specification, specify the presence of stated features, steps, operations, devices, components and / or combinations thereof, but do not preclude the presence or addition of one or more other features, steps, operations, devices, components and / or combinations thereof.

[0063] As shown in the figure, an EIS online detection system for a high-power fuel cell system includes a main loop and a control system; the main loop adopts a two-way parallel Boost converter topology structure, and small signal excitation injection is performed through an auxiliary DC / DC converter; the control system includes a sampling system, a PWM driving circuit, a communication module and an embedded controller; an equivalent circuit model of the battery is established;

[0064] The main loop includes a fuel cell, a DC / DC converter and a DC load connected in sequence; the control system includes a sampling system, a PWM driving circuit, a communication module and an embedded controller; the sampling system includes a CVM patrol detector, a voltage sensor and a high-precision current Hall sensor; the CVM patrol detector integrated in the sampling system is used for real-time processing and calculation of the collected fuel cell voltage, and data of single cell voltage consistency state, average voltage and total voltage are calculated; the voltage sensor measures the total voltage of the fuel cell output; the current sensor measures the output current of the fuel cell; the communication module is connected to the upper computer;

[0065] The main loop adopts a two-way parallel Boost converter topology structure, one way is used for output power transmission to meet the load demand, and the other way is used for injection of the multi-frequency sinusoidal periodic excitation signal, and the direct current bias is low to avoid the detection difficulty problem caused by the high direct current bias of the high-power fuel cell being submerged, thereby improving the detection precision of the small signal excitation.

[0066] The detection method of the EIS online detection system for the high-power fuel cell system, when the high-power fuel cell system is running, a multi-frequency sinusoidal periodic signal optimized by a non-dominated sorting genetic algorithm based on Tent chaotic mapping is injected into the output current of the auxiliary DC / DC converter; the single cell or stack voltage is sampled through the fuel cell voltage patrol CVM, and the current data is measured by combining elements such as a Hall current sensor, the voltage drop corresponding to the internal resistance of the fuel cell is calculated according to the open circuit voltage and output voltage of the fuel cell, the time domain signal is converted into a frequency domain signal through fast Fourier transform; the amplitude and phase fitting at a specific frequency point are obtained to obtain the impedance information of the battery in the modified Fouquet equivalent circuit model; the characteristic parameter data is imported into a least square support vector machine (LSSVM) for training and testing, and online in-situ detection and fault diagnosis are performed.

[0067] The multi-frequency sinusoidal periodic excitation signal in the diagnostic system is optimized by a non-dominated sorting genetic algorithm based on Tent chaotic mapping, the peak factor of the combined multi-frequency sinusoidal signal is reduced by optimizing the phase of the different frequency sinusoidal signals, and the problem of reduced EIS measurement accuracy caused by excessive amplitude superposition of the excitation signal is avoided.

[0068] The high-power fuel cell is improved by an equivalent circuit model, and the Fouquet model composed of four elements of ohmic internal resistance R ohm , activation internal resistance R ct , Warburg diffusion resistance Z w and constant phase Z CPE is improved, the series Foster circuit is used to replace the Warburg diffusion resistance Z w to further accurately simulate the low-frequency and high-frequency characteristics of the high-power fuel cell, and the double-layer distributed capacitance C dl is used to replace the CPE element to reduce the fitting difficulty, and the modified Fouquet equivalent circuit model composed of the ohmic internal resistance R ohm , the activation internal resistance R ct , the concentration resistance R w , the double-layer distributed capacitance C dl and the concentration capacitance C w is formed in series and parallel; wherein, the ohmic internal resistance R ohm , the activation internal resistance R ct indicate the degree of membrane dry fault, the concentration resistance R w indicates the degree of waterlogging fault, and the double-layer distributed capacitance C dl indicates the electrical characteristics generated by the stack bipolar plate and the electrolyte.

[0069] The online detection method of the detection method is used to identify the normal, membrane dry and waterlogging states of the fuel cell stack;

[0070] The data for LSSVM model training and model verification are derived from experimental samples with clear conclusions and contain state labels; the training set data and the test set data are randomly divided in a ratio of 3:1;

[0071] The least square support vector machine (LSSVM) takes the ohmic resistance, activation resistance and impedance at 2, 4, 8, 16, 32, 64, 128, 256 and 1024 frequency points in three states as characteristic vectors, and takes the normal, membrane dry and waterlogging states as label values, which correspond to 1, 2 and 3 respectively.

[0072] In the detection method, the sampling system collects voltage and current information, and converts analog signals into digital signals in real time through an A / D converter; one-way signals are input into a communication module and embedded into a controller for pre-data processing and calculation, and the other way is input into a DC / DC controller for PID operation and through a PWM drive circuit to realize stable voltage or current control.

[0073] The communication module uses CAN field bus protocol to transmit real-time data to the host computer and information storage medium for display and storage respectively.

[0074] The fuel cell voltage inspection CVM directly interacts with the communication module without passing through the A / D converter.

[0075] The method comprises the following steps:

[0076] Step S1, the initial population is valued by the NSGA-II genetic algorithm based on the Tent chaotic mapping improvement, and the algorithm convergence ability is improved; the phase of the sinusoidal excitation signal under different frequencies is optimized using the improved algorithm; wherein the optimization target is the wave crest factor, that is, the ratio of the signal peak value to twice the effective value;

[0077] The frequency values of each sinusoidal component of the sinusoidal excitation signal are 2, 4, 8, 16, 32, 64, 128, 256 and 1024 respectively, and the amplitude values of each sinusoidal component are all 4A, and the phase value is 0 rad by default.

[0078] Step S2, a modified Fouquet equivalent circuit model describing the Nyquist characteristics of the fuel cell is established; wherein the internal resistance of the fuel cell is simplified as a circuit composed of ohmic resistance R ohm , activation resistance R ct , concentration resistance R w , double-layer distributed capacitance C dl and concentration capacitance C w in series and parallel connection;

[0079] Step S3, the optimized multi-frequency sinusoidal excitation signal is superimposed on the output current of the stable working fuel cell system auxiliary DC / DC converter through the PWM drive circuit.

[0080] Step S4, after injecting the excitation signal, the sampling system synchronously collects the discrete sequences of response voltage and excitation current, and then converts them into frequency domain voltage complex sequence and frequency domain current complex sequence through fast Fourier transform; the impedance corresponding to the excitation frequency is calculated, that is, the response voltage of each frequency sinusoidal component of the excitation current signal and the internal resistance under each frequency sinusoidal excitation, the amplitude and phase information of the impedance corresponding to the frequency point are obtained, and the Nyquist curve is fitted; the voltage signal of the single battery collected by the CVM is directly input into the communication module and uploaded to the upper computer;

[0081] Step S5, the calculated impedance information is used as the original data set for fault diagnosis training and testing, wherein there are 20 characteristic data and 1 label column; the data set is input into the LSSVM vector machine to complete fault diagnosis.

[0082] Step S1 specifically includes the following steps:

[0083] The function of the multi-frequency sinusoidal excitation signal is:

[0084]

[0085] In the formula: f(t) is the frequency of the excitation model at time t.

[0086] The initialization population of the NSGA-II genetic algorithm is obtained by Tent chaotic mapping.

[0087] The optimization target of the NSGA-II genetic algorithm is defined as the crest factor representing the uniformity of signal amplitude distribution in the entire time domain, and the calculation formula is:

[0088]

[0089] In the formula: I ac,max is the maximum value of the multi-frequency sinusoidal excitation signal; I ac,min is the minimum value of the multi-frequency sinusoidal excitation signal; I eff is the effective value of the multi-frequency sinusoidal excitation signal. The expression of the multi-frequency sinusoidal excitation signal effective value I eff is:

[0090]

[0091] Further, step S2 specifically includes the following steps:

[0092] The impedance of the modified Fouquet equivalent circuit model can be expressed as:

[0093]

[0094] The real and imaginary parts of the impedance are calculated based on the above formula. The Nyquist curve is then plotted in the complex plane based on the relationship between the real and imaginary parts of the Fouquet equivalent circuit model at various frequencies to represent the electrochemical impedance spectrum of the fuel cell.

[0095] Step S4 specifically includes the following steps:

[0096] The sampling system includes key components such as a CVM (Cellular Dynamic Monitoring Unit), Hall current sensor, and voltage sensor, as well as sampling components such as precision resistors and analog-to-digital converters (ADCs). The CVM monitor can monitor the internal structure of the battery stack, measure the current operating status of individual cells, and transmit real-time data to the main controller via a fieldbus protocol. The detailed voltage data measured by the CVM can be used for fault diagnosis. Furthermore, when voltage imbalance, cell failure, or other problems occur, a fault alarm can be directly issued, triggering and recording the voltage data at the time of the fault. Step S5 specifically includes the following steps:

[0097] The LSSVM simplifies the standard SVM by converting the constraints into a squared loss function, thus simplifying the optimization process.

[0098] According to the SRM criterion, the objective function and constraints of LSSVM are as follows:

[0099]

[0100] In the formula, ξ i ξ is the prediction error; c is the regularization parameter or penalty parameter, used to control the impact on the prediction error ξ. i The degree of sample penalty. LSSVM handles the error by minimizing the squared loss, avoiding the introduction of slack variables; by constructing a Lagrange transform polynomial, the following transformation is performed:

[0101]

[0102] For the above equation w, b, e k α k By taking the partial derivatives and setting them to zero, we obtain the system of linear equations. Solving this system yields the calculation formula for the LSSVM algorithm:

[0103]

[0104] In the formula, K(x,x) i ) is the kernel function.

[0105] Example 1

[0106] This example presents an EIS online detection system and method for high-power fuel cell systems, referencing... Figs. 1 to 6The system includes a main circuit, control system, PWM drive circuit, and sampling module. An equivalent circuit model of the battery is established. During high-power fuel cell system operation, a multi-frequency sinusoidal periodic signal optimized by a non-dominated sorting genetic algorithm based on Tent chaotic mapping is superimposed on the fuel cell output current. This signal is then used to excite the fuel cell under test via the PWM drive circuit. Voltage data of each unit in the fuel cell is obtained through online sampling using fuel cell voltage monitoring (CVM), combined with current data measured by components such as Hall current sensors. The voltage drop corresponding to the internal resistance is calculated based on the fuel cell open-circuit voltage and output voltage. The time-domain signal is converted to a frequency-domain signal using a fast Fourier transform. Battery impedance information is obtained by fitting the amplitude and phase at specific frequency points. The feature parameter data is imported into a least squares support vector machine (LSSVM) for training and testing, enabling online in-situ detection and fault diagnosis.

[0107] The main circuit includes a fuel cell, a main DC / DC converter-1, an auxiliary DC / DC converter-2, and a DC load connected in sequence; the control system includes a sampling system, a PWM drive circuit, a communication module, and an embedded controller; the sampling system includes a CVM detector, a voltage sensor, and a high-precision current Hall sensor; the input terminal of the PWM drive circuit is connected to the DC / DC controller, and the output terminal of the PWM drive circuit is connected to the MOSFET in the DC / DC converter; the input terminals of the voltage sensor and the high-precision current Hall sensor of the sampling system are both connected to the DC / DC converter, and their output terminals are connected to the A / D conversion module.

[0108] The main circuit consists of a fuel cell, two parallel Boost converters, and a load. The system employs an integrated method for testing the internal resistance of the fuel cell using both DC / DC converters. The two DC / DC converters are used to transmit the fuel cell's output power and superimpose the excitation signal, respectively. High-frequency and mid-to-low-frequency excitation signals are applied to the voltage and current loops of the auxiliary DC / DC converter-2 to acquire the PEMFC's output voltage and compare it with the set desired output voltage. The output of the PWM drive circuit is connected to the gate of the DC / DC converter's switching transistor. Based on the comparison result, the PWM duty cycle of the MOSFET is adjusted, thereby regulating the DC / DC converter's output signal. The main DC / DC converter-1 ensures the stability of the system's output voltage, while the auxiliary DC / DC converter-2 provides the necessary excitation signal to the battery.

[0109] The communication module uses the CAN fieldbus protocol to transmit real-time data to a host computer and information storage medium for display and storage. The fuel cell voltage monitoring CVM can interact directly with the communication module without going through an A / D converter. The communication module is connected to the embedded controller.

[0110] In the main circuit, the DC / DC converter is connected to the fuel cell stack; the voltage and current output terminals of the fuel cell stack are directly connected to the sampling system; the output terminal of the excitation signal module is connected to the input terminal of the PWM drive circuit; and the output terminal of the PWM drive circuit is connected to the switching transistor of the DC / DC converter.

[0111] The sampling system in the system mainly acquires the voltage and current signals of the fuel cell without requiring additional hardware circuitry. Considering the sampling accuracy and the frequency range of the disturbance signals, the battery management system's sampling system should preferably use an analog front-end chip with a high sampling frequency.

[0112] The control system module is used for system control, signal processing, impedance spectrum calculation, fault diagnosis, etc., and can be a digital controller or microcomputer in the battery management system.

[0113] The information storage device is used to store the impedance spectrum information obtained from the initial and subsequent measurements of each battery, providing historical information as a basis for battery diagnosis. It can take the form of various products that can be programmed by computers, such as disk storage, CD-ROM, flash memory, and optical storage.

[0114] A DC / DC converter circuit and a modified Fouquet equivalent circuit model of the fuel cell are built in Simulink or other fitting software. Multiple sinusoidal superposition signals with frequencies of 2, 4, 8, 16, 32, 64, 128, 256, and 1024 Hz are added. Using the method described in this invention, voltage and current data of the battery stack and individual cells are obtained through a sampling system with a sampling frequency of 2500 Hz.

[0115] The operating parameters of the fuel cell stack were adjusted to induce membrane dryness and water flooding states. Using the same detection conditions and methods as healthy fuel cells, EIS information was obtained under both membrane dryness and water flooding states. After data preprocessing, curve feature point and feature value extraction, the changes in the spectrum can be clearly observed by comparing it with the EIS of normal cells.

[0116] like Figs. 4 to 5 As shown, by setting test operating conditions under different fault states, the parameters of each component in the fuel cell equivalent model can be obtained, and the parameter values ​​of the Fouquet equivalent circuit model can be corrected in the simulation. By simulating the changes in the battery's ohmic resistance, activation resistance, and impedance at various frequency points, and using the same detection conditions and methods, the EIS information estimated by the algorithm after impedance changes can be obtained. By intuitively comparing the changes in battery EIS after changes in ohmic resistance and charge transfer resistance, the impact of these parameters on battery performance can be clearly observed.

[0117] Furthermore, the impedance information at different frequency points in the impedance spectrum can effectively characterize different fault processes, especially reflecting changes in the water content inside the fuel cell stack. This allows for a deeper understanding of the battery's performance under different operating conditions through comparative analysis of the impedance spectra of fuel cells in different water content states, providing strong data support and a basis for fault diagnosis.

[0118] like Fig. 4 As shown, when the ohmic resistance increases, the EIS image shifts to the right along the real axis relative to the EIS image of a healthy cell, which may indicate a membrane dryness fault. Membrane dryness faults impede the ability of water molecules inside the stack to act as proton carriers, thus significantly increasing the ohmic impedance. Because the timescale of ohmic conduction is small, this change typically manifests as an increase in impedance at high frequencies in the impedance spectrum.

[0119] like Fig. 5 As shown, when the activation resistance increases, the height of the semicircle in the EIS image rises along the imaginary axis, and the diameter of the semicircle increases, which may indicate that the battery has experienced a flooding failure. During a flooding failure, liquid water hinders the effective supply of reactants, leading to a significant increase in mass transfer losses. Due to the long timescale of the mass transfer process, this change is usually reflected in the low-frequency range of the impedance spectrum, manifesting as an increase in low-frequency impedance, which in turn leads to an expansion of the impedance spectrum radius.

[0120] like Fig. 6 As shown, voltage and current signals were obtained under three states: normal, membrane dry, and flooded, with 40 sets of data collected for each state. After processing these voltage and current signals using a digital controller, ohmic resistance, polarization resistance, and impedance information at each frequency point were obtained. The collected impedance information and corresponding feature data, totaling 20 categories, were used as the raw data for health status identification. Simultaneously, a label matrix was established based on 120 sets of data, with labels 1, 2, and 3 representing the normal, membrane dry, and flooded states, respectively. The LSSVM method used a Gaussian radial basis kernel function to calculate the Euclidean distance between input data points and employed high-dimensional space mapping to allow data to be classified through a hyperplane segmentation, thereby improving fault diagnosis capabilities. Subsequently, 90 training samples and 30 test samples were randomly generated for training and testing the LSSVM model, respectively, to complete the diagnosis of fuel cell faults. This method effectively identifies the performance of fuel cells under different fault states, providing reliable data support for fault diagnosis.

[0121] Example 2

[0122] This example provides an EIS online detection system and method for high-power fuel cell systems. It proposes a modified Fouquet equivalent circuit model and adopts a two-way parallel Boost converter topology. During operation, it monitors and calculates the internal state information of the fuel cell stack, injects a multi-frequency sinusoidal periodic signal optimized by a non-dominated sorting genetic algorithm based on Tent chaotic mapping into the auxiliary DC / DC converter, and then uses EIS feature data to characterize some faults, providing a new approach for proactive fault diagnosis and early warning.

[0123] To achieve the above objectives, this example adopts the following technical solution: An online EIS detection system and method for a high-power fuel cell system is provided, including a main circuit and a control system. The main circuit employs a two-parallel Boost converter topology, with small-signal excitation injected through an auxiliary DC / DC converter-2. The control system includes a sampling system, a PWM drive circuit, a communication module, and an embedded controller; an equivalent circuit model of the battery is established; during high-power fuel cell system operation, a multi-frequency sinusoidal periodic signal optimized by a non-dominated sorting genetic algorithm based on Tent chaotic mapping is injected into the output current of the auxiliary DC / DC converter-2; the voltage of a single cell or stack is sampled through a fuel cell voltage monitoring CVM, and current data is measured by components such as Hall current sensors. The voltage drop corresponding to the internal resistance is calculated based on the fuel cell open-circuit voltage and output voltage, and the time-domain signal is converted to a frequency-domain signal through a fast Fourier transform. Battery impedance information is obtained by fitting the amplitude and phase at specific frequency points; the characteristic parameter data is imported into a least squares support vector machine (LSSVM) for training and testing, enabling online in-situ detection and fault diagnosis.

[0124] The main circuit includes a fuel cell, a main DC / DC converter-1, an auxiliary DC / DC converter-2, and a DC load connected in sequence. The control system includes a sampling system, a PWM drive circuit, a communication module, and an embedded controller. The sampling system includes a CVM (Cell Voltage Monitor), a voltage sensor, and a high-precision current Hall sensor. The input of the PWM drive circuit is connected to the DC / DC controller, and the output of the PWM drive circuit is connected to the MOSFET in the DC / DC converter. The inputs of the voltage sensor and the high-precision current Hall sensor in the sampling system are both connected to the DC / DC converter, and their outputs are connected to the A / D conversion module. The fuel cell voltage monitor (CVM) in the sampling system can directly interact with the communication module without going through an A / D converter. The communication module is connected to the embedded controller.

[0125] In a preferred embodiment, both the main DC / DC converter-1 and the auxiliary DC / DC converter-2 are composed of filter capacitors, inductors, and MOSFETs to form a basic Boost-type boost circuit, used for transmitting fuel cell output power or superimposing excitation signals; the communication module adopts the CAN fieldbus protocol, which can transmit real-time data to embedded controllers or other host computers, information storage media, etc., to perform data output, calculation, display, and storage functions.

[0126] Step S1 involves selecting the phase of each single-frequency sinusoidal signal using an NSGA-II genetic algorithm based on the Tent chaotic map, which yields a time-domain signal with a lower crest factor than that obtained with random phase. The optimization objective is the crest factor (the ratio of the signal's peak value to its effective value). The frequencies of the sinusoidal components of the sinusoidal excitation signal are 2, 4, 8, 16, 32, 64, 128, 256, and 1024 rad, respectively, with an amplitude of 4 Å for each component and a default phase value of 0 rad.

[0127] Step S2: Establish a modified Fouquet equivalent circuit model describing the Nyquist characteristics of the fuel cell. The internal resistance of the fuel cell is simplified to the ohmic internal resistance R. ohm Activation internal resistance R ct Concentration resistance R w Double-layer distributed capacitance C dl and concentration capacitance C w Circuits composed of series and parallel connections;

[0128] Step S3: The optimized multi-frequency sinusoidal excitation signal is superimposed on the output current of the stable PEMFC system through the PWM drive circuit;

[0129] Step S4: After the excitation signal is injected, the sampling system synchronously acquires the discrete sequences of response voltage and excitation current, and then converts them into frequency domain voltage complex sequences and frequency domain current complex sequences through fast Fourier transform; the impedance corresponding to the excitation frequency is calculated through impedance spectrum, that is, the response voltage of each frequency sinusoidal component of the excitation current signal and the internal resistance under sinusoidal excitation at each frequency, to obtain the amplitude and phase information of the impedance corresponding to the frequency point, and fits the Nyquist curve; the single cell voltage signal acquired by CVM is directly sent to the host computer via the communication module;

[0130] Step S5: The calculated impedance information is used as the original dataset for fault diagnosis training and testing, which contains 20 feature data and 1 label column; the dataset is input into the LSSVM vector machine to complete the fault diagnosis.

Claims

1. An EIS online monitoring system for a high-power fuel cell system, characterized in that: It includes the main circuit and the control system; the main circuit adopts a two-way parallel Boost converter topology, and small signal excitation is injected through an auxiliary DC / DC converter; the control system includes a sampling system, PWM drive circuit, communication module and embedded controller; an equivalent circuit model of the battery is established; The main circuit includes a fuel cell, a DC / DC converter, and a DC load connected in sequence; the control system includes a sampling system, a PWM drive circuit, a communication module, and an embedded controller; the sampling system includes a CVM detector, a voltage sensor, and a high-precision current Hall sensor; wherein, the CVM detector integrated in the sampling system is used to process and calculate the collected fuel cell voltage in real time, and calculate the data of single cell voltage consistency status, average voltage, and total voltage; the voltage sensor measures the total output voltage of the fuel cell; the current sensor measures the output current of the fuel cell; the communication module is connected to a host computer; The main circuit adopts a two-way parallel Boost converter topology. One way transmits output power to meet load requirements, and the other way injects multi-frequency sinusoidal periodic excitation signals. Its DC bias is low to avoid the detection difficulties caused by being overwhelmed by the high DC bias of high-power fuel cells, thereby improving the detection accuracy of small signal excitation. The detection method of the EIS online detection system for high-power fuel cell systems is as follows: When the high-power fuel cell system is running, a multi-frequency sinusoidal periodic signal optimized by a non-dominated sorting genetic algorithm based on Tent chaotic mapping is injected into the output current of the auxiliary DC / DC converter; the voltage of a single cell or stack is sampled by the fuel cell voltage inspection CVM, and the current data is measured by the Hall current sensor element. The voltage drop corresponding to the internal resistance of the fuel cell is calculated based on the open-circuit voltage and output voltage. The time-domain signal is converted into a frequency-domain signal by fast Fourier transform; the battery impedance information in the corrected Fouquet equivalent circuit model is obtained by fitting the amplitude and phase at a specific frequency point; the characteristic parameter data is imported into the least squares support vector machine LSSVM for training and testing, and online in-situ detection and fault diagnosis are performed.

2. The EIS online detection system for high-power fuel cell systems according to claim 1, characterized in that: The multi-frequency sinusoidal periodic excitation signal in the diagnostic system is optimized using a non-dominated sorting genetic algorithm based on Tent chaotic mapping. By optimizing the phase of sinusoidal signals of different frequencies, the peak factor of the combined multi-frequency sinusoidal signal is reduced, thereby avoiding the problem of reduced EIS measurement accuracy caused by excessive superposition of excitation signal amplitudes.

3. The EIS online detection system for high-power fuel cell systems according to claim 2, characterized in that: The high-power fuel cell, through an equivalent circuit model, is subjected to an internal resistance of ohmic R. ohm Activation internal resistance R ct Warburg diffusion resistance Z w and constant phase Z CPE The Fouquet model, consisting of four components, was improved by replacing the Warburg diffusion resistor Z with a series Foster circuit. w To further accurately simulate the low-frequency and high-frequency characteristics of high-power fuel cells, a double-layer distributed capacitor C is employed. dl Replacing the CPE element to reduce fitting difficulty, forming a structure with ohmic internal resistance R ohm Activation internal resistance R ct Concentration resistance R w Double-layer distributed capacitance C dl and concentration capacitance C w The modified Fouquet equivalent circuit model consists of series and parallel connections; where the ohmic internal resistance R... ohm Activation internal resistance R ct Indicates the degree of membrane dryness failure, concentration resistance R w Indicates the degree of flooding fault; double-layer distributed capacitance C dl This indicates the electrical characteristics generated by the bipolar plates and electrolyte in the fuel cell stack.

4. The EIS online detection system for high-power fuel cell systems according to claim 3, characterized in that: The online detection method is used to identify three states of fuel cell stacks: normal, membrane dry, and water flooding. The data used for LSSVM model training and model validation are derived from experimental samples with clear conclusions and contain state labels; the training set and test set data are randomly divided in a 3:1 ratio. The least squares support vector machine (LSSVM) uses the battery ohmic resistance, activation resistance, and impedance at frequency points of 2, 4, 8, 16, 32, 64, 128, 256, and 1024 in three states as feature vectors. The label values ​​are normal, membrane dry, and flooded, which correspond to the numbers 1, 2, and 3, respectively.

5. The EIS online detection system for high-power fuel cell systems according to claim 3, characterized in that: In the detection method, the sampling system collects voltage and current information, and converts the analog signal into a digital signal in real time through an A / D converter. One signal is input to the embedded controller through the communication module for pre-data processing and calculation, and the other signal is input to the DC / DC controller for PID calculation and voltage or current regulation control through the PWM drive circuit. The communication module uses the CAN fieldbus protocol to transmit real-time data to the host computer and information storage medium for display and storage, respectively.

6. The EIS online detection system for a high-power fuel cell system according to claim 5, characterized in that: The fuel cell voltage monitoring (CVM) interacts directly with the communication module, without the need for an A / D converter.

7. The EIS online detection system for a high-power fuel cell system according to claim 3, characterized in that: The method includes the following steps: Step S1: The initial population is assigned values ​​using the NSGA-II genetic algorithm based on the Tent chaotic mapping to improve the algorithm's convergence ability; the phase of the sinusoidal excitation signal at different frequencies is optimized using the improved algorithm; the optimization objective is the crest factor, which is the ratio of the signal peak value to twice the effective value. The frequencies of each sinusoidal component of the sinusoidal excitation signal are 2, 4, 8, 16, 32, 64, 128, 256, and 1024, respectively. The amplitude of each sinusoidal component is 4A, and the phase value is 0 rad by default. Step S2: Establish a modified Fouquet equivalent circuit model describing the Nyquist characteristics of the fuel cell; wherein, the internal resistance of the fuel cell is simplified to the ohmic internal resistance R. ohm Activation internal resistance R ct Concentration resistance R w Double-layer distributed capacitance C dl and concentration capacitance C w Circuits composed of series and parallel connections; Step S3: The optimized multi-frequency sinusoidal excitation signal is superimposed on the output current of the auxiliary DC / DC converter of the fuel cell system that is operating stably through the PWM drive circuit; Step S4: After the excitation signal is injected, the sampling system synchronously acquires the discrete sequences of response voltage and excitation current, and then converts them into frequency domain voltage complex sequences and frequency domain current complex sequences through fast Fourier transform; the impedance corresponding to the excitation frequency is calculated through impedance spectrum, that is, the response voltage of each frequency sinusoidal component of the excitation current signal and the internal resistance under sinusoidal excitation at each frequency, to obtain the amplitude and phase information of the impedance corresponding to the frequency point, and fits the Nyquist curve; the single cell voltage signal acquired by CVM is directly sent to the host computer via the communication module; Step S5: The calculated impedance information is used as the original dataset for fault diagnosis training and testing, which contains 20 feature data and 1 label column; the dataset is input into the LSSVM vector machine to complete the fault diagnosis.

8. The EIS online detection system for a high-power fuel cell system according to claim 7, characterized in that: Step S1 specifically includes the following steps: The function of the multi-frequency sinusoidal excitation signal is: In the formula: f(t) is the frequency of the excitation model at time t; The initial population for the NSGA-II genetic algorithm is initialized by the Tent chaos map; The optimization objective of the NSGA-II genetic algorithm is defined as the crest factor, which represents the uniformity of signal amplitude distribution throughout the time domain. Its calculation formula is as follows: In the formula: I ac,max I represents the maximum value of the multi-frequency sinusoidal excitation signal. ac,min The minimum value of the multi-frequency sinusoidal excitation signal; I eff The effective value of the multi-frequency sinusoidal excitation signal is I; where, the effective value of the multi-frequency sinusoidal excitation signal is I. eff The expression is: Furthermore, step S2 specifically includes the following steps: The impedance of the modified Fouquet equivalent circuit model is expressed as: The real and imaginary parts of the impedance are calculated based on the above formula. The Nyquist curve is then plotted in the complex plane based on the relationship between the real and imaginary parts of the Fouquet equivalent circuit model at various frequencies to represent the electrochemical impedance spectrum of the fuel cell.

9. The EIS online detection system for a high-power fuel cell system according to claim 8, characterized in that: Step S4 specifically includes the following steps: The sampling system includes key components such as a CVM (Cellular Dynamic Monitoring Unit), Hall current sensor, and voltage sensor, as well as precision resistors and components required for ADC (Analog-to-Digital Converter) sampling. The CVM monitors the internal structure of the battery stack, measures the current operating status of individual cells, and transmits real-time data to the main controller via a fieldbus protocol. The detailed voltage data measured by the CVM can be used for fault diagnosis. Furthermore, when voltage imbalance or cell failure occurs, a fault alarm can be directly issued, triggering and recording the voltage data at the time of the fault. Step S5 specifically includes the following steps: The LSSVM simplifies the standard SVM by converting the constraints into a squared loss function, thus simplifying the optimization process. According to the SRM criterion, the objective function and constraints of LSSVM are as follows: In the formula, ξ i ξ is the prediction error; c is the regularization parameter or penalty parameter, used to control the impact on the prediction error ξ. i The degree of sample penalty; LSSVM handles the error by minimizing the squared loss, avoiding the introduction of slack variables; By constructing a Lagrange transformation polynomial, the following transformation is performed: For the above equation w, b, e k α k By taking the partial derivatives and setting them to zero, we obtain the system of linear equations. Solving this system yields the calculation formula for the LSSVM algorithm: In the formula, K(x,x) i ) is the kernel function.

Images