Fuel cell fault diagnosis method, equipment and medium

Abstract

The embodiment of the invention discloses a fuel cell fault diagnosis method and device and a medium, belongs to the technical field of fuel cells, and solves the problem that the fuel cell fault diagnosis accuracy is low under the condition that historical data does not completely cover all faults. Comprising the following steps: performing cycle period dimension increase processing on a two-dimensional impedance spectrum corresponding to preset sample fuel cell impedance data to obtain a three-dimensional impedance field; performing data fitting on the three-dimensional impedance field to generate a mirror image impedance field, and training the mirror image impedance field through an interlaced learning method to obtain an interlaced learning impedance field expansion model; inputting the impedance spectrum of the fuel cell to be measured into the cross learning impedance field expansion model to generate a life cycle impedance field; performing coordinate plane projection processing on the life cycle impedance field of the fuel cell to be measured to obtain an impedance density map; and classifying the impedance density map to determine fault information of the to-be-detected fuel cell. Through the method, the accuracy of fault diagnosis is improved.

Description

Technical Field

[0001] The present application relates to the field of fuel cell technology, and in particular to a fuel cell fault diagnosis method, device and medium. Background Art

[0002] At present, the commonly used fuel cell status detection and fault diagnosis methods include: non-destructive testing technology based on electromagnetic field changes, non-destructive testing theory and methods based on magnetic field imaging, electrochemical impedance spectroscopy testing technology, data-driven fault diagnosis methods, and fault diagnosis based on deep learning.

[0003] However, fuel cell fault diagnosis in the prior art mostly relies on historical fault data, which is easily affected by data quality. When the historical data does not completely cover all faults, the fault diagnosis accuracy is low. Summary of the invention

[0004] The embodiments of the present application provide a fuel cell fault diagnosis method, device and medium for solving the following technical problems: fuel cell fault diagnosis in the prior art mostly relies on historical fault data and is easily affected by data quality. When historical data does not completely cover all faults, the fault diagnosis accuracy is low.

[0005] The present application embodiment adopts the following technical solutions: The embodiment of the present application provides a fuel cell fault diagnosis method. The method includes: performing a cyclic dimension increase process on the two-dimensional impedance spectrum corresponding to the preset sample fuel cell impedance data to obtain a three-dimensional impedance field corresponding to the preset sample fuel cell impedance data; performing data fitting on the three-dimensional impedance field corresponding to the preset sample fuel cell impedance data through an equivalent circuit to generate a mirror impedance field, and training the mirror impedance field through an interleaved learning method to obtain a cross-learning impedance field extension model; inputting the acquired impedance spectrum of the fuel cell to be tested into the cross-learning impedance field extension model, and generating a life cycle impedance field of the fuel cell to be tested through the cross-learning impedance field extension model; performing coordinate plane projection processing on the life cycle impedance field of the fuel cell to be tested to obtain an impedance density map; and classifying the impedance density map based on a preset neural network model to determine the fault information of the fuel cell to be tested through the classification result.

[0006] Through knowledge transfer, the embodiment of the present application adds the dimension of fuel cell life cycle on the basis of two-dimensional impedance spectrum, converts the two-dimensional impedance spectrum into a three-dimensional image, and then diagnoses fuel cell faults based on impedance field and impedance density, thereby improving the accuracy and efficiency of fault diagnosis. At the same time, the impedance density image can also better display the abnormal performance gain caused by the oxygen supply interruption of the fuel cell, and assist in mechanism analysis. The use of the staggered forward and backward learning method can fully mine the information in different time directions of the data, improve the utilization rate of the data, improve the understanding of the impedance changes throughout the life cycle, better capture the relationship between the impedance field and the fault, and more accurately determine the internal fault mechanism of the fuel cell.

[0007] In one implementation of the present application, a two-dimensional impedance spectrum corresponding to preset sample fuel cell impedance data is processed to increase the cycle dimension to obtain a three-dimensional impedance field corresponding to the preset sample fuel cell impedance data, specifically including: generating a two-dimensional battery impedance spectrum based on the preset sample fuel cell impedance data; drawing an impedance curve waterfall chart corresponding to impedance curves of different cycle periods based on the two-dimensional battery impedance spectrum to increase the cycle dimension of the two-dimensional battery impedance spectrum; performing color gradient processing on the impedance curve waterfall chart; performing surface fitting on the impedance curve waterfall chart after color gradient processing by using the cubic spline interpolation method, and smoothing the interpolation result by Gaussian filtering to obtain a three-dimensional impedance field corresponding to the preset sample fuel cell impedance data.

[0008] In one implementation of the present application, a three-dimensional impedance field corresponding to preset sample fuel cell impedance data is fitted through an equivalent circuit to generate a mirror impedance field, specifically including: performing component parameter fitting based on the first cycle period and the last cycle period corresponding to the preset sample fuel cell impedance data; wherein the component parameters include at least one of a series resistance, a constant phase component and a polarization resistance; through a linear interpolation function, based on the component parameter values ​​corresponding to the first cycle period and the last cycle period, respectively, the simulation circuit component parameters in the intermediate cycle period are determined to generate a complete simulation equivalent circuit component parameter value; based on the complete simulation equivalent circuit component parameter value, a mirror impedance field is constructed; wherein the mirror impedance field is used to reflect the impedance change trend of the battery under different cycle periods.

[0009] In one implementation of the present application, the simulation circuit element parameters in the intermediate cycle are determined based on the element parameter values ​​corresponding to the first cycle and the last cycle respectively through a linear interpolation function, specifically including: based on the function:

[0010] The slope between the component parameter value of the first cycle and the component parameter value of the last cycle is obtained; based on the point slope equation of the straight line, the component parameter value of the nth cycle is obtained:

[0011] Based on the component parameters of the first cycle and the last cycle, the simulation equivalent circuit component parameter values ​​of any intermediate cycle are determined:

[0012] in, is the component parameter value of the last cycle; is the component parameter value of the first cycle; is the parameter value of the simulation equivalent circuit element of any intermediate cycle; k is the slope; n is a cyclic sequence, where 1< n < m .

[0013] In one implementation of the present application, the mirror impedance field is trained by an interlaced learning method to obtain a cross-learning impedance field extension model, specifically including: acquiring mirror impedance field data, analyzing and modeling the mirror impedance field data by stepwise linear regression, and optimizing the obtained stepwise linear regression model by an interlaced learning method.

[0014] In one implementation of the present application, the stepwise linear regression model is:

[0015] in, Z is the impedance value; CPE T is the first element parameter in the equivalent circuit; CPE P is the second element parameter in the equivalent circuit; Rs is the parameter of the third element in the equivalent circuit; R is the fourth element parameter in the equivalent circuit; n is a cyclical sequence; is the preset coefficient; is the coefficient corresponding to the first element parameter; is the coefficient corresponding to the second element parameter; is the coefficient corresponding to the third element parameter; is the coefficient corresponding to the fourth element parameter; is the coefficient corresponding to the fifth element parameter; is the product of the last component parameter and the corresponding coefficient.

[0016] In one implementation of the present application, the obtained stepwise linear regression model is optimized through an interleaved learning method, specifically including: performing Gaussian noise processing on the impedance density image corresponding to the preset sample fuel cell impedance data to obtain an expanded data set; dividing the expanded data set into a training set and a validation set according to a preset ratio; training based on the training set, the validation set and the preset ResNet-50 neural network, and using the interleaved learning method to optimize the stepwise linear regression model.

[0017] In one implementation of the present application, the life cycle impedance field of the fuel cell to be tested is projected onto a coordinate plane to obtain an impedance density map, specifically including: projecting the life cycle impedance field of the fuel cell to be tested onto the xy, yz and xz planes to obtain a scatter plot; optimizing the scatter plot into a contour map, and performing gradient processing between the contour maps; filling the contour map through the contourf function, and performing image boundary smoothing processing on the filled contour map through the pcolormesh function to obtain an impedance density map.

[0018] An embodiment of the present application provides a fuel cell fault diagnosis device, comprising: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: perform cyclic period dimension increase processing on the two-dimensional impedance spectrum corresponding to the preset sample fuel cell impedance data to obtain the three-dimensional impedance field corresponding to the preset sample fuel cell impedance data; perform data fitting on the three-dimensional impedance field corresponding to the preset sample fuel cell impedance data through an equivalent circuit to generate a mirror impedance field, and train the mirror impedance field through an interleaved learning method to obtain a cross-learning impedance field extension model; input the acquired impedance spectrum of the fuel cell to be tested into the cross-learning impedance field extension model, and generate the life cycle impedance field of the fuel cell to be tested through the cross-learning impedance field extension model; perform coordinate plane projection processing on the life cycle impedance field of the fuel cell to be tested to obtain an impedance density map; and classify the impedance density map based on a preset neural network model to determine the fault information of the fuel cell to be tested through the classification result.

[0019] A non-volatile computer storage medium provided in an embodiment of the present application stores computer executable instructions, wherein the computer executable instructions are configured to: perform cyclic period dimension increase processing on a two-dimensional impedance spectrum corresponding to preset sample fuel cell impedance data to obtain a three-dimensional impedance field corresponding to the preset sample fuel cell impedance data; perform data fitting on the three-dimensional impedance field corresponding to the preset sample fuel cell impedance data through an equivalent circuit to generate a mirror impedance field, and train the mirror impedance field through an interleaved learning method to obtain a cross-learning impedance field extension model; input the acquired impedance spectrum of the fuel cell to be tested into the cross-learning impedance field extension model, and generate a life cycle impedance field of the fuel cell to be tested through the cross-learning impedance field extension model; perform coordinate plane projection processing on the life cycle impedance field of the fuel cell to be tested to obtain an impedance density map; and classify the impedance density map based on a preset neural network model to determine fault information of the fuel cell to be tested through the classification results.

[0020] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects: Through knowledge transfer, the embodiments of the present application add the dimension of fuel cell life cycle on the basis of two-dimensional impedance spectrum, convert the two-dimensional impedance spectrum into a three-dimensional image, and then diagnose fuel cell faults based on impedance field and impedance density, thereby improving the accuracy and efficiency of fault diagnosis. At the same time, the impedance density image can also better display the abnormal performance gain caused by the oxygen supply interruption of the fuel cell, and assist in mechanism analysis. The method of interleaved forward and backward learning can fully mine the information in different time directions of the data, improve the utilization rate of the data, improve the understanding of impedance changes throughout the life cycle, better capture the relationship between the impedance field and the fault, and more accurately determine the internal fault mechanism of the fuel cell. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. In the drawings: Figure 1 A flow chart of a fuel cell fault diagnosis method provided in an embodiment of the present application; Figure 2 A schematic diagram of a fuel cell fault diagnosis method flow chart provided in an embodiment of the present application; Figure 3 A schematic diagram of an impedance spectrum of a cycle initial state provided in an embodiment of the present application; Figure 4 A full-cycle impedance waterfall diagram provided in an embodiment of the present application; Figure 5 A schematic diagram of a full-cycle impedance evolution process provided in an embodiment of the present application; Figure 6 A schematic diagram of preliminary generation of an impedance field provided in an embodiment of the present application; Figure 7 A schematic diagram of an impedance field provided in an embodiment of the present application; Figure 8 A comparison diagram of an impedance field and a mirror impedance field provided in an embodiment of the present application; Fig. 9 A linear regression schematic diagram provided for an embodiment of the present application; Fig.10 A schematic diagram of interleaved learning provided in an embodiment of the present application; Fig.11 A schematic diagram of a 1% Gaussian noise image under normal conditions provided in an embodiment of the present application; Fig.12 A schematic diagram of a 10% Gaussian noise image under normal conditions provided in an embodiment of the present application; Fig.13 A schematic diagram of a 20% Gaussian noise image under normal conditions provided in an embodiment of the present application; Fig.14 A schematic diagram of a 1% Gaussian noise image in a dry film state provided in an embodiment of the present application; Fig.15 A schematic diagram of a 10% Gaussian noise image in a dry film state provided in an embodiment of the present application; Fig.16 A schematic diagram of a 20% Gaussian noise image in a dry film state provided in an embodiment of the present application; Fig.17 A schematic diagram of generating an impedance density image is provided in an embodiment of the present application, wherein: Fig.17 (a) is a scatter plot. Fig.17 (b) is a contour map. Fig.17 (c) in the middle shows a gradient image added between the contour lines. Fig.17 (d) in the middle is a map that reduces the obvious boundaries between color blocks. Fig.17 (e) is a smooth color transition diagram. Fig.17 Middle (f) is the impedance density map; Fig.18 A ResNet-50 training progress diagram provided in an embodiment of the present application, wherein: Fig.18 (a) is a schematic diagram of the loss. Fig.18 (b) is a diagram of accuracy. Fig.18 (c) is a comparison chart between prediction and reality; Fig.19A schematic diagram of the structure of a fuel cell fault diagnosis device provided in an embodiment of the present application.

[0022] Reference numerals: 200 fuel cell fault diagnosis device, 201 processor, 202 memory. DETAILED DESCRIPTION

[0023] Embodiments of the present application provide a fuel cell fault diagnosis method, device, and medium.

[0024] In order to enable those skilled in the art to better understand the technical solutions in this application, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are only part of the embodiments of this application, not all of them. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0025] The technical solution proposed in the embodiments of the present application is described in detail below with reference to the accompanying drawings.

[0026] Figure 1 A fuel cell fault diagnosis method flow chart provided in an embodiment of the present application is as follows: Figure 1 As shown, the fuel cell fault diagnosis method includes the following steps: Step 101 : performing a cyclic dimension increasing process on the two-dimensional impedance spectrum corresponding to the preset sample fuel cell impedance data to obtain a three-dimensional impedance field corresponding to the preset sample fuel cell impedance data.

[0027] In one embodiment of the present application, a two-dimensional battery impedance spectrum is generated based on preset sample fuel cell impedance data. Based on the two-dimensional battery impedance spectrum, an impedance curve waterfall diagram corresponding to the impedance curves of different cycle periods is drawn to increase the cycle dimension of the two-dimensional battery impedance spectrum. The impedance curve waterfall diagram is subjected to color gradient processing. The impedance curve waterfall diagram after color gradient processing is subjected to surface fitting by a cubic spline interpolation method, and the interpolation result is smoothed by a Gaussian filter to obtain a three-dimensional impedance field corresponding to the preset sample fuel cell impedance data.

[0028] Specifically, Figure 2 A schematic diagram of a fuel cell fault diagnosis method provided in an embodiment of the present application is shown in FIG. Figure 2As shown, taking the preset sample fuel cell as a lithium battery as an example, the lithium battery status is classified, a large amount of lithium battery impedance data under different cycle periods is collected, a waterfall chart of the impedance curve in the cycle process is drawn, the impedance curve in different cycle processes is color-gradient processed, the cubic spline interpolation method is used for fitting, and the interpolation result is smoothed using the Gaussian filtering method to generate a three-dimensional "impedance field".

[0029] Further, Figure 3 A schematic diagram of an impedance spectrum of a cycle initial state provided in an embodiment of the present application, Figure 4 A full-cycle impedance waterfall diagram provided in an embodiment of the present application, Figure 5 A schematic diagram of a full-cycle impedance evolution process provided in an embodiment of the present application is shown in FIG. Figure 6 A schematic diagram of preliminary generation of an impedance field provided in an embodiment of the present application, Figure 7 A schematic diagram of an impedance field provided in an embodiment of the present application. Figure 3 , Figure 4 , Figure 5 , Figure 6 as well as Figure 7 As shown in Figure 2, the specific process of impedance field generation is: Step 1.1: Generate a battery impedance spectrum, such as Figure 3 As shown, the lower curve represents the impedance curve of the last cycle, and the upper curve represents the impedance curve of the first cycle; Step 1.2: Generate an impedance curve waterfall diagram based on the impedance curves of different cycle states, such as Figure 4 As shown in the figure, the two-dimensional impedance spectrum is increased to a three-dimensional waterfall diagram, introducing the dimension of the cycle period of the lithium battery; Step 1.3: Perform color gradient processing based on the impedance data from the first cycle to the last cycle of the lithium battery to obtain an image such as Figure 5 As shown; Step 1.4: Flip Re(z), use the griddata function to perform cubic spline interpolation, and perform surface fitting on the full cycle impedance data of the lithium battery to form a three-dimensional impedance field such as Figure 6 As shown in , if there are abnormal points on the edge of the surface, Gaussian filtering is used to smooth the interpolation result, and finally the following is obtained: Figure 7 The three-dimensional impedance field is shown.

[0030] Step 102: Perform data fitting on the three-dimensional impedance field corresponding to the preset sample fuel cell impedance data through an equivalent circuit to generate a mirror impedance field, and train the mirror impedance field through an interleaved learning method to obtain a cross-learning impedance field expansion model.

[0031] In one embodiment of the present application, the element parameters are fitted based on the first cycle and the last cycle corresponding to the preset sample fuel cell impedance data; wherein the element parameters include at least one of the series resistance, the constant phase element and the polarization resistance. Through the linear interpolation function, based on the element parameter values ​​corresponding to the first cycle and the last cycle, the simulation circuit element parameters in the intermediate cycle are determined to generate the complete simulation equivalent circuit element parameter values. Based on the complete simulation equivalent circuit element parameter values, a mirror impedance field is constructed; wherein the mirror impedance field is used to reflect the impedance change trend of the battery under different cycle periods.

[0032] Specifically, for fuel cell fault diagnosis, such as Figure 2 As shown on the right, because fuel cells can usually only obtain one impedance curve, and a single impedance curve is difficult to reflect the impedance changes of its entire life cycle, and cannot provide sufficient information for accurate fault diagnosis, the relationship between the "mirror impedance field" generated by the lithium battery and the original impedance field of the lithium battery is used for machine learning. After the model is trained, the model is used to generate the life cycle impedance field of the fuel cell. In this process, the interleaved learning method is used to use the historical data set to expand the impedance field forward to the future prediction and use the future data set to expand the impedance field backward to the historical prediction. Stepwise linear regression training is used to capture the change process of the mirror impedance field, so as to realize the expansion of the fuel cell impedance curve to the impedance field of the entire life cycle.

[0033] Further, using Rs ( CPE-Rp ) Equivalent circuit fitting to the battery impedance data.

[0034] The first cycle and the 350th cycle (the last cycle) are selected for component parameter fitting, where the component parameters include Rs (Series resistance), CPE (Constant Phase Element) and Rs (polarization resistance), these parameter values ​​reflect the chemical and physical properties inside the battery.

[0035] Analyze the equivalent circuit component parameters of the first and last cycles. Since the cycle period used is shorter than the overall service life of the battery, the component parameters can be considered to be linear changes to a certain extent. On this basis, use the linear interpolation function to process the component parameters within the cycle period, that is, according to the component parameter values ​​of the first and last cycles, calculate the simulation circuit component parameters in the intermediate cycle period, and automatically generate the complete simulation equivalent circuit component parameter values. Therefore, let the component parameter value of the first cycle period be P 1. The component parameter value of the last cycle is P m, to calculate the n Cycle period (1< n < m ) component parameter values P n .

[0036] In one embodiment of the present application, based on the function:

[0037] Obtaining the slope between the component parameter value of the first cycle and the component parameter value of the last cycle; Based on the point-slope equation of a straight line, we get n Component parameter values ​​for a cycle:

[0038] Based on the component parameters of the first cycle and the last cycle, the simulation equivalent circuit component parameter values ​​of any intermediate cycle are determined:

[0039] in, is the component parameter value of the last cycle; is the component parameter value of the first cycle; is the parameter value of the simulation equivalent circuit element of any intermediate cycle; k is the slope; n is a cyclic sequence, where 1< n < m .

[0040] According to the above-mentioned component parameters and algorithms, a virtual impedance field is constructed which can reflect the impedance variation trend of the battery under different cycle periods. This virtual impedance field is the mirror impedance field in the embodiment of the present application. Figure 8 A comparison diagram of an impedance field and a mirror impedance field provided in an embodiment of the present application is shown in FIG. Figure 8 It can be seen that the changing trends of the mirror impedance field and the impedance field in different periods.

[0041] In one embodiment of the present application, mirror impedance field data is acquired, the mirror impedance field data is analyzed and modeled by stepwise linear regression, and the obtained stepwise linear regression model is optimized by an interleaved learning method.

[0042] Specifically, based on the basic data provided by the mirror impedance field, the mirror impedance field data is analyzed and modeled by stepwise linear regression, and a model that can accurately describe the variation law of the impedance field is obtained. Z is the impedance value, CPE-T , CPE-P , Rs, R are the component parameters in the equivalent circuit, n is an independent variable such as the number of cycles, the stepwise linear regression model can be expressed as:

[0043] in, Z is the impedance value; CPE T is the first element parameter in the equivalent circuit; CPE P is the second element parameter in the equivalent circuit; Rs is the parameter of the third element in the equivalent circuit; R is the fourth element parameter in the equivalent circuit; n is a cyclical sequence; is the preset coefficient; is the coefficient corresponding to the first element parameter; is the coefficient corresponding to the second element parameter; is the coefficient corresponding to the third element parameter; is the coefficient corresponding to the fourth element parameter; is the coefficient corresponding to the fifth element parameter; is the product of the last component parameter and the corresponding coefficient.

[0044] Fig. 9 A linear regression diagram is provided for the embodiment of the present application, such as Fig. 9 As shown, the stepwise linear regression method achieved good accuracy when training the model, and the fuel cell data was successfully expanded into an impedance field.

[0045] In one embodiment of the present application, Gaussian noise processing is performed on the impedance density image corresponding to the preset sample fuel cell impedance data to obtain an expanded data set. The expanded data set is divided into a training set and a validation set according to a preset ratio. Training is performed based on the training set, the validation set, and the preset ResNet-50 neural network, and the stepwise linear regression model is optimized by adopting an interleaved learning method.

[0046] Specifically, 1% - 100% Gaussian noise is added to the impedance density image of the preset sample fuel cell impedance data to expand the data set. The expanded battery impedance density image data set is divided into a training set and a validation set in a ratio of 7:3. The ResNet-50 neural network is used for training. Interleaved learning is used to further optimize the stepwise linear regression model training process to ensure that a high accuracy can be achieved in the field of lithium battery state classification.

[0047] Fig.10A schematic diagram of interleaved learning provided in an embodiment of the present application is shown in FIG. Fig.10 As shown, the data sets are moved alternately, and the historical and future data sets are used to expand the impedance field forward and backward respectively, so that the model can better learn the intrinsic connection of the data.

[0048] Step 103: input the acquired impedance spectrum of the fuel cell to be tested into the cross-learning impedance field extension model, and generate the life cycle impedance field of the fuel cell to be tested through the cross-learning impedance field extension model.

[0049] In one embodiment of the present application, the impedance field of the impedance curve of the fuel cell to be tested is expanded over the entire life cycle through a model trained with preset sample fuel cells to obtain impedance density images of the fuel cell to be tested under different states.

[0050] Fig.11 This is a schematic diagram of a 1% Gaussian noise image under normal conditions provided in an embodiment of the present application. Fig.12 This is a schematic diagram of a 10% Gaussian noise image under normal conditions provided in an embodiment of the present application. Fig.13 A schematic diagram of a 20% Gaussian noise image under normal conditions provided in an embodiment of the present application; Fig.14 This is a schematic diagram of a 1% Gaussian noise image in a dry film state provided in an embodiment of the present application. Fig.15 This is a schematic diagram of a 10% Gaussian noise image in a dry film state provided in an embodiment of the present application. Fig.16 A schematic diagram of a 20% Gaussian noise image in a dry film state provided in an embodiment of the present application; Fig.11 and Fig.14 In comparison, this is the Gaussian noise density image generated after adding 1% Gaussian noise. Fig.12 and Fig.15 In comparison, this is the Gaussian noise density image generated after adding 10% Gaussian noise. Fig.13 and Fig.16 In comparison, the Gaussian noise density image is generated after adding 20% ​​Gaussian noise. By comparison, it can be seen that in the dry state of the membrane, as the number of life cycles increases, the impedance will continue to increase, which is completely different from the normal state.

[0051] Step 104 , performing coordinate plane projection processing on the life cycle impedance field of the fuel cell to be tested to obtain an impedance density map.

[0052] In one embodiment of the present application, the life cycle impedance field of the fuel cell to be tested is projected onto the xy, yz and xz planes to obtain a scatter plot. The scatter plot is optimized into a contour plot, and gradient processing is performed between the contour plots. The contour plot is filled by the contourf function, and the image boundary is smoothed by the pcolormesh function to obtain an impedance density map.

[0053] Specifically, Fig.17 A schematic diagram of generating an impedance density image is provided in an embodiment of the present application, wherein: Fig.17 (a) is a scatter plot. Fig.17 (b) is a contour map. Fig.17 (c) in the middle shows a gradient image added between the contour lines. Fig.17 (d) in the middle is a map that reduces the obvious boundaries between color blocks. Fig.17 (e) is a smooth color transition diagram. Fig.17 (f) is the impedance density diagram; Fig.17 As shown, the generated impedance field is projected onto the xy, yz and xz planes to obtain a scatter plot, such as Fig.17 As shown in (a), the colors of the scatter plot represent the z value, x value, and y value respectively. The scatter plot is optimized to a contour plot, a gradient is added between the contour plots, the contour plot is processed using the contourf function, and finally the image boundary is smoothed using the pcolormesh function. The process is as follows: Fig.17 Middle (b) to Fig.17 The final impedance density image is shown in (e). Fig.17 As shown in (d).

[0054] Step 105: classify the impedance density map based on the preset neural network model to determine the fault information of the fuel cell to be tested through the classification result.

[0055] In one embodiment of the present application, a ResNet-50 neural network is used to perform fault diagnosis on the impedance density image of the fuel cell to be tested. The training process of the ResNet-50 neural network is as follows: the impedance density map of the historical fuel cell is used as an input sample, the fuel cell fault information corresponding to the input sample is used as an output sample, and the preset neural network model is trained to obtain the ResNet-50 neural network.

[0056] in, Fig.18 A ResNet-50 training progress diagram provided in an embodiment of the present application, wherein: Fig.18 (a) is a schematic diagram of the loss. Fig.18 (b) is a diagram of accuracy. Fig.18 (c) is a comparison chart between prediction and actual. Fig.18 As shown, the ResNet-50 neural network is selected, the training set is 70%, the validation set is 30%, and the loss curve is as follows Fig.18 As shown in (a), the horizontal axis is the period and the vertical axis is the loss value; the accuracy curve is as follows Fig.18 As shown in (b), the horizontal axis is the period and the vertical axis is the accuracy; the confusion matrix is ​​as follows Fig.18 As shown in (c), the horizontal axis is the predicted value and the vertical axis is the actual value. Fig.18 It can be seen that the trained ResNet-50 neural network has high accuracy and meets the requirements of actual use.

[0057] Fig.19 This is a schematic diagram of the structure of a fuel cell fault diagnosis device provided in an embodiment of the present application. Fig.19 As shown, the fuel cell fault diagnosis device 200 includes: at least one processor 201; and a memory 202 that is communicatively connected to the at least one processor 201; wherein the memory 202 stores instructions that can be executed by the at least one processor 201, and the instructions are executed by the at least one processor 201 so that the at least one processor 201 can: perform cyclic period dimension increase processing on the two-dimensional impedance spectrum corresponding to the preset sample fuel cell impedance data to obtain the three-dimensional impedance field corresponding to the preset sample fuel cell impedance data; perform data fitting on the three-dimensional impedance field corresponding to the preset sample fuel cell impedance data through an equivalent circuit to generate a mirror impedance field, and train the mirror impedance field through an interleaved learning method to obtain a cross-learning impedance field extension model; input the acquired impedance spectrum of the fuel cell to be tested into the cross-learning impedance field extension model, and generate the life cycle impedance field of the fuel cell to be tested through the cross-learning impedance field extension model; perform coordinate plane projection processing on the life cycle impedance field of the fuel cell to be tested to obtain an impedance density map; classify the impedance density map based on a preset neural network model to determine the fault information of the fuel cell to be tested through the classification result.

[0058] A non-volatile computer storage medium provided in an embodiment of the present application stores computer executable instructions, wherein the computer executable instructions are configured to: perform cyclic period dimension increase processing on a two-dimensional impedance spectrum corresponding to preset sample fuel cell impedance data to obtain a three-dimensional impedance field corresponding to the preset sample fuel cell impedance data; perform data fitting on the three-dimensional impedance field corresponding to the preset sample fuel cell impedance data through an equivalent circuit to generate a mirror impedance field, and train the mirror impedance field through an interleaved learning method to obtain a cross-learning impedance field extension model; input the acquired impedance spectrum of the fuel cell to be tested into the cross-learning impedance field extension model, and generate a life cycle impedance field of the fuel cell to be tested through the cross-learning impedance field extension model; perform coordinate plane projection processing on the life cycle impedance field of the fuel cell to be tested to obtain an impedance density map; and classify the impedance density map based on a preset neural network model to determine fault information of the fuel cell to be tested through the classification results.

[0059] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0060] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the embodiments of the present application may have various changes and variations. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A fuel cell fault diagnosis method, characterized in that: The method comprises: Performing a cyclic dimension increase process on the two-dimensional impedance spectrum corresponding to the preset sample fuel cell impedance data to obtain a three-dimensional impedance field corresponding to the preset sample fuel cell impedance data; Performing data fitting on the three-dimensional impedance field corresponding to the preset sample fuel cell impedance data through an equivalent circuit to generate a mirror impedance field, and training the mirror impedance field through an interleaved learning method to obtain a cross-learning impedance field expansion model; Inputting the acquired impedance spectrum of the fuel cell to be tested into the cross-learning impedance field extension model, and generating the life cycle impedance field of the fuel cell to be tested through the cross-learning impedance field extension model; Performing coordinate plane projection processing on the life cycle impedance field of the fuel cell to be tested to obtain an impedance density map; The impedance density map is classified based on a preset neural network model, so as to determine the fault information of the fuel cell to be tested through the classification result.

2. A fuel cell fault diagnosis method according to claim 1, characterized in that: The step of performing a cyclic dimension increasing process on the two-dimensional impedance spectrum corresponding to the preset sample fuel cell impedance data to obtain a three-dimensional impedance field corresponding to the preset sample fuel cell impedance data specifically includes: generating a two-dimensional battery impedance spectrum based on the preset sample fuel cell impedance data; Based on the two-dimensional battery impedance spectrum, an impedance curve waterfall diagram corresponding to the impedance curves of different cycle periods is drawn to increase the cycle dimension of the two-dimensional battery impedance spectrum; Performing color gradient processing on the impedance curve waterfall chart; The impedance curve waterfall diagram after color gradient processing is surface fitted by a cubic spline interpolation method, and the interpolation result is smoothed by Gaussian filtering to obtain a three-dimensional impedance field corresponding to the preset sample fuel cell impedance data.

3. A fuel cell fault diagnosis method according to claim 1, characterized in that: The method of performing data fitting on the three-dimensional impedance field corresponding to the preset sample fuel cell impedance data through an equivalent circuit to generate a mirror impedance field specifically includes: Based on the first cycle and the last cycle corresponding to the preset sample fuel cell impedance data, element parameter fitting is performed; wherein the element parameter includes at least one of a series resistance, a constant phase element and a polarization resistance; By means of a linear interpolation function, based on the component parameter values ​​corresponding to the first cycle and the last cycle respectively, the simulation circuit component parameters in the intermediate cycle are determined to generate complete simulation equivalent circuit component parameter values; Based on the parameter values ​​of the complete simulation equivalent circuit elements, the mirror impedance field is constructed; wherein the mirror impedance field is used to reflect the impedance change trend of the battery under different cycle periods.

4. A fuel cell fault diagnosis method according to claim 3, characterized in that: Determining the simulation circuit component parameters in the intermediate cycle period based on the component parameter values ​​corresponding to the first cycle period and the last cycle period by using a linear interpolation function specifically includes: Function-based: Obtaining the slope between the component parameter value of the first cycle and the component parameter value of the last cycle; Based on the point-slope equation of a straight line, we get n Component parameter values ​​for a cycle: Based on the component parameters of the first cycle and the last cycle, the simulation equivalent circuit component parameter values ​​of any intermediate cycle are determined: in, is the component parameter value of the last cycle; is the component parameter value of the first cycle; is the parameter value of the simulation equivalent circuit element of any intermediate cycle; k is the slope; n is a cyclic sequence, where 1< n < m .

5. A fuel cell fault diagnosis method according to claim 1, characterized in that: The method of training the mirror impedance field by an interleaved learning method to obtain a cross-learning impedance field extension model specifically includes: Mirror impedance field data are obtained, analyzed and modeled by stepwise linear regression, and the obtained stepwise linear regression model is optimized by an interleaved learning method.

6. A fuel cell fault diagnosis method according to claim 5, characterized in that: The stepwise linear regression model is: ； in, Z is the impedance value; CPE T is the first element parameter in the equivalent circuit; CPE P is the second element parameter in the equivalent circuit; Rs is the parameter of the third element in the equivalent circuit; R is the fourth element parameter in the equivalent circuit; n is a cyclical sequence; is the preset coefficient; is the coefficient corresponding to the first element parameter; is the coefficient corresponding to the second element parameter; is the coefficient corresponding to the third element parameter; is the coefficient corresponding to the fourth element parameter; is the coefficient corresponding to the fifth element parameter; is the product of the last component parameter and the corresponding coefficient.

7. A fuel cell fault diagnosis method according to claim 5, characterized in that: The stepwise linear regression model obtained by the interleaved learning method is optimized, specifically including: Performing Gaussian noise processing on the impedance density image corresponding to the preset sample fuel cell impedance data to obtain an expanded data set; Dividing the expanded data set into a training set and a validation set according to a preset ratio; Training is performed based on the training set, the validation set, and a preset ResNet-50 neural network, and an interleaved learning method is used to optimize a stepwise linear regression model.

8. A fuel cell fault diagnosis method according to claim 1, characterized in that: The process of performing coordinate plane projection processing on the life cycle impedance field of the fuel cell to be tested to obtain an impedance density map specifically includes: Projecting the life cycle impedance field of the fuel cell to be tested onto the xy, yz and xz planes to obtain a scatter plot; Optimizing the scatter plot into a contour plot, and performing gradient processing between the contour plots; The contour map is filled using the contourf function, and the filled contour map is smoothed by the pcolormesh function to obtain the impedance density map.

9. A fuel cell fault diagnosis device, characterized in that: The device comprises a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the device is triggered to execute the method according to any one of claims 1 to 8.

10. A non-volatile computer storage medium storing computer executable instructions, characterized in that: The computer executable instructions can execute the method according to any one of claims 1 to 8.

Images