Hydrogen discharge control method and device of battery system and storage medium

Abstract

The invention provides a hydrogen discharge control method and device of a battery system and a storage medium. The method comprises the following steps: determining the opening time and the opening period of a hydrogen discharge valve according to relevant parameters of an electric pile of the battery system and a neural network model; and the hydrogen discharging valve is controlled to be opened or closed according to the opening time and the opening period. According to the invention, pile related parameters of a battery system influencing hydrogen discharge control from a theoretical dimension are input into an output layer of a neural network model, and then are processed by an abstraction layer of the neural network model influencing hydrogen discharge control in combination with dimensions except the theoretical dimension; then the accurate opening time and the opening period of the hydrogen discharging valve are intelligently output from an output layer of the neural network model, and finally the opening or closing of the hydrogen discharging valve is intelligently and accurately controlled according to the opening time and the opening period, so that the hydrogen in the battery system can be accurately and intelligently discharged, the utilization rate of the hydrogen is increased, and the energy consumption is reduced. And the potential safety hazard of hydrogen emission is reduced.

Description

Technical Field

[0001] The present invention relates to the field of battery technology, and in particular to a hydrogen discharge control method, device and storage medium for a battery system. Background Art

[0002] Hydrogen discharge control is very important for the hydrogen utilization rate and safety of the battery system. In related technologies, a mathematical model is established based on physical laws to calculate the concentration of impurity gases in the battery system. When the impurity gas concentration reaches a threshold, hydrogen discharge begins.

[0003] However, the above-mentioned method for hydrogen discharge has problems such as inaccurate timing, unsafe and unintelligent hydrogen discharge, and reduced hydrogen utilization. For example, mathematical models have calculation errors and cannot take into account complex actual operating conditions. The timing of hydrogen discharge is not necessarily the time when the hydrogen concentration is the lowest, resulting in excessive hydrogen discharge, reduced hydrogen utilization, and safety hazards. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems existing in the prior art.

[0005] To this end, an object of the present invention is to propose a method that enables the hydrogen in the battery system to be discharged accurately and intelligently, thereby improving the utilization rate of hydrogen and reducing the safety hazards of hydrogen emissions.

[0006] Therefore, a second object of the present invention is to provide a hydrogen discharge control device for a battery system.

[0007] To this end, a third object of the present invention is to provide a computer-readable storage medium.

[0008] In order to achieve the above-mentioned objectives, an embodiment of the first aspect of the present invention proposes a hydrogen discharge control method for a battery system, the method comprising: determining the opening time and opening cycle of a hydrogen discharge valve based on the battery stack-related parameters and the neural network model of the battery system; and controlling the opening or closing of the hydrogen discharge valve based on the opening time and the opening cycle.

[0009] According to the hydrogen discharge control method of the battery system of an embodiment of the present invention, the stack-related parameters of the battery system that affect the hydrogen discharge control from a theoretical dimension are input into the output layer of the neural network model, and then processed by the abstract layer of the neural network model that combines dimensions other than the theoretical ones that affect the hydrogen discharge control. Then, the precise opening time and opening cycle of the hydrogen discharge valve are intelligently output from the output layer of the neural network model. Finally, the opening or closing of the hydrogen discharge valve is intelligently and accurately controlled according to the opening time and the opening cycle, so that the hydrogen in the battery system can be discharged accurately and intelligently, thereby improving the utilization rate of hydrogen and reducing the safety hazards of hydrogen emissions.

[0010] In some embodiments, determining the opening time and opening period of the hydrogen exhaust valve based on the fuel cell stack related parameters and the neural network model includes: inputting the fuel cell stack related parameters into the neural network model to determine the operating parameters of the hydrogen exhaust valve; and determining the opening time and opening period of the hydrogen exhaust valve based on the operating parameters and a preset amplification factor.

[0011] In some embodiments, before determining the opening time and the opening period of the hydrogen exhaust valve, it also includes: obtaining the battery stack related parameters, the initial neural network model and the actual operating parameters of the battery system; determining the initial operating parameters based on the battery stack related parameters and the initial neural network; and determining the neural network model based on the initial operating parameters and the actual operating parameters.

[0012] In some embodiments, determining the neural network model based on the initial operating parameters and the actual operating parameters includes: updating the weight coefficients and bias coefficients of neurons in the initial neural network model based on the initial operating parameters and the actual operating parameters until the initial operating parameters and the actual operating parameters meet preset conditions, thereby determining the neural network model.

[0013] In some embodiments, updating the weight coefficients and bias coefficients of neurons in the initial neural network model includes: obtaining a coefficient matrix of weight coefficients and bias coefficients; and updating the weight coefficients and bias coefficients of neurons in the initial neural network model according to the coefficient matrix.

[0014] In some embodiments, obtaining the coefficient matrix of the weight coefficients and the bias coefficients includes: obtaining the partial derivative matrix of the weight coefficients and the bias coefficients; and determining the coefficient matrix according to the partial derivative matrix and preset coefficients.

[0015] In some embodiments, when determining the coefficient matrix based on the partial derivative matrix and the preset coefficients, the following formula is used:

[0016] Among them, the is the coefficient matrix, is the partial derivative matrix, is a positive real number, is the weight coefficient, is the coefficient of variation.

[0017] In some embodiments, when determining the opening time and the opening period of the hydrogen exhaust valve according to the operating parameters and the preset amplification factor, the following formula is used:

[0018] Among them, the is the operating parameter, is the weight coefficient, is the coefficient of variation, is the opening time of the hydrogen exhaust valve, is the opening period of the hydrogen exhaust valve, and stated is the amplification factor.

[0019] In order to achieve the above-mentioned purpose, an embodiment of the second aspect of the present invention proposes a hydrogen discharge control device for a battery system, which includes: a determination module for determining the opening time and opening period of the hydrogen discharge valve based on the battery stack-related parameters and the neural network model of the battery system; and a control module for controlling the opening or closing of the hydrogen discharge valve according to the opening time and the opening period.

[0020] According to an embodiment of the present invention, a hydrogen discharge control device for a battery system discharges hydrogen from a battery system by inputting stack-related parameters of the battery system that affect hydrogen discharge control from a theoretical dimension into an output layer of a neural network model. The parameters are then processed by an abstract layer of the neural network model that combines dimensions other than theoretical dimensions that affect hydrogen discharge control. The output layer of the neural network model then intelligently outputs a precise opening time and opening cycle of a hydrogen discharge valve. Finally, the opening or closing of the hydrogen discharge valve is intelligently and accurately controlled based on the opening time and opening cycle, so that the hydrogen in the battery system can be accurately and intelligently discharged, thereby improving the utilization rate of hydrogen and reducing the safety hazards of hydrogen emissions.

[0021] In order to achieve the above-mentioned objectives, an embodiment of the third aspect of the present invention proposes a computer-readable storage medium, on which a hydrogen discharge control program for a battery system is stored. When the hydrogen discharge control program for the battery system is executed by a processor, a device installed with the hydrogen discharge control program for the battery system implements the hydrogen discharge control method for the battery system as described in the above-mentioned embodiment.

[0022] Additional aspects and advantages of the present invention will be set forth in part in the description which follows and, in part, will be obvious from the description which follows, or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments with reference to the accompanying drawings, in which: Figure 1 is a flow chart of a hydrogen discharge control method for a battery system according to one embodiment of the present invention; Figure 2 is a neural network topology diagram of a battery system according to one embodiment of the present invention; Figure 3 is a flow chart of a hydrogen discharge control method for a battery system according to another embodiment of the present invention; Figure 4 FIG. 4 is a block diagram of a hydrogen discharge control device for a battery system according to an embodiment of the present invention.

[0024] Reference numerals: Input layer 50; Abstraction layer 51; Output layer 52; A hydrogen discharge control device 100 for a battery system; Determination module 101; Control module 102. DETAILED DESCRIPTION

[0025] The embodiments described with reference to the drawings are exemplary, and embodiments of the present invention are described in detail below.

[0026] Fuel cell systems lack proper hydrogen concentration feedback during hydrogen discharge, leading to blind discharge. In recent years, the development of artificial intelligence has led to the development of a series of large-scale models that have demonstrated outstanding performance in various applications. In the field of automatic control, artificial intelligence has also developed numerous basic control algorithms. Hydrogen discharge strategies for fuel cell engines have always been a difficult problem to solve in the hydrogen energy industry, and traditional algorithms struggle to accurately address this issue. This is consistent with the advantages demonstrated by artificial intelligence across various industries.

[0027] In related technologies, for example, a mathematical model is established based on physical laws to calculate the concentration of impurity gases. When the concentration of impurity gases reaches a threshold, hydrogen begins to be discharged.

[0028] However, in the above-mentioned hydrogen discharge methods, the mathematical models used have calculation errors and cannot take into account the complex operating conditions of the book. The timing of hydrogen discharge is not necessarily the time when the hydrogen concentration is the lowest, resulting in excessive hydrogen discharge and reduced hydrogen utilization.

[0029] For example, hydrogen is discharged periodically and regularly. When the anode of the fuel cell system removes impurity gases, the hydrogen that is discharged at the same time is inevitable. This part of hydrogen should be a mixture of hydrogen and impurity gases in the same cavity of the anode.

[0030] However, in the above-mentioned hydrogen discharge method, it is difficult to separate hydrogen and other gases and discharge them, which does not make full use of hydrogen.

[0031] Therefore, by using the hydrogen discharge control method of the battery system in the embodiment of the present application, the opening time and opening period of the hydrogen discharge valve are intelligently and accurately output from the output layer of the neural network model after the battery system stack related parameters affecting the hydrogen discharge control from the theoretical dimension are input into the output layer of the neural network model and then processed by the abstraction layer of the neural network model combining the dimensions affecting the hydrogen discharge control other than the theoretical dimension, and finally the opening or closing of the hydrogen discharge valve is intelligently and accurately controlled according to the opening time and opening period, so that the hydrogen in the battery system can be accurately and intelligently discharged, thereby improving the utilization rate of hydrogen and reducing the safety hazards of hydrogen discharge.

[0032] The battery system hydrogen discharge control method in the embodiment of the present application will be described below in combination with Figures 1-3 The battery system hydrogen discharge control method in the embodiment of the present application will be described below in combination with

[0033] As Figure 1 shown, it is a flow chart of the battery system hydrogen discharge control method in an embodiment of the present application. The battery system hydrogen discharge control method in the embodiment of the present application at least includes steps S1 and S2.

[0034] Step S1, determining the opening time and opening period of the hydrogen discharge valve according to the battery system stack related parameters and the neural network model.

[0035] In the embodiment, the battery system stack related parameters are intention parameters of whether to open the hydrogen discharge valve, including: stack voltage average difference intention , stack voltage variance intention , stack minimum voltage intention , stack low voltage number intention , tail hydrogen concentration intention and stack voltage drop intention ; the operation parameters of the hydrogen discharge valve are the intention parameters of whether to open the hydrogen discharge valve obtained by the neural network model, which are set as P. The neural network model is a trained model.

[0036] As Figure 2Figure 2 shows a neural network topology diagram of a battery system according to an embodiment of the present invention. The battery system's stack-related parameters are input into the input layer 50 of the neural network model. After processing at the abstraction layer 51, the output layer 52 outputs the hydrogen drain valve's opening time P1 and opening period P2. It is understood that the parameters input into the input layer 50 are theoretical parameters that influence whether the hydrogen drain valve is open, enabling a theoretical decision on whether to open the valve. Factors other than the theoretical parameters are incorporated into the processing at the abstraction layer 51 as weight coefficients and deviation coefficients to determine whether the valve is open. This allows the output layer to output a more accurate result that combines theory and practice. Furthermore, the opening time and opening period of the hydrogen drain valve are output by the neural network model, enabling real-time determination of whether hydrogen drain is necessary without relying on specialized technicians, saving labor costs and making the process more intelligent.

[0037] It is understandable that Figure 2 The neural network topology in the figure is just an example. In actual use, other complex topological structures can be used according to needs. The principles are the same. Figure 2 There are two outputs P1 and P2 in the output layer, but it can also be simplified to one output by multiplying the output parameters by their respective coefficients. For example, simplifying the output to the actual operation intention P of the experimenter, multiplying P by two different coefficients W1 and W2 can also obtain P1 and P2.

[0038] Step S2: Control the opening or closing of the hydrogen exhaust valve according to the opening time and the opening period.

[0039] In an embodiment, the opening or closing of the hydrogen exhaust valve is controlled according to the opening time and the opening cycle. For example, the opening cycle P2 is 6 minutes and the opening time P1 is 1 minute, then the closing time is the opening cycle P2-opening time P1=6 minutes-1 minute=5 minutes, that is, the hydrogen exhaust valve is controlled to be opened for 1 minute and closed for 5 minutes, so as to achieve precise control of the opening or closing of the exhaust valve, so that the engine will not discharge hydrogen blindly, and will only discharge hydrogen when it is detected that hydrogen discharge is needed, thereby improving the utilization rate of hydrogen and reducing the safety hazards of hydrogen emissions.

[0040] According to the hydrogen discharge control method of the battery system of an embodiment of the present invention, the stack-related parameters of the battery system that affect the hydrogen discharge control from a theoretical dimension are input into the output layer of the neural network model, and then processed by the abstract layer of the neural network model that combines dimensions other than the theoretical ones that affect the hydrogen discharge control. Then, the precise opening time and opening cycle of the hydrogen discharge valve are intelligently output from the output layer of the neural network model. Finally, the opening or closing of the hydrogen discharge valve is intelligently and accurately controlled according to the opening time and the opening cycle, so that the hydrogen in the battery system can be discharged accurately and intelligently, thereby improving the utilization rate of hydrogen and reducing the safety hazards of hydrogen emissions.

[0041] In some embodiments, the opening time and opening period of the hydrogen exhaust valve are determined according to the electric pile related parameters and the neural network model, including: inputting the electric pile related parameters into the neural network model to determine the operation parameters of the hydrogen exhaust valve; and determining the opening time and opening period of the hydrogen exhaust valve according to the operation parameters and a preset amplification coefficient.

[0042] In embodiments, the electric pile related parameters of the battery system are intention parameters of whether to open the hydrogen exhaust valve, including: electric pile voltage average difference intention , electric pile voltage variance intention , electric pile minimum voltage intention , electric pile low voltage number intention , tail hydrogen exhaust concentration intention , and electric pile voltage drop intention ; the operation parameters of the hydrogen exhaust valve are intention parameters of whether to open the hydrogen exhaust valve obtained through the neural network model, and are set as P.

[0043] As shown in Figure 2 , the electric pile related parameters are input into the neural network model to determine the operation parameters of the hydrogen exhaust valve, for example, first, the weight coefficient and the bias coefficient of the abstract layer trained through the neural network model are determined, the weight coefficient and the bias coefficient are determined according to the requirements and the actual operation of experienced engineers when the neural network model starts training, the weight coefficient and the bias coefficient in the training process are not accurate enough, so the parameters (including the weight coefficient and the bias coefficient) used in the training need to be corrected each time until the hydrogen exhaust control of the neural network model can make the performance indicators of the battery system reach the requirements, the training is ended, and the weight coefficient and the bias coefficient of the abstract layer trained through the neural network model are obtained. The weight coefficient corresponds to each parameter in the electric pile related parameters of the battery system, including: the electric pile voltage average difference intention corresponds , and , the weights are , and , respectively; the electric pile voltage variance intention corresponds , and , the weights are , and , respectively; the electric pile minimum voltage intention corresponds , and , the weights are , and ; Intended number of low voltage fuel cell stacks correspond 、 and The weights are 、 and ; Tail exhaust hydrogen concentration intention correspond 、 and The weights are 、 and ; Stack voltage drop intention correspond 、 and The weights are 、 and ; correspond and The weights are and ; correspond and The weights are and ; correspond and The weights are and ; Coefficient of deviation Includes: Corresponding 、 and The deviations are 、 and ;correspond and The deviations are and ; Secondly, input the battery stack related parameters of the battery system in the input layer: the average difference of the battery stack voltage 、Stack voltage variance intention 、The minimum voltage intention of the battery stack 、Intended number of low-voltage battery stacks 、Tail exhaust hydrogen concentration intention and stack voltage drop intention .

[0044] Secondly, the input stack voltage difference intention 、Stack voltage variance intention 、The minimum voltage intention of the battery stack 、Intended number of low-voltage fuel cell stacks 、Tail exhaust hydrogen concentration intention and stack voltage drop intention The stack-related parameters are sent to the input layer of the neural network model. After the input layer processes the stack-related parameters, it will output the results to the abstract layer.

[0045] Secondly, at the abstraction layer, there are 、 and Three neurons are used to receive the processing results of the battery stack related parameters input by the input layer, make decisions on the received battery stack related parameters, and output the decision results to and Further decision making is done in two neurons, and The decision result will be output to the output layer, specifically: By the weight coefficient in the above abstract layer , coefficient of deviation The decision-making process of the abstract layer is obtained by obtaining the stack-related parameters of the battery system input by the input layer:

[0046] Finally, in the output layer 52, the abstract layer receives and The decision result is to use the sigmoid() function to determine the opening time and opening cycle of the hydrogen exhaust valve. In the sigmoid() function, the weight coefficient Also includes: and The weights are and ; Coefficient of deviation Also includes: ;The weight coefficient in the output layer , coefficient of deviation And the sigmoid() function is used to obtain the operating parameter P of the hydrogen exhaust valve:

[0047] The preset amplification factor is a proportional amplification factor of the operating parameter P determined according to demand and experimental calibration; let the opening time of the hydrogen discharge valve be P1, and the corresponding preset amplification factor be W1; let the opening period of the hydrogen discharge valve be P2, and the corresponding preset amplification factor be W2; according to the operating parameter P and the preset amplification factors W1 and W2, the opening time P1 and opening period P2 of the hydrogen discharge valve are determined as follows:

[0048] The opening time and opening period of the hydrogen exhaust valve are determined according to the operating parameters and the preset amplification factor. For example, the opening period P2 is 6 minutes and the opening time P1 is 1 minute according to the above parameters and the calculation formula. Then the closing time is: opening period - opening time = P2 - P1 = 6 minutes - 1 minute = 5 minutes, so as to realize the precise control of the opening or closing of the exhaust valve through the neural network model.

[0049] In some embodiments, before determining the opening time and opening cycle of the hydrogen exhaust valve, it also includes: obtaining the battery stack related parameters, initial neural network model and actual operating parameters of the battery system; determining the initial operating parameters based on the battery stack related parameters and the initial neural network; and determining the neural network model based on the initial operating parameters and the actual operating parameters.

[0050] In the embodiment, the battery system is such as a fuel cell; the actual stack-related parameters of the battery system are parameters that affect the hydrogen emission decision, which are selected by the experimenter based on experience and experimental calibration, including: the actual stack voltage difference, which is the difference between the actual average single-cell voltage and the lowest single-cell voltage of the fuel cell, set to ; The actual stack voltage variance is the actual single-chip voltage variance, set to ; The actual minimum voltage of the battery stack is the minimum single-chip voltage of the actual battery stack, set to ; The actual number of low voltage cells is the number of cells with low voltage. ; The actual tail exhaust hydrogen concentration is set to ; Actual stack voltage drop, the actual stack voltage drop is the voltage drop when the actual stack single chip average voltage deviates from the normal average voltage, set as .

[0051] Correspondingly, each battery system's battery stack related parameters have corresponding calibration values, which are determined by the test personnel based on experience and experimental calibration, and are divided into: calibration battery stack voltage difference, set as ; Calibrate the stack voltage variance, set as ; Calibrate the lowest voltage of the battery stack, set to ; Calibrate the low voltage quantity of the battery stack, set to ; Calibrate the tail exhaust hydrogen concentration and set it as This parameter is the "negative emotion feedback" parameter. When the actual value is greater than the calibration value, it means "not wanting" to open the hydrogen exhaust valve; calibrate the stack voltage drop, set it to .

[0052] The stack-related parameters of the battery system are intention parameters for opening the hydrogen exhaust valve, and the stack-related parameters of the battery system are obtained from the actual stack-related parameters of the battery system and the corresponding calibration values of the stack power generation conditions. Each parameter is 0 or 1, i.e. binary, 0 represents that the experimenter "does not want" to open the hydrogen exhaust valve, and 1 represents that the experimenter "wants" to open the hydrogen exhaust valve. The stack-related parameters of the battery system are used as input parameters of the input layer of the neural network. Specifically, the stack-related parameters of the battery system and the obtaining process include: Stack voltage average difference intention is ; Stack voltage variance intention is ; Stack minimum voltage intention ; Stack low voltage number intention is ; Tail hydrogen concentration intention is ; Stack voltage drop intention is ; As shown in Figure 2 , the initial neural network model includes an input layer 50, an abstract layer 51 and an output layer 52. The abstract layer 51 is an input layer perception unit, and the above-mentioned stack-related parameters of the battery system are used as input parameters of the input layer of the neural network. The abstract layer 51 is a judgment unit of the neural network. This layer simulates the complex decision of the professional personnel on the hydrogen exhaust valve. Because the operation of the actual operator on the hydrogen exhaust valve depends not only on the input parameters of the input layer, but also on some unknown, abstract and subtle other environmental factors, etc., but these factors are difficult to model and statistics. The role of this layer is to simulate unknown complex environmental factors through a complex network, and finally these factors will reach the purpose of intelligent decision through continuous training of the neural network in the laboratory. The actual operation parameters are the deviation coefficients in the abstract layer 51 of the trained neural network obtained after continuous training of the initial neural network model, and the weight coefficients corresponding to each input parameter, which are used to represent the influence of some unknown, abstract and subtle other environmental factors on the hydrogen exhaust decision.

[0053] The stack-related parameters of the battery system, the initial neural network model and the actual operation parameters are acquired to realize calibration of the selected stack-related parameters of the battery system affecting the hydrogen emission decision, the stack-related parameters of the battery system are determined from the actual stack-related parameters and the calibration values thereof, the intention of whether to open the hydrogen emission valve is obtained, the stack-related parameters of the battery system representing the intention of whether to open the hydrogen emission valve are taken as the input layer of the neural network, the training result of the neural network is affected from the theoretical parameters, the actual operation parameters representing the uncertain factors other than the theoretical parameters are taken as the parameters of the abstract layer, the training result of the neural network is affected from the factors other than the theoretical parameters, and preparation is made for obtaining the accurate opening time and opening period of the hydrogen emission valve.

[0054] The initial operation parameters are determined according to the stack-related parameters and the initial neural network, for example, the initial operation parameters correspond to the actual operation parameters, are the bias coefficients in the abstract layer 51 of the neural network which has not been trained , and the weight coefficients corresponding to the respective input parameters , and the initial operation parameters are inaccurate at this time, like an experimental personnel with insufficient experience, which cannot well control the hydrogen emission valve, so the actual operation of the experienced engineers in the laboratory is taken as the training input, and the parameters are continuously corrected to prepare for determination of the actual operation parameters.

[0055] The neural network model is determined according to the initial operation parameters and the actual operation parameters to realize training of the initial neural network model according to the initial operation parameters, and to obtain the accurate actual operation parameters after the training is completed.

[0056] In some embodiments, the neural network model is determined according to the initial operation parameters and the actual operation parameters, including updating the weight coefficients and the bias coefficients of the neurons in the initial neural network model according to the initial operation parameters and the actual operation parameters until the initial operation parameters and the actual operation parameters satisfy a preset condition, and determining the neural network model.

[0057] In an embodiment, the preset condition is that the opening period and the start time of the hydrogen emission valve determined by the experimental personnel according to the requirements, experience and experimental calibration, etc. all reach the performance indicators of the fuel cell system; the actual operation parameters are the bias coefficients in the abstract layer 51 of the neural network which has been trained , and the weight coefficients corresponding to the respective input parameters , to represent the influence of some unknown, abstract and subtle other environmental factors, etc. on the hydrogen emission decision; the initial operation parameters correspond to the actual operation parameters, are the bias coefficients in the abstract layer 51 of the neural network which has not been trained , and the weight coefficients corresponding to the respective input parameters At this time, the initial operating parameters are inaccurate, just like an inexperienced experimenter who cannot control the hydrogen exhaust valve well. Therefore, it is necessary to use the actual operation of experienced engineers in the laboratory as training input, and continuously correct the parameters to prepare for determining the actual operating parameters.

[0058] like Figure 2 As shown, in the abstract layer 51, the weight coefficient The parameters corresponding to the battery system's stack related parameters include: stack voltage mean difference intention correspond 、 and The weights are 、 and ; Stack voltage variance intention correspond 、 and The weights are 、 and ; Minimum voltage intention of the battery stack correspond 、 and The weights are 、 and ; Intended number of low voltage fuel cell stacks correspond 、 and The weights are 、 and ; Tail exhaust hydrogen concentration intention correspond 、 and The weights are 、 and ; Stack voltage drop intention correspond 、 and The weights are 、 and ; correspond and The weights are and ; correspond and The weights are and ; correspond and The weights are and ; Coefficient of deviation Includes: Corresponding 、 and The deviations are 、 and ;correspond and The deviations are and .

[0059] In the output layer 52, the output layer uses the sigmoid() function to determine the opening time and opening period of the hydrogen exhaust valve. In the function, the weight coefficient Also includes: and The weights are and ; Coefficient of deviation Also includes: .

[0060] With the above The included parameters correspond to each other and will not be repeated here; With the above The included parameters correspond to each other and will not be repeated here.

[0061] Update the weight coefficients of neurons in the above initial neural network model according to the initial operating parameters and actual operating parameters and coefficient of variation , until the initial operating parameters and actual operating parameters meet the experimental personnel's needs, experience and experimental calibration, etc., the opening cycle and start time of the hydrogen exhaust valve are determined, and all performance indicators of the fuel cell system are achieved. The neural network model is determined, and the training is completed to achieve a more accurate and intelligent trained neural network model.

[0062] The updating method uses, for example, the cost function of the learning algorithm:

[0063] in, is the weight coefficient of all the above , is the coefficient of variation for all the above , n is the number of training times, is the actual value of the operating parameter P of the hydrogen exhaust valve after training, and a is the value of the operating parameter P of the hydrogen exhaust valve predicted based on experience and experimental calibration.

[0064] In some embodiments, updating the weight coefficients and bias coefficients of neurons in the initial neural network model includes: obtaining a coefficient matrix of the weight coefficients and bias coefficients; and updating the weight coefficients and bias coefficients of neurons in the initial neural network model according to the coefficient matrix.

[0065] In an embodiment, a coefficient matrix of weight coefficients and bias coefficients is obtained. For example, the partial derivative matrix is ​​first determined by the weight coefficients and bias coefficients, and then the coefficient matrix is ​​determined by the partial derivative matrix to determine the weight coefficients and bias coefficients to be updated; the weight coefficients and bias coefficients used in the next training in the initial neural network model are updated according to the weight coefficients and bias coefficients obtained from the coefficient matrix to realize the training of the neural network model.

[0066] In some embodiments, obtaining a coefficient matrix of weight coefficients and bias coefficients includes: obtaining a partial derivative matrix of weight coefficients and bias coefficients; and determining the coefficient matrix based on the partial derivative matrix and preset coefficients.

[0067] In an embodiment, the partial derivative matrix of the weight coefficient and the bias coefficient is obtained , in preparation for determining the coefficient matrix; the preset coefficient is a small positive real number determined according to experimental calibration and requirements, set as ; According to the partial derivative matrix and preset coefficients Determination coefficient matrix , in order to determine the weight coefficient and bias coefficient used in the next training.

[0068] In some embodiments, when determining a coefficient matrix based on a partial derivative matrix and preset coefficients, the following formula is used:

[0069]

[0070] in, is the coefficient matrix, is the partial derivative matrix, is a positive real number, is the weight coefficient, is the coefficient of variation.

[0071] In the embodiment, the deviation coefficient in the abstract layer 51 of the trained neural network obtained by continuous training of the initial neural network model is , and the weight coefficients corresponding to each input parameter , used to represent the impact of some unknown, abstract, subtle other environmental factors on hydrogen emission decisions. Weight coefficient In the abstract layer 51, the parameters corresponding to the battery stack related parameters of the battery system include: stack voltage mean difference intention correspond 、 and The weights are 、 and ; Stack voltage variance intention correspond 、 and The weights are 、 and ; Minimum voltage intention of the battery stack correspond 、 and The weights are 、 and ; Intended number of low voltage fuel cell stacks correspond 、 and The weights are 、 and ; Tail exhaust hydrogen concentration intention correspond 、 and The weights are 、 and ; Stack voltage drop intention correspond 、 and The weights are 、 and ; correspond and The weights are and ; correspond and The weights are and ; correspond and The weights are and ; Coefficient of deviation In the abstract layer 51, including: 、 and The deviations are 、 and ;correspond and The deviations are and . Weight coefficient In the output layer 52, it also includes: and The weights are and ; Coefficient of deviation In the output layer 52, it also includes: .

[0072] According to the coefficient matrix, the weight coefficient of the next update of the initial neural network model is obtained, that is, the weight coefficient in the initial operation parameter and the coefficient of variation in the initial operating parameters for:

[0073] Among them, in each update training, the battery stack related parameters of the input battery system are judged ( 、 、 、 、 and ), the actual operating intention of the test personnel is the actual operating parameter P of the hydrogen exhaust valve; the test personnel perform multiple operations, each operation is a group, and each group of operations has input ( 、 、 、 、 and ) and the actual operating parameters P of the hydrogen exhaust valve. The weight coefficient and deviation coefficient can be updated once for each set of data input, and multiple learning is performed until the laboratory personnel determine that the performance of the fuel cell system meets the requirements, that is, the learning is completed.

[0074] In some embodiments, when determining the opening time and opening period of the hydrogen exhaust valve according to the operating parameters and the preset amplification factor, the following formula is used:

[0075] in, is the operation parameter, is the weight coefficient, is the coefficient of variation, is the opening time of the hydrogen discharge valve, is the opening cycle of the hydrogen exhaust valve, and is the magnification factor.

[0076] In an embodiment, Figure 2 As shown, in the output layer 52, the output layer uses the sigmoid() function to determine the opening time of the hydrogen exhaust valve. and open cycle , in the function, the weight coefficient Also includes: and The weights are and ; Coefficient of deviation Also includes: ; and Determined according to requirements and experimental calibration; when calculating P1 and P2, the weight coefficient and coefficient of variation Both are included in the calculation of P, that is, the weight coefficients when calculating P1 and P2 and coefficient of variation The reason why they are equal is that the "intention" to open the hydrogen discharge valve is only necessary for the experimenters on the actual system to judge once. By amplifying the coefficients W1 and W2 in the same proportion, the purpose of controlling the periodic opening of the hydrogen discharge valve can be achieved. W1 and W2 are the time amplification coefficients for the periodic opening of the hydrogen discharge valve, and they are also for protecting the electrical characteristics of the hydrogen discharge valve.

[0077] For example, based on the above parameters and calculation formula, the opening period P2 is 6 minutes and the opening time P1 is 1 minute. Then the closing time is: opening period - opening time = P2 - P1 = 6 minutes - 1 minute = 5 minutes, so as to achieve precise control of the opening or closing of the exhaust valve through the neural network model.

[0078] Reference below Figure 3 The hydrogen discharge control method of the battery system according to the embodiment of the present invention is described in detail.

[0079] like Figure 3 FIG. 1 is a flow chart of a method for controlling hydrogen discharge from a battery system according to another embodiment of the present invention. The method for controlling hydrogen discharge from a battery system according to the embodiment of the present invention comprises at least steps S10 to S18.

[0080] Step S10, obtaining the battery stack related parameters, initial neural network model and actual operation parameters of the battery system.

[0081] Step S11, determining initial operating parameters based on battery stack related parameters and the initial neural network.

[0082] Step S12, obtaining the partial derivative matrix of the weight coefficient and the bias coefficient.

[0083] Step S13: Determine a coefficient matrix based on the partial derivative matrix and preset coefficients.

[0084] Step S14: updating the weight coefficients and bias coefficients of the neurons in the initial neural network model according to the coefficient matrix.

[0085] Step S15: until the initial operating parameters and the actual operating parameters meet the preset conditions, the neural network model is determined.

[0086] Step S16: inputting the relevant parameters of the fuel cell stack into the neural network model to determine the operating parameters of the hydrogen exhaust valve.

[0087] Step S17: determining the opening time and opening period of the hydrogen exhaust valve according to the operating parameters and the preset amplification factor.

[0088] Step S18: Control the opening or closing of the hydrogen exhaust valve according to the opening time and the opening period.

[0089] According to the method of an embodiment of the present invention, the stack-related parameters of the battery system that affect hydrogen discharge control from a theoretical dimension are input into the output layer of the neural network model, and then processed by the abstract layer of the neural network model that combines dimensions other than theoretical ones that affect hydrogen discharge control. Then, the precise opening time and opening cycle of the hydrogen discharge valve are intelligently output from the output layer of the neural network model. Finally, the opening or closing of the hydrogen discharge valve is intelligently and accurately controlled according to the opening time and opening cycle, so that the hydrogen in the battery system can be discharged accurately and intelligently, thereby improving the utilization rate of hydrogen and reducing the safety hazards of hydrogen emissions.

[0090] Reference below Figure 4 A hydrogen discharge control device for a battery system according to an embodiment of the present invention will be described.

[0091] like Figure 4 Figure 1 is a block diagram of a hydrogen discharge control device for a battery system according to one embodiment of the present invention. The hydrogen discharge control device 100 for a battery system according to this embodiment of the present invention includes: a determination module 101 for determining the opening time and opening cycle of a hydrogen discharge valve based on battery stack parameters and a neural network model; and a control module 102 for controlling the opening or closing of the hydrogen discharge valve based on the opening time and opening cycle.

[0092] The hydrogen discharge control device 100 of the battery system according to the embodiment of the present application, when discharging hydrogen from the battery system, inputs the stack-related parameters of the battery system affecting the hydrogen discharge control from the theoretical dimension to the output layer of the neural network model, then processes through the abstraction layer of the neural network model combining the dimensions other than the theory affecting the hydrogen discharge control, and then intelligently outputs the accurate opening time and opening period of the hydrogen discharge valve from the output layer of the neural network model. Finally, the opening or closing of the hydrogen discharge valve is intelligently and accurately controlled according to the opening time and opening period, so that the hydrogen in the battery system can be accurately and intelligently discharged, thereby improving the utilization rate of hydrogen and reducing the safety hazards of hydrogen discharge.

[0093] In some embodiments, the determining module 101 is configured to determine the opening time and opening period of the hydrogen discharge valve according to the stack-related parameters and the neural network model, including: inputting the stack-related parameters to the neural network model to determine the operation parameters of the hydrogen discharge valve; and determining the opening time and opening period of the hydrogen discharge valve according to the operation parameters and a preset amplification coefficient.

[0094] In some embodiments, the determining module 101 is further configured to, before determining the opening time and opening period of the hydrogen discharge valve, acquire the stack-related parameters of the battery system, an initial neural network model, and actual operation parameters; determine initial operation parameters according to the stack-related parameters and the initial neural network; and determine the neural network model according to the initial operation parameters and the actual operation parameters.

[0095] In some embodiments, the determining module 101 is configured to determine the neural network model according to the initial operation parameters and the actual operation parameters, including: updating the weight coefficients and bias coefficients of the neurons in the initial neural network model according to the initial operation parameters and the actual operation parameters until the initial operation parameters and the actual operation parameters satisfy a preset condition, and determining the neural network model.

[0096] In some embodiments, the determining module 101 is configured to update the weight coefficients and bias coefficients of the neurons in the initial neural network model, including: acquiring a coefficient matrix of the weight coefficients and the bias coefficients; and updating the weight coefficients and the bias coefficients of the neurons in the initial neural network model according to the coefficient matrix.

[0097] In some embodiments, the determining module 101 is configured to acquire the coefficient matrix of the weight coefficients and the bias coefficients, including: acquiring a partial derivative matrix of the weight coefficients and the bias coefficients; and determining the coefficient matrix according to the partial derivative matrix and a preset coefficient.

[0098] In some embodiments, the determining module 101 is configured to determine the coefficient matrix according to the partial derivative matrix and the preset coefficient, and the following formula is used:

[0099]

[0100] in, is the coefficient matrix, is the partial derivative matrix, is a positive real number, is the weight coefficient, is the coefficient of variation.

[0101] In some embodiments, the determination module 101 is used to determine the opening time and opening period of the hydrogen exhaust valve according to the operating parameters and the preset amplification factor by substituting the following formula:

[0102] in, is the operation parameter, is the weight coefficient, is the coefficient of variation, is the opening time of the hydrogen discharge valve, is the opening cycle of the hydrogen exhaust valve, and is the magnification factor.

[0103] According to an embodiment of the present invention, a hydrogen discharge control device 100 for a battery system discharges hydrogen from a battery system by inputting stack-related parameters of the battery system that affect hydrogen discharge control from a theoretical dimension into the output layer of a neural network model. The parameters are then processed by an abstract layer of the neural network model that combines parameters that affect hydrogen discharge control from dimensions other than theoretical dimensions. The output layer of the neural network model then intelligently outputs a precise opening time and opening cycle of a hydrogen discharge valve. Finally, the opening or closing of the hydrogen discharge valve is intelligently and accurately controlled based on the opening time and opening cycle, so that hydrogen in the battery system can be accurately and intelligently discharged, thereby improving the utilization rate of hydrogen and reducing the safety hazards of hydrogen emissions.

[0104] The following describes a computer-readable storage medium according to an embodiment of the present invention.

[0105] The computer-readable storage medium of an embodiment of the present invention stores a hydrogen discharge control program for a battery system. When the hydrogen discharge control program for the battery system is executed by a processor, a device installed with the hydrogen discharge control program for the battery system implements the hydrogen discharge control method for the battery system as described in the above embodiment.

[0106] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "illustrative embodiments," "example," "specific example," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example.

[0107] While embodiments of the application have been shown and described, it is to be understood that the application is not limited to the details of the embodiments described, since the scope of the application will be defined with respect to the claims and their equivalents.

Claims

1. A method for controlling hydrogen discharge of a battery system, characterized in that: include: Determine the opening time and opening cycle of the hydrogen exhaust valve based on the battery stack related parameters and neural network model of the battery system; The opening or closing of the hydrogen exhaust valve is controlled according to the opening time and the opening period.

2. The hydrogen discharge control method of the battery system according to claim 1, characterized in that: The determining the opening time and opening cycle of the hydrogen exhaust valve according to the relevant parameters of the fuel cell stack and the neural network model includes: Inputting the fuel cell stack related parameters into the neural network model to determine the operating parameters of the hydrogen exhaust valve; The opening time and the opening period of the hydrogen exhaust valve are determined according to the operating parameters and a preset amplification factor.

3. The hydrogen discharge control method of the battery system according to claim 1 or 2, characterized in that: Before determining the opening time and the opening period of the hydrogen exhaust valve, the method further includes: Obtaining battery stack related parameters, an initial neural network model, and actual operating parameters of the battery system; Determining initial operating parameters according to the battery stack related parameters and the initial neural network; The neural network model is determined according to the initial operating parameters and the actual operating parameters.

4. The hydrogen discharge control method of the battery system according to claim 3, characterized in that: Determining the neural network model according to the initial operating parameters and the actual operating parameters includes: The weight coefficients and bias coefficients of the neurons in the initial neural network model are updated according to the initial operating parameters and the actual operating parameters until the initial operating parameters and the actual operating parameters meet preset conditions, thereby determining the neural network model.

5. The hydrogen discharge control method of the battery system according to claim 4, characterized in that: The updating of the weight coefficients and bias coefficients of neurons in the initial neural network model includes: Get the coefficient matrix of weight coefficients and bias coefficients; The weight coefficients and bias coefficients of the neurons in the initial neural network model are updated according to the coefficient matrix.

6. The hydrogen discharge control method of the battery system according to claim 5, characterized in that: The coefficient matrix for obtaining the weight coefficient and the deviation coefficient includes: Get the partial derivative matrix of weight coefficients and bias coefficients; The coefficient matrix is ​​determined according to the partial derivative matrix and preset coefficients.

7. The hydrogen discharge control method of the battery system according to claim 6, characterized in that: When determining the coefficient matrix according to the partial derivative matrix and the preset coefficients, the following formula is introduced: Among them, the is the coefficient matrix, is the partial derivative matrix, is a positive real number, is the weight coefficient, is the coefficient of variation.

8. The hydrogen discharge control method of the battery system according to claim 2, characterized in that: When determining the opening time and the opening period of the hydrogen exhaust valve according to the operating parameters and the preset amplification factor, the following formula is used: Among them, the is the operating parameter, is the weight coefficient, is the coefficient of variation, is the opening time of the hydrogen exhaust valve, is the opening period of the hydrogen exhaust valve, and stated is the amplification factor.

9. A hydrogen discharge control device for a battery system, characterized in that: include: A determination module, used to determine the opening time and opening cycle of the hydrogen exhaust valve according to the battery stack related parameters and the neural network model of the battery system; A control module is used to control the opening or closing of the hydrogen exhaust valve according to the opening time and the opening period.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a hydrogen discharge control program for a battery system. When the hydrogen discharge control program for the battery system is executed by a processor, a device installed with the hydrogen discharge control program for the battery system implements the hydrogen discharge control method for the battery system according to any one of claims 1 to 8.

Images