Battery discharge end point prediction method

Abstract

A battery discharge end point prediction method based on an interpretable physical driving neural network model, current data and voltage data during battery discharge are collected; an interpretable physical driving neural network based on a second-order RC equivalent circuit model is established; the collected current data are taken as features, and the voltage data are taken as labels, and the interpretable physical driving neural network is trained; after the training of the interpretable physical driving neural network is completed, given current data, the trained interpretable physical driving neural network outputs a voltage prediction curve, and the discharge end is predicted based on the voltage prediction curve.

Description

Technical Field

[0001] This invention belongs to the field of lithium battery health management, and in particular, it is a method for predicting the end point of battery discharge based on an interpretable physical-driven neural network. Background Technology

[0002] Data-driven approaches have been applied across various fields, and using artificial intelligence to solve scientific problems has become a new paradigm. In battery health management, numerous data-driven methods are used for battery health assessment, state of charge estimation, and discharge termination prediction. While these methods achieve high accuracy in predicting battery discharge termination, their operational mechanisms are unclear, they lack interpretability, and require large amounts of data for training. Therefore, a model that can be trained with limited data and is interpretable is needed.

[0003] The information disclosed in the background section is only intended to enhance the understanding of the background of the present invention, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention proposes a battery discharge termination prediction method based on an interpretable physical-driven neural network model. Based on an equivalent circuit model, the second-order RC equivalent circuit model is expanded into a physical-driven neural network model, thus making the described physical-driven neural network interpretable. To improve the prediction results, a physically meaningful boundary condition loss is designed to constrain the optimization process of the neural network.

[0005] The objective of this invention is achieved through the following technical solution: a method for predicting the end point of battery discharge based on an interpretable physical-driven neural network model includes:

[0006] Step S100: Collect current and voltage data during the battery discharge process;

[0007] Step S200: Establish an interpretable physical driving neural network based on a second-order RC equivalent circuit model;

[0008] Step S300: Use the collected current data as features and the voltage data as labels to train the interpretable physical drive neural network.

[0009] In step S400, after the interpretable physical driving neural network is trained, given current data, the trained interpretable physical driving neural network outputs a voltage prediction curve, and the discharge is predicted to end based on the voltage prediction curve.

[0010] In the battery discharge termination prediction method based on an interpretable physical-driven neural network model, in step S100, the current and voltage data collected during the battery discharge process are distributed as c = [c1, c2, ..., c...]. n ] and v = [v1, v2, ..., v n ], where n represents its length, the dataset consisting of current data and voltage data is represented as: Where the subscript i represents the i-th sample, and N represents the number of samples collected.

[0011] In the battery discharge termination prediction method based on an interpretable physical driving neural network model, step S200, which involves establishing an interpretable physical driving neural network based on a second-order RC equivalent circuit model, includes the following sub-steps:

[0012] Step S201: Establish a second-order RC equivalent circuit model and construct a recursive calculation graph based on the current and voltage control equations;

[0013] Step S202: Replace the unknown parameters in the constructed recursive computation graph with a neural network to construct an interpretable physical driving neural network based on a second-order RC equivalent circuit model.

[0014] In the battery discharge termination prediction method based on an interpretable physical-driven neural network model, in step S201, the voltage control equation of the second-order RC equivalent circuit model is:

[0015] Where: C b C sp and C s Both represent capacitors, used to capture open-circuit voltage, nonlinear voltage drop caused by surface overpotential, and ohmic voltage drop, respectively. q b q sp and q s They represent C respectively b C sp and C s The amount of charge in, V b It is the open-circuit voltage, V sp V represents the nonlinear voltage drop. s This represents the ohmic voltage drop, and V represents the battery's output voltage. C can be... b Represented as a third-order polynomial function of SOC (state of charge).

[0016] C b =α0·σ 3 +α1·σ 2 +α2·σ+α3, where: σ represents SOC, α0, α1, α2 and α3 represent unknown coefficients, and the formula for calculating SOC is: q max Indicates the maximum chargeable capacity, C max This indicates the nominal capacity. According to Ohm's law, the current relationship is as follows:

[0017] Among them: I p I b I sp and I s These represent the flow through R. p C b C sp and C s The current, I represents the load current; V p R represents p Voltage across the terminals; R s and R p It is a resistor with a fixed resistance value, R s Capture ohmic voltage drop, R s R represents the parasitic resistance that causes battery self-discharge. sp The nonlinear voltage drop caused by surface overpotential is captured, and its resistance is expressed as a nonlinear function of SOC:

[0018] R sp =γ0+γ1exp(γ2(1-σ)), where: γ0, γ1 and γ2 represent unknown coefficients.

[0019] Since the amount of electric charge can be expressed as the integral of current over time, the relationship between the various variables at time t and time t-1 can be established, using symbols... and Representing the capacitance C at time t respectively b C sp and C s The amount of stored charge and the magnitude of the current flowing through it. According to the formula above, the discharge voltage V at time t... t Based on h at time t-1 t-1 Calculations show that h t-1 It can be expressed as the integral of current i over time from 0 to t-1, while the current i at time t-1 is... t-1 And the charge h at time t-2 t-2 and the current I at time t-1 t-1 This is relevant. Therefore, the discharge voltage variation curve can be obtained using recursive calculation. The constructed recursive calculation graph is shown below. Figure 3 As shown.

[0020] In the battery discharge termination prediction method based on an interpretable physical-driven neural network model, in step S202, since the coefficients in the equivalent circuit model are unknown in actual use, a multilayer perceptron is used to model the SOC and V.b and 1 / R sp This allows the model to learn the unknown coefficients from the data. Where f1(·), f2(·), and f3(·) represent three multilayer perceptrons. The learnable multilayer perceptron model, combined with the computational process in step S201, constitutes an interpretable physical-driven neural network model. Besides SOC and V... b and 1 / R sp Apart from the variables, the remaining variables are calculated according to the formulas in step S201. Since the calculation formulas for all variables follow physical laws and the calculation process is consistent with the equivalent circuit model, the neural model architecture is physically driven and interpretable.

[0021] In the battery discharge termination prediction method based on an interpretable physical driving neural network model, step S300, training the interpretable physical driving neural network, includes the following sub-steps:

[0022] Step S301: Transfer the dataset The current data input can be interpreted to drive the neural network model to perform forward propagation and obtain the voltage prediction value, denoted as:

[0023]

[0024] in: This indicates the predictive voltage of the physically driven neural network model that can be explained. This refers to the interpretable physical-driven neural network model described above;

[0025] S302: Based on the predicted voltage and the actual voltage, design a weighted prediction error loss function. The calculation formula is as follows:

[0026]

[0027]

[0028] Where: v i and Representing the actual voltage and predicted voltage, β is a hyperparameter, w = [w i w2, ..., w n ] indicates weight.

[0029] The weight w is a vector consisting of n points taken at equal intervals in the interval [1, n / 5] (n>5), expressed as:

[0030] w = linspace(n / 5, 1, n),

[0031] S303: Design the boundary condition loss function based on the upper and lower voltage limits of the battery.

[0032]

[0033] Among them, V0 and V EOD These represent the charging cutoff voltage and the discharging cutoff voltage, respectively, and ReLU(·) represents the linear rectification function;

[0034] S304: Combine the prediction error loss function and the boundary condition loss function to construct the overall optimization objective function of the physical driving neural network, calculate the loss value, and perform backpropagation;

[0035] S305: Repeat steps S301 to S304. When the number of iterations reaches the set maximum number of iterations, the training of the physical-driven neural network model can be explained as complete.

[0036] In the battery discharge termination prediction method based on an interpretable physics-driven neural network model, in step S304, the overall optimization objective function is expressed as:

[0037] Compared with the prior art, the present invention has the following advantages: the present invention constructs an interpretable physical driving neural network for predicting the discharge end point based on the equivalent circuit model; it considers the physical boundary condition loss, requiring fewer training samples; and it combines the equivalent circuit model to improve the prediction accuracy and enhance the robustness of the model. Attached Figure Description

[0038] Various other advantages and benefits of the present invention will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiments below. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0039] In the attached diagram:

[0040] Figure 1 This is a flowchart illustrating a battery discharge termination prediction method based on an interpretable physical-driven neural network model, provided in one embodiment of the present invention.

[0041] Figure 2 This is a schematic diagram of a second-order RC equivalent circuit model provided in one embodiment of the present invention;

[0042] Figure 3This is a recursive calculation diagram for constructing a second-order RC equivalent circuit model provided in one embodiment of the present invention;

[0043] Figure 4 This is a schematic diagram of an interpretable physical-driven neural network model provided in one embodiment of the present invention;

[0044] Figure 5 This is a schematic diagram of three multilayer perceptrons inside an interpretable physical-driven neural network provided in one embodiment of the present invention.

[0045] The present invention will be further explained below with reference to the accompanying drawings and embodiments. Detailed Implementation

[0046] The following will refer to the appendix. Figures 1 to 5 Specific embodiments of the invention will be described in more detail below. While specific embodiments of the invention are shown in the accompanying drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0047] It should be noted that certain terms are used in the specification and claims to refer to specific components. Those skilled in the art will understand that different terms may be used to refer to the same component. This specification and claims do not distinguish components based on differences in terminology, but rather on differences in function. The terms "comprising" or "including" used throughout the specification and claims are open-ended and should be interpreted as "comprising but not limited to." The following descriptions are preferred embodiments for carrying out the invention; however, these descriptions are for the purpose of understanding the general principles of the specification and are not intended to limit the scope of the invention. The scope of protection of this invention is determined by the appended claims.

[0048] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. The accompanying drawings do not constitute a limitation on the embodiments of the present invention.

[0049] like Figures 1 to 5 As shown, the battery discharge end point prediction method based on an interpretable physics-driven neural network model includes:

[0050] Step S100: Collect current and voltage data during the battery discharge process;

[0051] Step S200: Establish an interpretable physical driving neural network based on a second-order RC equivalent circuit model;

[0052] Step S300: Use the collected current data as features and the voltage data as labels to train the interpretable physical drive neural network.

[0053] In step S400, after the interpretable physical driving neural network is trained, given current data, the trained interpretable physical driving neural network outputs a voltage prediction curve, and the discharge is predicted to end based on the voltage prediction curve.

[0054] In a preferred embodiment of the battery discharge termination prediction method based on an interpretable physical-driven neural network model, in step S100, the current and voltage data collected during the battery discharge process are distributed as c = [c1, c2, ..., c...]. n ] and v = [v1, v2, ..., v n ], where n represents its length, the dataset consisting of current data and voltage data is represented as: Where the subscript i represents the i-th sample, and N represents the number of samples collected.

[0055] In a preferred embodiment of the battery discharge termination prediction method based on an interpretable physical driving neural network model, step S200, which involves establishing an interpretable physical driving neural network based on a second-order RC equivalent circuit model, includes the following sub-steps:

[0056] Step S201: Establish a second-order RC equivalent circuit model and construct a recursive calculation graph based on the current and voltage control equations;

[0057] Step S202: Replace the unknown parameters in the constructed recursive computation graph with a neural network to construct an interpretable physical driving neural network based on a second-order RC equivalent circuit model.

[0058] In a preferred embodiment of the battery discharge termination prediction method based on an interpretable physical-driven neural network model, in step S201, the voltage control equation of the second-order RC equivalent circuit model is:

[0059] Where: C b C sp and C s Both represent capacitors, used to capture open-circuit voltage, nonlinear voltage drop caused by surface overpotential, and ohmic voltage drop, respectively. q b q sp and q s They represent C respectively b C sp and C s The amount of charge in, V b It is the open-circuit voltage, V sp V represents the nonlinear voltage drop. sThis represents the ohmic voltage drop, and V represents the battery's output voltage. C can be... b Represented as a third-order polynomial function of SOC (state of charge).

[0060] C b =α0·σ 3 +α1·σ 2 +α2·σ+α3, where: σ represents SOC, α0, α1, α2 and α3 represent unknown coefficients, and the formula for calculating SOC is: q max Indicates the maximum chargeable capacity, C max This indicates the nominal capacity. According to Ohm's law, the current relationship is as follows:

[0061] Among them: I p I b I sp and I s These represent the flow through R. p C b C sp and C s The current, I represents the load current; V p R represents p Voltage across the terminals; R s and R p It is a resistor with a fixed resistance value, R s Capture ohmic voltage drop, R s R represents the parasitic resistance that causes battery self-discharge. sp The nonlinear voltage drop caused by surface overpotential is captured, and its resistance is expressed as a nonlinear function of SOC:

[0062] R sp =γ0 + γ1exp(γ2(1-σ)), where γ0, γ1, and γ2 represent unknown coefficients. Based on the above formula and the equivalent circuit model, a recursive calculation graph is constructed as follows: Figure 3 As shown.

[0063] Since the amount of electric charge can be expressed as the integral of current over time, the relationship between the various variables at time t and time t-1 can be established. (Using symbols...) and Let C represent the capacitance at time t. b C sp and C s The amount of stored charge and the magnitude of the current flowing through it. According to the formula above, the discharge voltage V at time t... t Based on h at time t-1 t-1 Calculations show that h t-1It can be expressed as the integral of current i over time from 0 to t-1, while the current i at time t-1 is... t-1 And the charge h at time t-2 t-2 and the current I at time t-1 t-1 This is relevant. Therefore, the discharge voltage variation curve can be obtained using recursive calculation. The constructed recursive calculation graph is shown below. Figure 3 As shown.

[0064] In a preferred embodiment of the battery discharge termination prediction method based on an interpretable physical-driven neural network model, in step S202, since the coefficients in the equivalent circuit model are unknown in actual use, a multilayer perceptron is used to model the SOC and V. b and 1 / R sp This allows the model to learn the unknown coefficients from the data. Where f1(·), f2(·), and f3(·) represent three multilayer perceptrons. The learnable multilayer perceptron model, combined with the computational process in step S201, constitutes an interpretable physical-driven neural network model. Besides SOC and V... b and 1 / R sp Apart from the variables, the remaining variables are calculated according to the formulas in step S201. Since the calculation formulas for all variables follow physical laws and the calculation process is consistent with the equivalent circuit model, this neural model architecture is physically driven and interpretable. Figure 4 As shown. Internally, within the interpretable physics-driven neural network model, the structure of the three multilayer perceptrons is as follows: Figure 5 As shown: f1(·) consists of two linear layers and one sigmoid layer, with q b As input, the output is SOC; the structure of f2(·) is as follows: Linear, ReLU, Linear, ReLU, Linear, Sigmoid, and its input is q. b And SOC, the output is V b f3(·) consists of three alternating linear layers and three ReLU layers, with SOC as input and 1 / R as output. sp .

[0065] In a preferred embodiment of the battery discharge termination prediction method based on an interpretable physical driving neural network model, step S300, training the interpretable physical driving neural network, includes the following sub-steps:

[0066] Step S301: Transfer the dataset The current data input can be interpreted to drive the neural network model to perform forward propagation and obtain the voltage prediction value, denoted as:

[0067]

[0068] in: This indicates the predictive voltage of the physically driven neural network model that can be explained. This refers to the interpretable physical-driven neural network model described above;

[0069] S302: Based on the predicted voltage and the actual voltage, design a weighted prediction error loss function. The calculation formula is as follows:

[0070]

[0071]

[0072] Where: v i and Representing the actual voltage and predicted voltage, β is a hyperparameter, w = [w1, w2, ..., w n ] indicates weight.

[0073] The weight w is a vector consisting of n points taken at equal intervals in the interval [1, n / 5] (n>5), expressed as:

[0074] w = linspace(n / 5, 1, n),

[0075] S303: Design the boundary condition loss function based on the upper and lower voltage limits of the battery.

[0076]

[0077] Among them, V0 and V EOD These represent the charging cutoff voltage and the discharging cutoff voltage, respectively, and ReLU(·) represents the linear rectification function;

[0078] S304: Combine the prediction error loss function and the boundary condition loss function to construct the overall optimization objective function of the physical driving neural network, calculate the loss value, and perform backpropagation;

[0079] S305: Repeat steps S301 to S304. When the number of iterations reaches the set maximum number of iterations, the training of the physical-driven neural network model can be explained as complete.

[0080] In a preferred embodiment of the battery discharge termination prediction method based on an interpretable physical-driven neural network model, in step S304, the overall optimization objective function is expressed as:

[0081] In one embodiment, a battery discharge end-point prediction method based on interpretable physics includes the following steps:

[0082] S100: Collects current and voltage data during battery discharge under any operating conditions;

[0083] In this step, the current signal and voltage signal are respectively represented by c = [c1, c2, ..., c...]. n ] and v = [v1, v2, ..., v n ] represents the current signal, and n represents its length. A dataset consisting of current and voltage signals is represented as:

[0084] Where the subscript i represents the i-th sample, and N represents the number of samples collected.

[0085] It is understandable that, when the battery is fully charged, different discharge strategies will result in different discharge times, which in turn will lead to different numbers of data points collected. In this disclosure, in order to standardize the length of the training samples, the first 60 data points of each sample are uniformly truncated.

[0086] S200: Establish an interpretable physical driving neural network based on a second-order RC equivalent circuit model.

[0087] S300: The collected current data is used as a feature and the voltage data is used as a label to train the physical driving neural network;

[0088] S400: After training, given the current data under any discharge strategy, the trained physical drive network outputs a voltage prediction curve to achieve the prediction of the discharge end point under any discharge strategy.

[0089] This embodiment predicts the end point of battery discharge by constructing a physical driving neural network based on an equivalent circuit model. First, a recursive computation graph is constructed based on the second-order RC equivalent circuit model and the voltage and current control equations. Then, the learnable parameters of the neural network are used to replace the unknown parameters in the equivalent circuit model. To make model training more efficient and stable, a physical boundary condition loss function determined by the upper and lower limits of the battery voltage is constructed. The overall model architecture is similar to the equivalent circuit model, and the loss function also has physical meaning. Therefore, fewer samples are needed to train the model, resulting in higher prediction accuracy and stronger robustness.

[0090] In another embodiment, step S200, establishing an interpretable physical-driven neural network, includes the following sub-steps:

[0091] S201: Establish a second-order RC equivalent circuit model, such as Figure 2 As shown. The voltage control equation of the second-order RC equivalent circuit model is:

[0092] V = V b -V sp -V s

[0093]

[0094]

[0095]

[0096] Where: C b C sp and C s Both represent capacitors, used to capture open-circuit voltage, nonlinear voltage drop caused by surface overpotential, and ohmic voltage drop, respectively. q b q sp and q s They represent C respectively b C sp and C s The amount of charge in, V b It is the open-circuit voltage, which is a nonlinear function of SOC (state of charge); V sp V represents the nonlinear voltage drop. s V represents the ohmic voltage drop, and V represents the battery's output voltage.

[0097] For simplicity, C can be... b Represented as a third-order polynomial function of SOC:

[0098] C b =α0·σ 3 +α1·σ 2 +α2·σ+α3

[0099] Where σ represents SOC, and α0, α1, α2, and α3 represent unknown coefficients. The formula for calculating SOC is:

[0100]

[0101] Where: q max Indicates the maximum chargeable capacity, C max Indicates the nominal capacity.

[0102] According to Ohm's law, the current relationship is as follows:

[0103]

[0104] I b =I p +I

[0105]

[0106]

[0107] Among them: I p I b I sp and I s These represent the flow through R. p C b C sp and C s The current, I represents the load current; V p R represents p Voltage across the terminals; R s and R p It is a resistor with a fixed resistance value, R s Capture ohmic voltage drop, R s R represents the parasitic resistance that causes battery self-discharge; sp The nonlinear voltage drop caused by surface overpotential is captured, and its resistance can be expressed as a nonlinear function of SOC:

[0108] R sp =γ0+γ1exp(γ2(1-σ))

[0109] Where γ0, γ1 and γ2 represent unknown coefficients.

[0110] Based on the above formulas and equivalent circuit models, a recursive calculation graph is constructed, such as... Figure 3 As shown.

[0111] S202: Since the coefficients in the equivalent circuit model are unknown in practical use, three small multilayer perceptrons are used to model the SOC and V. b and 1 / R sp This allows the model to learn the unknown coefficients from the data.

[0112] σ=f1(q b )

[0113] V b =f2(q b ,σ)

[0114]

[0115] Where f1(·), f2(·), and f3(·) represent three small multilayer perceptrons. The structure of the interpretable physical-driven neural network is as follows: Figure 4 As shown.

[0116] This embodiment constructs a recursive computation graph based on a second-order RC equivalent circuit model and designs a physics-driven interpretable neural network architecture.

[0117] In another embodiment, step S300, training the physical driving neural network includes the following sub-steps:

[0118] S301: Input the current data into the model for forward propagation to obtain the predicted voltage value, denoted as:

[0119]

[0120] in: This represents the predicted voltage of the model. This represents the physical-driven neural network model.

[0121] S302: Based on the predicted voltage and the actual voltage, design a weighted prediction error loss function. The calculation formula is as follows:

[0122]

[0123]

[0124] Where: v i and Representing the actual voltage and predicted voltage, β is a hyperparameter, w = [w1, w2, ..., w n [] indicates weight.

[0125] The weight w is a vector consisting of n points taken at equal intervals in the interval [1, n / 5] (n>5), expressed as:

[0126] w = linspace(n / 5, 1, n)

[0127] S303: Design the boundary condition loss function based on the upper and lower voltage limits of the battery.

[0128]

[0129] Among them, V0 and V EOD These represent the charging cutoff voltage and the discharging cutoff voltage, respectively, and ReLU(·) represents the linear rectification function.

[0130] S304: Combine the prediction error loss function and the boundary condition loss function to construct the overall optimization objective function of the physical driving neural network, calculate the loss value, and perform backpropagation;

[0131] In this step, the overall optimization objective function is expressed as:

[0132]

[0133] S305: Repeat steps S301 to S304. When the number of iterations reaches the set maximum number of iterations, the training of the physical driving neural network is completed.

[0134] Regarding the training steps described above, and in conjunction with the preceding text, the designed model architecture and loss function both have physical meaning. By adopting the method of this embodiment, the model can be trained with fewer training samples and a high-precision prediction value of the battery discharge end point can be obtained.

[0135] Figure 2 This is a second-order RC equivalent circuit model. This model uses a constant voltage source and multiple resistors and capacitors to represent a battery. Adjusting the values ​​of different resistors and capacitors can simulate batteries with different properties. In the diagram, C... b R represents the voltage source, which stores all the charge that the battery can hold. sp -C sp R is used to capture the nonlinear voltage drop caused by surface overpotential. s -C s Capture ohmic voltage drop, R p This represents the parasitic resistance caused by battery self-discharge. The current and voltage control equations for the equivalent circuit model have been given previously.

[0136] Figure 3 It is a recursive calculation graph constructed based on the equivalent circuit model, which is for... Figure 2 This further illustrates the relationship between current and voltage. Since charge can be expressed as the integral of current over time, the relationship between various variables at time t and time t-1 can be established. Based on the charge at time t-1... The SOC at time t can be calculated, and further obtained and According to the amount of charge and They can be calculated separately. and At this point, the current output voltage V of the model can be calculated. t Current I t The input model is used to update the model's hidden state, i.e. and Updated to and Next, the model enters the next computational loop. Repeating the above process constructs the recursive computation graph. This recursive computation graph contains several unknown parameters that are unknowable in practical use.

[0137] Figure 4 This is a proposed interpretable physics-driven neural network architecture. It replaces... Figure 3The unknown parameters in the data allow the model to spontaneously learn internal parameters and battery properties from the data, thus enabling accurate prediction of the discharge termination point when encountering discharge strategies under unknown operating conditions. The model's input is current I, and its output is voltage V. This represents the hidden state. Since the entire neural network architecture is derived from an equivalent circuit model and the computation process follows physical laws, the model is physically driven and interpretable.

[0138] Figure 5 This diagram illustrates the three multilayer perceptrons within the proposed interpretable physical-driven neural network architecture, where (a), (b), and (c) represent f1(·), f2(·), and f3(·), respectively. f1(·) consists of two linear layers and one sigmoid layer, with a frequency q... b As input, the output is SOC; the structure of f2(·) is as follows: Linear, ReLU, Linear, ReLU, Linear, Sigmoid, and its input is q. b And SOC, the output is V b f3(·) consists of three alternating linear layers and three ReLU layers, with SOC as input and 1 / R as output. sp .

[0139] The technical solution of this disclosure will be further illustrated below through comparative experiments with Long Short-Term Memory Network (LSTM) and Convolutional Neural Network (CNN).

[0140] Specifically, the main evaluation indicators used in this comparative experiment are mean absolute error (MAE), mean relative error (MAPE), and root mean square error (RMSE).

[0141] Random walk discharge experiments were conducted on eight lithium nickel cobalt manganese oxide batteries, and current and voltage data were collected during the discharge process. Since the number of data points collected varied under different discharge strategies, the first 60 points were uniformly selected as the training set for this experiment; the length of the test set was unknown and could be arbitrary. To demonstrate that the proposed physical-driven neural network model can be trained with fewer training samples, the training set for this experiment consisted of only 30 samples.

[0142] To ensure the fairness of the experiment, all methods used the same hyperparameter settings, and the prediction accuracy of the three methods is shown in Table 1.

[0143] Table 1

[0144]

[0145] Although embodiments of the present invention have been described above in conjunction with the accompanying drawings, the present invention is not limited to the specific embodiments and application fields described above. The specific embodiments described above are merely illustrative and instructive, and not restrictive. Those skilled in the art can make many other forms based on the guidance of this specification and without departing from the scope of protection of the claims of the present invention, and all of these are within the scope of protection of the present invention.

Claims

1. A method for predicting the end point of battery discharge based on an interpretable physics-driven neural network model, characterized in that, It includes the following steps, Step S100: Collect current and voltage data during the battery discharge process; Step S200: Establish an interpretable physical driving neural network based on a second-order RC equivalent circuit model; Step S200 includes: Step S201: Establish a second-order RC equivalent circuit model and construct a recursive calculation graph based on the current and voltage control equations; Step S202: Replace the unknown parameters in the constructed recursive computation graph with a neural network to construct an interpretable physical driving neural network based on a second-order RC equivalent circuit model. In step S201, the voltage control equation of the second-order RC equivalent circuit model is: ,in: , and Both refer to capacitors, used to capture open-circuit voltage, nonlinear voltage drop caused by surface overpotential, and ohmic voltage drop, respectively. , and They represent , and The amount of charge in it, It is open-circuit voltage. This represents a nonlinear voltage drop. Indicates the ohmic voltage drop. Indicates the battery's output voltage. ,in: Indicates SOC, and The unknown coefficient, SOC, is calculated using the following formula: , Indicates the maximum chargeable capacity. This indicates the nominal capacity. According to Ohm's law, the current relationship is as follows: ,in: , , and They represent the flow through , , and The current, Indicates the load current; express The voltage across the two ends; and It is a resistor with a fixed resistance value. Capture ohmic voltage drop, This represents the parasitic resistance that causes battery self-discharge. Capturing the nonlinear voltage drop caused by surface overpotential: ,in: and Represents unknown coefficients. Since the amount of charge is expressed as the integral of current over time, it is possible to establish the relationship between various variables. Time and The connection of moments, using symbols and They represent Time Capacitor , and The amount of stored charge and the magnitude of the current flowing through it. Discharge voltage at time according to Moment Calculations show that Represented as current From time 0 The integral within, and Current at any moment And with Charge at time and Current at any moment The change curve of the discharge voltage was obtained by recursive calculation. In step S202, a multilayer perceptron is used to model the SOC. and This allows the model to learn the unknown coefficients from the data. ,in: , and These represent three multilayer perceptrons. The multilayer perceptrons form an interpretable physical-driven neural network model, where... It consists of two linear layers and one sigmoid layer, which is based on As input, the output is SOC; The structure consists of a linear layer, a ReLU layer, another linear layer, a ReLU layer, a linear layer, and a Sigmoid layer, with its input being... And SOC, the output is ; It consists of alternating combinations of 3 linear layers and 3 ReLU layers, with SOC as input and output as output. ; Step S300: Use the collected current data as features and the voltage data as labels to train the interpretable physical drive neural network. In step S400, after the interpretable physical driving neural network is trained, given current data, the trained interpretable physical driving neural network outputs a voltage prediction curve, and the discharge is predicted to end based on the voltage prediction curve.

2. The method for predicting the end point of battery discharge based on an interpretable physics-driven neural network model according to claim 1, characterized in that, In step S100, the current and voltage data collected during the battery discharge process are distributed as follows: and , The length of the dataset, consisting of current and voltage data, is represented as follows: , where subscript Indicates the first One sample, This indicates the number of samples collected.

3. The method for predicting the end point of battery discharge based on an interpretable physics-driven neural network model according to claim 1, characterized in that, Step S300, training the interpretable physical-driven neural network includes the following sub-steps: Step S301: Transfer the dataset The current data input can be interpreted to drive the neural network model to perform forward propagation and obtain the voltage prediction value, denoted as: ， in: This indicates the predictive voltage of the physically driven neural network model that can be explained. This refers to the interpretable physical-driven neural network model described above; S302: Based on the predicted voltage and the actual voltage, design a weighted prediction error loss function. The calculation formula is as follows: ， in: and Indicates the actual voltage and the predicted voltage. For hyperparameters, Indicates weight, The weights mentioned It is the interval [1, / 5] Taken at medium intervals A vector composed of points, >5, represented as: ， S303: Design the boundary condition loss function based on the upper and lower voltage limits of the battery. : ， in, and These represent the charging cutoff voltage and the discharging cutoff voltage, respectively. Represents a linear rectified function; S304: Combine the prediction error loss function and the boundary condition loss function to construct the overall optimization objective function of the physical driving neural network, calculate the loss value, and perform backpropagation; S305: Repeat steps S301 to S304. When the number of iterations reaches the set maximum number of iterations, the training of the physical-driven neural network model can be explained as complete.

4. The method for predicting the end point of battery discharge based on an interpretable physics-driven neural network model according to claim 3, characterized in that, In step S304, the overall optimization objective function is expressed as: .

Images