Lithium ion battery degradation prediction method for safe intelligent decision

Abstract

The invention discloses a lithium ion battery degradation prediction method for safety intelligent decision, which comprises the following steps: acquiring an original measurement signal, and extracting deep representative features hidden in the original measurement signal by adopting a one-dimensional convolutional neural network module; the battery capacity is estimated through a bidirectional long-short-term memory module; deriving the probability density of each battery cycle stage prediction point by using a kernel density estimation module; maintenance decisions based on probability prediction information assess costs of different decisions in each battery cycle stage, and decisions with lower costs are selected. According to the invention, uncertainty in battery capacity prediction and maintenance decision making is processed by constructing an integrated model combining a one-dimensional convolutional neural network algorithm, a bidirectional long-short-term memory technology and a kernel density estimation method, and intelligent analysis decision making of safe operation of the lithium ion battery based on probability prediction information is realized. The accuracy of lithium ion battery degradation prediction is improved, and the reliability of maintenance decision is improved.

Description

Technical Field

[0001] The present invention belongs to the field of electrochemical energy storage system operation and maintenance, and specifically relates to a lithium-ion battery degradation prediction method for safe and intelligent decision-making. Background Art

[0002] Lithium-ion batteries are rechargeable batteries that have been widely used in the field of electric vehicles and energy storage systems due to their excellent properties such as long cycle life, high energy density, and low cost. However, as the number of charge and discharge cycles increases, the internal structure of the battery will undergo significant changes, such as the loss of recoverable lithium, the thickening or decomposition of the solid electrolyte interface film, the decomposition of the electrolyte, and the corrosion of the collector. These major changes will cause the battery performance to continue to decline. When the performance degradation accumulates to a certain extent, equipment failure or catastrophic events may occur. In addition, there is capacity regeneration and local fluctuations during the battery cycle, which may lead to deviations in the prediction. The existing maintenance decision-making methods have limited health management capabilities for battery energy storage systems because they ignore prediction information or do not consider the uncertainty of the prediction.

[0003] Therefore, reliable and accurate Li-ion battery degradation prediction is of great practical significance, which enables managers to understand the health status of the battery in advance and helps to take informed maintenance decisions. Summary of the invention

[0004] Purpose of the invention: To address the problem that existing maintenance decision-making methods ignore prediction information or do not consider prediction uncertainty, resulting in limited health management capabilities of battery energy storage systems, a lithium-ion battery degradation prediction method for safe and intelligent decision-making is provided. The designed degradation prediction model has the characteristics of uncertainty quantification and can be used for predictive maintenance and decision-making.

[0005] Technical solution: To achieve the above purpose, the present invention provides a lithium-ion battery degradation prediction method for safe and intelligent decision-making, comprising the following steps:

[0006] S1: Obtain the original measurement signal and use a one-dimensional convolutional neural network module to extract deep representative features hidden in the original measurement signal;

[0007] S2: Estimate the battery capacity through a bidirectional long short-term memory module based on the deep representative features extracted in step S1;

[0008] S3: Based on the battery capacity estimated in step S2, the probability density of the prediction point of each battery cycle stage is derived using the kernel density estimation module;

[0009] S4: Maintenance decision based on probabilistic prediction information evaluates the cost of different decisions at each battery cycle stage and selects the decision with lower cost.

[0010] Furthermore, the original measurement signal in step S1 includes battery voltage, current, and temperature measurement data, and the extraction of deep representative features is specifically as follows:

[0011] The output of the m+1th convolutional layer for:

[0012]

[0013] In the formula, and Represent the weight and bias of convolutional layer m respectively, denotes the i-th input of convolutional layer m, ReLu(·) denotes the general ReLu function, and N and K are the number of input and convolution kernels.

[0014] Furthermore, in step S2, the battery capacity is estimated by using a bidirectional long short-term memory module, which is specifically expressed as:

[0015]

[0016] In the formula, z t Indicates the estimated battery capacity, and They represent the weights from the forward layer to the output layer and from the backward layer to the output layer, respectively, and b h represents the bias of the output layer; and are the outputs of forward and backward memory respectively.

[0017] Furthermore, the output of the forward memory in step S2 and the output of the backward memory The expression is as follows:

[0018]

[0019] In the formula, “→” and “←” represent the forward and backward processes respectively. Represents a general LSTM function, t represents the current sampling time, t-1 represents the previous sampling time, and t+1 represents the next sampling time.

[0020] Furthermore, the probability density of each battery cycle stage prediction point in step S3 is The calculation of is obtained as follows:

[0021]

[0022] In the formula, z i represents the i-th quantile of the estimated battery capacity, there are n quantiles in total, w represents the window width, and K(·) represents the general kernel density function.

[0023] The window width w is given by:

[0024]

[0025] In the formula, represents the standard deviation of the estimate and is given by:

[0026]

[0027] Furthermore, the decision in step S4 includes performing a fault replacement and performing a preventive replacement when a fault occurs.

[0028] The specific process of obtaining the decision in step S4 is as follows:

[0029] A1: In the current inspection cycle, determine whether a fault occurs. If a fault occurs, replace the lithium-ion battery. If no fault occurs, proceed to step A2.

[0030] A2: Calculate the cost rate R of replacing the battery and the cost rate DN of not taking any action. If the cost rate R is less than the cost rate DN, perform preventive replacement; otherwise, the lithium-ion battery enters the next inspection cycle.

[0031] Furthermore, the cost rate R in step A2 is calculated as follows:

[0032] Given the current inspection cycle h, if a decision is made to replace the battery, the cost rate R (i.e., the cost per unit time) is

[0033]

[0034] Where, d r Represents the cost of preventive battery replacement.

[0035] The calculation of the cost rate DN in step A2 is as follows:

[0036] If the battery is not replaced, there is a probability of failure risk P in the next cycle; in the event of failure, a corrective replacement cost d including downtime cost will be incurred c , the battery life is h+1; therefore, the cost rate DN of deciding not to take any action is given by

[0037]

[0038] In the formula, C 0 represents the measured capacity of a new battery, and η represents the threshold of the battery health state, which can be 70% or 80%, as well as any other value that does not guarantee the reliability of the battery.

[0039] Beneficial effects: Compared with the prior art, the present invention handles the uncertainty in battery capacity prediction and maintenance decision-making by constructing an integrated model that combines a one-dimensional convolutional neural network algorithm, a bidirectional long short-term memory technology, and a kernel density estimation method, thereby realizing intelligent analysis and decision-making for the safe operation of lithium-ion batteries based on probabilistic prediction information, improving the accuracy of lithium-ion battery degradation prediction, and enhancing the reliability of maintenance decisions, so that the safe operation of lithium-ion batteries can be effectively guaranteed. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 Schematic diagram of the structure of lithium-ion battery degradation prediction model for safe and intelligent decision-making;

[0041] Figure 2 This is a maintenance decision flow chart based on probabilistic prediction information. DETAILED DESCRIPTION

[0042] The present invention is further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0043] like Figure 1 As shown, the present invention provides a lithium-ion battery degradation prediction model for safe intelligent decision-making, which includes a battery measurement data input module, a one-dimensional convolutional neural network module, a bidirectional long short-term memory module, a kernel density estimation module, and a battery capacity data output module. The input data V, I, and T of the battery measurement data input module represent the voltage, current, and temperature of the lithium-ion battery during charging measured by the sensor, respectively. The one-dimensional convolutional neural network module is used to extract deep representative features hidden in the original measurement signal; the bidirectional long short-term memory module is used to estimate the battery capacity; and the kernel density estimation module is used to derive the probability density of the prediction point of each battery cycle stage.

[0044] In this embodiment, the above lithium-ion battery degradation prediction model is applied to lithium-ion battery degradation prediction, and a lithium-ion battery degradation prediction method for safe and intelligent decision-making is provided, including the following steps:

[0045] S1: Obtain the original measurement signal and use a one-dimensional convolutional neural network module to extract deep representative features hidden in the original measurement signal;

[0046] The battery measurement input data V, I, T are input into the one-dimensional convolutional neural network module through the battery measurement data input module, and the output of the m+1th convolutional layer can be obtained. for

[0047]

[0048] In the formula, and Represent the weight and bias of convolutional layer m respectively, represents the i-th input of convolutional layer m,

[0049] ReLu(·) represents the general ReLu function, N and K are the number of input and convolution kernels. That is the deep representative feature.

[0050] S2: Estimate the battery capacity through a bidirectional long short-term memory module based on the deep representative features extracted in step S1;

[0051] The battery capacity is estimated through a bidirectional long short-term memory module, which is specifically expressed as:

[0052]

[0053] In the formula, z t Indicates the estimated battery capacity, and They represent the weights from the forward layer to the output layer and from the backward layer to the output layer, respectively, and b h represents the bias of the output layer; and are the outputs of forward and backward memory respectively.

[0054] In the bidirectional long short-term memory, the forward memory is dedicated to learning the forward time-related information, while the backward memory is the opposite; according to the output y of the one-dimensional convolutional neural network algorithm conv , the output of the forward memory and the output of the backward memory The expression is as follows:

[0055]

[0056] In the formula, “→” and “←” represent the forward and backward processes respectively. Represents a general LSTM function, t represents the current sampling time, t-1 represents the previous sampling time, and t+1 represents the next sampling time.

[0057] S3: Based on the battery capacity estimated in step S2, the probability density of the prediction point of each battery cycle stage is derived using the kernel density estimation module;

[0058] Probability density of prediction points at each battery cycle stage The calculation of is obtained as follows:

[0059]

[0060] In the formula, z irepresents the i-th quantile of the estimated battery capacity, there are n quantiles in total, w represents the window width, and K(·) represents the general kernel density function;

[0061] The window width w is given by:

[0062]

[0063] In the formula, represents the standard deviation of the estimate and is given by:

[0064]

[0065] S4: Maintenance decision based on probabilistic prediction information evaluates the cost of different decisions at each battery cycle stage and selects the decision with lower cost.

[0066] like Figure 2 As shown in Figure 1, maintenance decisions based on probability prediction information include fault replacement and preventive replacement when a fault occurs. The specific process of obtaining the decision is:

[0067] A1: In the current inspection cycle, determine whether a fault occurs. If a fault occurs, replace the lithium-ion battery. If no fault occurs, proceed to step A2.

[0068] A2: Calculate the cost rate R of replacing the battery and the cost rate DN of not taking any action. If the cost rate R is less than the cost rate DN, perform preventive replacement; otherwise, the lithium-ion battery enters the next inspection cycle.

[0069] The cost rate R is calculated as follows:

[0070] Given the current inspection cycle h, if a decision is made to replace the battery, the cost rate R (i.e., the cost per unit time) is

[0071]

[0072] Where, d r Represents the cost of preventive battery replacement.

[0073] The cost rate DN is calculated as follows:

[0074] If the battery is not replaced, there is a probability of failure risk P in the next cycle; in the event of failure, a corrective replacement cost d including downtime cost will be incurred c , the battery life is h+1; therefore, the cost rate DN of deciding not to take any action is given by

[0075]

[0076] In the formula, C 0represents the measured capacity of a new battery, and η represents the threshold of the battery health state, which can be 70% or 80%, as well as any other value that does not guarantee the reliability of the battery.

Claims

1. A lithium-ion battery degradation prediction method for safe and intelligent decision-making, characterized in that: The steps include: S1: Obtain the original measurement signal and use a one-dimensional convolutional neural network module to extract deep representative features hidden in the original measurement signal; S2: Estimate the battery capacity through a bidirectional long short-term memory module based on the deep representative features extracted in step S1; S3: Based on the battery capacity estimated in step S2, the probability density of the prediction point of each battery cycle stage is derived using the kernel density estimation module; S4: Maintenance decision based on probabilistic prediction information evaluates the cost of different decisions at each battery cycle stage and selects the decision with lower cost.

2. The lithium-ion battery degradation prediction method for safe and intelligent decision-making according to claim 1 is characterized in that: The original measurement signal in step S1 includes battery voltage, current, and temperature measurement data, and the extraction of deep representative features is specifically as follows: The output of the m+1th convolutional layer for: In the formula, and Represent the weight and bias of convolutional layer m respectively, denotes the i-th input of convolutional layer m, ReLu(·) denotes the general ReLu function, and N and K are the number of input and convolution kernels.

3. The lithium-ion battery degradation prediction method for safe and intelligent decision-making according to claim 2, characterized in that: In step S2, the battery capacity is estimated by using a bidirectional long short-term memory module, which is specifically expressed as: In the formula, z t Indicates the estimated battery capacity, and They represent the weights from the forward layer to the output layer and from the backward layer to the output layer, respectively, and b h represents the bias of the output layer; and are the outputs of forward and backward memory respectively.

4. The lithium-ion battery degradation prediction method for safe and intelligent decision-making according to claim 3 is characterized in that: The output of the forward memory in step S2 and the output of the backward memory The expression is as follows: In the formula, "→" and "←" represent the forward and backward processes respectively. Represents a general LSTM function, t represents the current sampling time, t-1 represents the previous sampling time, and t+1 represents the next sampling time.

5. The lithium-ion battery degradation prediction method for safe and intelligent decision-making according to claim 3 is characterized in that: The probability density of each battery cycle stage prediction point in step S3 The calculation of is obtained as follows: In the formula, z i represents the i-th quantile of the estimated battery capacity, there are n quantiles in total, w represents the window width, and K(·) represents the general kernel density function.

6. A lithium-ion battery degradation prediction method for safe and intelligent decision-making according to claim 5, characterized in that: The window width w is given by: In the formula, represents the standard deviation of the estimate and is given by:

7. The lithium-ion battery degradation prediction method for safe and intelligent decision-making according to claim 5, characterized in that: The decision in step S4 includes performing a fault replacement when a fault occurs and performing a preventive replacement.

8. The lithium-ion battery degradation prediction method for safe and intelligent decision-making according to claim 7, characterized in that: The specific process of obtaining the decision in step S4 is as follows: A1: In the current inspection cycle, determine whether a fault occurs. If a fault occurs, replace the lithium-ion battery. If no fault occurs, proceed to step A2. A2: Calculate the cost rate R of replacing the battery and the cost rate DN of not taking any action. If the cost rate R is less than the cost rate DN, perform preventive replacement; otherwise, the lithium-ion battery enters the next inspection cycle.

9. The lithium-ion battery degradation prediction method for safe and intelligent decision-making according to claim 8, characterized in that: The calculation of the cost rate R in step A2 is as follows: Given the current inspection cycle h, if a decision is made to replace the battery, the cost rate R is Where, d r Represents the cost of preventive battery replacement.

10. The lithium-ion battery degradation prediction method for safe and intelligent decision-making according to claim 8, characterized in that: The calculation of the cost rate DN in step A2 is as follows: If the battery is not replaced, there is a probability of failure risk P in the next cycle; in the event of failure, a corrective replacement cost d including downtime cost will be incurred c , the battery life is h+1; therefore, the cost rate DN of deciding not to take any action is given by Where C0 represents the measured capacity of a new battery and η represents the threshold of the battery health status.

Images