CN113283632A

CN113283632A - Early battery fault warning method, system, device and storage medium

Info

Publication number: CN113283632A
Application number: CN202110396666.3A
Authority: CN
Inventors: 黄沛丰; 曾港辉; 白中浩
Original assignee: Hunan University
Current assignee: Hunan University
Priority date: 2021-04-13
Filing date: 2021-04-13
Publication date: 2021-08-20
Anticipated expiration: 2041-04-13
Also published as: CN113283632B

Abstract

The invention discloses a battery early fault early warning method, a system, a device and a storage medium, wherein the method comprises the steps of obtaining battery capacity data based on a micro overcharge cycle experiment; smoothing the battery capacity data by utilizing a Savitzky-Golay filtering algorithm; inputting the battery capacity data after smoothing into a trained long-short term memory network model; outputting predicted future battery capacity data by the trained long-term and short-term memory network model; calculating a root mean square error value between the future battery capacity data and the normal battery actual capacity data; and comparing the root mean square error value with a threshold value, and judging whether the battery is a fault battery. The invention can realize the early diagnosis of the fault battery suffering from micro abuse such as micro overcharge abuse, and effectively prevent the occurrence of battery accidents. The invention can be widely applied to the technical field of fault battery diagnosis.

Description

Early battery fault warning method, system, device and storage medium

Technical Field

The invention relates to the technical field of fault battery diagnosis, in particular to a battery early fault early warning method, a system, a device and a storage medium.

Background

The use of this feature allows for the diagnosis of faulty cells that are subject to minor abuse, such as micro-overcharge abuse, with a difference in their rate of capacity fade, compared to cells under normal charge and discharge cycles, at the middle and later stages of the full life cycle. The existing fault diagnosis algorithm can effectively diagnose faults with obvious characteristics, but the robustness of tiny fault characteristics caused by weak abuse such as improper quick charging, mechanical extrusion deformation and the like is relatively low, and one fault diagnosis algorithm can only act on specific faults, namely only can diagnose the specific faults.

Disclosure of Invention

The present invention is directed to solving at least one of the problems of the prior art. Therefore, the invention provides a battery early fault early warning method, a system, a device and a storage medium.

The technical scheme adopted by the invention is as follows:

in one aspect, an embodiment of the present invention includes a method for early warning of a battery early failure, including:

acquiring battery capacity data based on a micro overcharge cycle experiment;

smoothing the battery capacity data by utilizing a Savitzky-Golay filtering algorithm;

inputting the battery capacity data after smoothing into a trained long-short term memory network model;

the trained long-term and short-term memory network model outputs predicted future battery capacity data;

calculating a root mean square error value between the future battery capacity data and the normal battery actual capacity data;

and comparing the root mean square error value with a threshold value, and judging whether the battery is a fault battery.

Further, the smoothing of the battery capacity data by the Savitzky-Golay filtering algorithm is performed by the following formula:

in the formula, h_iDenotes a smoothing coefficient, H denotes a width of a sliding window, + w denotes an upper limit of the sliding window, -w denotes a lower limit of the sliding window, x_k+1Representing raw battery capacity data, x_k,smoothThe battery capacity data after the smoothing process is shown.

Further, after the smoothing processing is performed on the battery capacity data by using the Savitzky-Golay filtering algorithm, the method further includes:

and adopting a maximum and minimum standardization method to carry out normalization processing on the battery capacity data after the smoothing processing.

Further, the normalization processing of the smoothed battery capacity data by the method of maximum and minimum normalization is performed by the following formula:

in the formula, C_scaledRepresenting normalized battery capacity data, C representing smoothed battery capacity data, C_maxRepresents the maximum battery capacity data, C, of the smoothed battery capacity data_minThe minimum battery capacity data among the smoothed battery capacity data is shown.

Further, the method further comprises training the long-short term memory network model to obtain the trained long-short term memory network model, wherein a mean square error is adopted as a loss function in the training process, and an expression of the loss function is as follows:

where MSE represents the loss function, n represents the number of samples,

representing the predicted value of the model calculation, y_iRepresenting the true value.

On the other hand, the embodiment of the invention also comprises a battery early-fault early-warning system, which comprises:

the acquisition module is used for acquiring battery capacity data based on a micro overcharge cycle experiment;

the smoothing processing module is used for smoothing the battery capacity data by utilizing a Savitzky-Golay filtering algorithm;

the input module is used for inputting the battery capacity data after smoothing into the trained long-short term memory network model;

the output module is used for outputting predicted future battery capacity data by the trained long-term and short-term memory network model;

the calculation module is used for calculating a root mean square error value between the future battery capacity data and the normal battery actual capacity data;

and the comparison and judgment module is used for comparing the root mean square error value with a threshold value and judging whether the battery is a fault battery.

Further, the smoothing module is executed by the following formula:

Further, the system further comprises:

and the normalization processing module is used for performing normalization processing on the smoothed battery capacity data by adopting a maximum and minimum normalization method.

On the other hand, the embodiment of the invention also comprises a battery early-fault early-warning device, which comprises:

at least one processor;

at least one memory for storing at least one program;

when the at least one program is executed by the at least one processor, the at least one processor is caused to implement the early battery failure warning method.

In another aspect, the embodiment of the present invention further includes a computer-readable storage medium, on which a program executable by a processor is stored, and the program executable by the processor is used for implementing the battery early warning method when being executed by the processor.

The invention has the beneficial effects that:

the method is based on a micro-overcharge cycle experiment to obtain battery capacity data; smoothing the battery capacity data by utilizing a Savitzky-Golay filtering algorithm; inputting the battery capacity data after smoothing into a trained long-short term memory network model; outputting predicted future battery capacity data by the trained long-term and short-term memory network model; calculating a root mean square error value between the future battery capacity data and the normal battery actual capacity data; by comparing the root mean square error value with the threshold value, the early diagnosis of the fault battery suffering from micro abuse such as micro overcharge abuse can be realized, and the occurrence of battery accidents can be effectively prevented.

Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.

Drawings

The above and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:

FIG. 1 is a flowchart illustrating steps of a method for early warning of battery failure according to an embodiment of the present invention;

fig. 2 is a schematic diagram illustrating data comparison after smoothing the battery capacity data by using a Savitzky-Golay filtering algorithm according to an embodiment of the present invention;

FIG. 3 is a schematic diagram illustrating a partitioning rule of a data set according to an embodiment of the present invention;

FIG. 4 is a schematic structural diagram of a basic unit of a long-term and short-term memory network according to an embodiment of the present invention;

FIG. 5 is a diagram illustrating the basic structure and parameters of a long term memory network model according to an embodiment of the present invention;

FIG. 6 is a schematic diagram of the predicted effect graph of the group 1 capacity and the RMSE values of different prediction intervals according to the embodiment of the present invention;

FIG. 7 is a schematic diagram of the group 6 capacity prediction effect graph and the RMSE values of different prediction intervals according to the embodiment of the present invention;

fig. 8 is a diagram illustrating the predicted effect on 4.25V data according to an embodiment of the present invention and the RMSE values (L-40, M-10) of different prediction intervals;

fig. 9 is a diagram illustrating the predicted effect on 4.25V data according to an embodiment of the present invention and the RMSE values (L-50, M-30) of different prediction intervals;

fig. 10 is a diagram illustrating a prediction effect graph on 4.35V data according to an embodiment of the present invention and RMSE values (L is 40, M is 10) of different prediction intervals;

fig. 11 is a diagram illustrating the predicted effect on 4.35V data according to an embodiment of the present invention and the RMSE values (L is 50, M is 30) of different prediction intervals;

fig. 12 is a graph showing the root mean square error (L50, M30) between the capacity prediction data and the normal battery capacity data for each 30 cycles of each experimental group according to the embodiment of the present invention;

fig. 13 is a diagram illustrating the root mean square error (L-40, M-10) between the capacity prediction data and the normal battery capacity data for each 10 cycles of each experimental group according to the embodiment of the present invention;

FIG. 14 is a flowchart of an early battery failure warning algorithm based on capacity fading acceleration prediction according to an embodiment of the present invention;

fig. 15 is a schematic structural diagram of the early warning device for battery failure according to the embodiment of the present invention.

Detailed Description

Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the accompanying drawings are illustrative only for the purpose of explaining the present invention, and are not to be construed as limiting the present invention.

In the description of the present invention, it should be understood that the orientation or positional relationship referred to in the description of the orientation, such as the upper, lower, front, rear, left, right, etc., is based on the orientation or positional relationship shown in the drawings, and is only for convenience of description and simplification of description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus, should not be construed as limiting the present invention.

In the description of the present invention, the meaning of a plurality of means is one or more, the meaning of a plurality of means is two or more, and larger, smaller, larger, etc. are understood as excluding the number, and larger, smaller, inner, etc. are understood as including the number. If the first and second are described for the purpose of distinguishing technical features, they are not to be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated or implicitly indicating the precedence of the technical features indicated.

In the description of the present invention, unless otherwise explicitly limited, terms such as arrangement, installation, connection and the like should be understood in a broad sense, and those skilled in the art can reasonably determine the specific meanings of the above terms in the present invention in combination with the specific contents of the technical solutions.

The embodiments of the present application will be further explained with reference to the drawings.

Referring to fig. 1, an embodiment of the present invention provides a battery early-fault warning method, including but not limited to the following steps:

s1, acquiring battery capacity data based on a micro overcharge cycle experiment;

s2, smoothing the battery capacity data by utilizing a Savitzky-Golay filtering algorithm;

s3, inputting the battery capacity data subjected to smoothing into the trained long-short term memory network model;

s4, outputting predicted future battery capacity data by the trained long-term and short-term memory network model;

s5, calculating a root mean square error value between the future battery capacity data and the normal battery actual capacity data;

and S6, comparing the root mean square error value with a threshold value, and judging whether the battery is a fault battery.

In this embodiment, a large amount of capacity self-recovery stages exist in the battery capacity data obtained based on the micro overcharge cycle experiment, and the data are relatively fluctuated. If the capacity prediction is carried out by using the LSTM model based on the original data, the model is difficult to converge in the training stage, and a satisfactory prediction effect cannot be obtained. Therefore, before the model training, the raw data needs to be smoothed, and in this embodiment, a Savitzky-Golay (S-G) filtering algorithm is used.

The S-G filtering algorithm is a filtering algorithm for fitting data in a sliding window range on a time domain by using a local polynomial least square method. The algorithm has the greatest advantage that the shape and the width of the data can be ensured not to change while the fluctuating data is smoothed. The main calculation formula is as follows:

in the formula, h_iDenotes a smoothing coefficient, H denotes a width of a sliding window, + w denotes an upper limit of the sliding window, -w denotes a lower limit of the sliding window, x_k+1Representing raw battery capacity data, x_k,smoothThe battery capacity data after the smoothing process is shown. The larger the sliding window is, the more remarkable the smoothing effect is; h can be calculated by fitting a polynomial by using a least square method_iThe smaller the polynomial order, the more pronounced the smoothing effect.

Taking an experimental group 3 (the charge cut-off voltage is 4.3V) as an example, the data processed by the S-G algorithm is shown in fig. 2, and as can be seen from fig. 2, the battery capacity data processed by the S-G filtering algorithm not only removes the local fluctuation part of the original capacity data, but also basically maintains the rough trend of the original capacity data, and has a better filtering effect.

In this embodiment, after step S2, that is, after the battery capacity data is smoothed by using the Savitzky-Golay filtering algorithm, in order to eliminate the influence of abnormal data on model training, accelerate the convergence rate of the model, and improve the calculation accuracy of the model, the maximum and minimum normalization methods are further adopted to normalize the smoothed capacity data. Mapping the capacity data onto [0,1] by the following formula:

In this example, the capacity fade rates of the batteries with the charge cut-off voltages of 4.4V and 4.4V or more exhibited a difference from that of the batteries under normal charge and discharge cycles at the initial stage of overcharge, and therefore, the difference in capacity change was directly recognized for such batteries. For the batteries with the charging cut-off voltage of 4.25V-4.35V, the capacity attenuation trend of the batteries at the initial stage is similar to that of the normal batteries, and the batteries are difficult to directly distinguish, and if the long-term and short-term memory network model is used for predicting the future capacity of the batteries, the fault diagnosis can be carried out in advance by means of the difference between the capacity attenuation of the normal batteries and the capacity attenuation of the batteries at the middle and later stages of the full life cycle of the batteries, and the faulty batteries can be identified. Therefore, the early battery failure warning method based on the capacity acceleration and decay prediction proposed by the embodiment is directed to the micro overcharge failure battery with the charge cut-off voltage between 4.25V and 4.35V.

In the subsequent training and testing of the long-short term memory network model, the training set and the testing set are derived from the battery capacity data of the experimental groups 1-6, and the total number of the training set and the testing set is 943. In the embodiment of the invention, the early battery fault warning method includes the steps of predicting future capacity data with the length of M based on historical capacity data with the length of L, and according to the criterion, the division rule of input data and label data of a training set is shown in figure 3, wherein the division rule represents battery capacity data of the ith cycle.

The capacity prediction task to be realized by the embodiment is a typical time series prediction problem, and the traditional Recurrent Neural Network (RNN) can inherit useful information obtained by calculating past data in the operation process of current data by means of an internal circulation mechanism, and better cope with the characteristic that the data before and after the time series has correlation. However, when the span of the time sequence is large, the gradient information calculated by the RNN approaches infinity or approaches 0 in the back propagation operation process, which causes gradient explosion or gradient disappearance, and thus the model cannot be converged or trained sufficiently. And a long-short term memory network (LSTM) is used as an improved version of the RNN, and the retention and forgetting of the past information are realized by introducing a door mechanism, so that the long-term dependence problem of the traditional RNN can be effectively solved. Fig. 4 shows the basic unit structure of the LSTM model, which is mainly composed of four parts, namely a forgetting gate, an input gate, an output gate, and a cell state.

The forgetting gate mainly plays a role in selectively forgetting historical calculation information. As shown in equation 1, the input of the hidden node output data h is the previous time_t-1And input data X of the current time_tAfter calculation of a single-layer network, output is limited to be 0-1 by means of a sigmoid activation function, namely f_tThe value range of (a). f. of_tAs cell state C at the previous time_t-1When it is 0, it means that C is completely forgotten_t-1And when 1 is taken out, it means that C is completely retained_t-1Thereby realizing selective forgetting of the history information.

f_t＝σ(W_f·[h_t-1，X_t]+b_f) (formula 1);

the input gate mainly functions to selectively retain the input information at the current moment. As shown in

equations

2 and 3, the input gate is composed of two parts, and the inputs are allh_t-1And X_t. Like a forgetting gate, the first part of the input gate obtains a coefficient i with a value between 0 and 1 through a similar calculation method_t(ii) a The second part of the input gate calculates a new candidate state vector through a neural network layer and a tanh activation function

Based on the nature of the tanh activation function,

the value range of (a) is-1 to 1. Finally pass through i_tAnd

selectively assigning the current time information to the current time cell state C_t。

i_t＝σ(W_i·[h_t-1，X_t]+b_i) (formula 2);

as shown in equation 4, based on the output of the forgetting gate and the input gate, the cell state C at the current time_tThe update can be implemented. Renewed cell state C_tWill continue into the LSTM cell at the next instant.

Equations 5 and 6 show the computation mechanism of the output gate, where O_tThe calculation method of (a) is as same as (f)_tAnd i_tConsistently, the value range is also 01. Renewed cell state C_tAfter processing by tanh activation function, with O_tPerforming product operation to obtain hidden node output h at the current moment_t。h_tOn the one hand, the predicted value of the current time is output, and on the other hand, the predicted value is input as the next time LSTM unitA fraction of the value.

O_t＝σ(W_o·[h_t-1，X_t]+b_o) (formula 5);

h_t＝O_t×tanh(C_t) (formula 6).

In this embodiment, the long and short term memory network model is built by using Python 3.7, and the neural network model adopts an application program interface of a deep learning framework Keras 2.3.1. Because the data volume is not large, a network model with a deep depth is not required to be built or a large number of nerve units is not required to be selected, but in order to ensure that the long-term and short-term memory network model has certain learning capacity, the model selects two layers of LSTM networks, the number of basic nerve units of each layer of LSTM network is 50, the activation functions in the two layers of LSTM networks use default tanh functions, and a Dropout layer is added behind each layer of LSTM network, so that nodes in the nerve networks are randomly inactivated with certain probability, the occurrence of overfitting is effectively avoided, the prediction accuracy of the model on a test set is ensured, and the random inactivation probability of the model is set to be 0.1. In addition, a full connection layer is connected to the last of the model to ensure the length and the dimension of the output vector of the model. Since the long-short term memory network model predicts future capacity data based on historical capacity data, the dimensionality of both the input vector and the output vector is 1; the length L of the input vector and the length M of the output vector are hyper-parameters that need to be adjusted manually, and the optimal value matching needs to be determined by combining prediction results obtained by calculation of different numerical combinations, for example, the length of the input vector 50 and the length of the output vector 20, and the basic structure and parameter quantity of the long-short term memory network model are shown in fig. 5.

During the training process, the model needs to be optimized based on the value of the loss function. The present embodiment selects Mean Square Error (MSE) as the loss function, as shown in the following equation:

where MSE represents the loss function, n represents the number of samples,

In the process of back propagation, the model needs to use a specific optimization algorithm to update the weight parameters and the deviation parameters. In the embodiment, an Adam optimization algorithm is selected, and the learning rate can be adaptively adjusted based on the gradient information of each parameter, so that the convergence speed of the model is greatly increased.

Referring to table 1, table 1 shows the state of health (SOH) for different experimental groups of battery capacities at different numbers of cycles. As can be seen from table 1, if the data of 4.25V charge cut-off voltage is regarded as the normal battery capacity data, when the number of cycles is between 40 and 60, the difference in SOH between the

experimental groups

1 and 2 and the other four groups is 1% or less, and the difference in SOH between the insides thereof is also around 0.7%, and it is impossible to identify the faulty battery directly from the difference in SOH. While the difference between SOH under different cut-off voltage conditions is already obvious when the number of cycles is between 80 and 100. Wherein the difference in SOH between

test groups

1 and 2 is 0.3% and between 2% and 3% and the difference in SOH between

test groups

3 and 4 and between 3% and the difference in SOH between test groups 5 and 6 at a cycle number of 80; the difference in SOH between

test groups

1 and 2 was 0.6% at 100 cycles, whereas it differed from the SOH of

test groups

3 and 4 by 5% to 7% and from the SOH of test groups 5 and 6 by around 10%. Therefore, when the number of cycles is around 80-100, the fault diagnosis can be directly performed based on the capacity data actually acquired, but the battery capacity is seriously attenuated at the moment, so that the potential risk is high, and the diagnosis is late. The capacity prediction model (long-short term memory network model) built by the embodiment can predict the capacity of a future period of time at the stage with the cycle number of 40-60 or at the later stage, and the judgment of fault diagnosis is made in advance by using the difference characteristics between the capacity decay speeds of the later stage. Based on the above analysis, therefore, the present embodiment selects 40, 50, 60 as candidates for the true step length L.

TABLE 1 health State values for different experimental groups of Battery capacities at different cycle numbers

The larger the value of the capacity prediction length, i.e., the length M of the model output vector, the longer the model can predict the capacity in the future, and the earlier the information for fault diagnosis can be provided. However, for the prediction model itself, a larger value of M means that the model is more difficult to converge, and the error value between the predicted capacity value and the actual value of the model is larger. Therefore, considering the above factors together, the present embodiment selects 10, 20, and 30 as candidates for the capacity prediction length M.

In this embodiment, for the discussion of the influence of the true step length L and the prediction length M on the prediction result, the capacity data of the

experiment groups

1, 2, 3, 5, and 6 are selected as the training set, and the capacity data of the experiment group 4 is used as the test set. The final prediction results in different combinations are shown in table 2, where a _ RMSE represents the mean of the root mean square error between the predicted capacity value and the actual capacity value with L as the starting point and the following multiple prediction lengths M, and the prediction advance step size is 1.

TABLE 2 influence of different truth step lengths L and prediction lengths M on the prediction result

As can be seen from table 2, the 1 st and 6 th groups are the best performing ones of the 9 groups, in view of the combination of the a _ MRSE value and the predicted length. The a _ RMSE value of group 1 is the smallest, 0.01073, with the best prediction accuracy in group 9, while also achieving capacity prediction based on the least historical data. However, this group of prediction lengths is only 10, and the failure warning will be given later in time compared to

groups

2 and 3 where M is 20 and 30; the A _ RMSE for group 6 was 0.01774, less than the A _ RMSE for groups 3 and 9 where M is 30, and greater than the A _ RMSE for group 1. But the prediction length of the 6 th group is 30, so that the early warning of the fault can be ensured. Fig. 6 and 7 show the prediction effect of the group 1 and the group 6 and the RMSE value of each prediction interval, respectively, wherein the predicted capacity curve is composed of a plurality of prediction data connecting lines with the length of M, and the RMSE value corresponding to the cycle number of X is the RMSE value between the predicted capacity of the prediction interval [ X, X + M ] and the real capacity after S-G filtering.

Further, this example was verified on the data sets of 4.25V and 4.35V for the charge cut-off voltage with the parameters of group 1 (L is 40, M is 10) and group 6 (L is 50, M is 30) in table 2.

When the prediction accuracy on the faulty battery with the charging cut-off voltage of 4.25V is verified, the capacity data of the

experimental groups

1, 3, 4, 5, 6 are used as a training set, and the experimental group 2 is used as a test set. After the same data processing and model training processes, the A _ RMSE value is 0.005132 when L is 40 and M is 10 through calculation; when L is 50 and M is 30, the A _ RMSE value is 0.01325. It can be seen from fig. 8 and 9 that the model with two sets of parameter combinations can also maintain good accuracy on the 4.25V data set.

When the prediction accuracy on the faulty battery with the charging cut-off voltage of 4.35V is verified, the capacity data of the

experimental groups

1, 2, 3, 4, 6 are used as a training set, and the experimental group 5 is used as a test set. After the same data processing and model training processes, the A _ RMSE value is 0.005452 when L is 40 and M is 10 through calculation; when L is 50 and M is 30, the A _ RMSE value is 0.01825. As can be seen from fig. 10 and 11, the model with two sets of parameter combinations can also maintain good accuracy on the 4.35V data set.

In order to determine whether the battery being diagnosed is a faulty battery, a calculation method is required to characterize the difference of the capacity fading speed, and a specific numerical value is also required to be determined as a threshold value for fault diagnosis. In this embodiment, the Root Mean Square Deviation (RMSD) between the battery capacity data of the future M cycles predicted by the model and the off-line capacity data (after S-G filtering) of the M cycles of the normal battery is used as an index for characterizing such a difference. In this example, the batteries of

experimental groups

1 and 2 were regarded as normal batteries. The process of selecting the threshold is as follows:

(1) and smoothing the battery capacity data of the experimental group 1 by using an S-G filtering algorithm, and regarding the battery capacity data as the off-line capacity data of a normal battery, namely the reference data for calculating the root mean square error with the battery to be diagnosed.

(2) And respectively inputting the capacity data of 50 cycles or 40 cycles of the

experimental groups

2, 3, 4, 5 and 6 into the capacity prediction model by taking the initial prediction starting point as 50 or 40 and the prediction advance step length as 1, and obtaining the capacity prediction data of the last 30 cycles or 10 cycles of each experimental group.

(3) The root mean square error value between the capacity data of 30 or 10 cycles obtained for each prediction and the capacity data of the corresponding cycle number of experimental group 1 was calculated, and the error data calculated for each group was plotted as shown in fig. 12 and 13. RMSE data corresponding to the predicted point x in fig. 12 indicates [ x, x +30 ]) in this interval, the root mean square error value between the predicted capacity data and the capacity data of experimental group 1. In fig. 13, RMSE data corresponding to the predicted point x represents [ x, x +10 ]) in this interval, the root mean square error value between the predicted capacity data and the capacity data of experimental group 1 is calculated.

(4) Tables 3 and 4 summarize the number of cycles for which the model reports a fault (3 consecutive times greater than the number of cycles for the threshold value) when two parameter combinations select different values as the threshold value. As can be seen from tables 3 and 4, the larger the threshold selection, the larger the number of cycles at which a fault is reported. Considering that a certain error exists between a predicted value and a real capacity value of a capacity prediction model, selecting a threshold value needs to ensure certain redundancy, and a system is prevented from misreporting faults; at the same time, the promptness is also ensured. In summary, considering that the accuracy of the capacity prediction model is more accurate when L is 40 and M is 10, it is the best choice to select 0.075 as the threshold, and four groups of batteries are diagnosed as faults around 80 cycles; when L is 50 and M is 30, 0.1 is selected as the optimal threshold and four batteries are diagnosed as faulty around 70 cycles.

Table 3 shows the number of cycles corresponding to the failure when different thresholds are selected (L50, M30)

Table 4 shows the number of cycles (L40, M10) corresponding to the failure when different thresholds are selected

In this embodiment, an optimal combination of the L value and the M value is not determined, and the combination of L being 40 and M being 10 has high accuracy of capacity prediction, but the fault diagnosis is not advanced enough, so that the method is suitable for a system with high requirement on capacity accuracy and low tolerance to false alarm. On the contrary, for L50 and M30, the capacity prediction accuracy is lower than that of the previous group, but the fault can be identified earlier, and the capacity prediction method is more suitable for a system with more importance on safety because the capacity prediction accuracy is 10 cycles earlier than that of the previous group on average.

Fig. 14 is a schematic flowchart summary of the early battery failure warning method based on capacity acceleration and decay prediction according to the present embodiment.

The early-stage battery fault early-warning method provided by the embodiment of the invention has the following technical effects:

the embodiment of the invention obtains the battery capacity data based on the micro overcharge cycle experiment; smoothing the battery capacity data by utilizing a Savitzky-Golay filtering algorithm; inputting the battery capacity data after smoothing into a trained long-short term memory network model; outputting predicted future battery capacity data by the trained long-term and short-term memory network model; calculating a root mean square error value between the future battery capacity data and the normal battery actual capacity data; by comparing the root mean square error value with the threshold value, the early diagnosis of the fault battery suffering from micro abuse such as micro overcharge abuse can be realized, and the occurrence of battery accidents can be effectively prevented.

This embodiment also provides a battery early failure early warning system, includes:

Referring to fig. 15, an embodiment of the present invention further provides a battery early-failure warning device 200, which specifically includes:

at least one processor 210;

at least one memory 220 for storing at least one program;

when executed by the at least one processor 210, causes the at least one processor 210 to implement the method as shown in fig. 1.

The memory 220, which is a non-transitory computer-readable storage medium, may be used to store non-transitory software programs and non-transitory computer-executable programs. The memory 220 may include high speed random access memory and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid state storage device. In some embodiments, memory 220 may optionally include remote memory located remotely from processor 210, and such remote memory may be connected to processor 210 via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

It will be understood that the device configuration shown in fig. 15 is not intended to be limiting of device 200, and may include more or fewer components than shown, or some components may be combined, or a different arrangement of components.

In the apparatus 200 shown in fig. 15, the processor 210 may retrieve the program stored in the memory 220 and execute, but is not limited to, the steps of the embodiment shown in fig. 1.

The above-described embodiments of the apparatus 200 are merely illustrative, and the units illustrated as separate components may or may not be physically separate, may be located in one place, or may be distributed over a plurality of network units. Some or all of the modules can be selected according to actual needs to achieve the purposes of the embodiments.

Embodiments of the present invention also provide a computer-readable storage medium, which stores a program executable by a processor, and the program executable by the processor is used for implementing the method shown in fig. 1 when being executed by the processor.

The embodiment of the application also discloses a computer program product or a computer program, which comprises computer instructions, and the computer instructions are stored in a computer readable storage medium. The computer instructions may be read by a processor of a computer device from a computer-readable storage medium, and executed by the processor to cause the computer device to perform the method illustrated in fig. 1.

It will be understood that all or some of the steps, systems of methods disclosed above may be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on computer readable media, which may include computer storage media (or non-transitory media) and communication media (or transitory media). The term computer storage media includes volatile and nonvolatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data, as is well known to those of ordinary skill in the art. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, Digital Versatile Disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can accessed by a computer. In addition, communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media as known to those skilled in the art.

The embodiments of the present invention have been described in detail with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.

Claims

1. A battery early failure early warning method is characterized by comprising the following steps:

acquiring battery capacity data based on a micro overcharge cycle experiment;

2. The early battery fault warning method according to claim 1, wherein the smoothing of the battery capacity data by the Savitzky-Golay filtering algorithm is performed according to the following formula:

3. The early battery fault warning method according to claim 1, wherein after the battery capacity data is smoothed by a Savitzky-Golay filtering algorithm, the method further comprises:

4. The early warning method for battery failure according to claim 3, wherein the normalization of the smoothed battery capacity data by the method of maximum-minimum normalization is performed according to the following formula:

5. The early warning method for battery faults according to claim 1, further comprising training a long-short term memory network model to obtain the trained long-short term memory network model, wherein a mean square error is used as a loss function in the training process, and an expression of the loss function is as follows:

where MSE represents the loss function, n represents the number of samples,

6. A battery early-fault warning system, comprising:

7. The early battery failure warning system of claim 6, wherein the smoothing module is implemented by the following formula:

8. The early battery failure warning system of claim 6, further comprising:

9. A battery early-fault warning device, comprising:

at least one processor;

at least one memory for storing at least one program;

when executed by the at least one processor, cause the at least one processor to implement the method of any one of claims 1-5.

10. Computer-readable storage medium, on which a processor-executable program is stored, which, when being executed by a processor, is adapted to carry out the method according to any one of claims 1-5.