CN114924198A - Battery cell detection method, device, equipment and system - Google Patents

Abstract

The invention provides a method, a device, equipment and a system for detecting electric cores, wherein a preset electric core detection model is used for processing time series data so as to obtain abnormal probability values of all electric cores, whether electric core abnormality exists and which electric core is positioned to be abnormal can be determined according to the abnormal probability values, so that the abnormal electric core can be accurately found and replaced and maintained in time, the safety and the reliability of an energy storage system are further ensured, and the model is obtained based on actual measurement data training so as to improve the detection accuracy of the abnormal electric core.

Description

Battery cell detection method, device, equipment and system

Technical Field

The invention relates to the field of fault location, in particular to a method, a device, equipment and a system for detecting a battery cell.

Background

The energy storage battery in the energy storage system is generally formed by connecting a plurality of battery cells in series, in parallel, or in a series/parallel combination. Because the physique of every electric core is different, although dispose battery management system BMS among the energy storage system, the BMS also hardly guarantees that every electric core all works at normal charge-discharge state.

When the battery cell is in an abnormal state, for example: internal short circuit, ageing, circulation lithium loss, positive pole active material damage, negative pole active material damage, when failing for a short time, can lead to electric core ageing fast, electric core capacity reduces scheduling problem to arouse uncontrollable thermal runaway and incident, reduce energy storage system's security and reliability.

Therefore, a method capable of monitoring an abnormal cell in time when the cell is abnormal is needed to ensure the safety and reliability of the energy storage system.

Disclosure of Invention

In view of this, the present invention provides a method, an apparatus, a device, and a system for detecting a battery cell, so as to solve the problem that a method capable of timely monitoring an abnormal battery cell when the battery cell is abnormal is urgently needed.

In order to solve the technical problems, the invention adopts the following technical scheme:

a cell detection method comprises the following steps:

acquiring time series initial data of the acquired energy storage battery, and determining time series data corresponding to the time series initial data;

calling a preset battery cell detection model to process the time sequence data to obtain the abnormal probability value of each battery cell in the energy storage battery; the preset battery cell detection model is obtained based on training data; the training data comprises time sequence samples of the energy storage battery and identifications corresponding to the time sequence samples;

and determining an abnormal cell detection result of the energy storage battery based on the abnormal probability values of the cells.

Optionally, determining time-series data corresponding to the time-series initial data comprises:

taking the time-series initial data as time-series data;

or calculating a characteristic value of the time series initial data under a preset derivative characteristic according to a preset characteristic calculation rule, and taking the combination of the time series initial data and the characteristic value or the characteristic value as time series data.

Optionally, the time-series initial data includes: cell voltage, cell current, and critical point temperature.

Optionally, before a preset cell detection model is called to process the time-series data and obtain the abnormal probability values of the cells in the energy storage battery, the method further includes:

and updating the preset battery cell detection model.

Optionally, the preset cell detection model is a deep learning neural network model.

Optionally, the training process of the preset cell detection model includes:

acquiring a first time sequence initial sample of the energy storage battery under the condition that at least one abnormal electric core exists in the energy storage battery, and determining a first time sequence sample corresponding to the first time sequence initial sample;

using the position information of the at least one abnormal electric core in the energy storage battery as the identifier of the first time sequence sample;

acquiring a second time series initial sample of the energy storage battery under the condition that no abnormal electric core exists in the energy storage battery, and determining a second time series sample corresponding to the second time series initial sample;

taking a preset normal electric core identifier as an identifier of the second time sequence sample;

taking the first time series sample and the second time series sample as time series samples of an energy storage battery, and taking the identifier of the first time series sample and the identifier of the second time series sample as identifiers corresponding to the time series samples;

and training a preset electric core detection model by using the time sequence sample of the energy storage battery and the identifier corresponding to the time sequence sample until a preset training stopping condition is met.

Optionally, determining an abnormal cell detection result of the energy storage battery based on the abnormal probability values of the cells includes:

screening out the maximum abnormal probability value based on the abnormal probability values of the battery cores, and taking the maximum abnormal probability value as a target abnormal probability value;

determining position marks of the target abnormal probability values in the abnormal probability values of the battery cores;

if the position identification is a preset position identification, determining that an abnormal electric core detection result of the energy storage battery is a first identification; the first identification represents that the energy storage battery has no abnormal battery cell;

if the position identification is not a preset position identification, determining that the abnormal cell detection result of the energy storage battery is a second identification and the position identification; the second identification represents that the energy storage battery has an abnormal battery cell; and the position identification represents the position information of the detected abnormal electric core in the energy storage battery.

Optionally, when the abnormal cell detection result is not a preset abnormal cell detection result, the method further includes:

and under the condition that the abnormal electric core corresponding to the abnormal electric core detection result is replaced by the normal electric core, returning to the step of acquiring the acquired time sequence initial data of the energy storage battery, and sequentially executing until the abnormal electric core detection result is the preset abnormal electric core detection result.

A cell detection apparatus, comprising:

the data acquisition module is used for acquiring acquired time series initial data of the energy storage battery and determining time series data corresponding to the time series initial data;

the model processing module is used for calling a preset battery cell detection model to process the time sequence data to obtain the abnormal probability value of each battery cell in the energy storage battery; the preset battery cell detection model is obtained based on training data; the training data comprises time sequence samples of the energy storage battery and identifications corresponding to the time sequence samples;

and the result determining module is used for determining the abnormal cell detection result of the energy storage battery based on the abnormal probability values of the cells.

A cell detection apparatus, comprising: a memory and a processor;

wherein the memory is used for storing programs;

and the processor calls a program and is used for executing the battery cell detection method.

A battery cell detection system comprises the battery cell detection equipment.

Optionally, the system further comprises a data acquisition device;

the data acquisition equipment is used for acquiring time series initial data of the energy storage battery and outputting the time series initial data to the battery cell detection equipment;

the time series initial data includes: cell voltage, cell current, and critical point temperature.

Optionally, a cloud computing device is also included;

the cloud computing equipment is used for receiving the time sequence data, training and updating a preset cell detection model of the cloud based on the time sequence data, and periodically outputting the updated preset cell detection model of the cloud to the cell detection equipment so as to update the preset cell detection model in the cell detection equipment.

Optionally, the cell detection system is deployed in an energy storage container or an energy storage power station, and is configured to detect an abnormal cell.

Compared with the prior art, the invention has the following beneficial effects:

the invention provides a method, a device, equipment and a system for detecting electric cores, wherein a preset electric core detection model is used for processing time series data so as to obtain abnormal probability values of all electric cores, whether electric core abnormality exists and which electric core is positioned to be abnormal can be determined according to the abnormal probability values, so that the abnormal electric core can be accurately found and replaced and maintained in time, the safety and the reliability of an energy storage system are further ensured, and the model is obtained based on actual measurement data training so as to improve the detection accuracy of the abnormal electric core.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the prior art descriptions will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.

Fig. 1 is a flowchart of a method of a battery cell detection method according to an embodiment of the present invention;

fig. 2 is a structural diagram of a battery set according to an embodiment of the present invention;

fig. 3 is a flowchart of another method for detecting a battery cell according to an embodiment of the present invention;

fig. 4 is a flowchart of a method of a battery cell detection method according to another embodiment of the present invention;

fig. 5 is a flowchart of a method of a battery cell detection method according to another embodiment of the present invention;

fig. 6 is a structural diagram of a preset cell detection model according to an embodiment of the present invention;

fig. 7 is a schematic structural diagram of a battery cell detection apparatus according to an embodiment of the present invention;

fig. 8 is a schematic structural diagram of a battery cell detection device according to an embodiment of the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.

The application range of chemical batteries such as lithium ions, sodium ions, lithium iron phosphate and the like is more and more extensive, and the chemical batteries are mainly applied to energy storage systems such as hydraulic power, firepower, wind power and solar power stations and the like, and a plurality of fields such as electric tools, electric bicycles, electric automobiles, military equipment, aerospace and the like.

The energy storage battery of the energy storage system is usually formed by connecting a plurality of battery cells in series, in parallel, or in a series/parallel combination. Because the physique of every electric core is different, although energy storage system's Battery Management System (BMS) can guarantee that whole energy storage system works at normal parameter within range to and protect when appearing unusually, for example the power failure protection when the temperature is too high, also hardly guarantee through BMS that every electric core all works at normal state, normal charge-discharge state promptly.

When the battery cell does not work in a normal state, problems of internal short circuit, aging, loss of circulating lithium, damage of positive active materials, damage of negative active materials, short-term failure and the like may occur, so that the battery cell is in an abnormal state, and overcharge, overdischarge, overhigh temperature and the like are easy to occur. The long-term work can lead to electric core rapid aging at this kind of abnormal condition's electric core for the capacity of electric core reduces, causes short circuit scheduling problem in the electric core, can arouse lithium ion battery's self-discharge, capacity decay, thereby local thermal runaway arouses uncontrollable thermal runaway and incident.

In order to solve the problems, the terminal voltage characteristics of the energy storage battery to be detected can be extracted, and the Euclidean distance between the terminal voltage characteristics and all samples of the data set can be calculated, so that the health condition of the sample to be detected can be judged. Although the principle of the method is simple, the method cannot cope with the energy storage system with variable states, a large amount of data needs to be acquired offline, and online application cannot be realized.

In order to locate the abnormal chip on line, the inventor finds that the time-series data of the energy storage system can be detected in real time by using a deep learning technology so as to calculate the position of the abnormal cell in the whole energy storage system. The whole process can ensure that the detection is finished in the operation of the energy storage system, and the operation of the system cannot be influenced. And the required calculation amount is low during detection, edge deployment can be performed, and the safety of local data is ensured.

The invention provides a method, a device, equipment and a system for detecting a battery cell, wherein a preset battery cell detection model is used for processing time series data so as to obtain abnormal probability values of all battery cells, whether battery cell abnormalities exist or not and which battery cell is positioned to be abnormal can be determined according to the abnormal probability values, so that the abnormal battery cell can be accurately found and replaced and maintained in time, the safety and the reliability of an energy storage system are further ensured, and the model is obtained based on actual measurement data training so as to improve the detection accuracy of the abnormal battery cell.

On the basis of the above content, an embodiment of the present invention provides a battery cell detection method, which is applied to a battery cell detection device, where the battery cell detection device may be deployed at an edge, so that the calculation amount of a main controller of an energy storage system may be reduced, and the data security may also be ensured. Referring to fig. 1, the cell detection method may include:

and S11, acquiring the acquired time series initial data of the energy storage battery, and determining the time series data corresponding to the time series initial data.

Wherein the time-series initial data includes: cell voltage, cell current, and key point temperature.

In practical application, when all the cells in the energy storage battery are connected in series, the cell current is a series current, and the critical point temperature may be: the temperature of the lugs of the battery cell, the temperature of each surface of the battery cell and the like.

Specifically, a typical arrangement of the energy storage system is as follows: an energy storage system container contains 4 battery pack racks, each battery pack rack contains about 36 battery pack racks, and each battery pack rack contains 12 cells. The battery cells in each pack are connected in series, and the battery pack packs in the packs are also connected in series. Therefore, for each pack group rack, all the cells are connected in series. A total of about 432 cells, sharing a series current. The BMS and the temperature control of the energy storage system can record the voltage of each battery cell, the series current and the temperature of a plurality of key positions (such as battery cell lugs and the surfaces of the battery cells) of each battery pack through various electric signals and temperature sensors, and the detected temperature is used as the temperature of a key point. In this embodiment, the battery pack set rack is simplified to facilitate display, and it is shown in fig. 2 that one battery pack set rack includes 4 battery pack racks, and each battery pack rack includes 5 electric cores. All the battery cores are connected in series, and each battery pack has three temperature measuring points which are t1-1, t1-2 … … t 4-3.

When the time-series initial data of the energy storage battery is acquired, the time-series length L (which may be based on the current time, and be determined in the training of the neural network) is acquired, the voltage of each cell in the energy storage battery (20 values, v1-1, v1-2,.., v4-5, for example, in fig. 2) needs to be adjusted to obtain high accuracy of the algorithm, or to balance calculation resources and accuracy of the algorithm, the series current of the cells of the energy storage battery (1 value, i), and the critical temperature values of all critical positions of the battery pack set rack in the energy storage battery (12 values, t1-1, t1-2,. t4-3) are acquired. In this embodiment, the acquired voltage value of each electric core in the energy storage battery, the acquired series current value of the electric core of the energy storage battery, and the acquired key temperature value of the key position in the energy storage battery are used as the time series initial data of the energy storage battery.

After the time series initial data of the energy storage battery is obtained, time series data corresponding to the time series initial data needs to be determined.

Specifically, in the present embodiment, there are various ways to determine the time-series data corresponding to the time-series initial data.

1. The first mode is as follows:

and taking the time series initial data as time series data.

Specifically, after the time-series initial data is acquired, the time-series initial data is used as time-series data.

2. The second mode is as follows:

and calculating a characteristic value of the time series initial data under a preset derivative characteristic according to a preset characteristic calculation rule, and taking the combination of the time series initial data and the characteristic value or the characteristic value as time series data.

In this embodiment, after the time series initial data is obtained, the preset derivative features of the time series initial data are calculated.

Specifically, the preset derivative feature may be a time series feature, and may include:

1. the rate of change of voltage (or current) over time.

2. Rate of change of temperature with time.

3. dQ/dV, which is the rate of change of charge with charge voltage, and dV/dQ, which is the rate of change of voltage with charge voltage.

4. The dispersion ratio of each cell above features (1, 2 and 3) to each other cell in the same battery pack collection rack.

After the feature value of the preset derived feature is obtained, a combination of the time series initial data and the feature value of the preset derived feature may be used as time series data, or the feature value under the preset derived feature may be directly used as the time series data.

The specific implementation chosen to determine the time series data may be based on actual usage scenario settings.

And S12, calling a preset battery cell detection model to process the time sequence data to obtain the abnormal probability value of each battery cell in the energy storage battery.

In this embodiment, the preset battery cell detection model may be a battery cell detection device that is already built in advance, or the preset battery cell detection model may be updated before use, that is, before step S12 is executed. Specifically, the updating mode may be a preset electrical core detection model that is periodically updated and sent by receiving external equipment.

The external device may be a cloud computing device, such as a cloud platform, in this embodiment, the cloud computing device may train, generate, and update the preset electric core detection model, such as the cloud computing device, based on the time series data, train and update the preset electric core detection model at the cloud end, and periodically output the updated preset electric core detection model at the cloud end to the electric core detection device, so as to update the preset electric core detection model in the electric core detection device.

In addition, the cell detection equipment can train, generate and update the preset cell detection model by itself.

In practical application, the preset electric core detection model is obtained by training based on training data, wherein the training data comprises a time sequence sample of the energy storage battery and an identifier corresponding to the time sequence sample. The abnormal probability value of each battery cell in the energy storage battery can be detected through a preset battery cell detection model obtained through training of training data.

In this embodiment, the preset cell detection model may be a deep learning neural network model, and an encoder part of the deep learning neural network model may be: a recurrent neural network, a one-dimensional convolutional neural network, a self-attention mechanism neural network, etc.

In particular, the deep learning neural network model may be a time series model. The invention adopts a time series neural network model because the current abnormal state of the battery pack, such as the internal short circuit state, is related to the battery state in a previous period (continuously, not suddenly changed), or the internal short circuit state of the battery does not suddenly occur at a certain moment.

And presetting the abnormal probability value of each electric core in the energy storage battery as the data output by the electric core detection model.

Taking 20 electric cores as an example, the number of the abnormal probability values of each electric core in the output energy storage battery is 21, the first 20 abnormal probability values are the abnormal probability values (arranged according to the electric core sequence) corresponding to the 20 electric cores respectively, and the last abnormal probability value is the probability value that all the electric cores are normal.

Can be as follows:

[0.01,0.01,0.01,0.00,0.01,

0.01,0.01,0.01,0.00,0.01,

0.01,0.01,0.01,0.87,0.01,

0.01,0.01,0.01,0.00,0.01,0.01]。

in the alternative, the first and second sets of the first and second sets of the first and second sets of the first and second sets of the first and second sets of the first and second sets of the second,

[0.01,0.01,0.01,0.01,0.00,

0.03,0.00,0.01,0.01,0.01,

0.01,0.01,0.01,0.03,0.01,

0.01,0.02,0.01,0.01,0.02,0.76]。

and S13, determining abnormal cell detection results of the energy storage battery based on the abnormal probability values of the cells.

Specifically, after the abnormal probability value of each battery cell is determined, the abnormal battery cells can be screened out according to the probability value, or all the battery cells are determined to be normal.

Specifically, referring to fig. 3, step S13 may include:

and S21, screening out the maximum abnormal probability value based on the abnormal probability values of the battery cores, and taking the maximum abnormal probability value as a target abnormal probability value.

Specifically, in this embodiment, the maximum abnormal probability value is screened out from the abnormal probability values of the electric cores output by the electric core detection model. Still taking the above 20 cells as an example, if the abnormal probability value is located in the first 20 cells, it indicates that the cell with the highest abnormal probability value is an abnormal cell.

If the abnormal probability value is located at the 21 st bit, since the 21 st bit indicates that all the battery cells are normal, it is indicated that all the battery cells are not abnormal battery cells.

In practical application, the maximum abnormal probability value can be used as a screening standard, and the difference between the maximum abnormal probability value and other abnormal probability values is larger than a preset threshold value and can be used as the screening standard.

And S22, determining the position identification of the target abnormal probability value in the abnormal probability values of the battery cores.

Specifically, since the abnormal probability values are arranged according to the order of the battery cells, the position identifiers of the target abnormal probability values in the abnormal probability values of the battery cells may be determined according to the order, and in this embodiment, the position identifiers may be the 1 st, the 2 nd, and the 3 rd … … st 21 st.

It should be noted that, since the number of the battery cells is 20, if the number of the battery cells is not the first 20, the battery cell is regarded as the 21 st.

S23, judging whether the position mark is a preset position mark; if yes, go to step S24; if not, step S25 is executed.

Specifically, the 21 st abnormal probability value is special, which indicates that all the battery cells are normal, and the first 20 th bits indicate that the specific battery cells are abnormal, the preset position may be identified as the 21 st bit, that is, in this embodiment, it is determined whether the position identification is the 21 st bit.

And S24, determining that the abnormal electric core detection result of the energy storage battery is a first identifier.

The first identification represents that the energy storage battery has no abnormal battery cell, namely if the battery cell is at the 21 st position, all the battery cells are normal.

Such as: the abnormal probability value of each battery cell is as follows:

[0.01,0.01,0.01,0.01,0.00,

0.03,0.00,0.01,0.01,0.01,

0.01,0.01,0.01,0.03,0.01,

0.01,0.02,0.01,0.01,0.02,0.76]。

since 0.76 is located at the 21 st bit and is much larger than the abnormal probability values of the other bits, that is, the difference value from the other abnormal probability values is larger than a preset threshold (e.g., 0.5), it indicates that all cells are normal.

And S25, determining that the abnormal electric core detection result of the energy storage battery is a second identifier and the position identifier.

The second identification represents that the energy storage battery has an abnormal battery cell, and the position identification represents the position information of the detected abnormal battery cell in the energy storage battery, so that the abnormal battery cell can be determined according to the position identification, and the specific position of the abnormal battery cell can be found.

Such as:

[0.01,0.01,0.01,0.00,0.01,

0.01,0.01,0.01,0.00,0.01,

0.01,0.01,0.01,0.87,0.01,

0.01,0.01,0.01,0.00,0.01,0.01]。

since 0.87 is located at the 14 th position and is far greater than the abnormal probability values of other positions, that is, the difference between the abnormal probability values and other abnormal probability values is greater than a preset threshold (for example, 0.5), it indicates that the 14 th electric core is an abnormal electric core, and the 14 th position corresponds to 3-4 electric cores, which indicates that the 3-4 electric cores are abnormal electric cores, and the specific abnormality may be internal short circuit, aging, circulating lithium loss, positive electrode active material damage, negative electrode active material damage, short-term failure, or the like.

That is to say, the abnormal cell detection result is a second identifier and a location identifier (for example, 14 th bit, or 3 to 4) that indicate that the energy storage battery has an abnormal cell.

And outputting the abnormal cell detection result under the condition that the abnormal cell detection result is not a preset abnormal cell detection result.

Specifically, in this embodiment, if the preset abnormal cell detection result is the first identifier, the preset abnormal cell detection result indicates that all the cells are normal.

If yes, the abnormal cell detection result is not a preset abnormal cell detection result, and it is indicated that at least one abnormal cell exists. At this moment, an abnormal cell detection result can be output to a mobile terminal of a maintenance worker, so that the corresponding maintenance worker can replace the abnormal cell.

If the abnormal cell detection result is the preset abnormal cell detection result, all the cells are normal, and the preset abnormal cell detection result can be output at the moment, so that maintenance personnel can know the cell state in time.

In the embodiment, the time series data are processed by using the preset electric core detection model, so that the abnormal probability value of each electric core is obtained, whether the electric core is abnormal or not and which electric core is positioned to be abnormal can be determined according to the abnormal probability value, the abnormal electric core can be accurately found and replaced and maintained in time, the safety and reliability of the energy storage system are further ensured, the model is obtained based on actual measurement data training, and the detection accuracy of the abnormal electric core can be improved.

In the above embodiment, the preset cell detection model is mentioned, and if the cell detection device trains, generates, and updates the preset cell detection model by itself. Referring to fig. 4, the training process of the preset cell detection model includes:

and S31, determining training data.

The training data comprises time series samples of the energy storage battery and identifications corresponding to the time series samples.

Specifically, referring to fig. 5, determining the training data may include:

s41, under the condition that at least one abnormal electric core exists in the energy storage battery, obtaining a first time sequence initial sample of the energy storage battery, and determining a first time sequence sample corresponding to the first time sequence initial sample.

Specifically, taking an abnormal internal short circuit as an example, when a time-series sample of the energy storage battery is acquired, at least one cell (in this embodiment, an abnormal cell is taken as an example, such as the cell 3-4 in fig. 2) with an internal short circuit is used to replace a normal cell in the rack set rack, then the position (such as 3-4 or the number of bits) of the abnormal cell is recorded, and data of continuous charging and discharging are recorded, including the voltage of each cell (20 values, v1-1, v1-2, v4-5, the series current (1 value, i) of the cells in the energy storage battery, and the temperature values (12 values, t1-1, t1-2, v 4-3) of all key positions of the rack set rack in the time-series length L.

In this embodiment, the acquired data is referred to as a first time-series initial sample.

Thereafter, the positions of the abnormal electric cores can be exchanged, and the initial samples of the first time sequence are continuously collected. The process of determining the first time-series sample corresponding to the first time-series initial sample is described with reference to the above corresponding description.

Specifically, taking an abnormal cell as an example, after the abnormal cell records a certain amount of data (time sequence length L) at a position, the position of the abnormal cell in the battery pack set rack is randomly replaced, and the data is continuously recorded. Theoretically, the abnormal cells do not need to traverse each position of the battery pack set rack, but the more the recorded positions are, the better the data volume is, the better the data distribution is, and the better the data points are distributed (the abnormal cells are ensured to appear at each position of the battery pack rack).

For the process of acquiring the first time-series initial sample with two or more abnormal cells, the process is similar to that of the first abnormal cell, please refer to the above corresponding description.

And S42, using the position information of the at least one abnormal electric core in the energy storage battery as the identifier of the first time series sample.

In this embodiment, the position of the abnormal electrical core is used as a mark value for neural network training, which is referred to as a mark in this embodiment, such as 3-4 or several digits.

And S43, under the condition that the energy storage battery does not have an abnormal battery cell, obtaining a second time series initial sample of the energy storage battery, and determining a second time series sample corresponding to the second time series initial sample.

Specifically, in addition to the above-mentioned first time sequence initial sample of the battery pack set rack with the abnormal electric core, it is also necessary to record the working data that the battery pack set rack is a normal electric core, that is, the data of the normal electric core is not replaced by the abnormal electric core. In this embodiment, the time-series sample is referred to as a second time-series initial sample, and then the second time-series sample corresponding to the second time-series initial sample is determined, and the specific implementation process refers to the above corresponding description.

And S44, taking a preset normal electric core mark as the mark of the second time series sample.

Specifically, the preset normal cell identifier may be None, and the preset normal cell identifier None is used as the identifier of the second time series sample.

In this embodiment, the abnormal cell identification problem is abstracted into a classification problem, where the number of the classes is +1(1 is a cell without abnormality, corresponding to the None) of the total number of cells in the battery pack rack of the energy storage system.

S45, taking the first time series sample and the second time series sample as the time series samples of the energy storage battery, and taking the identification of the first time series sample and the identification of the second time series sample as the corresponding identifications of the time series samples.

The first time series of samples and the second time series of samples may be combined and used as time series of samples of an energy storage battery, and the identifier of the first time series of samples and the identifier of the second time series of samples may be combined and used as a corresponding identifier of the time series of samples. And then, storing the time sequence samples of the energy storage battery and the identifications corresponding to the time sequence samples into a database of abnormal electric core positions of the energy storage system.

And S32, training a preset battery cell detection model by using the training data until a preset training stop condition is met.

In this embodiment, the preset electric core detection model is a deep learning neural network model, and the specific structure refers to fig. 6. The neural network model includes an encoder, a fully connected network FC and Softmax. The encoder may be a recurrent neural network. In fig. 6, the number of encoders is 3, i.e., 3 time points (t) are acquired _-2 \t _-1 \t ₀ ) The data of each time point are respectively input into an encoder.

The time series data are input into an encoder of the deep learning neural network model, and a common encoder can be a time series encoder, and the specific structure is divided into three types:

1) based on the structure of the recurrent neural network. A Recurrent Neural Network (RNN) is a type of Recurrent Neural Network in which sequence data is input, recursion is performed in the direction of evolution of the sequence, and all nodes (Recurrent units) are connected in a chain. The recurrent neural network considers the information of the time dimension, transfers the parameters of the time dimension through one node after another, and retains the information important to the result. Common recurrent neural networks include Long Short Term Memory (LSTM), Gated Recurrent Units (GRU), etc., which can be used as encoders in embodiments of the present invention.

2) Neural networks based on one-dimensional convolution. Convolution operations in neural networks can identify very good features in the data. The single-layer convolution identifies a simple paradigm of the data, and the superposition of multiple layers of convolution can generate a simpler paradigm in a single layer into a more complex paradigm in a higher level layer. The one-dimensional convolution can be applied to data analysis of time series, and high-level features which are helpful to prediction results in data fragments are extracted to be used as semantic coding vectors of contexts.

3) Neural networks based on the self-attention mechanism. The autoflight mechanism is an improvement on the attention mechanism, which reduces reliance on external information and is more adept at capturing internal correlations of the input data. In the time series data processing, the self-attention mechanism mainly extracts features which are helpful for predicting results in data by calculating the cross-correlation coefficient of nodes at different moments. The neural network of the self-attention mechanism loses position information because the interrelation of unit nodes is calculated. Therefore, it is necessary to perform position coding on the input data at different times and add the position coded data to the original data.

In practical applications, taking the time-series data as only time-series initial data and the structure of the energy storage battery as an example in fig. 2, the input of the neural network is a series of time-series data, and at each time point t, the dimension of the data is 33, which contains 20 voltage values of v1-1, v1-2,. multidot.v 4-5, and 12 temperature values of 1 current value i, t1-1, t1-2,. multidot.t 4-3. L33-dimensional data vectors (20+1+12) are input to the time-series neural network model and processed as a classification problem for deep learning. The output of the neural network model is a 21-dimensional vector corresponding to the positions of 20 cells and None (i.e., no abnormal cell). The sum of the 21 digits of the output vector is 1. Each number of the output vector represents the probability that the cell at the position is an abnormal cell or the probability that the abnormal cell is None. If the system has no abnormal cell, the trained deep learning neural network model outputs a probability value of the 21 st class (i.e., None class) which is far greater than the probability values of cells at other positions. If an abnormal cell appears at a certain position in the system, the trained neural network model outputs a probability value at the position, which is far greater than probability values (including None) of cells at other positions.

More specifically, the input of the encoder is a 33-dimensional data vector (20+1+12) with a time sequence length L, and the output of the encoder is a vector (H in fig. 6), where the vector H is referred to as a semantic coding vector of a context, and represents an intrinsic pattern (hidden pattern) in past time sequence data, which is a result of neural network training and is also an important factor for calculating a current abnormal cell position. The vector H is passed through a fully connected network (FC in the figure, fullyconnected layer, a connection layer of a neural network model) to output a 21-dimensional vector. And performing Softmax operation on the 21-dimensional vector, namely mapping a real number domain output by the model to a vector of which the [0,1] represents probability distribution, namely the abnormal probability value of each battery cell. The vector is also 21-dimensional, and the probability that 20 cells in one battery pack rack are respectively abnormal cells and the probability that the energy storage battery has no abnormal cells are represented. The operational formula of Softmax is characterized as follows.

And taking the maximum value of the output vector as the probability value of the abnormal battery cell of the battery cell at the position. For example, the output 21-dimensional vector is

[0.01,0.01,0.01,0.00,0.01,

0.01,0.01,0.01,0.00,0.01,

0.01,0.01,0.01,0.87,0.01,

0.01,0.01,0.01,0.00,0.01,0.01]

The probability that the cell 3-4 is an abnormal cell is indicated to be 87%.

If the output 21-dimensional vector is:

[0.01,0.01,0.01,0.01,0.00,

0.03,0.00,0.01,0.01,0.01,

0.01,0.01,0.01,0.03,0.01,

0.01,0.02,0.01,0.01,0.02,0.76]

then it indicates that no cell is an abnormal cell, and this probability is 76%.

The deep learning neural network model of fig. 6 is a model structure constructed when time-series data is time-series initial data. If the time series data are the combination of the time series initial data and the characteristic value or the characteristic value, the internal structure of the deep learning neural network model can be adjusted according to the actual situation so as to be suitable for different inputs.

In addition, in a normal case, n battery pack assembly racks are included in one container, data of the n battery pack assembly racks are simultaneously acquired, and are simultaneously input into a deep learning neural network model (as a three-dimensional vector tenor) as n data points of one batch of the deep learning neural network model, that is, abnormal cell positions of the n battery pack assembly racks can be simultaneously predicted. If the battery cell detection equipment runs in an energy storage container, the abnormal battery cell positions of n battery pack collection racks in the container can be predicted; if the cell detection equipment runs at the cloud end or the station control end of the power station, data of m x n battery pack collection racks of all m containers of one energy storage power station can be collected simultaneously, and then the position of an abnormal cell of the m x n battery pack collection racks is predicted simultaneously by an algorithm. Therefore, the invention trains the neural deep learning network model according to the collected data of one battery pack assembly rack, and can predict abnormal cell positions in a plurality of battery pack assembly racks and even in the battery pack assembly racks of a plurality of groups of energy storage system containers on line.

After the deep learning neural network model is trained, the model can be deployed in an edge calculation module of the energy storage system to locate the abnormal cell online. And positioning of the abnormal battery cell can be carried out under the condition of ensuring the normal operation of the energy storage system. In addition, the system can be deployed in the container of the energy storage system to detect the abnormal electric core of one energy storage container, and also can be deployed at the control end of the energy storage power station to detect the abnormal electric core position of each energy storage container of the whole energy storage power station.

In this embodiment, the deep learning neural network model is obtained through training of a large amount of training data, and the precision and the accuracy of the deep learning neural network model obtained through training can be guaranteed, so that when the deep learning neural network model is used for detecting abnormal battery cells, higher accuracy is achieved.

In another implementation manner of the present invention, if an abnormal electrical core is detected by the neural network model, a maintenance worker may replace the abnormal electrical core with a normal electrical core. In order to ensure that the electric core inside the energy storage battery is normal after the electric core is replaced, the detection of the abnormal electric core can be further carried out by using a neural network model, and the repeated execution is carried out until all the electric cores are normal electric cores. Namely:

the abnormal cell detection result is not under the condition of predetermineeing abnormal cell detection result, still include:

and under the condition that the abnormal electric core corresponding to the abnormal electric core detection result is replaced by the normal electric core, returning to the step of acquiring the acquired time sequence initial data of the energy storage battery, and sequentially executing until the abnormal electric core detection result is the preset abnormal electric core detection result.

In this embodiment, for example, only one abnormal electrical core is set during training, and only one abnormal electrical core can be detected during each abnormal electrical core detection. When a plurality of abnormal cells occur in a battery pack at a certain moment or within a certain time period, the single classification result of the judgment given by the neural network is certain one of the plurality of abnormal cells. And the maintenance personnel can replace the battery cell at the corresponding position according to the judgment result of the neural network model. And when one battery cell is replaced, the neural network model performs single classification again, and the result is certain one of the remaining one or more abnormal battery cells. The method for eliminating one abnormal battery cell at a time is used for replacing the battery cells one by one, so that all the abnormal battery cells in one battery pack assembly rack can be eliminated or replaced, and the condition that a plurality of abnormal battery cells exist in one battery pack assembly rack can be met.

It should be noted that if at least one abnormal battery cell is set during training, all the abnormal battery cells can be directly detected through the neural network model, and all the abnormal battery cells are replaced. And after the replacement is finished, detecting again by using the neural network model so as to ensure that all the battery cells are normal battery cells.

In this embodiment, through the circulation that detects, changes, redetect unusual electric core, can guarantee that the electric core among the energy storage battery is normal electric core, guarantee energy storage system's security and reliability.

Optionally, on the basis of the embodiment of the battery cell detection method, another embodiment of the present invention provides a battery cell detection apparatus, and with reference to fig. 7, the battery cell detection apparatus may include:

the data acquisition module 11 is configured to acquire acquired time series initial data of the energy storage battery, and determine time series data corresponding to the time series initial data;

the model processing module 12 is configured to call a preset cell detection model to process the time series data, so as to obtain an abnormal probability value of each cell in the energy storage battery; the preset battery cell detection model is obtained based on training data; the training data comprises time sequence samples of the energy storage battery and identifications corresponding to the time sequence samples;

and a result determining module 13, configured to determine an abnormal cell detection result of the energy storage battery based on the abnormal probability values of the cells.

Further, when the data obtaining module 11 is configured to determine time-series data corresponding to the time-series initial data, the data obtaining module is specifically configured to:

taking the time-series initial data as time-series data;

or calculating a characteristic value of the time series initial data under a preset derivative characteristic according to a preset characteristic calculation rule, and using a combination of the time series initial data and the characteristic value or the characteristic value as time series data.

Further, the time-series initial data includes: cell voltage, cell current, and key point temperature.

Further, still include:

and the model updating module is used for updating the preset battery cell detection model.

Further, the preset battery cell detection model is a deep learning neural network model.

Further, still include the model generation module, the model generation module includes:

the first sample determining submodule is used for acquiring a first time sequence initial sample of the energy storage battery under the condition that at least one abnormal electric core exists in the energy storage battery, and determining a first time sequence sample corresponding to the first time sequence initial sample;

a first identifier determining submodule, configured to use position information of the at least one abnormal electrical core in the energy storage battery as an identifier of the first time series sample;

the second sample determining submodule is used for acquiring a second time series initial sample of the energy storage battery under the condition that no abnormal battery core exists in the energy storage battery, and determining a second time series sample corresponding to the second time series initial sample;

a second identifier determining submodule, configured to use a preset normal cell identifier as an identifier of the second time series sample;

the data determination submodule is used for taking the first time series sample and the second time series sample as time series samples of an energy storage battery, and taking the identification of the first time series sample and the identification of the second time series sample as corresponding identifications of the time series samples;

and the model training submodule is used for training a preset electric core detection model by using the time sequence sample of the energy storage battery and the identifier corresponding to the time sequence sample until a preset training stopping condition is met.

Further, the result determination module 13 includes:

the probability value calculation submodule is used for screening out the maximum abnormal probability value based on the abnormal probability values of all the battery cores and taking the maximum abnormal probability value as a target abnormal probability value;

a third identifier determining submodule, configured to determine a location identifier of the target abnormal probability value in the abnormal probability values of the battery cells;

the detection result determining submodule is used for determining that the abnormal cell detection result of the energy storage battery is a first identifier if the position identifier is a preset position identifier; the first identification represents that the energy storage battery has no abnormal battery cell; if the position identification is not a preset position identification, determining that the abnormal cell detection result of the energy storage battery is a second identification and the position identification; the second identification represents that the energy storage battery has an abnormal battery cell; and the position identification represents the position information of the detected abnormal electric core in the energy storage battery.

Further, the data acquisition module 11 is further configured to acquire the acquired time series initial data of the energy storage battery under the condition that the abnormal cell detection result is not the preset abnormal cell detection result and the abnormal cell corresponding to the abnormal cell detection result is replaced with the normal cell, and stop until the abnormal cell detection result is the preset abnormal cell detection result.

In the embodiment, the time series data are processed by using the preset electric core detection model, so that the abnormal probability value of each electric core is obtained, whether the electric core is abnormal or not and which electric core is positioned to be abnormal can be determined according to the abnormal probability value, the abnormal electric core can be accurately found and replaced and maintained in time, the safety and reliability of the energy storage system are further ensured, the model is obtained based on actual measurement data training, and the detection accuracy of the abnormal electric core can be improved.

It should be noted that, for the working processes of each module and each sub-module in this embodiment, please refer to the corresponding description in the above embodiments, which is not described herein again.

Optionally, on the basis of the embodiments of the battery cell detection method and apparatus, another embodiment of the present invention provides a battery cell detection device, including: a memory and a processor;

wherein the memory is used for storing programs;

and the processor calls a program and is used for executing the battery cell detection method.

Optionally, on the basis of the embodiment of the battery cell detection apparatus, another embodiment of the present invention provides a battery cell detection system, which includes the battery cell detection apparatus. The battery cell detection system can be deployed in an energy storage container or an energy storage power station and is used for detecting abnormal battery cells.

Referring to fig. 8, in another implementation manner of the present invention, the battery cell detection system further includes a data acquisition device;

the data acquisition equipment is used for acquiring time series initial data of the energy storage battery and outputting the time series initial data to the battery cell detection equipment.

The time series initial data includes: cell voltage, cell current, and critical point temperature.

In this embodiment, the data acquisition device may be a temperature sensor, an electrical signal detection device, or the like, so as to detect a voltage value of each electric core in the energy storage battery, a serial current value of the electric core of the energy storage battery, and a temperature value of a preset temperature detection position in the energy storage battery.

In another implementation manner of the present invention, the battery cell detection system further includes a cloud computing device;

the cloud computing equipment is used for receiving the time sequence data, training and updating a preset cell detection model at the cloud end based on the time sequence data, and periodically outputting the updated preset cell detection model at the cloud end to the cell detection equipment so as to update the preset cell detection model in the cell detection equipment.

The time-series data may be sent by the cell detection device.

Specifically, in an operating energy storage system, various electric signal detection devices and temperature sensors can acquire the voltage value of each battery cell, the current value of the battery cells connected in series, and the temperature values of several preset temperature detection positions of the battery pack assembly rack in real time. The time series initial data collected in real time are input into a data processing unit of the battery cell detection equipment, the collected time series initial data are processed to become time series data required by input of a neural network model, on one hand, the time series data are uploaded to cloud computing equipment (cloud end) through a data transmission module, on the other hand, the time series initial data are input into a trained neural network model in an edge computing module, namely, the position of an abnormal battery cell in a battery pack collection rack can be predicted in real time, and alarming and intelligent operation and maintenance are carried out. The data uploaded to the cloud computing device has two functions, namely, the data are stored and stored at the cloud end and used by other data-driven algorithms, and the neural network model is trained at the cloud end. As more and more data is collected over time, the neural network obtained by cloud training becomes more and more powerful and more accurate. And updating the neural network model in the edge calculation module through the data transmission module.

In the embodiment, the time series data are processed by using the preset electric core detection model, so that the abnormal probability value of each electric core is obtained, whether the electric core is abnormal or not and which electric core is positioned to be abnormal can be determined according to the abnormal probability value, the abnormal electric core can be accurately found and replaced and maintained in time, the safety and reliability of the energy storage system are further ensured, the model is obtained based on actual measurement data training, and the detection accuracy of the abnormal electric core can be improved.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (14)

1. A battery cell detection method is characterized by comprising the following steps:

acquiring time series initial data of the acquired energy storage battery, and determining time series data corresponding to the time series initial data;

calling a preset battery cell detection model to process the time sequence data to obtain the abnormal probability value of each battery cell in the energy storage battery; the preset battery cell detection model is obtained based on training data; the training data comprises time sequence samples of the energy storage battery and identifications corresponding to the time sequence samples;

and determining an abnormal cell detection result of the energy storage battery based on the abnormal probability values of the cells.

2. The cell detection method according to claim 1, wherein determining time-series data corresponding to the time-series initial data includes:

taking the time series initial data as time series data;

or calculating a characteristic value of the time series initial data under a preset derivative characteristic according to a preset characteristic calculation rule, and using a combination of the time series initial data and the characteristic value or the characteristic value as time series data.

3. The cell detection method of claim 1, wherein the time-series initial data includes: cell voltage, cell current, and critical point temperature.

4. The battery cell detection method according to claim 1, before invoking a preset battery cell detection model to process the time-series data and obtaining an abnormal probability value of each battery cell in the energy storage battery, further comprising:

and updating the preset battery cell detection model.

5. The cell detection method according to claim 1, wherein the preset cell detection model is a deep learning neural network model.

6. The cell detection method according to claim 1, wherein the training process of the preset cell detection model includes:

acquiring a first time sequence initial sample of the energy storage battery under the condition that at least one abnormal electric core exists in the energy storage battery, and determining a first time sequence sample corresponding to the first time sequence initial sample;

using the position information of the at least one abnormal electric core in the energy storage battery as the identifier of the first time sequence sample;

acquiring a second time series initial sample of the energy storage battery under the condition that no abnormal electric core exists in the energy storage battery, and determining a second time series sample corresponding to the second time series initial sample;

taking a preset normal electric core identifier as an identifier of the second time sequence sample;

taking the first time series sample and the second time series sample as time series samples of an energy storage battery, and taking the identifier of the first time series sample and the identifier of the second time series sample as identifiers corresponding to the time series samples;

and training a preset electric core detection model by using the time sequence sample of the energy storage battery and the identifier corresponding to the time sequence sample until a preset training stopping condition is met.

7. The battery cell detection method according to claim 1, wherein determining the abnormal cell detection result of the energy storage battery based on the abnormal probability values of the battery cells comprises:

screening out the maximum abnormal probability value based on the abnormal probability values of the battery cores, and taking the maximum abnormal probability value as a target abnormal probability value;

determining position marks of the target abnormal probability values in the abnormal probability values of the battery cores;

if the position identification is a preset position identification, determining that an abnormal electric core detection result of the energy storage battery is a first identification; the first identification represents that the energy storage battery has no abnormal battery core;

if the position identification is not a preset position identification, determining that the abnormal cell detection result of the energy storage battery is a second identification and the position identification; the second identification represents that the energy storage battery has an abnormal battery cell; and the position identification represents the position information of the detected abnormal electric core in the energy storage battery.

8. The battery cell detection method according to claim 1, wherein when the abnormal battery cell detection result is not a preset abnormal battery cell detection result, the method further comprises:

and under the condition that the abnormal electric core corresponding to the abnormal electric core detection result is replaced by the normal electric core, returning to the step of acquiring the acquired time sequence initial data of the energy storage battery, and sequentially executing until the abnormal electric core detection result is the preset abnormal electric core detection result.

9. The utility model provides a battery cell detection device which characterized in that includes:

the data acquisition module is used for acquiring the acquired time series initial data of the energy storage battery and determining time series data corresponding to the time series initial data;

the model processing module is used for calling a preset battery cell detection model to process the time sequence data to obtain the abnormal probability value of each battery cell in the energy storage battery; the preset battery cell detection model is obtained based on training data; the training data comprises time sequence samples of the energy storage battery and identifications corresponding to the time sequence samples;

and the result determining module is used for determining the abnormal cell detection result of the energy storage battery based on the abnormal probability values of the cells.

10. The utility model provides a battery cell detection equipment which characterized in that includes: a memory and a processor;

wherein the memory is used for storing programs;

the processor calls a program and is used for executing the cell detection method according to any one of claims 1 to 8.

11. A cell inspection system, characterized by comprising the cell inspection apparatus of claim 10.

12. The cell detection system of claim 11, further comprising a data acquisition device;

the data acquisition equipment is used for acquiring time series initial data of the energy storage battery and outputting the time series initial data to the battery cell detection equipment;

the time series initial data includes: cell voltage, cell current, and critical point temperature.

13. The cell detection system of claim 11, further comprising a cloud computing device;

the cloud computing equipment is used for receiving the time sequence data, training and updating a preset cell detection model at the cloud end based on the time sequence data, and periodically outputting the updated preset cell detection model at the cloud end to the cell detection equipment so as to update the preset cell detection model in the cell detection equipment.

14. The cell detection system according to any of claims 11 to 13, wherein the cell detection system is deployed in an energy storage container or an energy storage power station, and is configured to detect an abnormal cell.

Images