A method, device, equipment and system for detecting an electric cell

Abstract

The present invention provides a battery cell detection method, device, equipment and system, which uses a preset battery cell detection model to process time series data to obtain an abnormality probability value for each battery cell. Based on the abnormality probability value, it can be determined whether there is a battery cell abnormality and which battery cell has the abnormality, so that the abnormal battery cell can be accurately found and replaced and maintained in time, thereby ensuring the safety and reliability of the energy storage system. In addition, the model is trained based on measured data, which can improve the detection accuracy of abnormal battery cells.

Description

Technical Field

[0001] The present invention relates to the field of fault location, and more particularly to a battery cell detection method, device, equipment and system. Background Art

[0002] Energy storage batteries in energy storage systems are typically composed of multiple cells connected in series, parallel, or a combination of series and parallel. Due to the varying physical properties of each cell, even with a battery management system (BMS) in place, it's difficult for the BMS to ensure that each cell is operating in a normal charging and discharging state.

[0003] When the battery cell is in an abnormal state, such as internal short circuit, aging, cycle lithium loss, damage to positive electrode active materials, damage to negative electrode active materials, and short-term exhaustion, it may cause rapid aging of the battery cell and reduced battery capacity, thereby causing uncontrollable thermal runaway and safety accidents, reducing the safety and reliability of the energy storage system.

[0004] Therefore, there is an urgent need for a method that can detect abnormal cells in a timely manner when the cells are abnormal, so as to ensure the safety and reliability of the energy storage system. Summary of the Invention

[0005] In view of this, the present invention provides a battery cell detection method, device, equipment and system to solve the problem of urgently needing a method that can promptly monitor abnormal battery cells when the battery cells are abnormal.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] A battery cell detection method, comprising:

[0008] Acquire the collected time series initial data of the energy storage battery, and determine the time series data corresponding to the time series initial data;

[0009] Calling a preset cell detection model to process the time series data to obtain an abnormality probability value of each cell in the energy storage battery; the preset cell detection model is trained based on training data; the training data includes time series samples of energy storage batteries and identifiers corresponding to the time series samples;

[0010] Based on the abnormality probability values of the respective battery cells, a detection result of abnormal battery cells of the energy storage battery is determined.

[0011] Optionally, determining the time series data corresponding to the initial time series data includes:

[0012] Using the time series initial data as time series data;

[0013] Alternatively, according to a preset feature calculation rule, a feature value of the initial time series data under a preset derived feature is calculated, and a combination of the initial time series data and the feature value or the feature value is used as the time series data.

[0014] Optionally, the time series initial data includes: battery cell voltage, battery cell current and key point temperature.

[0015] Optionally, before calling a preset battery cell detection model to process the time series data to obtain an abnormality probability value of each battery cell in the energy storage battery, the method further includes:

[0016] Updated the preset battery cell detection model.

[0017] Optionally, the preset battery cell detection model is a deep learning neural network model.

[0018] Optionally, the training process of the preset battery cell detection model includes:

[0019] Acquire a first time series initial sample of the energy storage battery when at least one abnormal cell exists in the energy storage battery, and determine a first time series sample corresponding to the first time series initial sample;

[0020] Using the location information of the at least one abnormal battery cell in the energy storage battery as an identifier of the first time series sample;

[0021] Acquire a second time series initial sample of the energy storage battery when no abnormal battery cell exists in the energy storage battery, and determine a second time series sample corresponding to the second time series initial sample;

[0022] Using a preset normal cell identifier as an identifier of the second time series sample;

[0023] Using the first time series sample and the second time series sample as time series samples of the energy storage battery, and using 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;

[0024] The preset cell detection model is trained using the time series samples of the energy storage battery and the identifiers corresponding to the time series samples until a preset training stop condition is met.

[0025] Optionally, determining the abnormal cell detection result of the energy storage battery based on the abnormality probability value of each cell includes:

[0026] Based on the abnormal probability values of the respective battery cells, a maximum abnormal probability value is screened out and used as a target abnormal probability value;

[0027] Determine a position identifier of the target abnormality probability value in the abnormality probability values of each battery cell;

[0028] If the location identifier is a preset location identifier, determining that the abnormal cell detection result of the energy storage battery is a first identifier; the first identifier indicates that there is no abnormal cell in the energy storage battery;

[0029] If the location identifier is not a preset location identifier, the abnormal cell detection result of the energy storage battery is determined to be the second identifier and the location identifier; the second identifier indicates that there is an abnormal cell in the energy storage battery; the location identifier indicates the location information of the detected abnormal cell in the energy storage battery.

[0030] Optionally, when the abnormal battery cell detection result is not a preset abnormal battery cell detection result, the method further includes:

[0031] When the abnormal cell corresponding to the abnormal cell detection result has been replaced with a normal cell, the process returns to the step of obtaining the collected initial time series data of the energy storage battery and executes the step sequentially until the abnormal cell detection result is a preset abnormal cell detection result.

[0032] A battery cell detection device, comprising:

[0033] A data acquisition module is used to acquire the collected time series initial data of the energy storage battery and determine the time series data corresponding to the time series initial data;

[0034] a model processing module, configured to call a preset cell detection model to process the time series data to obtain an abnormality probability value for each cell in the energy storage battery; the preset cell detection model is trained based on training data; the training data includes time series samples of the energy storage battery and identifiers corresponding to the time series samples;

[0035] The result determination module is used to determine the abnormal cell detection result of the energy storage battery based on the abnormal probability value of each cell.

[0036] A battery cell detection device, comprising: a memory and a processor;

[0037] Wherein, the memory is used to store programs;

[0038] The processor calls the program and is used to execute the above-mentioned battery cell detection method.

[0039] A battery cell detection system includes the above-mentioned battery cell detection device.

[0040] Optionally, it also includes a data acquisition device;

[0041] The data acquisition device is used to collect the time series initial data of the energy storage battery and output the time series initial data to the battery cell detection device;

[0042] The time series initial data includes: battery cell voltage, battery cell current and key point temperature.

[0043] Optionally, cloud computing equipment is also included;

[0044] The cloud computing device is used to receive time series data, train and update the preset battery cell detection model in the cloud based on the time series data, and regularly output the updated preset battery cell detection model in the cloud to the battery cell detection device to update the preset battery cell detection model in the battery cell detection device.

[0045] Optionally, the battery cell detection system is deployed in an energy storage container or an energy storage power station to detect abnormal battery cells.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention provides a battery cell detection method, device, equipment and system, which uses a preset battery cell detection model to process time series data to obtain an abnormality probability value for each battery cell. Based on the abnormality probability value, it can be determined whether there is a battery cell abnormality and which battery cell has the abnormality, so that the abnormal battery cell can be accurately found and replaced and maintained in time, thereby ensuring the safety and reliability of the energy storage system. In addition, the model is trained based on measured data, which can improve the detection accuracy of abnormal battery cells. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0049] Figure 1 A flow chart of a battery cell detection method provided by an embodiment of the present invention;

[0050] Figure 2 A structural diagram of a battery pack assembly provided by an embodiment of the present invention;

[0051] Figure 3 A flow chart of another battery cell detection method provided by an embodiment of the present invention;

[0052] Figure 4 A flow chart of another battery cell detection method provided by an embodiment of the present invention;

[0053] Figure 5 A flow chart of another battery cell detection method provided by an embodiment of the present invention;

[0054] Figure 6 A structural diagram of a preset battery cell detection model provided in an embodiment of the present invention;

[0055] Figure 7 A schematic structural diagram of a battery cell detection device provided by an embodiment of the present invention;

[0056] Figure 8 A schematic structural diagram of a battery cell detection device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0058] The application scope of chemical batteries such as lithium-ion, sodium-ion, and lithium iron phosphate is becoming increasingly wide. They are mostly used in energy storage systems such as hydropower, thermal power, wind power and solar power stations, as well as power tools, electric bicycles, electric vehicles, military equipment, aerospace and other fields.

[0059] Energy storage systems typically consist of multiple cells connected in series, parallel, or a combination of these. Due to the varying physical properties of each cell, while the battery management system (BMS) ensures the entire system operates within normal parameters and provides protection against anomalies, such as power outages due to overheating, it is difficult for the BMS to guarantee that each cell is operating normally, i.e., charging and discharging properly.

[0060] When a battery cell is not operating in a normal state, it may experience internal short circuits, aging, loss of cyclic lithium, damage to the positive active material, damage to the negative active material, short-term exhaustion and other problems. This can cause the battery cell to be in an abnormal state and prone to overcharge, overdischarge, overheating, etc. Long-term operation of the battery cell in this abnormal state will lead to rapid aging of the battery cell, resulting in reduced battery capacity and causing internal short circuits. This may cause the lithium-ion battery to self-discharge, capacity decay, and local thermal runaway, leading to uncontrollable thermal runaway and safety accidents.

[0061] To address this issue, we can extract the terminal voltage characteristics of the energy storage battery to be tested and calculate the Euclidean distance with all samples in the dataset to determine the health of the tested sample. Although this method is simple in principle, it cannot cope with the highly variable state of energy storage systems. It also requires offline collection of large amounts of data, making it difficult to implement online applications.

[0062] To locate abnormal chips online, the inventors discovered that deep learning technology can be used to perform real-time monitoring of the energy storage system's time series data to calculate the abnormal cell's location within the entire system. This entire process ensures that the detection is completed while the energy storage system is operating, without disrupting system operations. Furthermore, the detection requires minimal computational effort, allowing for edge deployment and ensuring local data security.

[0063] Specifically, the present invention provides a battery cell detection method, device, equipment and system, which uses a preset battery cell detection model to process time series data to obtain an abnormality probability value for each battery cell. According to the abnormality probability value, it can be determined whether there is a battery cell abnormality and which battery cell has the abnormality, so that the abnormal battery cell can be accurately found and replaced and maintained in time, thereby ensuring the safety and reliability of the energy storage system. In addition, the model is trained based on measured data, which can improve the detection accuracy of abnormal battery cells.

[0064] Based on the above content, an embodiment of the present invention provides a cell detection method, which is applied to a cell detection device. The cell detection device can be deployed at the edge, which can reduce the computational workload of the main controller of the energy storage system and ensure data security. Figure 1 , the battery cell detection methods may include:

[0065] S11. Acquire the collected time series initial data of the energy storage battery, and determine the time series data corresponding to the time series initial data.

[0066] The time series initial data includes: battery cell voltage, battery cell current and key point temperature.

[0067] In practical applications, when all cells in the energy storage battery are connected in series, the cell current is the series current, and the key point temperatures can be: the cell tab temperature, the surface temperature of each cell, etc.

[0068] Specifically, a typical arrangement of the energy storage system is as follows: a container of an energy storage system contains 4 battery racks, each battery rack contains approximately 36 battery packs, and each battery pack contains 12 battery cells. The battery cells in each pack are connected in series, and the battery packs in the rack are also connected in series. Therefore, for each battery rack, all the battery cells are connected in series. There are a total of about 432 battery cells, sharing one series current. The BMS and temperature control of the energy storage system will record the voltage of each battery cell, the series current, and the temperature of several key positions of each battery pack (such as the battery cell tabs, the various surfaces of the battery cells, etc.) through various electrical signals and temperature sensors, and use the detected temperature as the key point temperature. In this embodiment, the battery rack is simplified for ease of display. Figure 2 The figure shows a battery rack consisting of four packs, each containing five cells. All cells are connected in series, and each pack has three temperature measurement points: t1-1, t1-2, ..., t4-3.

[0069] When collecting the initial time series data of the energy storage battery, the time series length L (which can be based on the current moment and within the historical L time period, the length L of the time series is an important adjustable parameter that needs to be determined during the training of the neural network. The length L needs to be adjusted to obtain high accuracy of the algorithm, or to balance computing resources and algorithm accuracy) will be collected. The voltage of each cell in the energy storage battery (in terms of Figure 2 For example, there are 20 values, v1-1, v1-2, ..., v4-5), the series current of the battery cells of the energy storage battery (1 value, i), and the key temperature values of all key locations of the battery pack set rack in the energy storage battery (12 values, t1-1, t1-2, ..., t4-3). In this embodiment, the voltage value of each battery cell in the energy storage battery, the series current value of the battery cells of the energy storage battery, and the key temperature values of the key locations in the energy storage battery collected above are used as the time series initial data of the energy storage battery.

[0070] After obtaining the initial time series data of the energy storage battery, it is necessary to determine the time series data corresponding to the initial time series data.

[0071] Specifically, in this embodiment, there are multiple ways to determine the time series data corresponding to the initial time series data.

[0072] 1. The first method:

[0073] The time series initial data is used as time series data.

[0074] Specifically, after the initial time series data is collected, the initial time series data is used as the time series data.

[0075] 2. The second method:

[0076] According to the preset feature calculation rules, the feature value of the initial time series data under the preset derived feature is calculated, and the combination of the initial time series data and the feature value or the feature value is used as the time series data.

[0077] In this embodiment, after the initial time series data is obtained, the preset derivative features of the initial time series data are calculated.

[0078] Specifically, the preset derived features may be time series features, which may include:

[0079] 1. The rate of change of voltage (or current) with time.

[0080] 2. The rate of change of temperature over time.

[0081] 3. dQ / dV and dV / dQ, dQ / dV is the rate of change of charge with charging voltage, and dV / dQ is the rate of change of voltage with charging charge.

[0082] 4. The discrete ratio of the above characteristics (1, 2, and 3) of each cell to every other cell in the same battery pack rack.

[0083] After obtaining the characteristic value of the preset derived feature, the combination of the time series initial data and the characteristic value of the preset derived feature can be used as the time series data, or the characteristic value under the preset derived feature can be directly used as the time series data.

[0084] The specific implementation method selected to determine time series data can be set based on the actual usage scenario.

[0085] S12: Calling a preset battery cell detection model to process the time series data to obtain an abnormality probability value of each battery cell in the energy storage battery.

[0086] In this embodiment, the preset cell detection model may be pre-built into the cell detection device, or may be updated before use, i.e., before executing step S12. Specifically, the updating method may be to receive a regularly updated preset cell detection model sent by an external device.

[0087] The external device can be a cloud computing device, such as a cloud platform. In this embodiment, the cloud computing device can train, generate and update the preset battery cell detection model. For example, the cloud computing device receives time series data, trains and updates the preset battery cell detection model in the cloud based on the time series data, and regularly outputs the updated preset battery cell detection model in the cloud to the battery cell detection device to update the preset battery cell detection model in the battery cell detection device.

[0088] In addition, the battery cell detection equipment can also train, generate and update the preset battery cell detection model by itself.

[0089] In practical applications, the preset cell detection model is trained based on training data, which includes time series samples of energy storage batteries and identifiers corresponding to the time series samples. The preset cell detection model trained with the training data can detect the abnormal probability value of each cell in the energy storage battery.

[0090] In this embodiment, the preset battery cell detection model can be a deep learning neural network model, and the encoder part of the deep learning neural network model can be: a recurrent neural network, a one-dimensional convolutional neural network, a self-attention mechanism neural network, etc.

[0091] Specifically, the deep learning neural network model can be a time series model. The present 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 over a period of time (continuous and cannot change suddenly), or the internal short circuit state of the battery does not occur suddenly at a certain moment.

[0092] The data output by the preset battery cell detection model is the abnormality probability value of each battery cell in the energy storage battery.

[0093] Taking 20 battery cells as an example, there are a total of 21 abnormal probability values for each battery cell in the output energy storage battery. The first 20 abnormal probability values are the abnormal probability values corresponding to the 20 battery cells (arranged in order of the battery cells), and the last abnormal probability value is the probability value that all battery cells are normal.

[0094] It can be:

[0095] [0.01,0.01,0.01,0.00,0.01,

[0096] 0.01,0.01,0.01,0.00,0.01,

[0097] 0.01,0.01,0.01,0.87,0.01,

[0098] 0.01,0.01,0.01,0.00,0.01,0.01].

[0099] Or,

[0100] [0.01,0.01,0.01,0.01,0.00,

[0101] 0.03,0.00,0.01,0.01,0.01,

[0102] 0.01,0.01,0.01,0.03,0.01,

[0103] 0.01,0.02,0.01,0.01,0.02,0.76].

[0104] S13. Determine a detection result of abnormal cells of the energy storage battery based on the abnormal probability value of each cell.

[0105] Specifically, after determining the abnormal probability value of each battery cell, the abnormal battery cells can be screened out according to the probability value, or all battery cells can be determined to be normal.

[0106] Specifically, refer to Figure 3 , step S13 may include:

[0107] S21. Based on the abnormality probability values of the respective battery cells, a maximum abnormality probability value is screened out and used as a target abnormality probability value.

[0108] Specifically, in this embodiment, the maximum abnormal probability value is selected from the abnormal probability values of each battery cell output by the preset battery cell detection model. Still using the aforementioned 20 battery cells as an example, if the abnormal probability value is within the top 20, then the battery cell with the maximum abnormal probability value is an abnormal battery cell.

[0109] If the abnormal probability value is at the 21st position, since the 21st position indicates that all battery cells are normal, it means that all battery cells are not abnormal batteries.

[0110] In practical applications, in addition to using the maximum abnormal probability value as a screening criterion, the maximum abnormal probability value and the difference between the maximum abnormal probability value and other abnormal probability values greater than a preset threshold value can also be used as a screening criterion.

[0111] S22: Determine the position identifier of the target abnormality probability value in the abnormality probability values of the respective battery cells.

[0112] Specifically, since the abnormal probability values are arranged in the order of the battery cells, the position identifier of the target abnormal probability value in the abnormal probability values of each battery cell can be determined in order. In this embodiment, the position identifier can be the 1st, 2nd, 3rd...21st.

[0113] It should be noted that since there are 20 battery cells, if it is not the first 20, it is considered to be the 21st.

[0114] S23. Determine whether the location identifier is a preset location identifier; if so, execute step S24; if not, execute step S25.

[0115] Specifically, since the 21st abnormal probability value is special, indicating that all battery cells are normal, and the first 20 digits indicate that a specific battery cell is abnormal, the preset position can be identified as the 21st position. That is, in this embodiment, it is determined whether the position identifier is the 21st position.

[0116] S24: Determine that the abnormal cell detection result of the energy storage battery is a first identification.

[0117] The first identifier indicates that there are no abnormal cells in the energy storage battery, that is, if it is the 21st bit, it means that all cells are normal.

[0118] For example, the abnormal probability value of each battery cell is:

[0119] [0.01,0.01,0.01,0.01,0.00,

[0120] 0.03,0.00,0.01,0.01,0.01,

[0121] 0.01,0.01,0.01,0.03,0.01,

[0122] 0.01,0.02,0.01,0.01,0.02,0.76].

[0123] Since 0.76 is located at the 21st position and is much larger than the abnormal probability values of other positions, that is, the difference with the other abnormal probability values is greater than the preset threshold (such as 0.5), it means that all battery cells are normal.

[0124] S25: Determine that the abnormal cell detection result of the energy storage battery is the second identifier and the position identifier.

[0125] The second identifier indicates that there is an abnormal cell in the energy storage battery, and the location identifier indicates the location information of the detected abnormal cell in the energy storage battery. The abnormal cell can be determined based on the location identifier, and the specific location of the abnormal cell can be found.

[0126] like:

[0127] [0.01,0.01,0.01,0.00,0.01,

[0128] 0.01,0.01,0.01,0.00,0.01,

[0129] 0.01,0.01,0.01,0.87,0.01,

[0130] 0.01,0.01,0.01,0.00,0.01,0.01].

[0131] Since 0.87 is located at the 14th position and is much larger than the abnormal probability values of other positions, that is, the difference with the other abnormal probability values is greater than the preset threshold (such as 0.5), it means that the 14th cell is an abnormal cell. The 14th position corresponds to the 3-4 cell, which means that the 3-4 cell is an abnormal cell. The specific abnormalities may be internal short circuit, aging, cycle lithium loss, positive electrode active material damage, negative electrode active material damage, short-term exhaustion, etc.

[0132] That is to say, the abnormal cell detection result is the second identifier and the position identifier (such as the 14th position, or 3-4) indicating that the energy storage battery has an abnormal cell.

[0133] In a case where the abnormal cell detection result is not a preset abnormal cell detection result, the abnormal cell detection result is output.

[0134] Specifically, in this embodiment, the preset abnormal battery cell detection result is the first indicator, and the preset abnormal battery cell detection result indicates that all battery cells are normal.

[0135] If the abnormal cell detection result is not the preset abnormal cell detection result, it means that there is at least one abnormal cell. At this time, the abnormal cell detection result will be output to the maintenance personnel's mobile terminal so that the corresponding maintenance personnel can replace the abnormal cell.

[0136] If the abnormal cell detection result is the preset abnormal cell detection result, it means that all cells are normal. At this time, the preset abnormal cell detection result can also be output so that maintenance personnel can understand the cell status in time.

[0137] In this embodiment, a preset battery cell detection model is used to process time series data to obtain an abnormality probability value for each battery cell. Based on the abnormality probability value, it can be determined whether there is a battery cell abnormality and which battery cell has the abnormality. Therefore, the abnormal battery cell can be accurately found and replaced and maintained in time, thereby ensuring the safety and reliability of the energy storage system. In addition, the model is trained based on measured data, which can improve the detection accuracy of abnormal battery cells.

[0138] The above embodiment mentions a preset cell detection model. If the cell detection device trains, generates and updates the preset cell detection model by itself, then refer to Figure 4 The training process of the preset battery cell detection model includes:

[0139] S31. Determine training data.

[0140] The training data includes time series samples of energy storage batteries and identifiers corresponding to the time series samples.

[0141] Specifically, refer to Figure 5 , determining the training data may include:

[0142] S41. Acquire a first time series initial sample of the energy storage battery when at least one abnormal cell exists in the energy storage battery, and determine a first time series sample corresponding to the first time series initial sample.

[0143] Specifically, taking the abnormality of internal short circuit as an example, when collecting time series samples of the energy storage battery, at least one cell with internal short circuit is determined (in this embodiment, one abnormal cell is taken as an example, such as Figure 2 The battery pack is set to a normal cell in the rack, and the position of the abnormal cell (such as 3-4, or the number) is recorded, and the continuous charge and discharge data is recorded, including the voltage of each cell in the time series length L (in Figure 2 For example, there are 20 values, v1-1, v1-2, ..., v4-5), the series current of the battery cells in the energy storage battery (1 value, i), and the temperature values of all key positions of the battery pack set rack (12 values, t1-1, t1-2, ..., t4-3).

[0144] In this embodiment, the collected data is referred to as the first time series initial sample.

[0145] After that, the position of the abnormal cell can be replaced and the first time series initial sample can be collected again. Please refer to the above description for the process of determining the first time series sample corresponding to the first time series initial sample.

[0146] Specifically, using an abnormal cell as an example, after recording a certain amount of data (time series length L) at one location, the abnormal cell's location in the battery pack rack is randomly changed and the data continues to be recorded. Theoretically, it's not necessary to record the abnormal cell at every location in the battery pack rack, but the more locations recorded, the better, the larger the data volume, and the more distributed the data points, the better (to ensure that the abnormal cell appears at every location in the battery pack rack).

[0147] The process of collecting the initial samples of the first time series with two or more abnormal cells is similar to that of the first abnormal cell. Please refer to the corresponding instructions above.

[0148] S42: Use the location information of the at least one abnormal battery cell in the energy storage battery as an identifier of the first time series sample.

[0149] In this embodiment, the position of the abnormal cell is used as a label value for neural network training, which is called an identifier in this embodiment, such as 3-4 or the number.

[0150] S43: Acquire a second time series initial sample of the energy storage battery when no abnormal battery cell exists in the energy storage battery, and determine a second time series sample corresponding to the second time series initial sample.

[0151] Specifically, in addition to collecting the first time series initial samples for the battery pack set rack containing abnormal cells, it is also necessary to record the operating data of the battery pack set rack containing all normal cells, that is, not replacing normal cells with abnormal cells. In this embodiment, this is referred to as the second time series initial samples. A second time series sample corresponding to the second time series initial samples is then determined. The specific implementation process is described above.

[0152] S44: Use the preset normal cell identifier as the identifier of the second time series sample.

[0153] 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.

[0154] In this embodiment, the abnormal cell identification problem is abstracted into a classification problem, and the number of categories is the total number of cells in the rack of the energy storage system battery pack + 1 (1 represents no abnormal cell, corresponding to None above).

[0155] S45: Use the first time series sample and the second time series sample as time series samples of the energy storage battery, and use 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.

[0156] The first time series sample and the second time series sample can be combined to form a time series sample of the energy storage battery, and the identifier of the first time series sample and the identifier of the second time series sample can be combined to form an identifier corresponding to the time series sample. The time series sample of the energy storage battery and the identifier corresponding to the time series sample are then stored in a database of abnormal battery cell locations in the energy storage system.

[0157] S32: Use the training data to train a preset cell detection model until a preset training stop condition is met.

[0158] In this embodiment, the preset cell detection model is a deep learning neural network model. Figure 6The neural network model includes an encoder, a fully connected network FC, and a softmax. The encoder can be a recurrent neural network. Figure 6 In the example, the number of encoders is 3, which means that 3 time points (t -2 \t -1 The data at each time point is input into an encoder respectively.

[0159] Time series data is input into the encoder of the deep learning neural network model. A common encoder can be a time series encoder, which has three specific structures:

[0160] 1) Structure based on recurrent neural network. Recurrent Neural Network (RNN) is a type of recurrent neural network that takes sequence data as input, recursively in the direction of sequence evolution, and all nodes (recurrent units) are connected in a chain. Recurrent neural networks take into account information in the time dimension, pass parameters in the time dimension through each node, and retain information that is important to the result. Common recurrent neural networks include long short-term memory networks (LSTM), gated recurrent units (GRU), etc., which can all be used as encoders in the embodiments of the present invention.

[0161] 2) Neural networks based on one-dimensional convolution. The convolution operation in neural networks is highly effective at identifying effective features in data. A single convolution layer identifies simple patterns in the data, while stacking multiple convolution layers can transform these simple patterns into more complex patterns in higher-level layers. One-dimensional convolution can be applied to time series data analysis, extracting high-level features from data segments that are helpful for prediction and serving as semantic encoding vectors for context.

[0162] 3) Self-attention-based neural networks. The self-attention mechanism is an improvement on the attention mechanism, reducing its reliance on external information and better capturing the internal correlations of input data. In time series data processing, the self-attention mechanism primarily calculates the interrelationships between nodes at different times to extract features from the data that are helpful for prediction. Because self-attention-based neural networks calculate the interrelationships between unit nodes, they lose positional information. Therefore, it is necessary to positionally encode the input data at different times before adding it to the original data.

[0163] In practical applications, the time series data is only the initial time series data, and the structure of the energy storage battery is Figure 2For example, the neural network input is a series of time series data. At each time point t, the data has a dimension of 33, including 20 voltage values (v1-1, v1-2, ..., v4-5), 1 current value (i), and 12 temperature values (t1-1, t1-2, ..., t4-3). L 33-dimensional data vectors (20+1+12) are input into the time series neural network model as a deep learning classification problem. The neural network model outputs a 21-dimensional vector, corresponding to the 20 battery cell locations and None (indicates no abnormal cell). The sum of the 21 numbers in the output vector is 1. Each number in the output vector represents the probability that the battery cell at that location is an abnormal cell, or the probability that the abnormal cell is None. If there are no abnormal cells in the system, the trained deep learning neural network model should output a significantly higher probability value for the 21st category (None) than for cells at other locations. If an abnormal cell appears at a certain location in the system, the probability value output by the trained neural network model at this location should be much greater than the probability values of cells at other locations (including None).

[0164] More specifically, the input of the encoder is a 33-dimensional data vector (20+1+12) of time series length L, and the output of the encoder is a vector ( Figure 6 This vector H is called the context semantic encoding vector. It represents the inherent patterns (hidden patterns) in past time series data. It is the result of neural network training and an important factor in calculating the current location of abnormal cells. Vector H is passed through a fully connected network (FC, a connection layer in a neural network model) to output a 21-dimensional vector. This 21-dimensional vector is then processed through a Softmax operation, which maps the real number domain output by the model to a vector representing a probability distribution in the range [0, 1], i.e., the abnormal probability value of each cell. This vector is also 21-dimensional and represents the probability of each of the 20 cells in a battery rack being an abnormal cell and the probability of the energy storage battery having no abnormal cells. The Softmax operation formula is as follows.

[0165]

[0166] The maximum value of the output vector is taken as the probability value of the cell at that location being an abnormal cell. For example, the output 21-dimensional vector is

[0167] [0.01,0.01,0.01,0.00,0.01,

[0168] 0.01,0.01,0.01,0.00,0.01,

[0169] 0.01,0.01,0.01,0.87,0.01,

[0170] 0.01,0.01,0.01,0.00,0.01,0.01]

[0171] This indicates that the probability that cell 3-4 is an abnormal cell is 87%.

[0172] If the output 21-dimensional vector is:

[0173] [0.01,0.01,0.01,0.01,0.00,

[0174] 0.03,0.00,0.01,0.01,0.01,

[0175] 0.01,0.01,0.01,0.03,0.01,

[0176] 0.01,0.02,0.01,0.01,0.02,0.76]

[0177] This indicates that no battery cell is an abnormal battery cell, and the probability is 76%.

[0178] It should be noted that Figure 6 The deep learning neural network model is based on a model structure built when the time series data is the initial time series data. If the time series data is a combination of the initial time series data and the eigenvalues, or the eigenvalues, the internal structure of the deep learning neural network model can be adjusted based on actual conditions to accommodate different inputs.

[0179] In addition, it should be noted that, under normal circumstances, a container contains n battery pack racks. The data of the n battery pack racks is collected simultaneously and used as n data points of a batch (batch processing) of the deep learning neural network model. (As a three-dimensional vector tensor), the deep learning neural network model is simultaneously input, that is, the abnormal cell locations of the n battery pack racks can be predicted simultaneously. If the battery cell detection equipment is running in an energy storage container, the abnormal cell locations of the n battery pack racks in this container can be predicted; if the battery cell detection equipment is running in the cloud or the station control end of the power station, the data of the m*n battery pack racks of all m containers of an energy storage power station can be collected simultaneously, and then the algorithm can simultaneously predict the locations of abnormal cells in the m*n battery pack racks. Therefore, the present invention trains a deep learning neural network model based on the collected data of a battery pack rack, and can online predict the abnormal cell locations in multiple battery pack racks or even battery pack racks of multiple energy storage system containers.

[0180] Once the deep learning neural network model is trained, it can be deployed in the energy storage system's edge computing module to locate abnormal cells online. This allows for the location of abnormal cells while ensuring the normal operation of the energy storage system. Furthermore, the model can be deployed inside the energy storage system's container to detect abnormal cells within a single container, or at the control end of the energy storage plant to detect the location of abnormal cells in every container throughout the plant.

[0181] In this embodiment, the deep learning neural network model is trained through a large amount of training data, which can ensure the precision and accuracy of the trained deep learning neural network model, and then when the deep learning neural network model is used to detect abnormal battery cells, it has higher accuracy.

[0182] In another implementation of the present invention, if an abnormal cell is detected by the above-mentioned neural network model, maintenance personnel can replace the abnormal cell with a normal cell. In order to ensure that the cells inside the energy storage battery are normal after the cell replacement, the neural network model can be further used to detect abnormal cells, and the process can be repeated until all cells are normal. That is:

[0183] In a case where the abnormal battery cell detection result is not a preset abnormal battery cell detection result, the method further includes:

[0184] When the abnormal cell corresponding to the abnormal cell detection result has been replaced with a normal cell, the process returns to the step of obtaining the collected initial time series data of the energy storage battery and executes the step sequentially until the abnormal cell detection result is a preset abnormal cell detection result.

[0185] In this embodiment, taking the example of setting only one abnormal cell during training, only one abnormal cell can be detected each time an abnormal cell detection is performed. When multiple abnormal cells appear simultaneously in a battery pack at a certain moment or within a certain time period, the single classification result given by the neural network must be one of these multiple abnormal cells. Maintenance personnel can replace the cell at the corresponding position based on the judgment result of the neural network model. When a cell is replaced, the neural network model performs a single classification again, and the result must be one of the remaining one or more abnormal cells. That is, by eliminating one abnormal cell at a time and replacing the cells one by one, all abnormal cells in a battery pack set rack can be eliminated or replaced, and the situation where there are multiple abnormal cells in a battery pack set rack can be handled.

[0186] It should be noted that if at least one abnormal cell is set during training, all abnormal cells can be directly detected through the neural network model and replaced. After the replacement is completed, the neural network model is used to re-test to ensure that all cells are normal.

[0187] In this embodiment, by performing a cycle of testing, replacing, and retesting abnormal cells, it is possible to ensure that all cells in the energy storage battery are normal cells, thereby ensuring the safety and reliability of the energy storage system.

[0188] Optionally, based on the embodiment of the above-mentioned battery cell detection method, another embodiment of the present invention provides a battery cell detection device, referring to Figure 7 , which may include:

[0189] The data acquisition module 11 is used to acquire the collected time series initial data of the energy storage battery and determine the time series data corresponding to the time series initial data;

[0190] a model processing module 12, configured to call a preset cell detection model to process the time series data to obtain an abnormality probability value for each cell in the energy storage battery; the preset cell detection model is trained based on training data; the training data includes time series samples of the energy storage battery and identifiers corresponding to the time series samples;

[0191] The result determination module 13 is configured to determine the abnormal cell detection result of the energy storage battery based on the abnormal probability value of each cell.

[0192] Furthermore, when the data acquisition module 11 is used to determine the time series data corresponding to the initial time series data, it is specifically used to:

[0193] Using the time series initial data as time series data;

[0194] Alternatively, according to a preset feature calculation rule, a feature value of the initial time series data under a preset derived feature is calculated, and a combination of the initial time series data and the feature value or the feature value is used as the time series data.

[0195] Furthermore, the time series initial data includes: battery cell voltage, battery cell current and key point temperature.

[0196] Furthermore, it also includes:

[0197] The model update module is used to update the preset battery cell detection model.

[0198] Furthermore, the preset battery cell detection model is a deep learning neural network model.

[0199] Furthermore, a model generation module is included, which includes:

[0200] a first sample determination submodule, configured to obtain a first time series initial sample of the energy storage battery when at least one abnormal cell exists in the energy storage battery, and determine a first time series sample corresponding to the first time series initial sample;

[0201] a first identification determination submodule, configured to use the location information of the at least one abnormal battery cell in the energy storage battery as an identification of the first time series sample;

[0202] A second sample determination submodule is configured to obtain a second time series initial sample of the energy storage battery when no abnormal cell exists in the energy storage battery, and determine a second time series sample corresponding to the second time series initial sample;

[0203] A second identification determination submodule, configured to use a preset normal cell identification as an identification of the second time series sample;

[0204] a data determination submodule, configured to use the first time series sample and the second time series sample as time series samples of the energy storage battery, and use 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;

[0205] The model training submodule is used to train a preset cell detection model using the time series samples of the energy storage battery and the identifiers corresponding to the time series samples until a preset training stop condition is met.

[0206] Furthermore, the result determination module 13 includes:

[0207] A probability value calculation submodule is used to screen out the maximum abnormal probability value based on the abnormal probability values of the respective battery cells and use it as the target abnormal probability value;

[0208] A third identifier determination submodule is used to determine a position identifier of the target abnormality probability value in the abnormality probability values of each battery cell;

[0209] The detection result determination submodule is used to determine, if the location identifier is a preset location identifier, that the abnormal cell detection result of the energy storage battery is a first identifier; the first identifier indicates that there are no abnormal cells in the energy storage battery; if the location identifier is not a preset location identifier, determine that the abnormal cell detection result of the energy storage battery is a second identifier and the location identifier; the second identifier indicates that there are abnormal cells in the energy storage battery; the location identifier indicates the location information of the detected abnormal cell in the energy storage battery.

[0210] Furthermore, the data acquisition module 11 is also used to obtain the collected time series initial data of the energy storage battery when the abnormal cell detection result is not a preset abnormal cell detection result and the abnormal cell corresponding to the abnormal cell detection result has been replaced with a normal cell, and stop when the abnormal cell detection result is a preset abnormal cell detection result.

[0211] In this embodiment, a preset battery cell detection model is used to process time series data to obtain an abnormality probability value for each battery cell. Based on the abnormality probability value, it can be determined whether there is a battery cell abnormality and which battery cell has the abnormality. Therefore, the abnormal battery cell can be accurately found and replaced and maintained in time, thereby ensuring the safety and reliability of the energy storage system. In addition, the model is trained based on measured data, which can improve the detection accuracy of abnormal battery cells.

[0212] It should be noted that, for the working process of each module and sub-module in this embodiment, please refer to the corresponding description in the above embodiment, which will not be repeated here.

[0213] Optionally, based on the above 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;

[0214] Wherein, the memory is used to store programs;

[0215] The processor calls the program and is used to execute the above-mentioned battery cell detection method.

[0216] Alternatively, based on the above embodiment of the battery cell detection device, another embodiment of the present invention provides a battery cell detection system, including the above battery cell detection device. The battery cell detection system can be deployed in an energy storage container or an energy storage power station to detect abnormal battery cells.

[0217] Reference Figure 8 ,In another implementation of the present invention, the battery cell detection system further includes a data acquisition device;

[0218] The data acquisition device is used to collect time series initial data of the energy storage battery and output the time series initial data to the battery cell detection device.

[0219] The time series initial data includes: battery cell voltage, battery cell current and key point temperature.

[0220] In this embodiment, the data acquisition device can be a temperature sensor, an electrical signal detection device, etc., to detect the voltage value of each battery cell in the energy storage battery, the series current value of the battery cells of the energy storage battery, and the temperature value of a preset temperature detection position in the energy storage battery.

[0221] In another implementation of the present invention, the battery cell detection system further includes a cloud computing device;

[0222] The cloud computing device is used to receive time series data, train and update the preset battery cell detection model in the cloud based on the time series data, and regularly output the updated preset battery cell detection model in the cloud to the battery cell detection device to update the preset battery cell detection model in the battery cell detection device.

[0223] The time series data can be sent by the battery cell testing equipment.

[0224] Specifically, in an operating energy storage system, various electrical signal detection devices and temperature sensors can collect real-time data on the voltage of each battery cell, the current of the series-connected battery cells, and the temperature at several preset temperature detection locations within the battery rack. This real-time collected initial time series data is input into the data processing unit of the battery cell detection device, where it is processed and converted into the time series data required for the neural network model input. This data is then uploaded to the cloud computing device (the cloud) via the data transmission module and fed into the trained neural network model in the edge computing module. This allows for real-time prediction of the location of abnormal battery cells within the battery rack, enabling alarms and intelligent operations and maintenance. The data uploaded to the cloud computing device serves two purposes: cloud storage for use by other data-driven algorithms and training of the neural network model in the cloud. As more data is collected over time, the neural network trained in the cloud becomes increasingly powerful and accurate. The updated neural network model in the cloud is then used to update the neural network model in the edge computing module via the data transmission module.

[0225] In this embodiment, a preset battery cell detection model is used to process time series data to obtain an abnormality probability value for each battery cell. Based on the abnormality probability value, it can be determined whether there is a battery cell abnormality and which battery cell has the abnormality. Therefore, the abnormal battery cell can be accurately found and replaced and maintained in time, thereby ensuring the safety and reliability of the energy storage system. In addition, the model is trained based on measured data, which can improve the detection accuracy of abnormal battery cells.

[0226] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A battery cell detection method, characterized in that: include: Acquire the collected time series initial data of the energy storage battery, and determine the time series data corresponding to the time series initial data; Calling a preset cell detection model to process the time series data to obtain an abnormality probability value of each cell in the energy storage battery; the preset cell detection model is trained based on training data; the training data includes time series samples of energy storage batteries and identifiers corresponding to the time series samples; Determining a detection result of abnormal cells of the energy storage battery based on the abnormal probability value of each cell; The training process of the preset battery cell detection model includes: Acquire a first time series initial sample of the energy storage battery when at least one abnormal cell exists in the energy storage battery, and determine a first time series sample corresponding to the first time series initial sample; Using the location information of the at least one abnormal battery cell in the energy storage battery as an identifier of the first time series sample; Acquire a second time series initial sample of the energy storage battery when no abnormal battery cell exists in the energy storage battery, and determine a second time series sample corresponding to the second time series initial sample; Using a preset normal cell identifier as an identifier of the second time series sample; Using the first time series sample and the second time series sample as time series samples of the energy storage battery, and using 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; The preset cell detection model is trained using the time series samples of the energy storage battery and the identifiers corresponding to the time series samples until a preset training stop condition is met.

2. The battery cell detection method according to claim 1, wherein: Determining time series data corresponding to the initial time series data includes: Using the time series initial data as time series data; Alternatively, according to a preset feature calculation rule, a feature value of the initial time series data under a preset derived feature is calculated, and a combination of the initial time series data and the feature value or the feature value is used as the time series data.

3. The battery cell detection method according to claim 1, wherein: The time series initial data includes: battery cell voltage, battery cell current and key point temperature.

4. The battery cell detection method according to claim 1, wherein: Before calling a preset cell detection model to process the time series data to obtain an abnormality probability value of each cell in the energy storage battery, the method further includes: Updated the preset battery cell detection model.

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

6. The battery cell detection method according to claim 1, characterized in that: Determining a detection result of an abnormal cell of the energy storage battery based on the abnormal probability value of each cell includes: Based on the abnormal probability values of the respective battery cells, a maximum abnormal probability value is screened out and used as a target abnormal probability value; Determine a position identifier of the target abnormality probability value in the abnormality probability values of each battery cell; If the location identifier is a preset location identifier, determining that the abnormal cell detection result of the energy storage battery is a first identifier; the first identifier indicates that there is no abnormal cell in the energy storage battery; If the location identifier is not a preset location identifier, the abnormal cell detection result of the energy storage battery is determined to be the second identifier and the location identifier; the second identifier indicates that there is an abnormal cell in the energy storage battery; the location identifier indicates the location information of the detected abnormal cell in the energy storage battery.

7. The battery cell detection method according to claim 1, characterized in that: In a case where the abnormal battery cell detection result is not a preset abnormal battery cell detection result, the method further includes: When the abnormal cell corresponding to the abnormal cell detection result has been replaced with a normal cell, the process returns to the step of obtaining the collected initial time series data of the energy storage battery and executes the step sequentially until the abnormal cell detection result is a preset abnormal cell detection result.

8. A battery cell detection device, characterized in that: include: A data acquisition module is used to acquire the collected time series initial data of the energy storage battery and determine the time series data corresponding to the time series initial data; a model processing module, configured to call a preset cell detection model to process the time series data to obtain an abnormality probability value of each cell in the energy storage battery; the preset cell detection model is trained based on training data; The training data includes time series samples of the energy storage battery and identifiers corresponding to the time series samples; wherein, the training process of the preset battery cell detection model includes: obtaining a first time series initial sample of the energy storage battery when there is at least one abnormal battery cell in the energy storage battery, and determining a first time series sample corresponding to the first time series initial sample; using the position information of the at least one abnormal battery cell in the energy storage battery as the identifier of the first time series sample; obtaining a second time series initial sample of the energy storage battery when there is no abnormal battery cell in the energy storage battery, and determining a second time series sample corresponding to the second time series initial sample; using a preset normal battery cell identifier as the identifier of the second time series sample; using the first time series sample and the second time series sample as the time series samples of the energy storage battery, and using the identifier of the first time series sample and the identifier of the second time series sample as the identifier corresponding to the time series sample; using the time series samples of the energy storage battery and the identifier corresponding to the time series sample to train the preset battery cell detection model until a preset training stop condition is met; The result determination module is used to determine the abnormal cell detection result of the energy storage battery based on the abnormal probability value of each cell.

9. A battery cell testing device, characterized in that: include: memory and processor; Wherein, the memory is used to store programs; The processor calls the program and is used to execute the battery cell detection method according to any one of claims 1 to 7.

10. A battery cell detection system, characterized in that: Comprising the battery cell detection device as claimed in claim 9.

11. The battery cell detection system according to claim 10, characterized in that: Also included are data acquisition devices; The data acquisition device is used to collect the time series initial data of the energy storage battery and output the time series initial data to the battery cell detection device; The time series initial data includes: battery cell voltage, battery cell current and key point temperature.

12. The battery cell detection system according to claim 10, characterized in that: It also includes cloud computing equipment; The cloud computing device is used to receive time series data, train and update the preset battery cell detection model in the cloud based on the time series data, and regularly output the updated preset battery cell detection model in the cloud to the battery cell detection device to update the preset battery cell detection model in the battery cell detection device.

13. The battery cell detection system according to any one of claims 10 to 12, characterized in that: The battery cell detection system is deployed in an energy storage container or an energy storage power station and is used to detect abnormal battery cells.

Images