Transferable battery thermal runaway risk assessment system based on autoencoder and adversarial learning

Abstract

This invention provides a transferable battery thermal runaway risk assessment system based on autoencoders and adversarial learning. The system comprises an autoencoder, a predictor, and a migrator. The system first encodes and decodes various types of data collected by internal and external sensors within the battery into the autoencoder. The decoded thermal runaway features are then fed into a fully connected network-based predictor, which maps the thermal runaway feature values to quantified thermal runaway risk values to predict the likelihood of thermal runaway in the battery within a certain period of time. For new battery models and application scenarios with limited data, model migration is performed using a migrator based on adversarial learning. The migrator applies the trained model and limited data to construct the autoencoder and predictor for the target domain, achieving model migration.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart batteries, and in particular relates to a transferable battery thermal runaway risk assessment system based on autoencoders and adversarial learning. Background Art

[0002] Current battery safety assessment algorithms rely on thresholds based on characteristic parameters of the battery management system (BMS). When a particular parameter reaches a threshold, a chain reaction has already occurred within the battery, and thermal runaway of the battery cells is irreversible. These models are only suitable for post-accident warnings and cannot effectively assess thermal runaway risks.

[0003] Furthermore, in battery systems, when operating scenarios and battery types change, there's no guarantee that the data sources for training and testing will remain identically distributed, rendering the existing model inapplicable. Furthermore, the amount of data in new scenarios is often small, and relabeling all the data is time-consuming and labor-intensive. This results in low accuracy and poor transferability of models trained using conventional methods. Summary of the Invention

[0004] In response to the above problems existing in the prior art, the present invention proposes a transferable battery thermal runaway risk assessment system based on autoencoders and adversarial learning. The purpose is to more efficiently extract thermal runaway features from sensor data, accurately predict the thermal runaway risk of batteries, and migrate existing models to new scenarios with limited existing data, thereby improving the accuracy of early warning before thermal runaway occurs.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A transferable battery thermal runaway risk assessment system based on autoencoders and adversarial learning, the system comprising:

[0007] Autoencoder: It includes at least a neural network encoder, a neural network decoder, and an output module, which is used to encode the input high-dimensional battery data into low-dimensional features, then perform a decoding operation, and output thermal runaway features for learning. Among them, at least one autoencoder loss function is set in the autoencoder to ensure the validity of the low-dimensional features by controlling the value of the at least one autoencoder loss function during operation;

[0008] Predictor: This includes at least a training module for predicting the risk of thermal runaway of the corresponding battery pack in the future by learning the thermal runaway features output by the autoencoder over a period of time;

[0009] Migrator: used to perform adversarial learning migration when the battery type and / or working conditions change, and migrate the autoencoder and predictor to the new scenario to work; the migrator includes at least a source domain feature extractor, a target domain feature extractor and a discriminator. When the battery type and / or working conditions change, the migrator first uses the source domain feature extractor to extract the thermal runaway features output by the autoencoder under the original working scenario or conditions, namely the source domain thermal runaway features, and then calls the predictor to perform source domain pre-training. After that, the migrator uses the target domain feature extractor to extract the target domain thermal runaway features. The discriminator performs adversarial learning migration based on the extracted source domain thermal runaway features and target domain thermal runaway features, updates the domain parameters of the autoencoder and predictor, and then uses the autoencoder and predictor after the updated domain parameters to perform target domain prediction.

[0010] Preferably, the autoencoder loss function is set as:

[0011]

[0012] Where x represents the input data tensor, Represents the extracted and output thermal runaway feature tensor. This function calculates the square of the difference between the input and output tensors.

[0013] Preferably, the high-dimensional battery data input to the autoencoder includes at least one of the following data: battery pack temperature, battery operating voltage, battery operating current, hydrogen concentration around the battery cell, and carbon monoxide concentration around the battery cell.

[0014] Preferably, the neural network encoder and neural network decoder of the autoencoder have the following network structure characteristics: the neural network encoder and the neural network decoder are both composed of several layers of fully connected neural networks, wherein the number of nodes in each layer of the encoder decreases from the input to the output layer; the number of nodes in each layer of the decoder increases from the input to the output layer; and wherein the number of nodes in the output layer of the encoder is the same as the number of nodes in the input layer of the decoder.

[0015] Preferably, the risk of thermal runaway of the corresponding battery pack predicted by the predictor includes at least one of the following:

[0016] Battery pack overheating, voltage drop, current surge, hydrogen content surge, carbon monoxide content increase, and relative humidity around the battery cell surge.

[0017] Preferably, the network structure characteristics of the predictor are:

[0018] The predictor consists of a fully connected input layer, a hidden layer, and an output layer, and a rectified linear unit (ReLU) layer is added after each layer.

[0019] Preferably, the workflow of the migrator includes at least: source domain pre-training, adversarial learning transfer, and target domain prediction.

[0020] Preferably, the neural network used by the migrator for adversarial learning transfer has the following structure and characteristics:

[0021] The feature extractor (source and target domains) consists of three long short-term memory (LSTM) neural network layers. The predictor consists of three fully connected layers, each followed by a Relu activation layer. The discriminator consists of three fully connected layers. The first two layers are each followed by a Relu activation layer. The final activation layer uses a sigmoid function and outputs a one-dimensional thermal runaway risk index.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] 1. The use of autoencoders for feature extraction can comprehensively analyze the various causes of thermal runaway, providing strong robustness. The encoder and prediction model utilize large amounts of data for supervised learning, effectively ensuring the accuracy of early warnings before thermal runaway occurs.

[0024] 2. Use adversarial learning to migrate thermal runaway prediction models and achieve adaption to the target domain. This is suitable for new scenarios with less data and offers greater portability and more reliable prediction results than large models for a single scenario.

[0025] 3. Map multi-dimensional battery data features to a one-dimensional thermal runaway risk index to intuitively reflect the possibility of thermal runaway in the battery in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Schematic diagram of the structure of the autoencoder of the present invention;

[0027] Figure 2 is an operational flow chart of the migrator of the present invention;

[0028] Figure 3 A schematic diagram of the network structure of the migrator counteracting the migration network of the present invention;

[0029] Figure 4 It is a schematic diagram of the model migration of the present invention. DETAILED DESCRIPTION

[0030] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be described in more detail below based on specific embodiments and in conjunction with the accompanying drawings.

[0031] It should also be noted here that, in order to avoid obscuring the present invention due to unnecessary details, only structures and / or processing steps closely related to the solutions of the present invention are shown in the specific embodiments, while other details that are not closely related to the present invention are omitted.

[0032] In addition, it should be noted that the terms "comprises", "includes" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus that includes a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or apparatus.

[0033] In the description of the present invention, it should be understood that the terms "upper", "lower", "bottom", "top", "front", "back", "inside", "outside", "left", "right", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention.

[0034] like Figures 1 to 4 As shown, the present invention provides a transferable battery thermal runaway risk assessment system based on autoencoder and adversarial learning, the system comprising:

[0035] An autoencoder includes at least a neural network encoder, a neural network decoder, and an output module, and is used to encode input high-dimensional battery data (e.g., sensor source data) into low-dimensional features, then perform a decoding operation to output thermal runaway features for learning. The autoencoder includes at least one autoencoder loss function, which is used to ensure the validity of the low-dimensional features by controlling the value of the at least one autoencoder loss function during operation.

[0036] Predictor: This includes at least a training module for predicting the risk of thermal runaway of the corresponding battery pack in the future by learning the thermal runaway features output by the autoencoder over a period of time;

[0037] Migrator: used to perform adversarial learning migration when the battery type and / or working conditions change, and migrate the autoencoder and predictor to the new scene to work; the migrator includes at least a source domain feature extractor, a target domain feature extractor and a discriminator. When the battery type and / or working conditions change, the migrator first uses the source domain feature extractor to extract the thermal runaway features output by the autoencoder under the original working scene or conditions, namely the source domain thermal runaway features, and then calls the predictor ( Figure 2The source domain predictor shown is the predictor whose domain parameters are not updated) is pre-trained in the source domain. After that, the migrator uses the target domain feature extractor to extract the target domain thermal runaway features. The discriminator performs adversarial learning migration based on the extracted source domain thermal runaway features and target domain thermal runaway features, updates the autoencoder and the predictor domain parameters, and then uses the autoencoder after the domain parameters are updated (i.e., Figure 4 The target domain autoencoder shown in Figure 4 The target domain predictor shown in is used to perform target domain prediction.

[0038] During operation, the predictor and autoencoder continuously iterate and fit the target battery pack characteristics based on the specified data.

[0039] Preferably, the autoencoder loss function is set as:

[0040]

[0041] Where x represents the input data tensor, Represents the extracted and output thermal runaway feature tensor. This function calculates the square of the difference between the input and output tensors.

[0042] Preferably, the high-dimensional battery data input to the autoencoder includes at least one of the following data: battery pack temperature, battery operating voltage, battery operating current, hydrogen concentration around the battery cell, and carbon monoxide concentration around the battery cell.

[0043] Preferably, the neural network encoder and neural network decoder of the autoencoder have the following network structure characteristics: the neural network encoder and the neural network decoder are both composed of several layers of fully connected neural networks, wherein the number of nodes in each layer of the encoder decreases from the input to the output layer; the number of nodes in each layer of the decoder increases from the input to the output layer; and wherein the number of nodes in the output layer of the encoder is the same as the number of nodes in the input layer of the decoder.

[0044] Preferably, the risk of thermal runaway of the corresponding battery pack predicted by the predictor includes at least one of the following:

[0045] Battery pack overheating, voltage drop, current surge, hydrogen content surge, carbon monoxide content increase, and relative humidity around the battery cell surge.

[0046] The network structure characteristics of the predictor are:

[0047] The predictor consists of a fully connected input layer, a hidden layer, and an output layer, and a rectified linear unit (ReLU) layer is added after each layer.

[0048] Figure 2 The operation flow chart of the migrator of the present invention is shown as follows: Figure 2 As shown in Figure 2, the workflow of its migrator includes at least: source domain pre-training, adversarial learning transfer, and target domain prediction.

[0049] In the source domain pre-training stage, the optimization goal of source domain pre-training is:

[0050]

[0051] Among them, X s and Y s Represents the source domain data distribution and source domain data label respectively, x s and y i Represents a sample and label from the source domain, n represents the number of source domain samples, represents the mean squared error loss, Es represents the feature extractor of the source domain, and P represents the thermal runaway risk index predictor.

[0052] See also Figure 3 The neural network used by the migrator for adversarial learning transfer has the following structure and characteristics:

[0053] The feature extractor (source and target domains) consists of three long short-term memory (LSTM) neural network layers. The predictor consists of three fully connected layers, each followed by a Relu activation layer. The discriminator consists of three fully connected layers. The first two layers are each followed by a Relu activation layer. The final activation layer uses a sigmoid function and outputs a one-dimensional thermal runaway risk index.

[0054] In the source domain pre-training and target domain detection stages, the feature extractor and predictor use RMSprop as the optimizer with a learning rate of 0.001. In the adversarial learning transfer stage, the SGD optimizer is used with a learning rate of 0.005, a momentum of 0.9, and a learning rate decay of 0.

[0055] See also Figure 4 In the source domain pre-training stage, the source domain autoencoder and source domain predictor (the autoencoder and predictor without updated domain parameters) use the source domain battery data to perform source domain pre-training on the source domain battery thermal runaway risk. Afterwards, the migrator uses the source domain features and the target domain features to perform adversarial learning transfer, and uses the target domain autoencoder and target domain predictor (the autoencoder and predictor after updating domain parameters) to predict the target domain battery thermal runaway risk.

[0056] Although the present invention is described herein with reference to specific embodiments, it should be understood that these embodiments are merely illustrative of the principles and applications of the invention. It should be understood that many modifications may be made to the illustrative embodiments, and that other arrangements may be devised, without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that the various dependent claims and features described herein may be combined in ways other than those described in the original claims. It should also be understood that features described in conjunction with individual embodiments may be employed in conjunction with other described embodiments.

Claims

1. A transferable battery thermal runaway risk assessment system based on autoencoder and adversarial learning, characterized in that: The system includes: Autoencoder: It includes at least a neural network encoder, a neural network decoder, and an output module, which is used to encode the input high-dimensional battery data into low-dimensional features, then perform a decoding operation, and output thermal runaway features for learning. Among them, at least one autoencoder loss function is set in the autoencoder to ensure the validity of the low-dimensional features by controlling the value of the at least one autoencoder loss function during operation. The autoencoder loss function is set to: Where x represents the input data tensor, Represents the extracted and output thermal runaway feature tensor; the high-dimensional battery data includes at least one of the following data: battery pack temperature, battery operating voltage, battery operating current, hydrogen concentration around the battery cell, and carbon monoxide concentration around the battery cell; Predictor: This includes at least a training module for predicting the risk of thermal runaway of the corresponding battery pack in the future by learning the thermal runaway features output by the autoencoder over a period of time; Migrator: used to perform adversarial learning migration when the battery type and / or working conditions change, and migrate the autoencoder and predictor to the new scenario to work; the migrator includes at least a source domain feature extractor, a target domain feature extractor and a discriminator. When the battery type and / or working conditions change, the migrator first uses the source domain feature extractor to extract the thermal runaway features output by the autoencoder under the original working scenario or conditions, namely the source domain thermal runaway features, and then calls the predictor to perform source domain pre-training. After that, the migrator uses the target domain feature extractor to extract the target domain thermal runaway features. The discriminator performs adversarial learning migration based on the extracted source domain thermal runaway features and target domain thermal runaway features, updates the domain parameters of the autoencoder and the predictor, and then uses the autoencoder and predictor after the updated domain parameters to perform target domain prediction.

2. The transferable battery thermal runaway risk assessment system based on autoencoder and adversarial learning according to claim 1, characterized in that: The neural network encoder and neural network decoder of the autoencoder have the following network structure characteristics: the neural network encoder and the neural network decoder are both composed of several layers of fully connected neural networks, where the number of nodes in each layer of the encoder decreases from the input to the output layer; the number of nodes in each layer of the decoder increases from the input to the output layer; and the number of nodes in the output layer of the encoder is the same as the number of nodes in the input layer of the decoder.

3. The transferable battery thermal runaway risk assessment system based on autoencoder and adversarial learning according to claim 2, characterized in that: The risk of thermal runaway of the corresponding battery pack predicted by the predictor includes at least one of the following: Battery pack overheating, voltage drop, current surge, hydrogen content surge, carbon monoxide content increase, and relative humidity around the battery cell surge.

4. The transferable battery thermal runaway risk assessment system based on autoencoder and adversarial learning according to claim 1, characterized in that: The network structure characteristics of the predictor are: The predictor consists of a fully connected input layer, a hidden layer, and an output layer, and a rectified linear unit layer is added after each layer.

5. The transferable battery thermal runaway risk assessment system based on autoencoder and adversarial learning according to claim 1, characterized in that: The workflow of the migrator includes at least: source domain pre-training, adversarial learning transfer, and target domain prediction.

6. The transferable battery thermal runaway risk assessment system based on autoencoder and adversarial learning according to claim 1, characterized in that: The neural network used by the migrator for adversarial learning transfer has the following structure and characteristics: Its feature extractor consists of three long short-term memory (LSTM) neural network layers, the predictor consists of three fully connected layers, each followed by a ReLU activation layer, and the discriminator consists of three fully connected layers. The first two fully connected layers are each followed by a ReLU activation layer, and the last activation layer uses a Sigmoid function to output a one-dimensional thermal runaway risk index.

Images