AI-Driven Solutions for EV Battery Thermal Management

Predicting thermal failure of new energy vehicle batteries using a time-varying twin converter model. The method involves converting the multi-element battery signal into time-series features using downsampling and segmented sampling techniques. These features are extracted using a time attention mechanism to capture both global and local signal variations. The extracted features are then fed into a neural network for predicting battery thermal failure.

Source

Method for predicting battery temperature in vehicles to improve battery performance and longevity. The method involves using a pre-trained battery temperature prediction model to accurately predict battery temperature at a future time based on current battery attributes. This allows better temperature control compared to just predicting linear models at the next time point. The prediction involves extracting features from the current battery attributes, generating an attention weight matrix based on those features, and using it to calculate the future temperature. This allows accurate temperature prediction even with longer time intervals.

Source

Method for predicting battery cell temperature using historical data from other cells. The method involves obtaining cell parameter information (like SOC, voltage, etc.) for a target cell at different steps in time. These parameters are then fed into separate models trained for current prediction and temperature prediction. By inputting the first cell's historical parameters into the current model and the second cell's historical parameters into the temperature model, the target cell's temperature after a set time can be predicted. This improves accuracy compared to just using a physical model as it avoids needing all the physical parameters and reduces modeling errors.

Source

Hard-pack lithium battery safety monitoring using deep learning to improve accuracy of detecting temperature abnormalities in individual cells. The method involves monitoring battery working state to extract temperature features at different steps: cell temperature difference, kernel PCA, homologous expansion, feature fusion, dynamic weighting, and vector conversion. This multi-step extraction improves understanding of cell states. A preset temperature analysis model processes the converted vectors to detect anomalies, then makes specific processing strategies for each cell based on the overall analysis.

Source

Estimating the surface temperature of a lithium ion battery using a deep neural network to improve battery monitoring and management. The method involves feeding inputs like voltage, current, charge state, and ambient temperature into a neural network trained on historical battery data. The network learns to predict the battery surface temperature. The neural network architecture uses long short-term memory (LSTM) gates to handle sequential inputs. The training involves finding an optimal filtering frequency for the input signals. The neural network is implemented on a battery management system.

Source

Battery thermal runaway prediction method to determine if a vehicle's battery is at risk of overheating and thermal runaway. The method involves a two-step process: (1) assessing if the vehicle meets conditions indicating thermal runaway risk, and (2) if so, using a trained machine learning model to predict the battery's risk of thermal runaway. This allows targeted prediction for vehicles with elevated risk rather than using a single model for all vehicles. The risk assessment involves scoring factors like driving time, battery power, temperature, and weather.

Source

A system for managing battery packs in electric vehicles using KNN machine learning to predict battery aging and optimize pack performance. The system monitors charging/discharging states of the pack cells and uses KNN to forecast aging changes. It then controls pack operation to mitigate aging and maintain optimal performance. By predicting cell aging and taking proactive measures, it aims to extend pack life beyond normal limits.

Source

Real-time prediction of lithium battery temperature using a multi-space-time width learning model. The model is constructed using a modified version of the breadth-first learning (BFL) algorithm. The modification involves adding a nonlinear space activation function to BFL to allow it to process nonlinear spatial information. This enables the derived Spatio-Temporal Breadth-First Learning (ST-BFL) model to accurately predict battery temperature in response to diverse operating conditions and space-time coupling characteristics.

Source

Estimating the temperature of large-size laminated batteries used in electric vehicles using a hybrid model that combines a thermal model and a neural network. The thermal model provides accurate temperature distribution inside the battery, but is complex and requires temperature sensors. The neural network estimates temperature without a physical model, but has high data requirements. The hybrid model leverages the strengths of both approaches to improve accuracy and generalization. It uses a thermal model to capture the battery's thermal behavior over a wide temperature range, and a neural network to fill in gaps where sensor data is insufficient.

Source

A method and device for predicting the temperature of lithium batteries in electric vehicles that improves the accuracy of temperature prediction compared to existing methods. The method involves collecting battery data at different aging levels, selecting relevant temperature indices using gray correlation analysis, clustering based on aging level, training a time convolution network (TCN) model on clustered subsets, optimizing the TCN weights using a modified chaos-based algorithm, and testing the optimized model. This involves using a clustering algorithm to divide the battery data into groups based on aging level, training separate TCN models for each aging group, optimizing the TCN weights using a modified chaos-based algorithm, and testing the optimized models on a separate set. This allows better modeling of the complex aging-temperature relationship.

Source

Battery thermal runaway prediction method for lithium ion batteries in energy storage systems using a hybrid model combining a battery model and neural network. The method involves obtaining battery parameters like voltage, current, and temperature through data acquisition. These parameters are fed into a battery electric heating coupling model to simulate the battery's internal temperature. The simulated internal temperature is then fed into an LSTM neural network to predict the surface temperature. The LSTM output is used as input to a pre-judging model to determine if thermal runaway is likely based on the predicted surface temperature.

Source

Online estimation of internal temperature of lithium batteries using a neural network and thermal network model. The method involves training a neural network using measured surface temperature and internal temperature data from a battery to learn the relationship between surface and internal temperatures. This neural network is then used to estimate the internal temperature based on the surface temperature in real-time during battery operation. The neural network is trained using a thermal network model that simulates the heat distribution inside the battery to provide ground truth internal temperatures for training. The thermal network model also helps improve the robustness and accuracy of the neural network estimates by providing a physical basis for the temperature relationship.

Source

Efficient method to predict the temperature inside a lithium battery cell using a physical information neural network. The method involves processing input parameters like battery current, voltage, and time through a neural network model containing an embedded internal temperature model of the battery. This avoids the need for large numbers of physical parameters and assumptions compared to traditional methods.

Source

Early warning system to detect abnormal heating in battery packs of energy storage systems. The system uses a trained neural network model to predict battery temperature based on historical operation data. During real-time operation, the predicted temperature is compared with the measured temperature to determine if the battery is overheating. This allows detecting potential thermal runaway before it becomes critical. The neural network model combines long short-term memory (LSTM) and convolutional neural networks (CNN) to accurately predict temperature by extracting time and space features from battery data.

Source

A method for estimating and limiting the maximum electrical power that an electric vehicle battery can deliver without overheating. The method involves using a supervised learning neural network with forward propagation to estimate the maximum power based on measured cell temperatures and battery state of charge. This allows precise and reliable power limiting without requiring temperature sensors inside the battery. The neural network is trained using historical temperature and SOC data to learn the relationship between cell temperatures and maximum power.

Source

Training a thermal management model for a battery pack of an electric vehicle using reinforcement learning to adaptively control the cooling/heating of the battery pack for optimal energy consumption in different environments. The method involves collecting battery pack data, creating a reward function based on battery temperature, pump energy, and adjustment energy, training an initial model using this reward, then further training it with real-time battery pack data. The trained model can predict optimal cooling/heating parameters for a given battery state.

Source

Estimating the charging time of an electric vehicle battery that takes into account thermal stress and environmental conditions. The method involves training a neural network with measured charging times and battery temperatures at different ambient temperatures. The neural network is used to estimate charging time given initial and target battery states and ambient temperature.

Source

Power battery thermal runaway diagnosis method and system using a hybrid prediction framework to accurately predict and monitor battery thermal runaway. It combines a recurrent neural network (RNN) called GRU for battery operation data prediction with a state space model to estimate internal temperature. The RNN predicted values are used as observations for the state space model. This integrates model driving (using physics) and data driving (using machine learning) to improve accuracy compared to just one approach alone.

Source

A temperature prediction method for batteries in vehicles that doesn't require internal sensors. The method involves obtaining the vehicle's current battery operating conditions and feeding them into a pre-trained neural network model to predict battery temperature. The model is trained using historical battery data and confirmed thermal resistance and capacity parameters. The method avoids the need for internal sensors by leveraging existing battery management system data.

Source

Method for accurately predicting and monitoring the internal temperature field of lithium ion battery packs in applications like electric vehicles and energy storage systems. The method involves using a high-precision mathematical model to calculate temperature changes based on battery operating conditions, and then training a neural network algorithm to predict temperatures quickly. This allows online real-time monitoring of the entire pack temperature field using a limited number of measurement points. The model is first verified by comparing calculated temperatures to experimental data.

Source