Method for estimation of state of charge and state of health in a lithium-ion battery pack, and system thereof
A machine learning method of estimating State of Charge (SOC) and State of Health (SOH) of a battery pack, the method including collecting time-series data of a battery pack; processing the collected data utilizing a reconstruction algorithm to reconstruct a first phase state space reconstruction; training a first neural network model using the first phase state space reconstruction for predicting the SOH; feeding the time-series data to the first neural network to predict an estimated SOH value; processing the time-series data utilizing the reconstruction algorithm to reconstruct a second phase state space reconstruction; training a second neural network model using the second phase state space reconstruction, taking into account the estimated SOH value, for predicting the SOC, and feeding the time-series data and the estimated SOH value to the second neural network to obtain an estimated SOC value. A system configured to perform the above method is also disclosed.
1 . A machine learning method of estimating State of Charge (SOC) and State of Health (SOH) of a battery pack, the method comprising:
collecting a plurality of first time-series data related to first operational parameters of a battery pack, wherein the first time-series data are collected from a battery management system (BMS) integrated with the battery pack, the BMS being configured to record the first operational parameters at predetermined intervals during charging and discharging cycles of the battery pack;
processing said collected first time-series data utilizing a Nonlinear State Space Reconstruction (NSSR) algorithm to reconstruct a first phase state space reconstruction;
training a first Long Short-Term memory (LSTM) neural network model using the first phase state space reconstruction for predicting SOH of said battery pack;
feeding said first time-series data to the first LSTM neural network to predict an estimated SOH value of the battery pack;
processing a plurality of second time-series data related to second operational parameters, utilizing said NSSR algorithm to reconstruct a second phase state space reconstruction;
training a second LSTM neural network model using the second phase state space reconstruction, taking into account said estimated SOH value, for predicting SOC of the battery pack; and
feeding the second time-series data and said estimated SOH value of the battery pack to the second LSTM neural network to obtain an estimated SOC value.
2 . The method according to claim 1 , wherein accuracy of said estimated SOH value and said estimated SOC value is evaluated using a testing dataset, wherein the training of the first and second LSTM models continues until the accuracy reaches a predetermined threshold.
3 . The method according to claim 1 , wherein the training step of each of the first and second LSTM neural network models involves validating the trained model using a testing dataset and calculating the root mean square error (RMSE) between said estimated SOH and SOC values and testing SOH and SOC values, and refining the LSTM neural network models based on the RMSE.
4 . The method according to claim 3 , wherein the step of validating is focused on the middle and late periods of the cycle life of the battery pack.
5 . The method according to claim 3 , wherein the step of validating is performed using a constant 1 C rate of charge/discharge.
6 . The method according to claim 1 , wherein the NSSR includes determining a delay time and an embedding dimension for reconstructing the phase state spaces based on the collected time-series data.
7 . The method according to claim 1 , further comprising normalizing the collected first or second time-series data prior to processing with the NSSR.
8 . The method according to claim 1 , wherein the NSSR algorithm reduces the interference of instantaneous noises through a delayed phase state space.
9 . The method according to claim 1 , wherein the SOC is estimated continuously in real-time during operation of the battery pack and the SOH is estimated at the start or end of a charging cycle.
10 . A system for estimating the State of Charge (SOC) and State of Health (SOH) of a battery pack using machine learning, comprising:
a data acquisition module configured to collect first time-series data related to first operational parameters and second time-series data related to second operational parameters of the battery pack;
a data processing module incorporating a Nonlinear State Space Reconstruction (NSSR) algorithm configured to reconstruct the first time-series data to a first reconstruct phase state space for SOH estimation and the second time-series data to a second reconstruct phase state space for SOC estimation;
an estimation module comprising a first Long Short-Term Memory (LSTM) neural network model configured for SOH estimation and a second LSTM neural network model configured for SOC estimation, based on the first and second reconstruct phase state space;
a training module for training the first and second LSTM neural network models, wherein:
the first LSTM neural network model is configured to be trained using the first reconstruct phase state space as input to predict an estimated SOH value;
the second LSTM neural network model is configured to be trained using the second reconstruct phase state space as input and taking into account said estimated SOH value, to predict an estimated SOC value;
wherein accuracy of said estimated SOH value and said estimated SOC value is evaluated using testing data, the training of the first and second LSTM models continues until the accuracy reaches a predetermined threshold.
11 . The system according to claim 10 further comprising a validation module configured to validate each of the trained model using a testing dataset and calculate the root mean square error (RMSE) between said estimated SOH and SOC values and testing SOH and SOC values, and refine the LSTM neural network models based on the RMSE.
12 . The system according to claim 11 , wherein the validation is performed using a constant 1 C rate of charge/discharge.
13 . The system according to claim 10 , wherein the NSSR algorithm is configured to determine a delay time and an embedding dimension for reconstructing the phase state spaces based on the collected time-series data.
14 . The system according to claim 10 , wherein the data processing module is configured to perform normalization of the first or second time-series data prior to processing with the NSSR algorithm to mitigate the influence of outliers.
15 . The system according to claim 10 , wherein the NSSR algorithm is configured to reduce the interference of instantaneous noises through a delayed phase state space.
16 . The system according to claim 10 , wherein the system is configured to estimate the SOC continuously in real-time during operation of the battery pack and to estimate the SOH at the start or end of a charging cycle.
17 . The system according to claim 10 , wherein the training module is configured to perform adjustment of the LSTM neural network model parameters based on environmental factors including the temperature of working condition.
18 . The system according to claim 10 , wherein the training module is configured to input a state vector comprising a plurality of SOC and/or SOH values from previous time steps into the first and the second LSTM neural network models.
19 . The method according to claim 1 , wherein the first operational parameters include one or more of mean voltage, minimum temperature, ratio of capacity increment to voltage increment, and number of cycles.
20 . The method according to claim 1 , wherein the second operational parameters include one or more of mean voltage, minimum temperature, current, ratio of capacity increment to voltage increment, and SOH, wherein the SOH is the estimated SOH from the first LSTM neural network.
21 . The system according to claim 10 , wherein the first operational parameters include one or more of mean voltage, minimum temperature, ratio of capacity increment to voltage increment, and number of cycles.
22 . The system according to claim 10 , wherein the second operational parameters include one or more of mean voltage, minimum temperature, current, ratio of capacity increment to voltage increment, and SOH, wherein the SOH is the estimated SOH from the first LSTM neural network.