A Lithium-ion Battery SOC Estimation Method Based on Time-Delay Convolutional Neural Network

Abstract

The present invention provides a lithium-ion battery SOC estimation method based on a time-delay convolutional neural network, and belongs to the technical field of lithium-ion batteries. The technical problem that time lag cannot be directly combined with a convolutional neural network is solved. The technical scheme is as follows: it includes the following steps: step 1) measuring the terminal voltage, current and temperature of the lithium-ion battery when the SOC changes from 1 to 0; step 2) preprocessing the measured data to construct a time-delay convolutional neural network training data set and a test data set for SOC estimation; step 3) using the time-delay convolutional neural network to train and test the data set to achieve real-time SOC estimation. The beneficial effects of the present invention are: the time-delay convolutional neural network in the present invention can consider more data and can have a higher estimation accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium-ion batteries, and in particular to a lithium-ion battery SOC estimation method based on a time-delay convolutional neural network. Background Art

[0002] Accurately estimating the state of charge (SOC) of lithium-ion batteries is important because it tells users how much "fuel" is left in a system or device powered by the battery, such as electric vehicles, drones, and cell phones.

[0003] Battery models can be divided into three categories: electrochemical models, equivalent circuit models, and machine learning models. Electrochemical models are generally considered to have high computational costs and are therefore not suitable for real-time estimation of lithium-ion battery SOC. Although electrochemical models can establish various reduced-order models through simplification and assumptions, the SOC estimation accuracy is similar to that of equivalent circuit models. Machine learning models have recently become increasingly popular in lithium-ion battery SOC estimation. Typical models include Gaussian processes, feedforward neural networks, time-delay neural networks, recurrent neural networks, convolutional neural networks, and long short-term memory neural networks. The time-delay recurrent neural network successfully applies the concept of time delay to recurrent neural networks. The time-delay recurrent neural network is explained in detail in the book "Time-delay Recurrent Neural Networks", which introduces many important issues, and some scholars have used this network to study lithium-ion battery SOC estimation. Unlike recurrent neural networks, the invention of convolutional neural networks is intended for image processing. Images can be regarded as inputs of a two-dimensional structure. Therefore, time delays cannot be directly combined with convolutional neural networks like time-delay recurrent neural networks. Combining the concept of time lag with convolutional neural networks and using them to estimate the SOC of lithium-ion batteries has become a technical challenge.

[0004] How to solve the above technical problems is the subject faced by the present invention. Summary of the invention

[0005] The purpose of the present invention is to provide a lithium-ion battery SOC estimation method based on a time-delay convolutional neural network, which obtains various battery parameters through battery charging and discharging experiments at different temperatures, then processes the data and constructs a training data set, and imports the trained data set into a time-delay convolutional neural network for training, which can finally be used for real-time estimation of the SOC of a lithium-ion battery.

[0006] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is specifically: a lithium-ion battery SOC estimation method based on a time-delay convolutional neural network, comprising the following steps:

[0007] Step 1) Fully charge a new lithium-ion battery under varying temperatures, and then repeatedly measure the terminal voltage, current, and temperature of the lithium-ion battery with SOC ranging from 1 to 0 through a constant current discharge test, a DST condition discharge test, a FUDS condition discharge test, and a US06 condition discharge test;

[0008] Step 2) preprocessing the measured data to construct a time-delay convolutional neural network training data set and a test data set for SOC estimation;

[0009] Step 3) Use the time-delay convolutional neural network to train and test the dataset to achieve the final SOC real-time estimation.

[0010] As a further optimization scheme of a lithium-ion battery SOC estimation method based on a time-delay convolutional neural network provided by the present invention, the step 2) specifically includes the following steps:

[0011] Step 2-1) The data does not have a unified dimension. When fitting the neural network algorithm, there will be problems such as slow convergence speed and high error, so the measured data needs to be normalized. The battery objects of various working condition experiments are the same, but the range of variation of voltage and current is different. Therefore, it is necessary to pre-process the data of various working conditions separately first, and use the maximum and minimum normalization method to normalize the input data. The output is the SOC of the lithium-ion battery. The range of SOC is between 0 and 1, so no normalization is required.

[0012] Step 2-2) After the data is normalized, the following method is used to construct a data set. First, the measured values ​​in a discharge cycle are truncated into a number of data of length n, as shown below:

[0013]

[0014] Among them, x ij represents the jth data of the i-th cycle in the input data set, x ij The corresponding output is t i_j The SOC value of the lithium-ion battery at the moment. U represents the voltage across the battery, I represents the current passing through the battery, T represents the temperature of the battery during operation, and t represents the discharge time. In the present invention, n is 10.

[0015] Step 2-3) For a discharge experiment with a total number of charge and discharge cycles N and a single discharge cycle m. In order to fully explore the connection between the input data before and after each discharge cycle, the padding method is used to expand the edge data, and then a total of N×m groups of input data with a size of 4×10 are generated, and the corresponding N×m SOC values ​​are output data. The input and output data of different working conditions are merged into the total data set.

[0016] Step 2-4) In order not to affect the training of the data set, the total data set is not shuffled and then the training and test data sets are extracted. The first 80% of the data of a discharge condition is selected as the training data set, and the last 20% of the data is selected as the test data set, which are used as the training and test data of the time-delay convolutional neural network.

[0017] As a further optimization scheme of a lithium-ion battery SOC estimation method based on a time-delay convolutional neural network provided by the present invention, the step 3) specifically comprises the following steps:

[0018] Step 3-1) The time-delay convolutional neural network architecture developed for lithium-ion battery SOC estimation includes: an input layer (Input), three convolutional layers C (Convolution), a fully connected layer F (Fully connected layer) and an output layer (Output);

[0019] Step 3-2) The time-delay operation of the time-delay convolutional neural network is specifically manifested in the convolution layer. For the input data xij When the convolution kernel of each convolution layer is convolved, the convolution result of the current i-th cycle data and the i-1th cycle x i-1j The weighted sum of the convolution results of the data is used as the final convolution layer output, and the convolution kernels of the two convolution operations have different weights, so that the time lag effect can be achieved.

[0020] Step 3-3) The convolutional layers and fully connected layers are all activated using the relu function, and the output layer is activated using the tanh function, and the estimated values ​​are mapped to 0 to 1.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] (1) The present invention combines the concept of time lag with convolutional neural network, proposes a time lag convolutional neural network algorithm and can be applied to lithium-ion battery SOC estimation.

[0023] (2) The present invention uses a time-delay convolutional neural network algorithm, which can consider the data of the previous cycle through a time-delay operation while considering the data of the past moments of the current cycle, and can have a higher estimation accuracy.

[0024] (3) The convolution operation of the time-delay convolutional neural network of the present invention in the convolutional layer is to use the weighted sum of the two convolution kernels processing the two data respectively as the final convolutional layer output. The convolutional neural network with multiple convolution kernels uses multiple convolution kernels to process the same data. Therefore, the time-delay convolutional neural network can consider more data and have higher estimation accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0026] Figure 1 This is the time-delay convolutional neural network framework of the present invention.

[0027] Figure 2 It is a schematic diagram of the time-delay convolution operation of the present invention.

[0028] Figure 3 It is the SOC estimation result of a whole discharge cycle in the constant current discharge experiment test set of the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. Of course, the specific embodiments described here are only used to explain the present invention and are not used to limit the present invention.

[0030] Example 1

[0031] See also Figures 1 to 3 The present invention provides a technical solution as follows: the present invention provides a lithium-ion battery SOC estimation method based on a time-delay convolutional neural network. This embodiment uses Panasonic lithium-ion battery NCR-18650B as the object for research, with a calibrated voltage of 3.7V and a battery capacity of 3400mAh; the battery is fully charged in a constant current-constant voltage charging mode, and after standing for 1 hour, the battery is in a fully charged state. The battery is subjected to discharge experiments in constant current discharge, DST conditions, FUDS conditions and US06 conditions, respectively, until the voltage drops to the discharge cut-off voltage, and the experiment is repeated.

[0032] In order to better achieve the purpose of the present invention, this embodiment is a lithium-ion battery SOC estimation method based on a time-delay convolutional neural network, which specifically includes the following steps:

[0033] Step 1) Fully charge a new lithium-ion battery under varying temperatures, and then repeatedly measure the terminal voltage, current, and temperature of the lithium-ion battery with SOC ranging from 1 to 0 through a constant current discharge test, a DST condition discharge test, a FUDS condition discharge test, and a US06 condition discharge test;

[0034] Step 2) preprocessing the measured data to construct a time-delay convolutional neural network training data set and a test data set for SOC estimation;

[0035] Step 3) Use the time-delay convolutional neural network to train and test the dataset to achieve the final SOC real-time estimation.

[0036] Preferably, the step 2) specifically comprises the following steps:

[0037] Step 2-1) The data does not have a unified dimension, which will lead to slow convergence and high error when fitting the neural network algorithm, so the measured data needs to be normalized; the battery objects of various working condition experiments are the same, but the range of variation of voltage and current is different, so it is necessary to pre-process the data of various working conditions separately, and use the maximum and minimum normalization method to normalize the input data. The output is the SOC of the lithium-ion battery. The range of SOC is between 0 and 1, so no normalization is required.

[0038] Step 2-2) After the data is normalized, the following method is used to construct a data set. First, the measured values ​​in a discharge cycle are truncated into a number of data of length n, as shown below:

[0039]

[0040] Among them, x ij represents the jth data of the i-th cycle in the input data set, x ij The corresponding output is t i_j The SOC value of the lithium-ion battery at the moment, U represents the voltage across the battery, I represents the current passing through the battery, T represents the temperature of the battery during operation, t represents the discharge time, and in the present invention, n is 10.

[0041] Step 2-3) For a discharge experiment with a total number of charge and discharge cycles N and a single discharge cycle m, in order to fully explore the relationship between the input data before and after each discharge cycle, the padding method is used to expand the edge data, and then a total of N×m groups of input data with a size of 4×10 are generated, and the corresponding N×m SOC values ​​are output data. The input and output data of different working conditions are merged into a total data set;

[0042] Step 2-4) In order not to affect the training of the data set, the total data set is not shuffled and then the training and test data sets are extracted. The first 80% of the data of a discharge condition is selected as the training data set, and the last 20% of the data is selected as the test data set, which are used as training and test data for the time-delay convolutional neural network.

[0043] Preferably, the step 3) specifically comprises the following steps:

[0044] Step 3-1) The time-delay convolutional neural network architecture developed for lithium-ion battery SOC estimation is as follows Figure 1 The figure shows: an input layer (Input), three convolutional layers C (Convolution), a fully connected layer F (Fully connected layer) and an output layer (Output);

[0045] Step 3-2) The time-delay operation of the time-delay convolutional neural network is specifically manifested in the convolution layer. For the input data x ij When the convolution kernel of each convolution layer is convolved, the convolution result of the current i-th cycle data and the i-1th cycle x i-1j The weighted sum of the convolution results of the data is used as the final convolution layer output, and the convolution kernels of the two convolution operations have different weights, such as Figure 2 As shown, this can achieve a time-lag effect.

[0046] Step 3-3) The convolutional layer and the fully connected layer are all activated using the relu function, and the output layer is activated using the tanh function, and the estimated value is mapped from 0 to 1. After training, the error evaluation index MAPE of the final network is 1.74%. Figure 3 It is the SOC estimation result of a whole discharge cycle in the constant current discharge experiment test set.

[0047] The present invention combines the concept of time lag with convolutional neural network, proposes a time lag convolutional neural network algorithm and can be applied to lithium-ion battery SOC estimation.

[0048] The present invention uses a time-delay convolutional neural network algorithm, which can consider the data of the previous cycle through a time-delay operation while considering the data of the past moments of the current cycle, and can have a higher estimation accuracy.

[0049] The convolution operation of the time-delay convolutional neural network of the present invention in the convolutional layer is to use the weighted sum of two data processed by two convolution kernels respectively as the final convolutional layer output. The convolutional neural network with multiple convolution kernels uses multiple convolution kernels to process the same data. Therefore, the time-delay convolutional neural network can consider more data and have higher estimation accuracy.

[0050] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A lithium-ion battery SOC estimation method based on a time-delay convolutional neural network, characterized in that: The following steps are involved: Step 1) Fully charge a new lithium-ion battery under varying temperatures, and then repeatedly measure the terminal voltage, current, and temperature of the lithium-ion battery with SOC ranging from 1 to 0 through a constant current discharge test, a DST condition discharge test, a FUDS condition discharge test, and a US06 condition discharge test; Step 2) preprocessing the measured data to construct a time-delay convolutional neural network training data set and a test data set for SOC estimation; Step 3) Use the time-delay convolutional neural network to train and test the data set to achieve the final SOC real-time estimation; The step 3) specifically comprises the following steps: Step 3-1) The time-delay convolutional neural network architecture developed for lithium-ion battery SOC estimation includes: an input layer, three convolutional layers C, a fully connected layer F and an output layer; Step 3-2) The time-delay operation of the time-delay convolutional neural network is specifically manifested in the convolution layer. For the input data x ij When the convolution kernel of each convolution layer is convolved, the convolution result of the current i-th cycle data and the i-1th cycle x i-1j The weighted sum of the convolution results of the data is used as the final convolution layer output, and the convolution kernels of the two convolution operations have different weights to achieve a time-delay effect; Step 3-3) The convolutional layers and fully connected layers are all activated using the relu function, and the output layer is activated using the tanh function, and the estimated values ​​are mapped to 0 to 1.

2. The lithium-ion battery SOC estimation method based on time-delay convolutional neural network according to claim 1 is characterized in that: The step 2) specifically comprises the following steps: Step 2-1) normalize the measured data. The battery objects of various working condition experiments are the same, and the voltage and current change ranges are different. The data of various working conditions are pre-processed separately first, and the input data is normalized using the maximum and minimum normalization method. The output is the SOC of the lithium-ion battery. The range of SOC is between 0 and 1, and no normalization is required; Step 2-2) After the data is normalized, the following method is used to construct a data set. First, the measured values ​​in a discharge cycle are truncated into a number of data of length n, as shown below: Among them, x ij represents the jth data of the i-th cycle in the input data set, x ij The corresponding output is t i_j The SOC value of the lithium-ion battery at the moment, U represents the voltage across the battery, I represents the current passing through the battery, T represents the temperature of the battery during operation, t represents the discharge time, and n is 10; Step 2-3) For a discharge experiment with a total number of charge and discharge cycles N and a single discharge cycle m, in order to fully explore the relationship between the input data before and after each discharge cycle, the padding method is used to expand the edge data, and then a total of N×m groups of input data with a size of 4×10 are generated, and the corresponding N×m SOC values ​​are output data. The input and output data of different working conditions are merged into a total data set; Step 2-4) In order not to affect the training of the data set, the total data set is not shuffled, and the training and test data sets are extracted. The first 80% of the data of a discharge condition is selected as the training data set, and the last 20% of the data is selected as the test data set, which are used as training and test data for the time-delay convolutional neural network.

Images