CN110658459A - Lithium ion battery state of charge estimation method based on bidirectional cyclic neural network - Google Patents

Abstract

The invention discloses a method for estimating the state of charge of a lithium ion battery based on a bidirectional cyclic neural network. The real-time state of charge value of the lithium ion battery is obtained by using the data generated in real time by the lithium ion battery and the trained bidirectional cyclic neural network. After the training of the bidirectional recurrent neural network is completed, the state of charge value can be estimated in real time, which is very convenient. The bidirectional recurrent neural network considers the characteristics of time series data and uses the data before and after the current result, which is suitable for the state of charge of lithium-ion batteries. value estimation field. The invention belongs to a data-driven method, does not require complicated electrochemical knowledge, can effectively extract historical data of lithium ion batteries, model the discharge characteristics of lithium ion batteries, obtain accurate state of charge estimation, and can process A complex nonlinear system with a large amount of data does not require information in the battery field, but only requires historical data of lithium-ion batteries.

Description

State-of-charge estimation method for lithium-ion battery based on bidirectional recurrent neural network

technical field

The invention relates to the technical field of battery management systems and deep learning, in particular to a method for estimating the state of charge of a lithium ion battery based on a bidirectional cyclic neural network.

Background technique

Lithium-ion batteries are currently the fastest growing and most promising battery technology. Compared with traditional batteries, lithium-ion batteries have the advantages of light weight, fast charging, high energy density, low self-discharge rate, and long service life.

State of Charge (SOC), one of the key states of battery monitoring, is defined as the percentage of the battery's remaining capacity to its maximum capacity. Reliable SOC estimation can accurately judge the current state of the battery, prevent possible dangers, and ensure the safe and stable operation of the battery. However, due to the nonlinear and time-varying characteristics of the SOC of Li-ion batteries, the SOC value cannot be directly observed, and the battery discharge characteristics are easily affected by factors such as battery aging and temperature changes, making SOC estimation challenging.

Battery SOC estimation methods are mainly divided into two categories, one is the SOC estimation method based on the characteristics of the battery itself, and the other is the data-driven SOC estimation method. The latter has attracted great attention in the past two years. It does not need to rely on the electrochemical methods in the traditional battery field at all, but only needs the historical data of battery charge and discharge, from which the end-to-end mapping relationship between battery characteristics and SOC values can be learned. This kind of method is mainly based on support vector machine and neural network, especially neural network. Because of its strong data fitting ability and can handle a large order of data, it has achieved good results in the field of battery SOC prediction. .

For example, the simplest fully-connected neural network can capture the nonlinear relationship between output and input, take manually selected battery features as the input of the fully-connected neural network, and use the battery SOC value as the output of the fully-connected neural network. The network can fit this relationship well, and the effect obtained has exceeded the support vector machine. However, the fully connected neural network simply grasps the relationship between input and output, and its effect in the field of battery SOC prediction still needs to be improved.

SUMMARY OF THE INVENTION

In view of this, the present invention provides a method for estimating the state of charge of a lithium ion battery based on a bidirectional cyclic neural network, so as to realize accurate estimation of the state of charge of the lithium ion battery.

Therefore, the present invention provides a method for estimating the state of charge of a lithium ion battery based on a bidirectional cyclic neural network, comprising the following steps:

S1: Obtain the battery voltage value, battery current value and battery surface temperature value of the lithium-ion battery at the current moment;

S2: perform data sampling processing, data standardization processing and data dimension change processing on the acquired data at the current moment;

S3: Input the processed data at the current moment into the trained bidirectional recurrent neural network to obtain the state of charge value of the lithium-ion battery at the current moment.

In a possible implementation manner, in the above-mentioned method for estimating the state of charge of a lithium-ion battery provided by the present invention, the training process of the bidirectional recurrent neural network includes the following steps:

S11: Manually select the features input to the bidirectional recurrent neural network, including battery voltage value, battery current value and battery surface temperature value;

S12: Collect battery voltage value, battery current value, battery surface temperature value and state of charge value under different working conditions;

S13: Perform data sampling processing on the collected battery voltage value, battery current value, battery surface temperature value and state-of-charge value, and perform data standardization processing on the battery voltage value, battery current value and battery surface temperature value after data sampling processing. Data dimension change processing;

S14: Initialize the bidirectional cyclic neural network, input the processed data of the battery voltage value, battery current value and battery surface temperature value under the same working condition as the data to be measured into the initialized bidirectional cyclic neural network, and use the time-based The back-propagation algorithm is used for training, and the network hyperparameters are continuously adjusted to obtain a trained bidirectional recurrent neural network.

In a possible implementation manner, in the above-mentioned method for estimating the state of charge of a lithium-ion battery provided by the present invention, step S13 is to perform data analysis on the collected battery voltage value, battery current value, battery surface temperature value and state of charge value. Sampling processing, performing data standardization processing and data dimension change processing on the battery voltage value, battery current value and battery surface temperature value after data sampling processing, which specifically includes the following steps:

S131: Under different working conditions, re-sample the battery voltage value, battery current value, battery surface temperature value and state of charge value, and set the data interval to 1s to generate a data point;

S132: Standardize the resampled battery voltage value, battery current value and battery surface temperature value, so that the battery voltage value, battery current value and battery surface temperature value are all distributed in the [0,1] interval. The formula is:

Among them, D represents any one of the battery voltage value, battery current value and battery surface temperature value, D _t represents the data at time t, D _min represents the smallest data point, and D _max represents the largest data point;

S133: Perform dimension change processing on the battery voltage value, battery current value and battery surface temperature value after the normalization process, and connect the battery voltage value, battery current value and battery surface temperature value at each time point in the normalized data as Vector [V, I, T], connecting the data points of k time steps as a sample input data of the bidirectional recurrent neural network [[V _t ,I _t ,T _t ],[V _t+1 ,I _{t +1} ,T _t+1 ],...,[V _t+k-1 ,I _t+k-1 ,T _t+k-1 ]], and finally get all sample input data [sample number, time step, feature number].

In a possible implementation manner, in the above-mentioned method for estimating the state of charge of a lithium-ion battery provided by the present invention, in step S14, the bidirectional recurrent neural network is initialized, and the processed battery voltage value, battery current value and battery surface The temperature value and the data to be measured are the data input under the same working conditions as the initialized bidirectional recurrent neural network, and the time-based back propagation algorithm is used for training, and the network hyperparameters are continuously adjusted to obtain the trained bidirectional recurrent neural network. It includes the following steps:

S141: Initialize each parameter value of the bidirectional cyclic neural network, and randomly set each parameter value to any value in the [0,1] interval;

S142: Input all sample input data [number of samples, time steps, number of features] into the bidirectional recurrent neural network, and through the forward propagation of the bidirectional recurrent neural network, calculate the predicted value of the state of charge at the current moment, and calculate the current moment The distance between the predicted value of the state of charge and the actual value of the state of charge at the current moment, to obtain the distance between the predicted value of the state of charge of all samples and the real value of the state of charge, the calculation formula is as follows:

表示荷电状态的预测值，m表示样本数量，i表示样本序号；where y represents the true value of the state of charge,

Represents the predicted value of the state of charge, m represents the number of samples, and i represents the sample number;

S143: Update each parameter value of the bidirectional recurrent neural network using gradient descent and backpropagation algorithms;

Steps S142 to S143 are repeated until the bidirectional recurrent neural network converges, and the training is completed.

In a possible implementation manner, in the above-mentioned method for estimating the state of charge of a lithium-ion battery provided by the present invention, in step S2, data sampling processing, data normalization processing and data dimension change processing are performed on the acquired data at the current moment. It includes the following steps:

S21: Re-sample the battery voltage value, battery current value and battery surface temperature value, and set the data interval to 1s to generate a data point;

S22: Standardize the sampled data, so that the battery voltage value, battery current value and battery surface temperature value are all distributed in the [0,1] interval. The formula for normalization is:

Among them, D represents any one of the battery voltage value, battery current value and battery surface temperature value, D _t represents the data at time t, D _min represents the smallest data point, and D _max represents the largest data point;

S23: Perform dimension change processing on the normalized data, connect the battery voltage value, battery current value and battery surface temperature value at each time point in the normalized data as a vector [V, I, T], connect k The data points of the time steps are connected as a sample input data of the bidirectional recurrent neural network [[V _t , I _t , T _t ], [V _t+1 , I _t+1 , T _t+1 ],  … ,[V _t+k-1 ,I _t+k-1 ,T _t+k-1 ]], and finally get all sample input data [sample number, time step, feature number].

The method for estimating the state of charge of the lithium ion battery provided by the present invention utilizes the data generated in real time by the lithium ion battery, and uses the trained bidirectional cyclic neural network model to obtain the real-time state of charge value of the lithium ion battery. The bidirectional cyclic neural network model is in After the training is completed, the state of charge value can be estimated in real time, which is very convenient. The two-way recurrent neural network can fully consider the characteristics of time series data, and use the data before the current result and the data after the current result, which is more effective than the one-way recurrent neural network. The effect of the network is more accurate, and the estimation of the state of charge of lithium-ion batteries has great potential, which is very suitable for the estimation of the state of charge of lithium-ion batteries. The invention belongs to a data-driven method, does not require complicated electrochemical-related knowledge, and completely starts from data, can effectively extract the information expressed by the historical data of the lithium ion battery, model the discharge characteristics of the lithium ion battery, and obtain accurate charge Electric state estimation results, and can process complex nonlinear systems with a large amount of data, do not need information in the battery field, only the historical data of lithium-ion batteries.

Description of drawings

1 is one of the flowcharts of a method for estimating the state of charge of a lithium ion battery based on a bidirectional cyclic neural network provided by the present invention;

2 is a schematic structural diagram of a bidirectional cyclic neural network in a method for estimating the state of charge of a lithium ion battery based on a bidirectional cyclic neural network provided by the present invention;

3 is one of the flow charts of the bidirectional cyclic neural network training process in a method for estimating the state of charge of a lithium ion battery based on a bidirectional cyclic neural network provided by the present invention;

4 is the second flow chart of a bidirectional cyclic neural network training process in a method for estimating the state of charge of a lithium ion battery based on a bidirectional cyclic neural network provided by the present invention;

5 is the third flowchart of the bidirectional cyclic neural network training process in a method for estimating the state of charge of a lithium ion battery based on a bidirectional cyclic neural network provided by the present invention;

6 is the second flowchart of a method for estimating the state of charge of a lithium-ion battery based on a bidirectional cyclic neural network provided by the present invention;

Fig. 7 is the effect diagram of applying the bidirectional recurrent neural network in the present invention to the US06 data set under the condition of 45°C;

8 is a graph of the mean absolute error of the bidirectional recurrent neural network in the present invention applied to the US06 data set under the condition of 45°C;

Fig. 9 is the effect diagram of the bidirectional recurrent neural network of the present invention applied to the BJDST data set under the condition of 45°C;

FIG. 10 is a graph of the mean absolute error of the bidirectional recurrent neural network of the present invention applied to the BJDST data set under the condition of 45°C.

Detailed ways

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are merely illustrative and not intended to limit the present invention.

The present invention selects the data set of the INR 18650-20R type lithium ion battery, the data set is the data collected under three temperature conditions of 0°C, 25°C and 45°C, and different simulations are applied to the lithium ion battery under the same temperature conditions. The loads of the car driving state, including US06, FUDS, DST and BJSDT, the details of lithium-ion batteries are shown in Table 1, and the details of the dataset are shown in Table 2.

Table 1 Specific parameters of INR 18650-20R lithium ion battery

Table 2 INR 18650-20R Li-ion battery dataset

Among them, each data set contains the full charging and full discharging process of the lithium-ion battery. There are two experiments under each data set. One is to apply a simulated load from the remaining 80% of the lithium-ion battery capacity to the end of discharge. The other is to apply a simulated load from the remaining 50% of the battery capacity to the end of discharge, using the former as training data and the latter as real-time data to test the effect of the present invention.

A method for estimating the state of charge of a lithium ion battery based on a bidirectional cyclic neural network provided by the present invention, as shown in FIG. 1 , includes the following steps:

S1: Obtain the battery voltage value, battery current value and battery surface temperature value of the lithium-ion battery at the current moment;

S2: perform data sampling processing, data standardization processing and data dimension change processing on the acquired data at the current moment;

S3: Input the processed data at the current moment into the trained bidirectional recurrent neural network to obtain the state of charge value of the lithium-ion battery at the current moment.

The method for estimating the state of charge of the lithium ion battery provided by the present invention utilizes the data generated in real time by the lithium ion battery, and uses the trained bidirectional cyclic neural network model to obtain the real-time state of charge value of the lithium ion battery. The bidirectional cyclic neural network model is in After the training is completed, the state of charge value can be estimated in real time, which is very convenient. The two-way recurrent neural network can fully consider the characteristics of time series data, and use the data before the current result and the data after the current result, which is more effective than the one-way recurrent neural network. The effect of the network is more accurate, and the estimation of the state of charge of lithium-ion batteries has great potential, which is very suitable for the estimation of the state of charge of lithium-ion batteries. The invention belongs to a data-driven method, does not require complicated electrochemical-related knowledge, and completely starts from data, can effectively extract the information expressed by the historical data of the lithium ion battery, model the discharge characteristics of the lithium ion battery, and obtain accurate charge Electric state estimation results, and can process complex nonlinear systems with a large amount of data, do not need information in the battery field, only the historical data of lithium-ion batteries.

In a specific implementation, in the above-mentioned method for estimating the state of charge of a lithium-ion battery provided by the present invention, the specific structure of the bidirectional recurrent neural network may be: an input layer, a hidden layer and an output layer; wherein, the hidden layer adopts a bidirectional long short-term memory The network (Long Short-Term Memory, LSTM) layer is followed by a full connection layer (Full Connection, FC); among them, the bidirectional LSTM structure can be superimposed, that is, the bidirectional LSTM layer can include several bidirectional LSTM network structures connected to each other. The structure is shown in Figure 2. In Figure 2, X _t , X _t+1 ...X _t+k are input data, and Y _t , Y _t+1 ...Y _t+k are output data. The specific parameter configuration of the bidirectional cyclic neural network is different due to the data under different working conditions. The present invention takes the time step as 40-50, the number of hidden units as 64, and the bidirectional LSTM structure stacking two layers as an example to illustrate.

During specific implementation, in the above-mentioned method for estimating the state of charge of a lithium-ion battery provided by the present invention, the trained bidirectional cyclic neural network can be obtained by using the historical data of the lithium-ion battery and using bidirectional cyclic neural network training. The training process of the recurrent neural network, as shown in Figure 3, can include the following steps:

S11: Manually select the features input to the bidirectional recurrent neural network, including battery voltage value, battery current value and battery surface temperature value;

S12: Collect battery voltage value, battery current value, battery surface temperature value and state of charge value under different working conditions;

Specifically, different working conditions may include different temperatures, different loads and other conditions; the collection of battery voltage value, battery current value, battery surface temperature value and state of charge value needs to be collected using special equipment respectively;

S13: Perform data sampling processing on the collected battery voltage value, battery current value, battery surface temperature value and state-of-charge value, and perform data standardization processing on the battery voltage value, battery current value and battery surface temperature value after data sampling processing. Data dimension change processing; in this way, the processed data can be directly input into the bidirectional recurrent neural network for training;

S14: Initialize the bidirectional cyclic neural network, input the processed battery voltage value, battery current value and battery surface temperature value under the same working conditions as the data to be measured into the initialized bidirectional cyclic neural network, and use the time-based post-processing The propagation algorithm is trained, and the network hyperparameters are continuously adjusted to obtain a trained bidirectional recurrent neural network.

In a specific implementation, in the above-mentioned method for estimating the state of charge of a lithium-ion battery provided by the present invention, in step S13, data sampling processing is performed on the collected battery voltage value, battery current value, battery surface temperature value and state of charge value, and the The battery voltage value, battery current value and battery surface temperature value after data sampling processing are subjected to data standardization processing and data dimension change processing, as shown in Figure 4, which may specifically include the following steps:

S131: Under different working conditions, re-sample the battery voltage value, battery current value, battery surface temperature value and state of charge value, and set the data interval to 1s to generate a data point;

S132: Standardize the resampled battery voltage value, battery current value and battery surface temperature value, so that the battery voltage value, battery current value and battery surface temperature value are all distributed in the [0,1] interval. The formula is:

Among them, D represents any one of the battery voltage value, battery current value and battery surface temperature value, D _t represents the data at time t, D _min represents the smallest data point, and D _max represents the largest data point;

Specifically, the battery voltage value, battery current value and battery surface temperature value are distributed in the [0,1] interval, which is beneficial to the training of the bidirectional recurrent neural network; at the same time, the standardized reference value can also be retained, so that when estimating in real time The same standardization method is adopted to ensure the consistency of data distribution;

S133: Perform dimension change processing on the battery voltage value, battery current value and battery surface temperature value after the normalization process, and connect the battery voltage value, battery current value and battery surface temperature value at each time point in the normalized data as A vector [V, I, T] connecting data points at k time steps as a sample input data for a bidirectional recurrent neural network [[V _t ,I _t ,T _t ],[V _t+1 ,I _t+1 ,T _t+1 ],...,[V _t+k-1 ,I _t+k-1 ,T _t+k-1 ]], and finally get all sample input data [number of samples, time steps, number of features] , so that it can be directly fed into the bidirectional recurrent neural network for training.

In specific implementation, in the above-mentioned method for estimating the state of charge of a lithium-ion battery provided by the present invention, step S14: initialize a bidirectional cyclic neural network, and compare the processed battery voltage value, battery current value and battery surface temperature value with the value to be measured. The data is the initialized bidirectional recurrent neural network with the data input under the same working conditions. The time-based back propagation algorithm is used for training, and the network hyperparameters are continuously adjusted to obtain the trained bidirectional recurrent neural network, as shown in Figure 5. The specific Can include the following steps:

S141: Initialize each parameter value of the bidirectional cyclic neural network, and randomly set each parameter value to any value in the [0,1] interval;

S142: Input all sample input data [sample number, time step, feature number] into the bidirectional cyclic neural network, and through the forward propagation of the bidirectional cyclic neural network, calculate the predicted value of the state of charge at the current moment, and calculate the state of charge at the current moment. The distance between the predicted value and the real value of the state of charge at the current moment, to obtain the distance between the predicted value of the state of charge of all samples and the real value of the state of charge, the calculation formula is as follows:

表示荷电状态的预测值，m表示样本数量，i表示样本序号；where y represents the true value of the state of charge,

Represents the predicted value of the state of charge, m represents the number of samples, and i represents the sample number;

Specifically, the distance between the predicted value of the state of charge at the current moment and the real value of the state of charge at the current moment is calculated, that is, the difference is squared; the real value of the state of charge at the current moment is the resampled state of charge value;

S143: Update each parameter value of the bidirectional recurrent neural network by using gradient descent and backpropagation algorithms;

Steps S142 to S143 are repeated until the bidirectional recurrent neural network converges and the training is completed.

Preferably, in order to speed up the training speed of the bidirectional recurrent neural network and improve the effect accuracy of the bidirectional recurrent neural network, the Adam optimization method and the mini-batch optimization method can be used, and the batch size can be set to 128. The number of times can be set to 2000 times.

In specific implementation, in the above-mentioned method for estimating the state of charge of a lithium-ion battery provided by the present invention, in step S2, data sampling processing, data normalization processing and data dimension change processing are performed on the acquired data at the current moment, as shown in FIG. 6 . , which may include the following steps:

S21: Re-sample the battery voltage value, battery current value and battery surface temperature value, and set the data interval to 1s to generate a data point;

S22: Standardize the sampled data, so that the battery voltage value, battery current value and battery surface temperature value are all distributed in the [0,1] interval. The formula for normalization is:

Among them, D represents any one of the battery voltage value, battery current value and battery surface temperature value, D _t represents the data at time t, D _min represents the smallest data point, and D _max represents the largest data point;

Specifically, the standardized benchmark values retained during network training can be used, so that the same standardized method can be used for network training and real-time estimation, thereby ensuring the consistency of data distribution;

S23: Perform dimension change processing on the normalized data, connect the battery voltage value, battery current value and battery surface temperature value at each time point in the normalized data as a vector [V, I, T], connect k The data points of the time steps are connected as a sample input data of the bidirectional recurrent neural network [[V _t ,I _t ,T _t ],[V _t+1 ,I _t+1 ,T _t+1 ],…,[ V _t+k-1 ,I _t+k-1 ,T _t+k-1 ]], and finally get all sample input data [number of samples, time steps, number of features]. In summary, using the same processing method as the historical data during network training to process the acquired data at the current moment, the distribution of the data input to the trained bidirectional recurrent neural network can be kept consistent, thereby ensuring lithium ion Accuracy of battery state-of-charge estimates.

Table 3 shows the effects of four data sets US06, FUDS, DST and BJSDT under three temperature conditions of 0°C, 25°C, and 45°C, and the evaluation standard is Mean Absolute Error (MAE).

Table 3 Effects of four datasets under three temperature conditions

It can be seen from Table 3 that the above-mentioned method for estimating the state of charge of lithium-ion batteries provided by the present invention has excellent effect. Less than 1%, especially at 45 °C, the MAE values of the four datasets US06, FUDS, DST and BJDST are all the smallest, and the effect is the best. As shown in Figures 7 and 9, the predicted values of the SOC of the two datasets US06 and BJDST almost coincide with the real values, as shown in Figures 8 and 10, the MAE values of the two datasets US06 and BJDST are both less than 1% , which shows that the above-mentioned lithium-ion battery state of charge estimation method provided by the present invention is very accurate on the two data sets US06 and BJDST, which has reached a very high standard in practical applications. Applicability in the field of battery SOC prediction.

The method for estimating the state of charge of the lithium ion battery provided by the present invention utilizes the data generated in real time by the lithium ion battery, and uses the trained bidirectional cyclic neural network model to obtain the real-time state of charge value of the lithium ion battery. The bidirectional cyclic neural network model is in After the training is completed, the state of charge value can be estimated in real time, which is very convenient. The two-way recurrent neural network can fully consider the characteristics of time series data, and use the data before the current result and the data after the current result, which is more effective than the one-way recurrent neural network. The effect of the network is more accurate, and the estimation of the state of charge of lithium-ion batteries has great potential, which is very suitable for the estimation of the state of charge of lithium-ion batteries. The invention belongs to a data-driven method, does not require complicated electrochemical-related knowledge, and completely starts from data, can effectively extract the information expressed by the historical data of the lithium ion battery, model the discharge characteristics of the lithium ion battery, and obtain accurate charge Electric state estimation results, and can process complex nonlinear systems with a large amount of data, do not need information in the battery field, only the historical data of lithium-ion batteries.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit and scope of the invention. Thus, provided that these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims (5)

1. A lithium ion battery state of charge estimation method based on a bidirectional cyclic neural network is characterized by comprising the following steps:

s1: acquiring a battery voltage value, a battery current value and a battery surface temperature value of the lithium ion battery at the current moment;

s2: performing data sampling processing, data standardization processing and data dimension change processing on the acquired data at the current moment;

s3: and inputting the processed data at the current moment into the trained bidirectional circulation neural network to obtain the charge state value of the lithium ion battery at the current moment.

2. The lithium ion battery state of charge estimation method of claim 1, wherein the training process of the bi-directional recurrent neural network comprises the steps of:

s11: manually selecting characteristics input to the bidirectional recurrent neural network, including a battery voltage value, a battery current value and a battery surface temperature value;

s12: under different working conditions, collecting a battery voltage value, a battery current value, a battery surface temperature value and a charge state value;

s13: carrying out data sampling processing on the acquired battery voltage value, battery current value, battery surface temperature value and state of charge value, and carrying out data standardization processing and data dimension change processing on the battery voltage value, battery current value and battery surface temperature value after the data sampling processing;

s14: initializing the bidirectional cyclic neural network, inputting data under the same working condition with the data to be measured in the processed battery voltage value, battery current value and battery surface temperature value into the initialized bidirectional cyclic neural network, training by utilizing a time-based back propagation algorithm, and continuously adjusting network hyper-parameters to obtain the trained bidirectional cyclic neural network.

3. The lithium ion battery state of charge estimation method of claim 2, wherein in step S13, the data sampling processing is performed on the collected battery voltage value, battery current value, battery surface temperature value and state of charge value, and the data standardization processing and data dimension change processing are performed on the battery voltage value, battery current value and battery surface temperature value after the data sampling processing, specifically comprising the following steps:

s131: under different working conditions, sampling the battery voltage value, the battery current value, the battery surface temperature value and the state of charge value again, and setting the data interval to be 1s to generate a data point;

s132: standardizing the battery voltage value, the battery current value and the battery surface temperature value after resampling to ensure that the battery voltage value, the battery current value and the battery surface temperature value are all distributed in a [0,1] interval, wherein the formula of the standardization is as follows:

wherein D represents any one of a battery voltage value, a battery current value and a battery surface temperature value_tData representing time t, D_minRepresents the smallest data point, D_maxRepresents the largest data point;

s133: dimension change processing is carried out on the battery voltage value, the battery current value and the battery surface temperature value after the standardization processing, and the battery voltage value, the battery current value and the battery surface temperature value at each time point in the data after the standardization processing are connected as a vector [ V, I, T]Connecting the data points of k time steps as a sample input data [ V ] of the bidirectional cyclic neural network_t,I_t,T_t],[V_t+1,I_t+1,T_t+1],……,[V_t+k-1,I_t+k-1,T_t+k-1]]Finally obtaining all sample input data [ sample number, time step, characteristic number]。

4. The lithium ion battery state of charge estimation method of claim 3, wherein step S14, initializing the bidirectional cyclic neural network, inputting the data of the processed battery voltage value, battery current value and battery surface temperature value under the same working condition as the data to be measured into the initialized bidirectional cyclic neural network, training by using a time-based back propagation algorithm, and continuously adjusting network hyper-parameters to obtain the trained bidirectional cyclic neural network, specifically comprising the following steps:

s141: initializing each parameter value of the bidirectional cyclic neural network, and randomly setting each parameter value as any numerical value in an interval of [0,1 ];

s142: inputting all sample input data [ sample number, time step and characteristic number ] into the bidirectional circulation neural network, calculating a predicted value of the SOC at the current moment through forward propagation of the bidirectional circulation neural network, calculating the distance between the predicted value of the SOC at the current moment and the actual value of the SOC at the current moment, and solving the distance between the predicted value of the SOC and the actual value of the SOC of all samples, wherein the calculation formula is as follows:

where y represents the true value of the state of charge,representing a predicted value of the state of charge, m representing the number of samples, and i representing the serial number of the samples;

s143: updating each parameter value of the bidirectional cyclic neural network by using a gradient descent and back propagation algorithm;

and repeating the step S142 to the step S143 until the bidirectional circulation neural network is converged, and finishing the training.

5. The method for estimating the state of charge of the lithium ion battery according to any one of claims 1 to 4, wherein the step S2 is to perform data sampling processing, data normalization processing and data dimension change processing on the acquired data at the current time, and specifically includes the following steps:

s21: sampling the battery voltage value, the battery current value and the battery surface temperature value again, and setting the data interval to be 1s to generate a data point;

s22: carrying out standardization processing on the sampled data to enable the battery voltage value, the battery current value and the battery surface temperature value to be distributed in a [0,1] interval, wherein the formula of the standardization processing is as follows:

wherein D represents any one of a battery voltage value, a battery current value and a battery surface temperature value_tData representing time t, D_minRepresents the smallest data point, D_maxRepresents the largest data point;

s23: performing dimension change processing on the normalized data, and connecting the battery voltage value, the battery current value and the battery surface temperature value of each time point in the normalized data into a vector [ V, I, T ]]Connecting the data points of k time steps as a sample input data [ V ] of the bidirectional cyclic neural network_t,I_t,T_t],[V_t+1,I_t+1,T_t+1],……,[V_t+k-1,I_t+k-1,T_t+k-1]]Finally obtaining all sample input data [ sample number, time step, characteristic number]。

Images