KR20230029088A - Battery SOC Estimation method and system using Artificial Neural Networks and Vehicle Simulator - Google Patents

Abstract

The present invention provides a method for estimating the SOC of a battery, which includes: a vehicle driving simulator for sensing the voltage and current of a battery which changes as the vehicle is driven, and generating and transmitting discharge data including the sensing data of the voltage and current of the battery; and a battery SOC estimator for estimating the battery SOC by learning by inputting the discharge data as parameters of an artificial neural network.

Description

Battery SOC Estimation method and system using Artificial Neural Networks and Vehicle Simulator

The present invention relates to a technology for estimating the SOC of a battery, and more particularly, to acquire discharge data consisting of a voltage and current of a battery that changes according to vehicle driving, and a required discharge time (T) in real time, and for the discharge data It relates to a vehicle driving simulator capable of improving battery SOC estimation accuracy by learning artificial neural networks, and a method and system for estimating SOC of a battery using artificial neural networks.

Currently, lithium-ion batteries have high voltage density, can maintain output for a long time, and are light in weight. It is used as a key driving force.

However, due to the characteristics of batteries that are frequently charged and discharged, overcharging and overdischarging may occur, and in this case, there is a risk of system down or battery explosion. In order to solve such a problem, research on a battery management system (BMS) that accurately grasps the state of the battery and manages it so that the battery can be operated efficiently and stably is underway.

State Of Charge (SOC), one of the parameters that can be used in the BMS, is an important measure indicating the remaining capacity of the battery. indicate Therefore, it is necessary to accurately predict the SOC for stable and efficient battery operation.

Conventional SOC estimation methods include a current integration method, an open circuit voltage (OCV), and a Kalman filter measurement method. The current integration method has an advantage in that it is easy to implement in a manner in which changes in charging and discharging currents are added after specifying an arbitrary initial SOC. However, if there is an error in the initial SOC setting, errors are continuously accumulated, making it difficult to accurately estimate the SOC. In addition, the OCV estimates the SOC by measuring the voltage in a stable state in which current does not flow to the battery, but it is difficult to estimate in real time because a long time must be waited for the measurement to reach an equilibrium state. In addition, the extended Kalman filter method is advantageous in estimating the state of a nonlinear model, but there is a problem in that the error increases significantly if the initial state setting is incorrect during linearization, and related research is still in progress.

And, as one of the recent SOC estimation methods, there is a big data-based method. As a representative machine learning technique, there is an SOC estimation method through artificial neural network (ANN) learning. The artificial neural network is a learning method designed by imitating the brain structure of a biological model, and is used in various fields such as pattern recognition, identification, and classification, and can efficiently learn input and output relationships. The estimation method using the artificial neural network has the advantage that it does not have to consider internal electrical and chemical changes of the battery, is advantageous for estimating a nonlinear model, and can operate even in a low-end processor.

However, it is difficult to estimate the SOC of a battery applied to a vehicle because the voltage and current of the battery change according to the vehicle driving time or driving environment.

Accordingly, development of a technology capable of accurately estimating the SOC of a battery has been required in the prior art even when the voltage and current of the battery change according to the driving time of the vehicle and the driving environment of the vehicle.

Korean Patent Publication No. 1020190081379 Korean Patent Publication No. 1020070043150 Korean Patent Publication No. 1020090042367

The present invention can improve the estimation accuracy of the battery SOC by acquiring discharge data composed of the voltage and current of the battery, which changes according to vehicle driving, and the required discharge time (T) in real time, and learning the artificial neural network for the discharge data. Its purpose is to provide a method and system for estimating the SOC of a battery using a vehicle driving simulator and an artificial neural network.

To this end, one embodiment according to the present invention, sensing the voltage and current of the battery that changes according to the vehicle driving; generating discharge data including sensing data of voltage and current of the battery; and estimating the SOC of the battery by inputting and learning the discharge data as a parameter of an artificial neural network.

Another embodiment according to the present invention is a vehicle driving simulator that senses voltage and current of a battery that changes according to vehicle driving to generate and transmit discharge data including the sensing data of the voltage and current of the battery; and a battery SOC estimator for estimating the battery SOC by learning by inputting the discharge data as a parameter of the artificial neural network.

In addition, the driving of the vehicle is characterized in that it is performed according to a previously determined driving cycle.

In addition, the discharge data further includes a required discharge time, characterized in that the discharge required time is defined as a value that increases in correspondence with the number of acquisitions of voltage and current sensing data.

In addition, the vehicle driving is performed multiple times according to a predetermined driving cycle until the fully charged battery is discharged.

Discharge data obtained as the fully charged battery is driven multiple times until it is discharged is input to an input layer of the artificial neural network and learned.

In addition, the artificial neural network is a multilayer neural network and further uses a backpropagation method,

The backpropagation method is characterized in that an error is reduced by updating weights for parameters in a direction from an output layer to an input layer.

In addition, the artificial neural network is a multilayer neural network,

It includes a plurality of hidden layers, and the activation function of the hidden layer is ReLU,

The learning label value is characterized in that the current data obtained by the discharge experiment is calculated through the current integration method after designating an arbitrary initial SOC.

The present invention described above acquires discharge data composed of the voltage and current of the battery, which changes according to vehicle driving, and the required discharge time (T) in real time, and improves the estimation accuracy of the battery SOC by learning the artificial neural network for the discharge data. It provides possible effects.

1 is a block diagram of an apparatus for estimating SOC of a battery using a vehicle driving simulator and an artificial neural network according to a preferred embodiment of the present invention.
2 is a diagram showing the configuration and embodiment of a vehicle driving simulator according to a preferred embodiment of the present invention.
3 is a diagram showing a travel cycle graph of FTP-75.
4 is a diagram showing the configuration of a battery SOC estimator 200 according to a preferred embodiment of the present invention.
5 is a flowchart of a method for estimating SOC of a battery according to a preferred embodiment of the present invention;
6 is a diagram illustrating a multilayer neural network according to a preferred embodiment of the present invention.
7 is a schematic representation of the concept of a backpropagation algorithm;
8 is a diagram illustrating an experimental environment of the present invention.
9(a) to (d) are graphs of the voltage change of each battery for one cycle discharge experiment.
10 is a diagram defining the required discharge time.
11 is a diagram showing the structure of a multilayer neural network model according to the present invention.
Fig. 12 shows a definition graph of ReLU.
13 is a diagram showing SOC errors of each model in which learning is performed without adding a discharge time (T) parameter.
14 is a diagram showing the SOC error of each model in which learning is performed by adding a discharge time (T) parameter as an input.
15 to 18 are diagrams showing SOC error graphs of batteries of a model using a discharge time (T) as an input parameter and a model that does not use a discharge time (T) as an input parameter.

The present invention creates a vehicle driving simulator to check the SOC change of the battery according to the output of the actual vehicle, and then FTP-75 (Federal Test Procedure-75), a certified driving cycle used to measure the fuel efficiency of a vehicle in the United States. ), the driving situation is implemented, and then the battery discharge simulation is performed. As a result, the voltage, current, and discharge time (T) of the battery that change according to vehicle driving are checked in real time and discharge data is acquired, and the battery voltage and current, discharge time (T) that changes according to vehicle driving The artificial neural network learns the information to estimate the battery SOC.

As described above, the present invention can improve performance by significantly reducing the SOC estimation error of the battery by estimating the SOC of the battery by learning the voltage, current, and discharge time information of the battery with an artificial neural network.

In addition, since the present invention estimates the SOC based on the current that changes according to the RPM of the vehicle driving and the battery voltage that changes accordingly, instead of discharging at a constant current, it can be applied to actual vehicle driving situations.

A method and system for estimating SOC of a battery using a vehicle driving simulator and an artificial neural network according to a preferred embodiment of the present invention will be described in detail with reference to the drawings.

<Configuration of battery SOC estimator using vehicle driving simulator and artificial neural network>

1 is a block diagram of an apparatus for estimating SOC of a battery using a vehicle driving simulator and an artificial neural network according to a preferred embodiment of the present invention. Referring to FIG. 1, it is composed of a battery unit 10, a vehicle 20, a vehicle driving simulator 100, and a battery SOC estimator 200.

The battery unit 10 is configured by connecting a plurality of batteries in series and supplies driving power to the vehicle 20 .

The vehicle 20 is provided with a vehicle driving simulator 100, and the vehicle driving simulator 100 senses in real time the voltage and current of the battery unit 10, and the required discharge time T, which change according to the driving of the vehicle. The obtained discharge data is provided to the battery SOC estimator 200. The vehicle driving simulator 100 implements a driving situation based on FTP-75 (Federal Test Procedure-75), a certified driving cycle used to measure fuel efficiency of a vehicle in an experimental or learning environment, thereby discharging the battery. run the simulation However, when applied to an actual vehicle, discharge simulation is performed according to the driving conditions of the vehicle, enabling simulation to indicate the driver's propensity of the vehicle.

The battery SOC estimator 200 calculates and outputs the SOC after learning discharge data provided from the vehicle driving simulator 100 through a multilayer neural network. In particular, the battery SOC estimator 200 may employ an arithmetic system such as a computer in an experimental environment or a learning environment. However, when applied to an actual vehicle, it may be configured to be calculated through the ECU (Electronic Control Unit) of the vehicle and guided to the driver through a display device.

<Configuration of vehicle driving simulator 100>

The configuration and embodiment of the vehicle driving simulator 100 will be described in more detail with reference to FIG. 2 .

FIG. 2 shows an example implemented in a RC Car (Radio Control Car) for experimentation and learning. FIG. 2 (a) is a configuration diagram of the vehicle driving simulator 100, and FIG. 2 (b) is a vehicle It shows an implementation example of the driving simulator 100.

The vehicle driving simulator 100 includes an RC Car (Radio Control Car) frame 102, a DC converter 104, a motor driver 106, first and second motors 108 and 110, a sensor 120, and a control device ( 122), a communication device 124, and wheels 112 to 118.

The frame 102 uses a square steel frame and an MC nylon plate to prevent each part of the vehicle driving simulator 100 from moving or losing balance due to vibration of the first and second motors 108100 during discharge simulation. do.

Some of the wheels 112 to 118 rotate by receiving driving force directly from the first and second motors 108 and 110 to model a 2WD vehicle.

The battery unit 10 is configured by connecting four fully charged 4.2V batteries in series. The DC converter 104 receives power from the battery 10 and converts it into a rated voltage for driving the first and second motors 108 and 110 serving as the DC motors and outputs the converted voltage. The motor driver 106 adjusts the RPM of the first and second motors 108 and 110 under the control of the controller 114 .

The first and second motors 108 and 110 are driven under the control of the motor driver 106 to provide driving force to some of the wheels 112 to 118 to rotate them.

The sensor 120 detects the current of the battery 10 that changes according to the driving of the first and second motors 108 and 110, the corresponding voltage, and the required discharge time, and provides them to the controller 122.

The controller 122 controls the RPM of the first and second motors 108 and 110 by controlling the motor driver 106 according to a predetermined driving period of the simulation. It is set up similarly to the FTP-75 used specifically for city and city driving cycles. The FTP-75 is defined by the US Environmental Protection Agency, and is also called CVS-75 (Constant Volume Sampler-75).

3 shows a travel cycle graph of the FTP-75. Referring to FIG. 3, the driving cycle includes three driving phases such as a cold start, a stabilized phase, and a hot start, and one soaking period.

In addition, the control device 122 operates the first and second motors 108 and 110 according to a predetermined driving cycle, and the current of the battery 10 that changes according to the driving of the first and second motors 108 and 110 and A battery SOC estimator 200 located outside by detecting the corresponding voltage and discharge time through the sensor 120 and generating discharge data with the current, voltage, discharge time, etc. detected from the sensor 120 provided by

Since the vehicle driving simulator 100 is manufactured so that it cannot drive real distances, the RPM of the first and second motors 108 and 110, the 3rd gear ratio and final reduction ratio of the actual vehicle, and tire specifications (section width, aspect ratio, wheel size) The actual vehicle speed is simulated according to the driving cycle of the FTP-75 using , and the vehicle speed in this case can be calculated according to Equation 1.

는 타이어의 단면폭(Section width),

는 타이어의 편평비(Flat ratio), s는 타이어의 크기(size),

는 차량의 기어비(Gear ratio), c는 차량의 종감속비(Comprehensive reduction cost)이다. 상기의 시뮬레이터의 모델로 사용한 차량은 현대자동차의 아반떼 스포츠(AD) 16년식이며, 타이어의 규격은225/40/18을 사용한다. 상기 아반떼 스포츠(AD)의 3단 기어비와 종감속비는 각각 1.294와 4.467이다. In Equation 1 above,

is the section width of the tire,

is the flat ratio of the tire, s is the size of the tire,

is the gear ratio of the vehicle, and c is the comprehensive reduction cost of the vehicle. The vehicle used as the model of the above simulator is Hyundai Motor Company's Avante Sports (AD) 16 year model, and the tire size is 225/40/18. The third gear ratio and final reduction ratio of the Avante Sport (AD) are 1.294 and 4.467, respectively.

<Configuration of battery SOC estimator 200>

4 is a diagram showing the configuration of a battery SOC estimator 200 according to a preferred embodiment of the present invention. Referring to FIG. 4 , the battery SOC estimator 200 includes a communication device 202, a control device 204, a memory unit 206, a display unit 208, and a user interface unit 210.

The communication device 202 is responsible for communication between the vehicle driving simulator 100 and the control device 204 .

The control device 204 not only controls each part of the battery SOC estimator 200 as a whole, but also receives the discharge data provided by the vehicle driving simulator 100 according to a preferred embodiment of the present invention to artificially The SOC of the battery is estimated by the neural network, and the result is output through the display unit 208 to guide the user.

The memory unit 206 stores various information including the control program of the control device 204, and particularly stores various information for SOC estimation according to the present invention.

The display unit 208 displays information according to the control of the control device 204 and guides the user.

The user interface unit 210 provides various types of information input by the user to the control device 204 .

<Procedure of battery SOC estimation method>

Now, a battery SOC estimation method according to the present invention will be described with reference to the drawings.

The present invention estimates the SOC of a battery using an artificial neural network and a vehicle driving simulator. Equipment used in the vehicle driving simulator includes four lithium ion batteries with a rated capacity of 1300 mAh connected in series, a power supply, a battery chamber, a vehicle driving simulation, a cell balancing module, a voltage sensor, and a current sensor Arduino. The power supply is used for fully charging the battery at 4.2V, and the battery chamber is used for minimizing safety accidents during charging and discharging.

5 is a flowchart of a method for estimating SOC of a battery according to a preferred embodiment of the present invention. Referring to FIG. 5, four batteries connected in series are fully charged (step 300). Thereafter, the vehicle driving simulator 100 discharges the battery by driving the first and second motors according to a predetermined driving cycle (step 302). While the battery is discharging, the vehicle driving simulator 100 senses the current, voltage, and discharging time of the battery, and configures discharging data by combining the sensed result data (step 304). Then, the vehicle driving simulator 100 transmits the discharge data to the battery SOC estimator 200 (step 306).

The battery SOC estimator 200 uses the discharge data as parameters of the artificial neural network to estimate the SOC of each battery with a total of four artificial neural network models (step 308).

<Multilayer Neural Network>

Since the apparatus 200 for estimating SOC of a battery according to the present invention estimates SOC using an artificial neural network model, the artificial neural network will be described in more detail.

In the present invention, first, discharge data obtained through battery discharge was input to a multilayer neural network among several neural networks and learned. The multilayer neural network is as shown in FIG. 6 . The multi-layer neural network is a neural network that compensates for the disadvantages of the perceptron, which can only perform limited learning due to the constraint that it is composed of only linear functions by adding one or more hidden layers between the input layer and the output layer in the single-layer perceptron. In the single-layer perceptron, weights are updated only by forward propagation. The forward propagation method means that parameters are sequentially calculated from the input layer to the output layer, and the weights are updated to obtain the result values. Unlike the single layer perceptron, the multilayer neural network uses forward propagation and back propagation methods, and the back propagation method gradually reduces errors while updating weights for parameters in the direction from the output layer to the input layer, contrary to the forward propagation method.

Here, the steps of the backpropagation method are largely classified into four steps.

First, the first step is to calculate the output value using the existing weights.

And step 2 is a step of subtracting the partial differential value of the error with each weight from the existing weight.

And step 3 is a step of performing step 2 for all weights.

Finally, step 4 is a step in which steps 1 to 3 are repeated as many times as the preset learning count.

7 is a schematic representation of the concept of the above backpropagation algorithm. Due to this learning method of the multilayer neural network, the multilayer neural network according to the present invention can be expressed more complexly in a nonlinear form, unlike the single layer perceptron, and is advantageous for classification and numerical prediction.

A process of updating the weights of the backpropagation algorithm follows Equations 2 to 6.

First, input values and output values used in the forward propagation calculation follow Equations 2 and 3.

는 노드

의 입력값이고,

는

번째 노드에 들어오는

번째 노드의 가중치이고,

는 활성화 함수이다. 상기

,

는

와

에 들어오는 이전 노드의 출력값이고

는 입력 바이어스 값이다. 오차에 따라 가중치를 업데이트할 때 필요한

의 값은 수학식 4 및 수학식 5에 따라 구할 수 있다. In Equations 2 and 3 above,

is the node

is the input value of

Is

entering the first node

is the weight of the th node,

is the activation function. remind

,

Is

and

is the output value of the previous node entering

is the input bias value. Required when updating weights based on errors

The value of can be obtained according to Equations 4 and 5.

는 라벨 값이고,

는 출력값이다. 상기의 식들로

를 사용하여 가중치를 업데이트 할 수 있으며 이는 수학식 6에 따른다.In Equation 4 and Equation 5 above,

is the label value,

is the output value. with the above formulas

The weights can be updated using Equation 6.

는 시간 인덱스,

는 학습률이다.In Equation 6 above

is the time index,

is the learning rate.

<Experimental process for battery SOC estimation method using vehicle driving simulator and artificial neural network>

Experimental procedures for the SOC estimation method of a battery using the vehicle driving simulator and artificial neural network according to the present invention described above will be described.

First, the actual experimental environment is as illustrated in FIG. 8 . In the experiment according to the present invention, four batteries with a rated capacity of 1300 mAh connected in series were used as a power source of the vehicle driving simulator. Although the same amount of current flows through all the batteries connected in series, the state of charge of the battery varies depending on the internal chemical characteristics and operating environment of each battery while charging and discharging several times, and charge imbalance occurs. Accordingly, in the experiment, a cell balancing module was used that equalizes the voltage of the series-connected batteries in order to prevent charge imbalance occurring during charging and discharging.

Before operating the vehicle driving simulator, four lithium ion batteries are fully charged with a constant voltage of 4.2V using a power supply, and this case is defined as 100% SOC. After completion of such charging, a stabilization time of 1 hour is given to create an equilibrium state. After that, the series-connected batteries are discharged using a vehicle driving simulator.

9(a) to (d) show graphs of voltage changes of each battery in a 1-cycle discharge experiment. As illustrated in (a) to (d) of FIG. 9 , a period from the starting point of the discharge experiment until some of the four batteries connected in series are discharged and the operation of the vehicle driving simulator stops is defined as one cycle. After that, the experiment process is repeated to sense and obtain voltage, current, and discharge time (T) from cycle 1 to cycle 10, and then transmit to a PC used as a battery SOC estimator and provide them as input parameters of the multilayer neural network. The discharge time T can be defined as shown in FIG. 10 as accumulated information of time passing during the discharge experiment. That is, the required discharge time (T) is defined as a value that increases in correspondence with the number of acquisitions of voltage and current sensing data.

In order to calculate the SOC to be used as the learning label value of the multilayer neural network, after individually charging the four batteries in a fully discharged state, an arbitrary initial SOC of each battery was designated, and the current data obtained by the discharge experiment was calculated through the current integration method. . The calculation formula according to the current integration method is shown in Equation 7.

In Equation 7, SOC(t) is the SOC at time t, I(t) is the current at time t, SOC(0) is the initial SOC, and C is the rated capacity of the battery.

11 shows the structure of a multilayer neural network model according to the present invention. Four multilayer neural network models consisting of 2, 3, 4, and 5 voltage and current data were used for the multilayer neural network model, and discharge time parameters (T) were used and not used in the four models. Classification was carried out.

The hidden layers of each of the multilayer neural network models have a total of two layers, and the number of nodes in the first hidden layer is 128 and the number of nodes in the second hidden layer is 64, and the ReLU (Rectified Linear Unit) function is used as the activation function of the hidden layer. . The ReLU is an activation function that is used most recently and is employed to improve the gradient loss problem of sigmoid and hyperbolic tangent.

12 is ReLU, which can be expressed linearly when a positive value and converges to 0 for a negative value, so calculation is very simple and learning speed is very fast. One node of the output layer of each model was used, and learning was repeated 15,000 times.

<Experiment result>

Now, the results of the above experiments will be described.

SOC estimation was performed with four models in which the number of voltage and current input parameters were configured differently in the multilayer neural network, and then SOC estimation was performed by additionally using a discharge time parameter (T).

Table 1 and FIG. 13 show the SOC error of each model that was trained without adding the discharge time parameter (T). The model using two of the above voltage and current parameters was labeled as Model 1, the model using three parameters as Model 2, the model using four parameters as Model 3, and the model using five parameters as Model 4.

Cell numbers
Input models Cell 1 Cell 2 Cell 3 Cell 4 Model 1 3.86% 3.94% 5.26% 3.88% Model 2 3.65% 3.55% 5.1% 2.92% Model 3 3.86% 3.66% 5.19% 3.58% Model 4 2.88% 2.71% 3.9% 2.52%

Table 1 is a table showing SOC errors according to the number of current input parameters.

The error of each of the batteries is calculated as MAE (Mean Absolute Error), which follows Equation 8.

은 계산할 데이터의 양을 의미하며

는 전류 적산법을 사용하여 계산된 SOC,

은 다층신경망 모델에 의해 추정된 SOC를 의미한다.In Equation 8 above,

is the amount of data to be calculated

is the SOC calculated using the current integration method,

denotes the SOC estimated by the multilayer neural network model.

The error of the model with two input parameters is 3.86% for battery 1, 3.94% for battery 2, 5.26% for battery 3, and 3.88% for battery 4. The error is 3.65% for battery 1, 3.55% for battery 2, 5.1% for battery 3, and 2.92% for battery 4. 3.66%; It was estimated at 2.52%. Model 4 in Table 1 is a model using five voltage and current parameters, and it can be seen that the error is generally small among the four models. The minimum error of the multilayer neural network model without using the discharge time is 2.52% and the maximum error is 5.26%.

Table 2 and FIG. 14 show the SOC error of each model that was trained by adding the discharge time (T) parameter as an input. As in Table 1 and FIG. 13, the error was calculated through MAE, and Model 1 used 2 models of voltage and current parameters, Model 2 used 3 models, Model 3 used 4 models, and Model 3 used 5 parameters respectively. was marked as Model 4.

Cell numbers
Input models Cell 1 Cell 2 Cell 3 Cell 4 Model 1 1.97% 1.83% 1.83% 1.99% Model 2 1.61% 2.06% 1.65% 2.15% Model 3 1.72% 1.79% 1.62% 1.76% Model 4 2.05% 1.84% 2.23% 1.82%

Table 2 shows the SOC error according to the number of voltage and current input parameters after adding the required discharge time (T) as a parameter.

When Table 1 and Table 2 were compared, the multilayer neural network with the discharge time (T) added because the model with the discharge time (T) added as an input parameter had a smaller overall error than the model that was trained without adding it. It can be determined that the model has relatively good estimation performance.

In addition, the error of the model with four input parameters of voltage and current is 1.72% for battery 1, 1.79% for battery 2, 1.62% for battery 3, and 1.76% for battery 4. Although the error showed a difference according to the number of models, it was confirmed that the error was generally small compared to the other number of models. In the case of a model with five input parameters, the number of parameters is relatively large compared to other models, and the distance between the data is large, so it is judged that the learning efficiency is rather low. The minimum error of the model using the discharge time (T) parameter is 1.61%, and the maximum error is 2.15%. 15 to 18 show SOC error graphs of batteries of a model using a discharge time (T) as an input parameter and a model that does not use a discharge time (T) as an input parameter.

That is, as a result of learning by adding the discharge time (T) as an input parameter to the multilayer neural network model, the lowest SOC error was 1.61% of the model using three voltage and current parameters, Jiechao Lv [Jiechao Lv, Baochen Jiang , Xiaoli Wang, Yirong Liu, and Yucheng Fu, "Estimation of the State of Charge of Lithium Batteries Based on Adaptive Unscented Kalman Filter Algorithm", Electronics, Vol. 9, no. 9, 1425, Sep. 2020. https://doi.org/10.3390/electronics9091425.]. and J. H. Park [J. H. Park, J. H. Lee, S. J. Kim, and I, S. Lee, "Real-Time State of Charge Estimation for Each Cell of Lithium Battery Pack Using Neural Networks", Applied Sciences, Vol. 10, no. 23, 8644, 8644, Dec. 2020. https://doi.org/10.3390/app10238644.] The SOC error of the model using voltage data as eight input parameters in the multilayer neural network is 2.8% for battery 1 and 3.1 for battery 2 at high temperature (40 degrees). %, battery 3 is 3.4%, battery 4 is 1.1%, and in Table 2 according to the present invention, the SOC errors of Model 3 are 1.72% for battery 1, 1.79% for battery 2, 1.62% for battery 3, and 1.76 for battery 4 %am.

And in Table 2 above, the SOC error of Model 3 is less than 2% for all four batteries, J. H. Park [J. H. Park, J. H. Lee, S. J. Kim, and I, S. Lee, "Real-Time State of Charge Estimation for Each Cell of Lithium Battery Pack Using Neural Networks", Applied Sciences, Vol. 10, no. 23, 8644, 8644, Dec. 2020. https://doi.org/10.3390/app10238644.] When compared with the model using voltage data as eight input parameters in the multilayer neural network, the overall error was small, so the actual SOC estimation could be more accurate.

Finally, among the multilayer neural networks using the discharge time (T) parameter, it is judged that the estimation performance of the one using three and four voltage and current parameters, respectively, is relatively excellent.

As described above, the present invention produced a vehicle driving simulator to estimate the SOC change according to the actual vehicle output, and conducted a discharge experiment using FTP-75, which is used as a driving cycle test mode in the city center, as the driving cycle of the simulator. . After that, through the acquired voltage and current data, it was determined that the number of input parameters of the multilayer neural network was 2, 3, 4, and 5, respectively, with and without adding the discharge time (T) parameter to the four models. A method of estimating SOC after classification and learning was proposed. First, the lowest error of the model without adding the discharge time (T) was confirmed to be 2.52%. It was confirmed that the minimum errors of the multilayer neural network model using the discharge time (T) as an input parameter and the model without it were 1.61% and 2.52%, respectively, with a difference of 0.91%. Also, when Tables 1 and 2 were compared, Since the SOC estimation error of all batteries was reduced in all models, it was confirmed through experiments that the SOC estimation performance was better when the discharge time (T) was used as an input.

Lastly, in the present invention, during the discharge experiment, the current that changes according to the RPM output of the simulator and the resulting change in battery voltage are confirmed and obtained, and the SOC is estimated through a multi-layer neural network, so actual vehicle driving is not a method of discharging at a constant constant current. It will be applicable to the situation.

As described above, the technical ideas described in the embodiments of the present invention may be implemented independently, or may be implemented in combination with each other. In addition, the present invention has been described through the embodiments described in the drawings and detailed description of the invention, but these are only exemplary, and various modifications and other equivalent embodiments may be made by those skilled in the art in the art to which the present invention belongs. possible. Therefore, the technical protection scope of the present invention should be defined by the appended claims.

10: battery part
20: vehicle
100: vehicle driving simulator
200: battery SOC estimator

Claims (8)

In the battery SOC estimation method,
sensing voltage and current of a battery that change according to vehicle driving;
generating discharge data including sensing data of voltage and current of the battery;
and estimating SOC of the battery by inputting the discharge data as a parameter of an artificial neural network and learning the result. According to claim 1,
The driving of the vehicle is
A method for estimating SOC of a battery, characterized in that it is performed according to a predetermined driving cycle. According to claim 1,
The discharge data further includes a discharge time,
The method for estimating SOC of a battery, characterized in that the discharge required time is defined as a value that increases in correspondence with the number of acquisitions of voltage and current sensing data. According to claim 1 or 3,
The driving of the vehicle is
It is performed multiple times according to a predetermined driving cycle until the fully charged battery is discharged.
The method of estimating the SOC of a battery, characterized in that the discharge data obtained as the fully charged battery is driven multiple times until it is discharged is input to the input layer of the artificial neural network and learned. In the battery SOC estimation system,
a vehicle driving simulator that senses the voltage and current of the battery, which change according to vehicle driving, and generates and transmits discharge data including the sensing data of the voltage and current of the battery; and
A SOC estimating system using a vehicle driving simulator and an artificial neural network, characterized in that it comprises a; battery SOC estimator for estimating the battery SOC by inputting the discharge data as a parameter of the artificial neural network and learning it. According to claim 5,
The driving of the vehicle is
SOC estimation system using a vehicle driving simulator and an artificial neural network, characterized in that it is performed according to a predetermined driving cycle. According to claim 5,
The discharge data further includes a discharge time,
The discharge time required is defined as a value that increases corresponding to the number of acquisitions of voltage and current sensing data. According to claim 5 or 7,
The driving of the vehicle is
It is performed multiple times according to a predetermined driving cycle until the fully charged battery is discharged.
The SOC estimation system using a vehicle driving simulator and an artificial neural network, characterized in that the discharge data obtained as the driving until the fully charged battery is discharged is performed a number of times is input to the input layer of the artificial neural network and learned.

Images