KR20220161981A - Smart battery management system using artificial intelligence module - Google Patents

Abstract

The present invention relates to a smart battery management system (S-BMS) using an artificial intelligence module. The smart battery management system integrates and manages one or more batteries and includes an artificial intelligence module estimating the SOC of a battery by using an artificial neural network (ANN) consisting of an input layer, a hidden layer, and an output layer, wherein each layer consists of nodes (neurons) and weights connecting the nodes (neurons). SOC(k) is obtained by building an input node selecting at least one from output SOC(k-1) of HKF or voltage difference V_L(k)-V_L(k-1) per cycle in voltage V_L(k), current I_L(k), and temperature T(k) measured at cycle k at an output terminal of a lithium battery pack by the ANN input layer. The present invention has the advantage of providing the S-BMS using an artificial intelligence module, which allows a BMS to be trusted by overcoming SOC estimation error limitations of an artificial neural network (ANN) applied to the BMS.

Description

Smart battery management system using artificial intelligence module

The present invention relates to a smart battery management system using an artificial intelligence module, and more specifically, to a smart battery management system that improves SOC prediction accuracy by pioneering an input layer of an artificial intelligence module ANN.

Currently, lithium-ion batteries (LiBs) are widely used in energy storage devices in smart grids and electric vehicles, and the battery state of charge (SOC) represents the usable battery capacity to optimize performance and extend the lifespan of LiBs. It is one of the most important conditions to monitor.

Energy storage has emerged as a major concern for modern smart grids and electric vehicles (EVs), and lithium-ion batteries (LiBs) are already commercially available due to their advantages of large capacity and long cycle life.

Mature lithium-ion batteries (LiBs) for consumer electronics applications are positioning themselves as the leading technology platform for plug-in hybrid electric vehicles (PHEVs) and all EVs, providing energy storage, load leveling and frequency regulation in the context of the smart grid. It is widely used in large facilities for support.

In systems such as electric vehicles, electric ships, and energy storage devices powered by lithium-ion batteries, BMS (Battery Management System) plays a role in ensuring the safety and reliability of batteries that supply power to the system.

In other words, for a lithium-ion battery to operate predictably, a battery management system (BMS) must be designed to monitor and control the battery's energy level.

BMS is generally defined as any electronic device that manages a rechargeable battery (cell or battery pack), the main goal of BMS is monitoring, calculation, communication, protection and optimization of the battery, and estimation of state of charge (SOC) of battery is BMS is one of the important features of

If the state of charge (SOC) continues to be too low or too high, the battery deteriorates faster than when the SOC is maintained at an intermediate level. Maintaining it is very important.

However, accurate estimation of SOC is very complicated because it depends on many nonlinear factors.

SOC is defined as the current usable capacity in a battery divided by its normal capacity. Many problems with LiBs, such as poor battery performance, increased maintenance costs, accelerated aging, catastrophic device failure, and even dangerous accidents, are related to incorrect SOC estimation. have.

Therefore, accurate estimation of SOC in the BMS is a very important issue for battery performance optimization, including extending battery life and preventing permanent damage to the battery.

Although accurate SOC plays an important role in reliable BMS, there is no way to directly measure SOC, and it can be estimated using battery-specific relationships or control theory using direct measurement parameters such as voltage, current, and temperature. have.

However, in practice, the battery SOC depends on many factors such as current, temperature, battery capacitance, internal resistance, battery life, open circuit voltage and service history, which are non-linear functions of all parameters, so accurate SOC estimation is very difficult. .

In order to solve this problem, various techniques have been proposed to estimate the SOC of a battery cell or battery pack, and the main techniques include a discharge test, a current integration method, an open circuit voltage measurement, a resistivity measurement, and an intelligent SOC estimation method.

In addition, computational intelligence technologies such as artificial neural networks (ANN), fuzzy logic, and Kalman filters (KF) have been developed to be applied to BMS for SOC estimation, and compared to other technologies, using simple current and voltage measurement values, more precisely SOC has the advantage of being able to estimate .

In addition, intelligent technology can be easily implemented in digital electronic systems and has the advantage of being able to respond to the time-invariant operation of batteries through self-learning capabilities.

As such technology, in "Battery state estimation method, device and computer program for performing the method (registration number: 10-2016715)", Ampere-hour Counting Method and Kalman Filter are combined (hybrid type A method capable of improving the accuracy of estimating the state of a battery using a Kalman filter method), a device for performing the method, and a computer program are disclosed.

To summarize what has been described in the above-mentioned literature, there is a conventional method of estimating the SOC of a battery using an ampere-hour counting method. The current integration method is a method of measuring battery SOC by converting an actual current value into a standard current value using the Peukert equation as shown in [Equation 1] below and integrating the value.

Here, SOC ₀ is the initial state of charge of the battery, Q _n is the nominal capacity of the battery, I is the current flowing through the battery in real time over time (t), and a is the current efficiency factor calculated and inferred by the Peukert equation (generally 1 is set).

As shown in FIG. 2 , a conventional battery state estimation device may be modeled as a Thevenin equivalent circuit by reflecting electrochemical characteristics of a battery.

In FIG. 1, V _oc is the open circuit voltage, I(t) is the current, R _o is the internal resistance of the battery, R _p is the polarization resistance, C _p is the polarization capacity, V _L is the output terminal voltage, I _L means the output stage current.

[Equation 1] below can be obtained by applying Kirchhoff's law and the current integration method to the equivalent circuit of the battery.

This current integration method has the advantage of being easily implemented, but has problems of initial value and accumulation of battery SOC estimation errors due to sensor noise, requiring periodic calibration such as resetting the battery SOC when the battery is fully charged.

In other words, since the battery SOC is estimated by the current measured by the current sensor installed in the battery's charge/discharge path, accurate sensing of the current sensor is very important, but the current sensor measures the actual current due to performance degradation or degradation. There are problems in that a current offset that is different from the measured current value occurs, and an error range between an actual battery SOC and an estimated battery SOC increases as time passes due to the accumulation of such current offset.

3 is a graph of SOC prediction results of an electric vehicle battery by the current integration method, and the maximum error of the estimated SOC is 17% and the average error is 2.6%.

In order to solve the above problems, a method using a Kalman filter (KF) algorithm has been studied.

The Kalman filter (KF) configures a battery equivalent circuit as shown in FIG. 2, calculates a system equation including the SOC of the battery and the polarization capacity and resistance of the battery equivalent circuit as state variables, and calculates the observed SOC of the battery as an output variable. A step of estimating the SOC of the battery at a predetermined cycle by calculating a measurement equation including a measurement current for the battery, applying the system equation and the measurement equation to a Kalman filter (KF) algorithm.

Here, the system equation and the measurement equation can be expressed as [Equation 2] and [Equation 3] below, respectively.

Here, S(k+1) and u _c (k+1) are SOC and polarization capacity of the battery at the current stage, that is, predicted values of SOC and polarization capacity of the battery, respectively, and S(k) and u _c (k) are These are the SOC and battery polarization capacity of the previous step, respectively. In addition, I(k) represents the previous battery current, w ₁ (k) and w ₂ (k) are the noise of the system, v(k) is the voltage measurement noise, and R ₁ , R ₂ and C are the battery equivalents, respectively. These are the internal resistance, polarization resistance and polarization capacity of the circuit.

The above "Battery State Estimation Method, Apparatus and Computer Program for Performing the Method (Korean Patent Registration No.: 10-2016715)" is a hybrid Kalman filter combining the current integration method and the Kalman filter to improve the accuracy of battery state estimation ( HKF) is a method of predicting battery SOC.

That is, the hybrid Kalman filter (HKF) applies the current integration method by further including the current observation noise to the measured current in the step of calculating the measurement equation among the SOC prediction methods using the Kalman filter (KF), and the following [Equation 4] It is to calculate the measurement equation such as

where z _k is the measurement equation, S(t ₀ ) is the initial SOC value, Q _n is the battery nominal capacity, and H(k) is the current measurement noise.

An artificial neural network (ANN) applied to BMS for SOC estimation is widely known as shown in Korean Patent Publication No. 10-2021-0045830 'Method and Apparatus for Monitoring State of Battery'.

However, although accurate estimation of the SOC as described above is an important factor in reliability of the BMS, the known technologies each have a problem in that they have a predetermined limit of estimation error.

Republic of Korea Patent Registration No. 10-2016715 'Method for estimating battery state, device and computer program for performing the method' Korean Patent Publication No. 10-2021-0045830 'Method and device for monitoring battery condition'

The present invention is to solve the above problems, and an object of the present invention is to provide a smart battery management system using an artificial intelligence module that makes the BMS reliable by overcoming the SOC estimation error limit of an artificial neural network (ANN) applied to the BMS.

In order to achieve the above object, the present invention provides an input layer, a hidden layer, and an output in a smart battery management system (S-BMS) for integrated management of one or more batteries. An artificial intelligence module for estimating the SOC of a battery using an artificial neural network (ANN) composed of layers, each of which is composed of nodes (neurons) and weights connecting each of the nodes (neurons). It is configured to include, and the voltage V _L (k), current I _L (k), temperature T (k), and voltage difference per cycle V _L (k) - The technical summary is a smart battery management system using an artificial intelligence module, which is composed of V _L (k-1) and calculates SOC (k) in the output layer.

Here, at the battery pack output stage, a Thevenin equivalent circuit is modeled with output current and voltage, and a hybrid method for estimating the SOC of the battery at a predetermined cycle by combining an ampere-hour counting method and a Kalman filter. Characterized in that a hybrid Kalman filter (HKF) is further included, and the full cycle output value SOC (k-1) of the hybrid Kalman filter (HKF) is further added to the ANN input layer It is desirable to be a smart battery management system using an artificial intelligence module.

The present invention is also composed of an input layer, a hidden layer, and an output layer in a smart battery management system (S-BMS) for integrated management of one or more batteries, and the Each of the layers includes an artificial intelligence module for estimating SOC(k) of the battery using an artificial neural network (ANN) composed of nodes (neurons) and weights connecting the nodes (neurons); It is connected to the output terminal of the battery pack, models a Thevenin equivalent circuit with output current and voltage, and combines the ampere-hour counting method and the Kalman filter to determine the SOC (k) of the battery at a preset cycle. It is configured to include a hybrid Kalman filter (HKF) that estimates; voltage V _L (k), current I _L (k), and temperature T measured by the ANN input layer at period k at the lithium battery pack output terminal. (k) and the full cycle output SOC (k-1) of the Hybrid Kalman Filter (HKF), and smart battery management using an artificial intelligence module characterized in that SOC (k) is obtained in the output layer. system as another technical point.

According to the present invention described above, there is an advantage in providing a smart battery management system using an artificial intelligence module that makes the BMS reliable by overcoming the SOC estimation error limit of the artificial neural network (ANN) applied to the BMS.

1 is a graph of SOC prediction results of an electric vehicle battery by the prior art current integration method
2 is a basic configuration diagram of an ANN neuron applied to the present invention
3 is a Thevenin equivalent circuit diagram of a battery
Figure 4 is a configuration diagram of one embodiment of the present invention in which 4 nodes are applied to the ANN input layer
5 is a graph of SOC prediction results of the electric vehicle battery according to FIG. 4
6 is a configuration diagram of another embodiment of the present invention in which 4 nodes are applied to the ANN input layer
7 is a graph of SOC prediction results of the electric vehicle battery according to FIG. 6
8 is a configuration diagram of another embodiment of the present invention in which 5 nodes are applied to the ANN input layer
9 is a graph of SOC prediction results of the electric vehicle battery according to FIG. 8

Hereinafter, an embodiment of the present invention will be described in detail based on the accompanying drawings FIGS. 1 to 9. On the other hand, in the drawings and detailed description, drawings and descriptions of configurations and functions that can be easily understood by those skilled in the art, such as general BMS, SOC, Kalman filter, and hybrid Kalman filter, are simplified or omitted.

In particular, in the drawings and detailed description, detailed descriptions and illustrations of specific technical configurations and actions of elements not directly related to the technical features of the present invention are omitted, and only the technical configurations related to the present invention are briefly shown or described. did

The present invention is composed of an input layer, a hidden layer, and an output layer in a smart battery management system (S-BMS) for integrated management of one or more batteries, and the layers Each includes an artificial intelligence module that estimates the SOC of the battery using an artificial neural network (ANN) composed of nodes (neurons) and weights connecting each of the nodes (neurons), and the ANN It is a technology that improves the SOC measurement accuracy of the output layer by pioneering the input layer.

The ANN is a computational model composed of artificial neurons, which are a series of individual processing units interconnected by weights, and is a general-purpose approximation capable of modeling any nonlinear function with desired accuracy.

The network is arranged in layers where the input layer receives input and the output layer generates output, and the middle layer is called a hidden layer because it is not connected to the outside world.

2 is a basic configuration of neurons in a network,

로 표시되고, The input signal vector is

is indicated by

로 표시되며, Neuron weights are

is indicated by

net is the multiplicative response of the weights to the input signal,

b1 is the activation function of an external parameter called bias,

ym is the neuron's output response.

As shown in FIG. 3, the input layer of the artificial neural network (ANN) for SOC estimation at the level of the prior art is the output voltage V _L , current I _L , and temperature T measured and input in the Thevenin equivalent circuit of the battery.

In the present invention, when the measured voltage V _L (k)-V _L (k-1) is greater than 0, the SOC estimation accuracy is improved by reflecting the fact that the voltage is being charged and is being discharged when it is smaller.

That is, as shown in FIG. 4, the voltage V _L (k), current I _L (k), and temperature T ( k) and the voltage difference per cycle V _L (k)-V _L (k-1) were input to obtain SOC(k) at the output layer.

5 is 1000 using four nodes including the voltage difference V _L (k)-V _L (k-1) per period of FIG. 4 in the input layer of the artificial neural network (ANN) technology applied to the BMS for SOC estimation. As the result obtained by learning the number of battery samples, the maximum error of the SOC estimated by ANN is 2% and the average error is 0.4%, and it can be seen that there is a greater error reduction effect than the estimation by the current integration method in FIG. can

The present invention is connected to the battery pack output terminal and models a Thevenin equivalent circuit with the output voltage V _L and current I _L as shown in FIG. , and configured to include a hybrid Kalman filter (100: Hybrid Kalman Filter, HKF) that estimates the SOC (k) of the battery at a preset period, and as shown in FIG. 6, the input layer of the ANN (200) is lithium Voltage V _L (k), current I _L (k), temperature T (k) measured at cycle k at the battery pack output stage, and full cycle output SOC (k-1) of the Hybrid Kalman Filter (HKF) SOC(k) can be obtained in the output layer using four nodes consisting of

7 is the result obtained by learning 1000 battery samples, and the maximum error of the SOC estimated by ANN is 2.1%, and the average error is 0.24%. , and it can be seen that there is a greater error reduction effect than the estimation by the current integration method of FIG. 1 .

Although the results of FIGS. 5 and 7 satisfy the SOC estimation target values at the current technical level, as described above, the accurate estimation of SOC is a very important factor in trusting the BMS, so these error limits still need to be overcome. can be regarded as the target value.

In order to overcome this error limit, the present invention provides the voltage difference V _L (k)-V _L (k-1) per cycle and the previous cycle to the input layer of the artificial neural network (ANN) for SOC estimation, as shown in FIG. The error was corrected using 5 nodes to which the output SOC value of the HKF of was added.

9 is the result of randomly testing 1000 battery samples related to this, and the maximum error of the SOC estimated by ANN is 0.78% and the average error is 0.09%. In comparison, it can be seen that there is a drastic error reduction effect.

As described above, although the smart battery management system for fire prevention according to an embodiment of the present invention is shown according to the above description and drawings, this is only an example and is within the scope of not departing from the technical spirit of the present invention. It will be appreciated by those skilled in the art that various changes and modifications are possible.

100: Hybrid Kalman Filter (HKF)
200: artificial neural network circuit ANN (artificial neural network)

Claims (3)

In a smart battery management system (S-BMS) that integrates and manages one or more batteries,
It consists of an input layer, a hidden layer, and an output layer, and each of the layers consists of nodes (neurons) and weights connecting the nodes (neurons). An artificial neural network (ANN) It is configured to include an artificial intelligence module that estimates the SOC of the battery using a network),
Voltage V _L (k), current I _L (k), temperature T (k), and voltage difference per cycle V _L (k)-V _L (k- 1), and a smart battery management system using an artificial intelligence module, characterized in that SOC (k) is obtained from the output layer. According to claim 1
At the output stage of the battery pack, a Thevenin equivalent circuit is modeled with output current and voltage, and a hybrid Kalman filter that estimates the SOC of the battery at a preset cycle by combining an ampere-hour counting method and a Kalman filter. (Hybrid Kalman Filter, HKF) is further included,
Smart battery management system using an artificial intelligence module, characterized in that the full cycle output value SOC (k-1) of the hybrid Kalman filter (HKF) is further added to the ANN input layer. In a smart battery management system (S-BMS) that integrates and manages one or more batteries,
It consists of an input layer, a hidden layer, and an output layer, and each of the layers is composed of nodes (neurons) and weights connecting the nodes (neurons). An artificial neural (ANN) An artificial intelligence module for estimating the SOC (k) of the battery using a network);
It is connected to the output terminal of the battery pack, models a Thevenin equivalent circuit with output current and voltage, and combines the ampere-hour counting method and the Kalman filter to determine the SOC (k) of the battery at a preset cycle. Estimating a hybrid Kalman filter (Hybrid Kalman Filter, HKF);
It consists of,
Voltage V _L (k), current I _L (k), temperature T (k) measured by the ANN input layer at period k at the lithium battery pack output stage, and full cycle output of the Hybrid Kalman Filter (HKF) It consists of SOC (k-1),
A smart battery management system using an artificial intelligence module, characterized in that SOC (k) is obtained in the output layer.

Images