Method for improving reliability of SOH / RUL estimation through SOH filtering

Abstract

The present invention relates to a method for improving the reliability of SOH / RUL estimation through SOH filtering. In cases where SOH is estimated on the basis of limited data provided by an electric vehicle or when SOH is calculated and provided using different criteria or methodologies, inconsistencies may arise in SOH or the reliability of SOH is low. To address this issue, the present invention relates to a method for improving the reliability of SOH estimation through SOH filtering, which improves the reliability of SOH estimation by utilizing the change characteristics and physical characteristics of SOH, and further enables accurate calculation of a remaining useful lifetime (RUL) on the basis of the SOH with improved reliability.

Description

A method to improve the reliability of SOH / RUL estimation through SOH filtering

[0001] The present invention relates to a method for improving the reliability of SOH / RUL estimation through SOH filtering, and more particularly, to a method for improving the reliability of SOH / RUL estimation through SOH filtering, which improves the reliability of SOH estimation by utilizing the change characteristics and physical characteristics of SOH, and further enables accurate calculation of RUL (remaining useful lifetime) based on SOH with improved reliability, since SOH is inconsistent or has low reliability when estimating SOH with limited data provided from an electric vehicle or when SOH is calculated and provided using different standards or methods.

[0002] As public concern about environmental pollution and energy conservation grows, electric vehicles, which emit almost no carbon and are more economical than internal combustion engine vehicles, are becoming more widespread.

[0003] Since these electric vehicles are equipped with batteries and use the power supplied by the batteries as their power source, a method to efficiently manage the batteries and improve the safety of the electric vehicles is necessary.

[0004] This is made possible by estimating and providing the state of health (SOH) or remaining useful lifetime (RUL) of the electric vehicle battery.

[0005] SOH is information indicating how much the current performance of the battery has deteriorated compared to the initial performance of the battery, and RUL is estimated based on SOH, which indicates the operating time (remaining life) until the battery needs to be replaced.

[0006] Through SOH or RUL, electric vehicle users can inspect or replace the battery before it malfunctions or stops operating, allowing them to efficiently manage the battery and operate the electric vehicle stably.

[0007] In particular, a device that predicts and provides the SOH of electric vehicles is being developed in line with the recent SDV (software defined vehicle) development trend.

[0008] In this case, SOH can be estimated using limited data or calculated (estimated) using different methods, which can lead to inconsistencies and low reliability. Furthermore, to accurately estimate RUL, the reliability of SOH needs to be improved.

[0009] Meanwhile, the capacity of a battery has a physical characteristic of decreasing gradually and almost linearly up to the knee point, and then decreasing rapidly when the knee point is exceeded. In other words, the SOH calculated using the capacity of the battery also has a characteristic of decreasing gradually and almost linearly up to the knee point, but then decreasing rapidly when the knee point is exceeded.

[0010] Based on the physical characteristics of SOH, the usable area of ​​the battery installed in an electric vehicle is set to the knee point, and the reason why SOH is usually set to 70% to 80% is also due to the knee point.

[0011] Accordingly, the present invention proposes a method for estimating and providing highly reliable RUL by improving the reliability of SOH estimation using the change characteristics and physical characteristics of SOH and estimating RUL using the SOH with improved reliability.

[0012] Next, we will briefly explain the prior art existing in the technical field of the present invention, and then describe the technical details that the present invention seeks to achieve differently from the prior art.

[0013] First, Korean Patent No. 2530223 (May 3, 2023) relates to a method and system for predicting the remaining useful life of a battery, which comprises measuring the operating current during the charge cycle and discharge cycle of the battery, and calculating the capacity decline of the battery based on the measured operating current, maximum SOC, minimum SOC, physical properties of the battery, and capacity in the previous cycle, thereby predicting the remaining useful life (RUL) of the battery.

[0014] That is, Korean Patent No. 2530223 predicts the remaining useful life of a battery by using the deterioration parameters including the maximum SOC, minimum SOC, and physical properties of the battery (limited state of charge) and the operating current of the battery.

[0015] On the other hand, the present invention estimates the RUL (remaining useful life) using the estimated SOH for the battery of an electric vehicle, and improves reliability by using the physical characteristics of the SOH to accurately estimate the RUL. Korean Patent No. 2530223 does not describe, suggest, or imply any technical features of the present invention.

[0016] In addition, Korean Patent No. 1777334 (September 5, 2017) relates to a battery SOH estimation device and method, which receives current sensing data, voltage sensing data, and temperature sensing data of a battery after charging or discharging the battery to calculate a first SOC, receives the current sensing data, voltage sensing data, and temperature sensing data after a subsequent discharge or charge after the charging or discharging to calculate a second SOC, and then calculates the SOH of the battery using the amount of change in the first and second SOC.

[0017] That is, Korean Patent Publication No. 1777334 only estimates SOH by receiving battery measurement data, and does not describe at all a method for improving the reliability of SOH.

[0018] On the other hand, the present invention improves the reliability of SOH by estimating SOH by receiving limited data or filtering SOH calculated by different standards or methods by utilizing the physical characteristics of battery SOH that decrease almost linearly and gradually up to the knee point, and at the same time, enables accurate estimation of RUL through SOH with improved reliability. Korean Patent Publication No. 1777334 does not describe, suggest, or imply any technical features of the present invention.

[0019] The present invention was created to solve the above problems, and its purpose is to provide a method for accurately estimating RUL based on SOH with improved reliability, as well as improving the reliability of SOH estimation by filtering SOH using the physical characteristics of SOH.

[0020] In addition, the present invention aims to provide a method for improving the reliability of SOH estimation by performing SOH filtering by calculating the degree of reflection of the SOH change rate in a long-term cycle section and the SOH change rate in a short-term cycle section according to the physical characteristics of SOH.

[0021] In addition, the present invention aims to provide a method for verifying and filtering the SOH by utilizing the change characteristics of the SOH that cannot increase as the number of cycles of the battery increases and the change characteristics of the SOH that does not change rapidly in short cycles until reaching the knee point.

[0022] In addition, the present invention aims to provide a method for improving the reliability of SOH estimation by filtering SOH by reflecting SOH filtered in a previous cycle, SOH verified by currently received SOH, SOH change rate weighted according to the degree of reflection of SOH change rate in short and long cycles, current cycle number, or a combination thereof.

[0023] In addition, the present invention aims to provide a method for more accurately estimating the RUL of a battery by using SOH whose reliability is improved by performing SOH filtering.

[0024] In addition, the present invention aims to provide a method for calculating a filtered SOH change rate for each cycle in a predetermined cycle range through a filtered SOH, and estimating RUL using the calculated SOH change rate and the SOH corresponding to the knee point of the battery.

[0025] A method for improving the reliability of SOH estimation through SOH filtering according to one embodiment of the present invention includes: an SOH information receiving step for receiving an SOH (state of health) and a cycle number for a battery of an electric vehicle; an SOH verification step for verifying the received SOH using a change characteristic of the SOH; and an SOH filtering step for filtering the verified SOH using a physical characteristic of the SOH, which is verified as satisfying the change characteristic of the SOH; wherein the SOH of the battery gradually decreases within a predetermined range until a knee point, and then rapidly decreases beyond the predetermined range when the knee point is passed, thereby improving the reliability of the SOH estimation.

[0026] The above SOH verification step outputs the received SOH as a verified SOH if the received SOH satisfies a predetermined SOH change characteristic, and outputs the verified SOH by performing a predetermined SOH correction if the received SOH does not satisfy the predetermined SOH change characteristic, and the SOH change characteristic is characterized in that the SOH cannot increase as the number of cycles increases, the SOH does not change more rapidly than a predetermined maximum change value in a predetermined short cycle before reaching a knee point, or a combination thereof.

[0027] The above SOH correction is characterized in that, considering a change characteristic in which the SOH cannot increase as the number of cycles increases, if the received SOH is greater than a previously received SOH, the received SOH is replaced with a previously received SOH, or, considering a change characteristic in which the SOH does not change more rapidly than a predetermined maximum change value in a predetermined short cycle before reaching a knee point, if a change between the received SOH and the previously received SOH is greater than the maximum change value, the received SOH is replaced with a previously received SOH, or a combination thereof.

[0028] In addition, the SOH filtering step further includes a weight estimation step for estimating a weight for the degree of reflection of the SOH change rate for the first cycle section between the initial cycle and the current cycle and the SOH change rate for the second cycle section between the previous cycle and the current cycle when filtering the SOH; and is characterized in that the verified SOH is filtered by calculating a weighted SOH change rate by reflecting the SOH change rates of the first cycle section and the second cycle section according to the estimated weight.

[0029] In addition, the weight estimation step inputs the SOH change rate for the first cycle section, the SOH change rate for the second cycle section, the current cycle number, or a combination thereof into a DNN model for weight estimation to estimate the weight, and the DNN model for weight estimation is characterized in that it is generated by learning each first learning data generated by labeling weights for each first feature data set including the SOH change rate for the first cycle section, the SOH change rate for the second cycle section, the current cycle number, the verified SOH, or a combination thereof.

[0030] In addition, the SOH filtering step is characterized in that the verified SOH is filtered through A x SOH_filtered[t-1], B x SOH[t], L(SOH[t] - ax cycle[t]) or a combination thereof, wherein the SOH_filtered[t-1] is the SOH filtered in the previous cycle, the SOH(t) is the verified SOH, a is the weighted SOH change rate calculated through the estimated weight, cycle[t] is the current cycle number, the A and B are constants, and the L is the gain of the SOH[t] and ax cycle[t].

[0031] In addition, the SOH filtering step further includes a coefficient estimation step of inputting the SOH filtered in the previous cycle, the verified SOH, the calculated weighted SOH change rate, the current cycle number, or a combination thereof into a coefficient estimation DNN model to estimate coefficients including A, B, and L; and the coefficient estimation DNN model is characterized in that it is generated by learning second learning data labeled with A, B, and L for each second feature data set including the SOH filtered in the previous cycle, the verified SOH, the weighted SOH change rate reflecting the SOH change rate for the first cycle section and the SOH change rate for the second cycle section, the current cycle number, or a combination thereof.

[0032] Meanwhile, a method for improving the reliability of RUL estimation through SOH filtering according to another embodiment of the present invention further includes an RUL estimation step for calculating the RUL (remaining useful life) of a battery by using the filtered SOH according to the method for improving the reliability of each SOH estimation, thereby estimating the RUL, wherein the RUL estimation step includes an SOH change rate calculation step for calculating an SOH change rate using the filtered SOH in a predetermined cycle range including a current cycle and at least one previous cycle; and a knee-point cycle number calculation step for calculating a cycle number corresponding to a knee-point of the battery through the calculated SOH change rate and the SOH corresponding to the knee-point of the battery in the predetermined cycle range; and the RUL is calculated by subtracting the current cycle number from the cycle number corresponding to the calculated knee-point.

[0033] Meanwhile, a device for improving the reliability of SOH estimation through SOH filtering according to another embodiment of the present invention is characterized by including: a memory for storing a program implementing a method for improving the reliability of SOH estimation through each SOH filtering; and a processor configured to execute the program stored in the memory.

[0034] In addition, according to another embodiment of the present invention, a device for improving the reliability of RUL estimation through SOH filtering further includes an RUL estimation step for calculating the RUL (remaining useful life) of a battery by using the filtered SOH through the method for improving the reliability of SOH estimation through each SOH filtering, and estimating the RUL, wherein the RUL estimation step includes an SOH change rate calculation step for calculating an SOH change rate using the filtered SOH in a predetermined cycle range including a current cycle and at least one previous cycle; and a knee-point cycle number calculation step for calculating a cycle number corresponding to a knee-point of the battery through the calculated SOH change rate and the SOH corresponding to the knee-point of the battery in the predetermined cycle range; and a memory for storing a program for implementing a method for improving the reliability of RUL estimation through SOH filtering by calculating the RUL by subtracting the current cycle number from the cycle number corresponding to the calculated knee-point; and a processor configured to execute the program stored in the memory.

[0035] As described above, the present invention improves reliability by receiving SOH from an electric vehicle and filtering the SOH by considering the physical characteristics of the SOH, and estimates RUL based on the SOH with improved reliability, thereby providing a user with accurate and reliable SOH / RUL, thereby enabling inspection or replacement of the battery before a problem occurs.

[0036] FIG. 1 is a drawing illustrating the physical characteristics of a battery according to one embodiment of the present invention.

[0037] FIG. 2 is a diagram illustrating a method for improving the reliability of SOH estimation through SOH filtering according to one embodiment of the present invention.

[0038] FIG. 3 is a diagram illustrating the operation of a device for improving the reliability of SOH estimation for filtering SOH according to one embodiment of the present invention.

[0039] FIG. 4 is a diagram illustrating a method for generating a DNN model for weight estimation according to one embodiment of the present invention.

[0040] FIG. 5 is a diagram illustrating a method for generating a DNN model for coefficient estimation according to one embodiment of the present invention.

[0041] FIG. 6 is a diagram illustrating a method for estimating RUL according to one embodiment of the present invention.

[0042] FIG. 7 is a diagram illustrating the operation of a device for improving the reliability of SOH / RUL estimation for estimating RUL according to one embodiment of the present invention.

[0043] FIG. 8 is a block diagram showing the configuration of a device for improving the reliability of SOH / RUL estimation through SOH filtering according to one embodiment of the present invention.

[0044] Fig. 9 is a block diagram showing the configuration of a DNN model generation device according to one embodiment of the present invention.

[0045] FIG. 10 is a flowchart illustrating a procedure for improving the reliability of SOH estimation by filtering SOH according to one embodiment of the present invention.

[0046] FIG. 11 is a flowchart illustrating a procedure for improving the reliability of RUL estimation by estimating RUL using filtered SOH according to one embodiment of the present invention.

[0047] Figure 12 is a flowchart illustrating a procedure for creating a DNN model according to one embodiment of the present invention.

[0048] [Description of symbols] 100: Device for improving reliability of SOH / RUL estimation through SOH filtering; 110: SOH information receiving unit; 120, 220: SOH verification unit; 130, 230: SOH filtering unit; 131: First SOH change rate calculation unit; 132: Weight estimation unit; 133: Weight-reflected SOH change rate calculation unit; 134: Coefficient estimation unit; 140: RUL estimation unit; 141: Second SOH change rate calculation unit; 142: Knee point cycle number calculation unit; 210: SOH information collection unit; 240: First learning data generation unit; 250: Second learning data generation unit; 260: First learning unit; 270: Second learning unit.

[0049] Hereinafter, with reference to the attached drawings, a preferred embodiment of a method for improving the reliability of SOH / RUL estimation through SOH filtering of the present invention will be described in detail. The same reference numerals in each drawing represent the same elements. In addition, specific structural and functional descriptions of the embodiments of the present invention are merely illustrative for the purpose of explaining the embodiments according to the present invention, and unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as generally understood by a person of ordinary skill in the art to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning they have in the context of the related art, and it is preferable not to interpret them in an ideal or excessively formal sense unless explicitly defined herein.

[0050] In addition, since the present invention improves the reliability of SOH through SOH filtering and calculates RUL based on SOH with improved reliability, calculating RUL with SOH with improved reliability ultimately has the same meaning as estimating RUL with improved reliability.

[0051] FIG. 1 is a drawing illustrating the physical characteristics of a battery according to one embodiment of the present invention.

[0052] As illustrated in FIG. 1, the capacity retention of a battery mounted on an electric vehicle according to an embodiment of the present invention exhibits a physical characteristic of gradually decreasing within a predetermined range due to deterioration as the number of charge and discharge cycles increases, and then rapidly decreasing beyond the predetermined range when the knee point is exceeded.

[0053] Even when looking at the battery capacity reduction (degradation data) through an actual battery cycling test and the prediction of capacity reduction through artificial intelligence, it can be seen that the battery capacity gradually decreases up to the knee point and then decreases rapidly after passing the knee point (see Figure 1).

[0054] Here, the AI-based prediction results utilize the Bacon-Watts model, which can be used to identify a battery's knee point. The Bacon-Watts model predicts battery capacity loss by considering linear and nonlinear loss mechanisms, and considers battery usage conditions such as temperature, voltage, current, and charge and discharge rates.

[0055] Additionally, the battery's capacity is used to estimate the State of Health (SOH). SOH can be calculated as the ratio of the current capacity to the battery's manufacturing capacity (i.e., the nominal or original capacity). Therefore, SOH also has the physical characteristic of gradually decreasing within a certain range up to the knee point, but then rapidly decreasing beyond the knee point.

[0056] Due to these characteristics, the SOH of batteries typically installed in electric vehicles is known to reach the knee point when used at approximately 80% of the manufacturing level.

[0057] Furthermore, because electric vehicles typically do not provide SOH, SOH is often estimated. Electric vehicles that estimate battery SOH, or the SOH information providers installed in electric vehicles, each use different criteria and methods for estimating SOH, and the data used for estimating SOH also vary, resulting in inconsistency and low reliability.

[0058] Therefore, the RUL estimated based on the SOH is also inherently unreliable. Low-reliability SOH / RUL means unreliable information about the battery's condition and lifespan. Providing such unreliable SOH / RUL data can expose users to risks due to inability to replace or inspect the battery at the appropriate time.

[0059] Accordingly, the present invention aims to improve the reliability of SOH estimation by filtering SOH by adjusting the degree of reflection of the physical characteristics of SOH and the rate of change of SOH between long-term and short-term cycles.

[0060] FIG. 2 is a diagram illustrating a method for improving the reliability of SOH estimation through SOH filtering according to one embodiment of the present invention.

[0061] As illustrated in FIG. 2, a method for improving the reliability of SOH / RUL estimation through SOH filtering according to an embodiment of the present invention is to improve the reliability of SOH by filtering SOH by reflecting the SOH change rate of the first cycle section (a_long) and the SOH change rate of the second cycle section (a_short) by utilizing the physical characteristics of SOH. Here, a cycle is counted including a charge cycle, a discharge cycle, or a combination thereof.

[0062] Meanwhile, when a device (100) for improving the reliability of SOH / RUL estimation configured to realize a method for improving the reliability of SOH / RUL estimation according to an embodiment of the present invention receives information including SOH and the number of cycles of a battery from an application (e.g., an electric vehicle), it must first verify the SOH using the change characteristics of the SOH.

[0063] The above SOH verification outputs the received SOH as a verified SOH if the received SOH satisfies a predetermined SOH change characteristic, and outputs the received SOH as a verified SOH if the received SOH does not satisfy the predetermined SOH change characteristic, by performing a predetermined SOH correction.

[0064] The above SOH change characteristic means that as the cycle number of the battery increases, the later SOH cannot increase more than the previous SOH, and the SOH does not change to decrease more rapidly than a predetermined maximum change value in a predetermined short cycle before reaching the knee point.

[0065] Therefore, the device (100) for improving the reliability of SOH / RUL estimation considers the change characteristic that SOH cannot increase as the number of cycles of the battery increases, and if the SOH received at a specific point in time is greater than the previously received SOH, replaces the SOH received at the specific point in time with the previously received SOH for correction. In addition, considering the change characteristic that the SOH does not change more rapidly than a predetermined maximum change value in a predetermined short cycle before reaching the knee point, and if the change in SOH is greater than the predetermined maximum change value as a result of comparing the SOH received at the specific point in time with the previously received SOH, replaces the SOH received at the specific point in time with the previously received SOH for correction. Meanwhile, when performing correction, correction may be performed using the previously filtered SOH.

[0066] A device (100) for improving the reliability of SOH / RUL estimation according to one embodiment of the present invention verifies whether the SOH received at a specific point in time does not meet the change characteristics and thus its accuracy is significantly reduced, and if the SOH change characteristics are not satisfied, the device corrects the SOH in advance to improve the accuracy.

[0067] The above verification process checks whether the received SOH violates basic requirements and corrects them. Filtering is performed on SOHs that pass this verification.

[0068] Next, we will explain the SOH change rate utilized in SOH filtering. First, the first cycle period refers to a long-term cycle period that is longer than the second cycle period and may include the period from the initial cycle to the current cycle. The second cycle period refers to a short-term cycle period and may include only one cycle period from the previous cycle to the current cycle.

[0069] The SOH change rate of the first and second cycle sections can be calculated using the following [Mathematical Formula 1].

[0070] [Mathematical Formula 1]

[0071] a_long = (SOH[1] - SOH[t]) / (cycle[1] - cycle[t]),

[0072] a_short = (SOH[t-1] - SOH[t]) / (cycle[t-1] - cycle[t]),

[0073] Here, SOH[t] refers to the SOH at the current time point (t), and cycle[t] refers to the number of cycles at the current time point (t).

[0074] That is, since the first cycle section refers to the section between the first cycle and the current cycle, the SOH change rate of the first cycle section (= slope of SOH change amount for the long-term cycle section) is calculated by calculating the ratio (a / b) of the result (a) of subtracting the SOH received and verified in the current cycle from the SOH received and verified in the first cycle and the result (b) of subtracting the current cycle number from the initial cycle number (i.e., 1).

[0075] In addition, the SOH change rate in the second cycle section (= slope of SOH change amount for the short-term cycle section) is calculated by calculating the ratio (c / d) of the result (c) of subtracting the SOH received and verified in the current cycle from the SOH received and verified in the previous cycle and the result (d) of subtracting the current cycle number from the previous cycle number.

[0076] In addition, the reliability improvement device (100) of the SOH / RUL estimation calculates a weighted SOH change rate reflecting the SOH change rate of the first cycle section and the second cycle section in order to filter the verified SOH.

[0077] The above weighted SOH change rate (a) is calculated using the following [Mathematical Formula 2].

[0078] [Equation 2]

[0079] a = (wx a_long + (1 ?? w) x a_short)

[0080] Here, w refers to a weight for the degree of reflection of the SOH change rate (a_long) of the first cycle section and the SOH change rate (a_short) of the second cycle section. Here, w has a value between 0 and 1.

[0081] That is, the larger the value of w, the higher the SOH change rate in the first cycle section (long-term cycle section) is reflected, and the smaller the value of w, the higher the SOH change rate in the second cycle section (short-term cycle section) is reflected.

[0082] The above weights are estimated through a DNN model for weight estimation, and the DNN model for weight estimation is described in detail with reference to Fig. 4.

[0083] When filtering SOH through the above weighted SOH change rate (a), the filtered SOH can be expressed by the following [Mathematical Formula 3].

[0084] [Equation 3]

[0085] SOH_filtered[t] = wx SOH[t] + (w ?? 1) xax cycle[t]

[0086] The above [Mathematical Formula 3] can be used by changing it into various formulas such as [Mathematical Formula 4] to [Mathematical Formula 7].

[0087] [Equation 4]

[0088] SOH_filtered[t] = wx SOH[t-1] + (w ?? 1) xax cycle[t].

[0089] Here, the current filtered SOH (SOH_filtered[t]) is calculated by adjusting the degree of reflection using the weights obtained by multiplying the SOH in the previous cycle (SOH[t-1]) and the weighted SOH change rate (a) by the current cycle number (cycle[t]). Here, the term obtained by multiplying the weighted SOH change rate (a) by the current cycle number (cycle[t]) represents the SOH change rate reflecting the long-term and short-term cycles. Therefore, the current filtered SOH (SOH_filtered[t]) adjusts the degree of reflection of the previous SOH and the SOH reflecting the long-term and short-term cycles by the weight w. In addition, since the SOH change rate (a_long, a_short) and the weighted SOH change rate (a) have a - value, the term (w-1) is used instead of (1-w).

[0090] [Equation 5]

[0091] SOH_filtered[t] = wx SOH[t] + (w ?? 1) xax cycle[t-1].

[0092] Here, the filtered SOH (SOH_filtered[t]) in the current cycle is calculated by multiplying the SOH (SOH[t]) in the current cycle and the weighted SOH change rate (a) by the number of previous cycles (cycle[t-1]) and adjusting the degree of reflection using another weight.

[0093] [Equation 6]

[0094] SOH_filtered[t] = wx SOH[t-1] + (w ?? 1) xax cycle[t].

[0095] Here, the current filtered SOH (SOH_filtered[t]) is calculated by multiplying the SOH in the previous cycle (SOH[t-1]) and the weighted SOH change rate (a) by the current cycle number (cycle[t]) and adjusting the degree of reflection using another weight.

[0096] [Equation 7]

[0097] SOH_filtered[t] = wx SOH[t-2] + (w ?? 1) xax cycle[t-1].

[0098] Here, the current filtered SOH (SOH_filtered[t]) is calculated by multiplying the SOH in the previous cycle (SOH[t-2]) and the weighted SOH change rate (a) by the number of previous cycles (cycle[t-1]) and adjusting the degree of reflection using another weight.

[0099] Meanwhile, if the voltage, current, and temperature of the battery are all present, the SOH can be estimated based on the equivalent circuit model (ECM) using a Kalman filter. However, since the present invention does not receive voltage, current, and temperature, but only the SOH and cycle count, the SOH is filtered using a Kalman filter using the physical characteristics of the SOH according to the knee point and the cycle count.

[0100] That is, if SOH gradually decreases within a predetermined range up to the knee point, and the verified SOH[t] is used as a measurement value and the ax cycle[t] is used as a prediction value, the reliability improvement device (100) of SOH / RUL estimation can filter the verified SOH by applying it to a Kalman filter or a general state observer as in the following [Mathematical Formula 8], thereby improving the reliability of SOH estimation.

[0101] [Equation 8]

[0102] SOH_filtered[t] = A x SOH_filtered[t-1] + B x SOH[t] + L(SOH[t] - ax cycle[t])

[0103] That is, the reliability of the filtered SOH (SOH_filtered[t]) can be improved by final filtering with A x SOH_filtered[t-1], B x SOH[t], L(SOH[t] - ax cycle[t]) or a combination thereof.

[0104] Here, SOH_filtered[t], SOH_filtered[t-1], SOH[t], a, and cycle[t] represent the SOH filtered in the current cycle (t) (SOH_filtered[t]), the SOH filtered in the previous cycle (t-1) (SOH_filtered[t-1]), the verified SOH (SOH[t]), the weighted SOH change rate (a), and the current cycle number (cycle[t]), respectively.

[0105] In [Mathematical Formula 8], the L(SOH[t] - ax cycle[t]) term can be expressed as L((SOH[t] - ax cycle[t]) if the weight w in (wx SOH[t] + (w - 1) xax cycle[t]) is determined as a specific value L between 0 and 1.

[0106] Therefore, A and B are arbitrary constants, and L represents the verified SOH and the gain for the ax cycle[t]. In addition, A, B, and L must be determined by interlinking the items that reflect the weighted SOH change rate (a) in SOH_filtered[t-1], SOH[t], and SOH[t], respectively.

[0107] The above SOH / RUL estimation reliability enhancement device (100) is configured to filter the verified SOH by estimating A, B, and L through a coefficient estimation DNN model. The coefficient estimation DNN model will be described in detail with reference to FIG. 4.

[0108] FIG. 3 is a diagram illustrating the operation of a device for improving the reliability of SOH / RUL estimation for filtering SOH according to one embodiment of the present invention.

[0109] As illustrated in FIG. 3, the reliability improvement device (100) for SOH / RUL estimation according to one embodiment of the present invention receives SOH information including SOH and cycle number from an SOH information providing device on the application side.

[0110] The reliability enhancement device (100) for the above SOH / RUL estimation first verifies the received SOH by considering the variation characteristics of the SOH. As described above, the verification is performed to improve the accuracy of an SOH that is incorrectly estimated or has low accuracy by considering the variation characteristics of the SOH.

[0111] The reliability enhancement device (100) for the above SOH / RUL estimation calculates the SOH change rate for the first cycle section and the second cycle section. The SOH change rate is calculated using the verified SOH. The first cycle section refers to a predetermined long-term cycle section, and the second cycle section refers to a predetermined short-term cycle section.

[0112] The reliability enhancement device (100) for the above SOH / RUL estimation estimates weights for the SOH change rate of the first cycle section and the SOH change rate of the second cycle section calculated above. The weights are estimated using a DNN model for weight estimation.

[0113] The reliability improvement device (100) of the above SOH / RUL estimation calculates a weighted SOH change rate (a) according to the degree of reflection of the SOH change rate of the first and second cycle sections using the estimated weight.

[0114] The reliability improvement device (100) of the above SOH / RUL estimation calculates the filtered SOH in the current cycle using [Mathematical Formula 8] including the calculated weighted SOH change rate (a), the number of cycles, the filtered SOH in the previous cycle, the verified SOH, or a combination thereof.

[0115] At this time, each coefficient of the above [Mathematical Formula 8] is estimated through a DNN model for coefficient estimation, as described above.

[0116] Below, we will first explain the DNN model for estimating weights (w), and then explain the DNN model for estimating coefficients (A, B, L).

[0117] FIG. 4 is a diagram illustrating a method for generating a DNN model for weight estimation according to one embodiment of the present invention.

[0118] As illustrated in Fig. 4, the DNN model for weight estimation of the present invention is generated through a DNN model generation device (200).

[0119] The above DNN model generation device (200) generates first learning data by labeling a weight (w) to a first feature data set including the SOH change rate of the first cycle section, the SOH change rate and cycle number of the second interval section, the verified SOH, or a combination thereof.

[0120] Here, the SOH change rate and verified SOH of each cycle section are calculated using the same method as described with reference to Fig. 3. That is, the DNN model generation device (200) collects SOH information from the SOH information providing device on the application side to generate first learning data.

[0121] That is, the DNN model generation device (200) generates the first learning data by labeling with weights for deriving a weighted SOH change rate according to the actual SOH, based on the SOH change rate, number of cycles, verified SOH, or a combination thereof for each cycle of the battery.

[0122] The above actual SOH refers to a standard SOH that is measured in advance and established based on the number of cycles for a specific battery of a specific manufacturer, and can be prepared and used in advance in the present invention even if not provided by the manufacturer.

[0123] In addition, the DNN model generation device (200) trains a pre-prepared DNN (deep neural network) (hereinafter referred to as the first DNN) using the first learning data generated above to generate a DNN model for weight estimation.

[0124] That is, when the DNN model generation device (200) filters the SOH received and verified from the application side, it generates a DNN model for weight estimation so that it can estimate weights for extracting the weight-reflected SOH change rate (a) by reflecting the SOH change rate of each cycle section.

[0125] The above first DNN is configured to include an input layer including a plurality of input nodes, a hidden layer including a plurality of hidden nodes, and an output layer including an output node.

[0126] Each input node of the input layer and each hidden node of the hidden layer, and each hidden node of the hidden layer and each output node of the output layer are each connected through a link having a predetermined weight.

[0127] The above input layer is input with input data including the SOH change rate of the first cycle section constituting the first learning data, the SOH change rate of the second cycle section, the number of cycles, the verified SOH, or a combination thereof, configured as an input vector.

[0128] For example, if the SOH change rate of the first cycle section is configured to be input to the first input node of the first DNN, the SOH change rate of the first cycle section in the first learning data is arranged to be input to the first input node.

[0129] The output node of the above output layer is configured to output a learning result (weight) according to the input, and since the DNN model generation device (200) already knows the result (weight) according to the input vector input to the input layer during the learning process, it updates (adjusts) the weight of the link so as to reduce the error between the learning result output during the learning process and the actual weight. Here, the weight of the link refers to the weight of the first DNN, and the weight output as the learning result refers to the weight of the degree of reflection of the SOH change rate of the first and second cycle sections.

[0130] The above learning is performed by updating the weights through the backpropagation method, which allows the error to be reduced by backpropagating the error to the first DNN.

[0131] At this time, the first DNN that has completed learning by learning all the first learning data becomes the DNN model for weight estimation.

[0132] In order to estimate the weight using the above DNN model for weight estimation, the SOH change rate of the first cycle section, the SOH change rate of the second cycle section, the number of cycles, the verified SOH, or a combination thereof are input into the DNN model for weight estimation, and the weight estimation result is output.

[0133] By applying the above estimated weight to [Mathematical Formula 2], the weight-reflected SOH change rate can be calculated.

[0134] Meanwhile, in the present invention, the weight of the degree of reflection of the SOH change rate of each cycle section is pre-generated to be labeled as a weight for deriving a weighted reflection SOH change rate according to the actual SOH.

[0135] In addition, in the present invention, the weights may be estimated and used through a DNN model for weight estimation, but the weights may also be extracted through computer simulation and then the SOH change rate reflecting the weights may be calculated.

[0136] FIG. 5 is a diagram illustrating a method for generating a DNN model for coefficient estimation according to one embodiment of the present invention.

[0137] As shown in Fig. 5, the DNN model for coefficient estimation of the present invention is generated through a DNN model generation device (200).

[0138] The above DNN model generation device (200) generates second learning data by labeling a second feature data set including a weighted SOH change rate, a cycle number, or a combination thereof according to the degree of reflection of the previously filtered SOH, the verified SOH, the SOH change rate of the first and second cycle sections, a coefficient (A) of the previously filtered SOH, a coefficient (B) of the verified SOH, and a coefficient (L) of the result of multiplying the verified SOH and the weighted SOH change rate by the cycle number.

[0139] That is, the DNN model generation device (200) generates second learning data by labeling each coefficient according to the actual SOH based on the previously estimated SOH, the verified SOH, the number of cycles, the weighted SOH change rate, or a combination thereof.

[0140] In addition, the DNN model generation device (200) trains another pre-prepared DNN (deep neural network) (hereinafter, second DNN) using the generated second learning data to generate a DNN model for coefficient estimation.

[0141] The above second DNN is configured to include an input layer including a plurality of input nodes, a hidden layer including a plurality of hidden nodes, and an output layer including an output node.

[0142] In the above second DNN, each input node constituting the input layer is configured to input different data constituting the second learning data.

[0143] Additionally, learning is performed by updating weights through backpropagation, a method that reduces errors by backpropagating them from the second DNN. At this time, the second DNN, which has completed training by learning all of the second training data, becomes the DNN model for coefficient estimation.

[0144] Here, the input of the DNN model for coefficient estimation is the previously filtered SOH, the verified SOH, the weighted SOH change rate according to the degree of reflection of the SOH change rate of each section, the number of cycles, or a combination thereof, and the output is each coefficient (A, B, L).

[0145] In addition, in the present invention, each coefficient may be estimated and used through the coefficient estimation DNN model, but each coefficient may be extracted through computer simulation and then the filtered SOH may be calculated.

[0146] FIG. 6 is a diagram illustrating a method for estimating RUL according to one embodiment of the present invention.

[0147] As illustrated in FIG. 6, a device (100) for improving the reliability of SOH / RUL estimation according to one embodiment of the present invention is configured to improve the reliability of the RUL estimation by estimating the RUL using the filtered SOH.

[0148] Since the above RUL is calculated based on the filtered SOH, the RUL can be estimated by knowing the capacity of the battery and the number of cycles changed in each cycle.

[0149] Since SOH is the ratio of the current capacity to the nominal capacity at the time of manufacturing, the current capacity in each cycle can be calculated using the verified SOH in each cycle. At this time, the number of cycles can also be calculated by multiplying the current capacity by an arbitrary variable (b) (cycle = b x capacity), so ultimately the final cycle (i.e., the number of cycles corresponding to the knee point (i.e., EOL (end of life))) can be calculated by multiplying the arbitrary variable (b) by the SOH corresponding to the knee point.

[0150] Therefore, RUL can be estimated by subtracting the current cycle number (the number of cycles received) from the final cycle number. Here, b can be calculated by dividing the change in capacity by the cycle change within a given cycle range. In this case, since SOH can be calculated as the ratio of the current capacity to the nominal capacity, b can be calculated by calculating the rate of change in SOH within the given cycle range.

[0151] That is, b refers to the SOH change rate (slope) calculated as delta SOH / delta cycle within a given cycle range. The SOH change rate for estimating RUL refers to the change rate for the filtered SOH.

[0152] To this end, the reliability improvement device (100) of the SOH / RUL estimation calculates the SOH change rate (b) in a predetermined cycle range by using the SOH filtered for each cycle in a predetermined cycle range.

[0153] At this time, if the nominal capacity of the battery installed in the electric vehicle is registered in advance, the reliability improvement device (100) for SOH / RUL estimation may calculate the capacity change rate instead of the SOH change rate.

[0154] The above-described cycle range includes the current cycle and at least one previous cycle. In Fig. 6, the cycle range is set to 3. When the cycle range is set to 3, the cycle consists of three cycles, including cycle[t-2], cycle[t-1], and cycle[t]. This cycle range can be set in various ways.

[0155] In addition, since the SOH or capacity corresponding to the knee point is constant for each cell or pack of the battery, the reliability improvement device (100) for SOH / RUL estimation calculates the final number of cycles corresponding to the knee point by multiplying the SOH change rate by the SOH corresponding to the EOL (knee point) (e.g., 0.8).

[0156] That is, the SOH corresponding to the above knee point refers to the maximum SOH according to the above knee point, so for example, if 80% is used, it is considered to have reached the end of its life.

[0157] The result of subtracting the number of cycles received from the number of cycles corresponding to the above-mentioned estimated knee point can be estimated as RUL.

[0158] FIG. 7 is a diagram illustrating the operation of a device for improving the reliability of SOH / RUL estimation for estimating RUL according to one embodiment of the present invention.

[0159] As illustrated in FIG. 7, the reliability improvement device (100) for SOH / RUL estimation according to one embodiment of the present invention improves the reliability of RUL estimation by estimating RUL using filtered SOH.

[0160] The reliability improvement device (100) of the above SOH / RUL estimation calculates the SOH change rate corresponding to a predetermined cycle range by using the SOH filtered for each cycle in a predetermined cycle range.

[0161] That is, the reliability improvement device (100) of SOH / RUL estimation estimates RUL by multiplying the SOH corresponding to the knee point by the SOH change rate to calculate the final number of cycles corresponding to the knee point.

[0162] FIG. 8 is a block diagram showing the configuration of a device for improving the reliability of SOH / RUL estimation through SOH filtering according to one embodiment of the present invention.

[0163] As shown in FIG. 8, a device (100) for improving the reliability of SOH / RUL estimation according to one embodiment of the present invention is configured to include an SOH information receiving unit (110), an SOH verification unit (120), an SOH filtering unit (130), an RUL estimation unit (140), and an alarm unit (150).

[0164] The above SOH information receiving unit (110) receives SOH information including the SOH and cycle count of the battery installed in the application from the application side (e.g., an electric vehicle). The SOH information may be provided from an SOH information providing device on the application side.

[0165] The above SOH verification unit (120) verifies the received SOH by considering the change characteristics of the SOH. The verification and correction of the SOH has been described with reference to FIGS. 2 and 3, and thus will be omitted here.

[0166] The above SOH filtering unit (130) is configured to filter the verified SOH according to the physical characteristics of the SOH in order to improve the reliability of the SOH estimation, and includes a first SOH change rate calculation unit (131), a weight estimation unit (132), a weight-reflecting SOH change rate calculation unit (133), and a coefficient estimation unit (134).

[0167] The above first SOH change rate calculation unit (131) calculates the SOH change rate of the first cycle section and the SOH change rate of the second cycle section based on the current cycle.

[0168] As described above, the first cycle section refers to the cycle section between the first cycle and the current cycle, and the second cycle section refers to the cycle section between the previous cycle and the current cycle.

[0169] Calculating the SOH change rate for each cycle section above is described with reference to Fig. 2, so it is omitted here.

[0170] The above weight estimation unit (132) estimates a weight for the degree of reflection of the SOH change rate for each cycle section when filtering the verified SOH.

[0171] The above weight estimation unit (132) configures the SOH change rate of the first cycle section, the SOH change rate of the second cycle section, the number of received cycles, the verified SOH, or a combination thereof as input data suitable for input to a DNN model for weight estimation, and inputs the configured input data into the DNN model for profit weight estimation to estimate the weight.

[0172] The above weighted SOH change rate calculation unit (133) calculates the weighted SOH change rate, which is the result of reflecting the SOH change rate of each cycle section using the estimated weight. The calculation of the weighted SOH change rate is described with reference to FIG. 2, so it is omitted here.

[0173] The above coefficient estimation unit (134) estimates the coefficient (A) of the filtered SOH for the previous cycle, the coefficient (B) of the verified SOH, and the coefficient (L) of the term obtained by multiplying the verified SOH and the weighted SOH change rate by the current cycle number when filtering the verified SOH in the current cycle using [Mathematical Formula 8].

[0174] The above coefficient estimation unit (134) estimates the coefficient by inputting the SOH verified for the previous cycle, the verified SOH, the weighted SOH change rate, the number of current cycles received, or a combination thereof into the coefficient estimation DNN model, and since it has been described with reference to FIGS. 2, 3, and 5, it will be omitted here.

[0175] Thereafter, the SOH filtering unit (130) improves the reliability of SOH estimation by filtering the verified SOH by applying the estimated coefficient according to [Mathematical Formula 8].

[0176] The above RUL estimation unit (140) is configured to estimate the RUL using the filtered SOH, and includes a second SOH change rate calculation unit (141) and a knee point cycle number calculation unit (142).

[0177] The above second SOH change rate calculation unit (141) calculates the SOH change rate using the SOH filtered for each cycle within a predetermined cycle range.

[0178] The above predetermined cycle range is preset, and calculating the SOH change rate in the cycle range is described with reference to Fig. 6, so it is omitted here.

[0179] The above-mentioned knee point cycle number calculation unit (142) calculates the final cycle number corresponding to the knee point using the SOH change rate calculated through the second SOH change rate calculation unit (141) and the SOH corresponding to the knee point. Calculating the final cycle number is omitted here as it has been described with reference to FIGS. 6 and 7.

[0180] Thereafter, the RUL estimation unit (140) estimates the RUL by subtracting the number of received cycles from the calculated final number of cycles.

[0181] In addition, the reliability estimation device (100) for SOH / RUL estimation provides filtered SOH, estimated RUL, or a combination thereof, so that a driver of an electric vehicle can check the condition of the battery.

[0182] Meanwhile, SOH filtering and RUL estimation are performed before SOH reaches the knee point.

[0183] Therefore, if the received SOH exceeds the knee point, an alarm should be provided to the electric vehicle so that the battery can be replaced quickly.

[0184] Accordingly, if the rate of change between consecutive SOHs exceeds a predetermined threshold rate of change continuously (e.g., three times) more than a predetermined threshold, the alarm unit (150) determines that the battery condition has exceeded the knee point and provides an alarm.

[0185] Fig. 9 is a block diagram showing the configuration of a DNN model generation device according to one embodiment of the present invention.

[0186] As illustrated in FIG. 9, a DNN model generation device (200) according to one embodiment of the present invention is configured to generate a DNN model for weight estimation and a DNN model for coefficient estimation, and includes an SOH information collection unit (210), an SOH verification unit (220), an SOH filtering unit (230), a first learning data generation unit (240), a second learning data generation unit (250), a first learning unit (260), and a second learning unit (270).

[0187] The SOH information collection unit (210) collects SOH information including the SOH and cycle count according to the battery cycle from the application side. The SOH information is collected for learning purposes and can be collected through an SOH provision device provided on the application side.

[0188] The above SOH verification unit (220) verifies the collected SOH according to the change characteristics of the SOH, and the SOH filtering unit (230) filters the verified SOH according to the physical characteristics of the SOH.

[0189] The above SOH verification unit (220) and SOH filtering unit (230) have the same configuration and perform the same function as the SOH verification unit (120) and SOH filtering unit (130) described with reference to FIG. 8.

[0190] The first learning data generation unit (240) generates first learning data by configuring a first feature data set including, for each cycle, the SOH change rate of the first cycle section, the SOH change rate of the second cycle section, the number of cycles, the verified SOH, or a combination thereof, and labeling the configured first feature data set with a weight according to the actual SOH.

[0191] The above first learning unit (260) trains the first DNN using the generated first learning data to create a DNN model for weight estimation.

[0192] The process of generating the first learning data and performing learning is described with reference to Fig. 4, so it is omitted here.

[0193] The second learning data generation unit (250) configures a second feature data set including, for each cycle, the filtered SOH from the previous cycle, the verified SOH, the weighted SOH change rate reflecting the SOH change rate between the first and second cycle sections, the number of cycles, or a combination thereof, and generates second learning data by labeling the configured second feature data set with the actual SOH and the coefficients (A, B, and L) according to [Mathematical Formula 8].

[0194] Meanwhile, in order to calculate the above reflection degree, the weight (w) must be estimated for each cycle, and when filtering the SOH, the SOH filtering unit (230) estimates the weight through the generated DNN model for weight estimation to calculate the reflection degree, thereby enabling the second learning data to be generated through the second learning data generation unit (250).

[0195] The above second learning unit (270) creates a DNN model for coefficient estimation by training the second DNN using the generated second learning data.

[0196] The process of generating the second learning data and performing learning is described with reference to Fig. 5, so it is omitted here.

[0197] In addition, in the present invention, weights and coefficients are labeled according to the actual SOH, and there is no limitation on the labeling method, and the first and second learning data can be configured and provided in advance.

[0198] The above first and second DNNs can be configured with various artificial intelligence learning networks such as a deep convolutional neural network (DCNN), a transformer, a temporal convolutional neural network (TCNN), a convolutional neural network (CNN), a recurrent neural network (RNN), a general regression neural network (GRNN), etc.

[0199] FIG. 10 is a flowchart illustrating a procedure for improving the reliability of SOH estimation by filtering SOH according to one embodiment of the present invention.

[0200] As illustrated in FIG. 10, in order to improve the reliability of SOH estimation according to one embodiment of the present invention, a procedure for filtering SOH first performs an SOH information receiving step in which a reliability estimation device (100) of SOH / RUL estimation receives SOH information including SOH and the number of cycles (S110).

[0201] Here, SOH information is received to provide improved reliability of SOH and RUL estimation according to the cycle of the battery mounted on the electric vehicle.

[0202] Next, the reliability estimation device (100) of the SOH / RUL estimation performs an SOH verification step of verifying the SOH according to the change characteristics of the SOH and correcting it as necessary (S120).

[0203] The above SOH verification step verifies and corrects according to the change characteristics of the SOH as needed, so that reliability can be improved even if the received SOH is incorrect. Since it has been described with reference to Fig. 2, it will be omitted here.

[0204] Additionally, the reliability estimation device (100) of the SOH / RUL estimation performs an SOH filtering step of filtering the verified SOH.

[0205] The above SOH filtering step first performs a first SOH change rate calculation step for calculating the SOH change rate of the first cycle section and the second cycle section based on the current cycle (S130).

[0206] As described above, the SOH change rate for each cycle section is calculated using the SOH verified for each cycle.

[0207] Next, the SOH filtering step performs a weight estimation step that estimates a weight for the degree of reflection of the SOH change rate for each cycle section calculated through the first SOH change rate calculation step (S140).

[0208] The above weight estimation step configures input data using the SOH change rate of the first cycle section, the SOH change rate of the second cycle section, the verified SOH, the number of cycles, or a combination thereof, and estimates the weight by inputting the configured input data into a DNN model for weight estimation.

[0209] Next, the SOH filtering step performs a weighted SOH change rate calculation step that calculates a weighted SOH change rate that reflects the SOH change rates of the first and second cycle sections according to the estimated weights (S150).

[0210] Estimating the above weight and calculating the SOH change rate reflecting the above weight are performed for each cycle.

[0211] Next, the SOH filtering step performs a coefficient estimation step that estimates a coefficient for a result obtained by subtracting the result of multiplying the filtered SOH, the verified SOH, the weighted SOH change rate and the number of cycles for the previous cycle according to [Mathematical Formula 8], or a combination thereof (S160).

[0212] The above coefficient estimation step configures input data using filtered SOH, verified SOH, weighted SOH change rate, number of cycles, or a combination thereof for the previous cycle, and inputs the configured input data into a DNN model for coefficient estimation to estimate each coefficient.

[0213] Next, the SOH filtering step finally filters the verified SOH through the result of multiplying the filtered SOH, the verified SOH, the weighted SOH change rate and the number of cycles for the previous cycle in which the coefficients were estimated, or a combination thereof (S170).

[0214] FIG. 11 is a flowchart illustrating a procedure for improving the reliability of RUL estimation by estimating RUL using filtered SOH according to one embodiment of the present invention.

[0215] As illustrated in FIG. 11, a procedure for improving the reliability of RUL estimation according to one embodiment of the present invention first performs a second SOH change rate calculation step in which the reliability estimation device (100) of SOH / RUL estimation calculates an SOH change rate for the filtered SOH for a predetermined cycle range including the current cycle (S210).

[0216] The above-described predetermined cycle range can be set in various ways. As described above, calculating the SOH change rate for the above-described cycle range utilizes SOH that has been filtered for each cycle to improve reliability.

[0217] Next, the reliability estimation device (100) of the SOH / RUL estimation performs a knee point cycle number calculation step of calculating the number of cycles corresponding to the knee point by using the SOH change rate for the filtered SOH calculated for a predetermined cycle range and the SOH corresponding to the knee point (S220).

[0218] As described above, the number of cycles corresponding to the above knee point is calculated by multiplying the SOH change rate by the SOH corresponding to the knee point.

[0219] Next, the reliability estimation device (100) of the SOH / RUL estimation performs the RUL estimation step of estimating the RUL by calculating the RUL using the number of cycles corresponding to the knee point (S230).

[0220] The above RUL is calculated by subtracting the number of cycles received from the number of cycles corresponding to the calculated knee point.

[0221] In addition, a device (100) for improving the reliability of SOH / RUL estimation according to one embodiment of the present invention is configured to include a memory for storing a program implementing a method for improving the reliability of SOH and RUL estimation, and a processor configured to execute the program stored in the memory.

[0222] Figure 12 is a flowchart illustrating a procedure for creating a DNN model according to one embodiment of the present invention.

[0223] As illustrated in FIG. 12, the procedure for generating a DNN model for weight estimation according to one embodiment of the present invention is as follows: first, the DNN model generation device (200) performs an SOH information collection step (S310) for collecting SOH and the number of cycles according to the cycle of the battery, an SOH verification step (S320) for verifying the collected SOH, and an SOH filtering step (S330) for filtering the verified SOH.

[0224] The above SOH information can be collected through an SOH information provision device on the application side, and the SOH verification step and the SOH filtering step are the same as the SOH verification step and the SOH filtering step described with reference to FIG. 10.

[0225] Next, the DNN model generation device (200) performs a first learning data generation step of generating first learning data by labeling weights on a first feature data set including the SOH change rate of the first and second cycle sections, the number of cycles, the verified SOH, or a combination thereof calculated in step S230 (S340).

[0226] Next, the DNN model generation device (200) performs a first learning step of generating a DNN model for weight estimation by training the first DNN using the generated first learning data (S350). The process of generating the DNN model for weight estimation has been described with reference to FIG. 4, and thus will be omitted here.

[0227] Next, the DNN model generation device (200) performs a second learning data generation step of generating second learning data to generate a DNN model for coefficient estimation.

[0228] The second learning data generation step first estimates a weight for the degree of reflection of the SOH change rate for each cycle section using a DNN model for weight estimation, and performs a weighted SOH change rate calculation step for calculating a weighted SOH change rate according to the reflection of the SOH change rate for each cycle section (S360).

[0229] Next, the second learning data generation step finally generates the second learning data by labeling the coefficients for the filtered SOH for the previous cycle, the verified SOH, the weighted SOH change rate, the number of cycles, or a combination thereof, in the second data set including the filtered SOH for the previous cycle, the verified SOH, the result of subtracting the result of multiplying the weighted SOH change rate and the number of cycles from the verified SOH, or a combination thereof according to [Mathematical Formula 8] (S370).

[0230] Next, the DNN model generation device (200) performs a second learning step of generating a DNN model for coefficient estimation by training the second DNN using the generated second learning data (S380).

[0231] As described above, the present invention has the effect of improving the reliability of SOH estimation by filtering using the physical characteristics of SOH, and at the same time, improving the reliability of RUL estimation by estimating RUL through the filtered SOH.

[0232] In addition, although the preferred embodiments of the present invention have been illustrated and described above, the present invention is not limited to the specific embodiments described above, and various modifications can be implemented by a person having ordinary skill in the art to which the invention pertains without departing from the gist of the present invention claimed in the claims, and such modifications should not be understood individually from the technical idea or prospect of the present invention.

[0233] As described above, the present invention improves reliability by receiving SOH from an electric vehicle and filtering the SOH by considering the physical characteristics of the SOH, and estimates RUL based on the SOH with improved reliability, thereby providing a user with accurate and reliable SOH / RUL, thereby enabling inspection or replacement of the battery before a problem occurs, and thus has industrial applicability.

Claims

1. SOH information receiving step for receiving SOH (state of health) and cycle number for the battery of an electric vehicle; An SOH verification step for verifying the received SOH using the change characteristics of the SOH; and It includes an SOH filtering step for filtering the verified SOH using the physical characteristics of the SOH, which is verified to satisfy the change characteristics of the SOH; A method for improving the reliability of SOH estimation through SOH filtering, characterized in that the reliability of SOH estimation is improved by filtering the verified SOH in consideration of the physical characteristics of the SOH of the battery, in which the SOH gradually decreases within a predetermined range until a knee point and then rapidly decreases beyond the predetermined range when the knee point is passed.

2. In claim 1, The above SOH verification step is: If the received SOH satisfies the predetermined SOH change characteristic, the received SOH is output as a verified SOH, and if it violates the predetermined SOH change characteristic, a predetermined SOH correction is performed and output as a verified SOH. The above SOH change characteristics are: A method for improving the reliability of SOH estimation through SOH filtering, characterized in that SOH cannot increase as the number of cycles increases, SOH does not change more rapidly than a predetermined maximum change value in a predetermined short cycle before reaching a knee point, or a combination thereof.

3. In claim 2, The above SOH correction is, Considering the change characteristic that the above SOH cannot increase as the number of cycles increases, if the received SOH is greater than the previously received SOH, the received SOH is replaced with the previously received SOH, or Considering the change characteristic that the SOH does not change more rapidly than a predetermined maximum change value in a predetermined short-term cycle before reaching the knee point, if the change between the received SOH and the previously received SOH is greater than the maximum change value, the received SOH is replaced with the previously received SOH, or A method for improving the reliability of SOH estimation through SOH filtering, characterized by including a combination of these.

4. In claim 1, The above SOH filtering step is, When filtering the above SOH, a weight estimation step is further included for estimating a weight for the degree of reflection of the rate of change of SOH for the first cycle section between the initial cycle and the current cycle and the rate of change of SOH for the second cycle section between the previous cycle and the current cycle; A method for improving the reliability of SOH estimation through SOH filtering, characterized in that the verified SOH is filtered by calculating a weighted SOH change rate reflecting the SOH change rate of the first cycle section and the second cycle section according to the estimated weight.

5. In claim 4, The above weight estimation step is, The SOH change rate for the first cycle section, the SOH change rate for the second cycle section, the current cycle number, or a combination thereof are input into a DNN model for weight estimation to estimate the weight, A method for improving the reliability of SOH estimation through SOH filtering, characterized in that the DNN model for weight estimation is generated by learning each first learning data generated by labeling weights for each first feature data set including the SOH change rate for the first cycle section, the SOH change rate for the second cycle section, the current cycle number, the verified SOH, or a combination thereof.

6. In claim 4, The above SOH filtering step is, Filter the verified SOH through A x SOH_filtered[t-1], B x SOH[t], L(SOH[t] - ax cycle[t]) or a combination thereof, A method for improving the reliability of SOH estimation through SOH filtering, characterized in that the above SOH_filtered[t-1] is the SOH filtered in the previous cycle, the above SOH(t) is the verified SOH, a is the weighted SOH change rate reflecting the estimated weight, cycle[t] is the current cycle number, the above A and B are constants, and the above L is the gain of the above SOH[t] and ax cycle[t].

7. In claim 6, The above SOH filtering step is, It further includes a coefficient estimation step for estimating coefficients including A, B, and L by inputting the SOH filtered in the previous cycle, the verified SOH, the weighted SOH change rate, the current cycle number, or a combination thereof into a DNN model for coefficient estimation; A method for improving the reliability of SOH estimation through SOH filtering, characterized in that the DNN model for the coefficient estimation is generated by learning second learning data labeled A, B, and L for each second feature data set including the SOH filtered in the previous cycle for each cycle, the verified SOH, the SOH change rate for the first cycle section and the SOH change rate for the second cycle section, the current cycle number, or a combination thereof.

8. Further comprising an RUL estimation step of estimating the RUL by calculating the RUL of the battery using the filtered SOH according to the method for improving the reliability of SOH estimation according to at least one of claims 1 to 7; The above RUL estimation step is, An SOH change rate calculation step for calculating an SOH change rate using the filtered SOH in a predetermined cycle range including the current cycle and at least one previous cycle; and A knee point cycle number calculation step for calculating the number of cycles corresponding to the knee point through the calculated SOH change rate and the SOH corresponding to the knee point of the battery in the above-mentioned predetermined cycle range; A method for improving the reliability of RUL estimation through SOH filtering, characterized in that the RUL is calculated by subtracting the current cycle number from the cycle number corresponding to the calculated knee point.

9. A memory storing a program implementing a method for improving the reliability of SOH estimation through SOH filtering according to any one of claims 1 to 7; and A device for improving the reliability of SOH estimation through SOH filtering, characterized in that it comprises a processor configured to execute a program stored in the above memory.

10. A method for improving the reliability of SOH estimation through SOH filtering according to any one of claims 1 to 7, further comprising an RUL estimation step for calculating the RUL of a battery by using the filtered SOH, and the RUL estimation step comprises an SOH change rate calculation step for calculating the SOH change rate using the filtered SOH in a predetermined cycle range including a current cycle and at least one previous cycle; and a knee-point cycle number calculation step for calculating the number of cycles corresponding to the knee-point of the battery through the calculated SOH change rate and the SOH corresponding to the knee-point of the battery in the predetermined cycle range; and a memory for storing a program for implementing a method for improving the reliability of RUL estimation through SOH filtering by calculating the RUL by subtracting the current cycle number from the calculated cycle number corresponding to the knee-point; and A device for improving the reliability of RUL estimation through SOH filtering, characterized in that it comprises a processor configured to execute a program stored in the above memory.

Images