Secondary battery degradation state estimation system, secondary battery degradation state estimation method, and storage medium

Abstract

Provided are a secondary battery degradation state estimation system, secondary battery degradation state estimation method, and storage medium that improve the accuracy of estimating secondary battery degradation using a learned model. The secondary battery degradation state estimation system comprises: a state variable measurement unit that measures the state variables of the secondary battery; a preprocessing unit that outputs input data; and a degradation state estimation unit that estimates the degradation state of the secondary battery. The preprocessing unit includes a state variable processing unit that sets a unit charge / discharge amount. The state variable processing unit sets a recent interval closest to the current time and one or more past intervals preceding the recent interval as desired intervals derived from the unit charge / discharge amount used in calculating the input data. For each of the set intervals, the state variable processing unit calculates a voltage change based on the state variable as interval data, and outputs the calculated interval data as part of the input data.

Description

Technical Field

[0001] The present invention relates to a degradation state estimation system, a degradation state estimation method and a storage medium. Background Art

[0002] There is known a technology for learning a learned model based on learning data that uses time series data related to the SOC (State Of Charge) of a battery from a first time point to a second time point later than the first time point and the SOH of the battery at the first time point as input data and the SOH at the second time point as output data (for example, refer to Japanese Patent No. 2019-168453). Summary of the Invention

[0003] When estimating the SOH of the secondary battery using the learned model learned as described above, the SOC is included in the input data. Since the SOC included in the input data is an estimated value, the accuracy of the SOH output by the learned model may be reduced.

[0004] The present invention has been made in consideration of such circumstances, and one object of the present invention is to provide a degradation state estimation system, a degradation state estimation method, and a storage medium capable of improving the accuracy of estimating degradation of a secondary battery using a learned model.

[0005] In order to solve the above-mentioned problems and achieve the object, the present invention adopts the following means.

[0006] (1): A degradation state estimation system according to one embodiment of the present invention comprises: a state variable measuring unit that measures state variables including output current and output voltage of a secondary battery in operation at each specified timing; a pre-processing unit that outputs input data calculated based on the state variables measured by the state variable measuring unit; and a degradation state estimation unit that uses the input data output by the pre-processing unit to estimate the degradation state of the secondary battery in operation through a learned degradation state model, wherein the pre-processing unit comprises a state variable processing unit that sets a unit charge and discharge amount, and the state variable processing unit sets a most recent interval closest to the current moment and one or more past intervals preceding the most recent interval as a desired interval obtained based on the unit charge and discharge amount used in calculating the input data, and calculates a voltage change amount obtained based on the state variable as interval data for each of the set intervals, and outputs the calculated interval data by including it in the input data.

[0007] (2): A degradation state estimation system according to one embodiment of the present invention comprises: a state variable measuring unit that measures state variables including output current and output voltage of a secondary battery in operation at each predetermined timing; a pre-processing unit that outputs input data calculated based on the state variables measured by the state variable measuring unit; and a degradation state estimation unit that estimates the degradation state of the secondary battery in operation through a learned degradation state model using the input data output by the pre-processing unit, wherein the pre-processing unit comprises a state variable processing unit that sets a large charge and discharge amount as a unit charge and discharge amount, and a unit charge and discharge amount that is smaller than the large charge and discharge amount. The state variable processing unit sets the most recent large interval and the most recent small interval closest to the current moment and one or more past large intervals and past small intervals that are earlier than the most recent large interval and the most recent small interval as the expected interval obtained based on the large charge and discharge amount and the small charge and discharge amount used in the calculation of the input data, and calculates the voltage change amount obtained based on the state variable as large interval data and small interval data according to each of the set large intervals and the small intervals, and outputs the calculated large interval data and the small interval data by including them in the input data.

[0008] (3): Based on the degradation state estimation system of the above-mentioned scheme (1) or (2), the state variable processing unit may set the period during which the current cumulative value obtained by accumulating the output current measured by the state variable measuring unit becomes a specified value as the desired interval.

[0009] (4): Based on the degradation state estimation system of any one of the above schemes (1) to (3), at least one of the calculated charge and discharge amount in each desired interval and the calculated voltage change amount in each desired interval may be a slope change rate calculated using the least squares method.

[0010] (5) In the degradation state estimation system according to any one of (1) to (4) above, the degradation state model may be configured as an RNN (recurrent neural network).

[0011] (6): Based on the degradation state estimation system of (5) above, the intermediate layer of the RNN may be constructed as an LSTM (Long Short-Term Memory) or a GRU (Gated Recurrent Unit).

[0012] (7) In the degradation state estimation system according to any one of (1) to (4) above, the degradation state model may be configured as a CNN (convolutional neural network).

[0013] (8): A degradation state estimation method according to one embodiment of the present invention is performed by a computer in a degradation state estimation system as follows: measuring state variables including output current and output voltage of a secondary battery in operation at each specified time; outputting input data calculated based on the measured state variables; using the output input data to estimate the degradation state of the secondary battery in operation through a learned degradation state model; setting a unit charge and discharge amount used in the calculation of the input data as a desired interval based on the unit charge and discharge amount, setting a most recent interval closest to the current moment and one or more past intervals preceding the most recent interval, calculating a voltage change amount based on the state variables as interval data for each of the set intervals, and outputting the calculated interval data by including it in the input data.

[0014] (9): A storage medium according to one embodiment of the present invention stores a program, wherein the program is used to cause a computer in a degradation state estimation system to perform the following processing: measuring state variables including output current and output voltage of a secondary battery in operation at each specified time; outputting input data calculated based on the measured state variables; using the output input data to estimate the degradation state of the secondary battery in operation through a learned degradation state model; setting a unit charge and discharge amount used in the calculation of the input data as a desired interval based on the unit charge and discharge amount, setting a most recent interval closest to the current moment and one or more past intervals preceding the most recent interval, calculating a voltage change amount based on the state variables as interval data for each of the set intervals, and outputting the calculated interval data by including it in the input data.

[0015] According to (1), (8), and (9), when estimating the degradation state of a secondary battery using a learned degradation state model, interval data based on charge and discharge amounts and voltage changes calculated corresponding to a plurality of desired intervals is used. This improves the accuracy of the secondary battery degradation estimation output by the learned degradation state model.

[0016] According to (2), when estimating the degradation state of a secondary battery using a learned degradation state model, large-interval data and small-interval data based on charge and discharge amounts and voltage changes calculated corresponding to a plurality of large-intervals and small-intervals are used. This improves the accuracy of the secondary battery degradation estimation output by the learned degradation state model.

[0017] According to (3), a large number of intervals can be set based on the current integrated value.

[0018] According to (4), the charge / discharge amount and the voltage change amount, which are a large amount of interval data, can be made into a high-precision charge / discharge amount and the voltage change amount with reduced noise.

[0019] According to (5), by using RNN for the degradation state model, a highly accurate estimation result can be expected.

[0020] According to (6), by setting the intermediate layer in the RNN of the degradation state model to LSTM, it is possible to expect an estimation result with high accuracy.

[0021] According to (7), by using CNN for the degradation state model, it is possible to expect an estimation result with high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a diagram showing a configuration example of the degradation state estimation device according to the present embodiment.

[0023] Figure 2 This is a diagram illustrating an example of a scenario for acquiring a large amount of section data according to the present embodiment.

[0024] Figure 3 This is a diagram illustrating the SOC-OCV characteristics of this embodiment.

[0025] Figure 4 This is a diagram for explaining the identification of SOH according to the relationship between the charge and discharge amount and the voltage change amount in this embodiment.

[0026] Figure 5 This is a diagram for explaining the identification of SOH according to the relationship between the charge and discharge amount and the voltage change amount in this embodiment.

[0027] Figure 6 This is a diagram for explaining the identification of SOH according to the relationship between the charge and discharge amount and the voltage change amount in this embodiment.

[0028] Figure 7 This is a flowchart showing an example of processing steps executed by the degradation state estimation device according to the present embodiment in connection with estimation of SOH. DETAILED DESCRIPTION

[0029] Hereinafter, embodiments of a degradation state estimation system, a degradation state estimation method, and a degradation state estimation storage medium according to the present invention will be described with reference to the accompanying drawings.

[0030] Figure 1 FIG1 shows an example of the overall configuration of a degradation state estimation device 100 according to this embodiment. The degradation state estimation device 100 estimates the state of hydration (SOH) of a secondary battery 200 as the degradation state of the secondary battery 200. The degradation state estimation device 100 shown in FIG1 includes a state variable measurement unit 101, a pre-processing unit 102, a second learned model 103 (an example of a degradation state model), and a degradation state estimation unit 104.

[0031] The state variable measurement unit 101 measures the output current and output voltage as state variables of the operating secondary battery 200 and outputs the measured output current Iout and output voltage Vout. The output voltage Vout can also be calculated based on the CCV (Closed Circuit Voltage) detected by a sensor in the secondary battery 200. The output voltage Vout can also be calculated based on the OCV (Open Circuit Voltage).

[0032] The pre-processing unit 102 calculates state variable processing data corresponding to the current time t using the output current Iout and output voltage Vout input from the state variable measuring unit 101 as pre-processing. The state variable data is used by the characteristic identification unit 123 to estimate the SOC-OCV characteristics. Data other than the power data P(t) in the state variable processing data is output as input data Din(t) corresponding to the current time t. The degradation state estimation unit 104 uses this input data Din(t) to estimate the SOH.

[0033] The pre-processing unit 102 includes a state variable processing unit 121 , a first learned model 122 (an example of a characteristic identification model), and a characteristic identification unit 123 .

[0034] The state variable processing unit 121 receives inputs of the output current Iout and the output voltage Vout from the state variable measuring unit 101. At each predetermined estimation timing, the state variable processing unit 121 calculates, based on the input output current Iout and output voltage Vout, a large amount of interval data (charge / discharge amounts LCA(t) to LCA(t-2), voltage variation LEV(t) to LEV(t-2)), a small amount of interval data (charge / discharge amount SCA(t), voltage variation LEV(t)), voltage data V(t), and power data P(t) corresponding to the current time t.

[0035] The voltage data V(t) may be CCV or OCV.

[0036] Power data P(t) can represent discharged power (consumed power) or charged power. Instead of power data P(t), information that can be used to understand the extent of power consumption and charged power, such as information indicating current flow, auxiliary equipment usage, and vehicle usage patterns, can also be used. Such power data P(t), or data that serves as a substitute for power data P(t), can be estimated from the cloud or unique information when characteristics do not change.

[0037] The interval data includes large-volume interval data and small-volume interval data. The large-volume interval data is the voltage variation calculated by the state variable processing unit 121 for each of a plurality of large-volume intervals corresponding to a unit large-volume charge / discharge amount and a constant unit large-volume charge / discharge amount. The small-volume interval data is the voltage variation calculated by the state variable processing unit 121 for each of a plurality of large-volume intervals corresponding to a unit small-volume charge / discharge amount and a constant unit small-volume charge / discharge amount.

[0038] Hereinafter, the “charge / discharge amount large range” is also referred to as the “large range range”, and the “charge / discharge amount small range” is also referred to as the “small range range”.

[0039] Reference Figure 2 , an example of a method for calculating a large amount of interval data is described. This figure shows the unit charge / discharge capacity Aprd of the secondary battery 200 obtained for the period from the current time t to a time t-3 past the current time t. The unit charge / discharge capacity Aprd shown in this figure is obtained by integrating the output current Iout.

[0040] The state variable processing unit 121 sets the period from the current time t to the time t-1 at which the predetermined unit current integrated value (the integrated value of the output current Iout) is obtained as the bulk interval closest to the current time t (the most recent bulk interval T1). The state variable processing unit 121 sets the period from the time t1 to the time t-1 at which the predetermined unit current integrated value is obtained as the first bulk interval past the most recent bulk interval T1 (the past bulk interval T2-1). The state variable processing unit 121 sets the period from the time t-1 to the time t-2 at which the predetermined unit current integrated value is obtained as the past bulk interval T2-2 past the past bulk interval T2-1.

[0041] In the following description, when there is no particular distinction between the past large-volume sections T2-1 and T2-2, they are referred to as past large-volume sections T2. When there is no particular distinction between the recent large-volume section T1 and the past large-volume section T2, they are referred to as large-volume sections T.

[0042] As described above, the large intervals T are set to obtain a predetermined large amount of charge and discharge based on each predetermined current integrated value. Therefore, the length of each large interval T may be different. The large intervals T may also be set based on a predetermined unit time.

[0043] The large number of intervals T may be as long as, for example, several tens to several hundreds of seconds.

[0044] The number of past large-volume intervals T2 set by the state variable processing unit 121 is not limited to two, and may be one or more.

[0045] In the example shown in this figure, three large-volume intervals T are set to be continuous over time. They can be made discontinuous by providing a gap between two temporally adjacent large-volume intervals T. When three or more large-volume intervals T are set, it is possible to set the intervals so that two large-volume intervals T preceding and following each other in a certain time are continuous, while other intervals T preceding and following each other in a different time are discontinuous.

[0046] The state variable processing unit 121 calculates a large interval T corresponding to obtaining a certain unit large charge and discharge amount LCA set as described above. Figure 2 As shown, the state variable processing unit 121 calculates the actual time of the most recent large interval T1 so that the unit large charge / discharge amount LCA = the charge / discharge amount LCA(t), calculates the actual time of the past large interval T2-1 so that the charge / discharge amount LCA(t-1) = the charge / discharge amount LCA(t), and calculates the actual time of the past large interval T2-2 so that the charge / discharge amount LCA(t-2) = the charge / discharge amount LCA(t-1). In other words, the state variable processing unit 121 finds the past times (t-1), (t-2), and (t-3) at which the unit large charge / discharge amount LCA = LCA(t) = LCA(t-1) = LCA(t-2), and calculates the actual times T1, T2-1, and T2-2. Alternatively, the most recent high-volume interval can be fixed at T1, i.e., T1 = T2-1 = T2-2. The charge / discharge amount LCA(t) is calculated for the most recent high-volume interval T1, the charge / discharge amount LCA(t-1) is calculated for the past high-volume interval T2-1, and the charge / discharge amount LCA(t-2) is calculated for the past high-volume interval T2-2. However, setting the unit high-volume charge / discharge amount LCA successfully improved the accuracy of the SOH estimation described later.

[0047] The state variable processing unit 121 can calculate at least one of the charge / discharge amounts LCA(t), LCA(t-1), and LCA(t-2) as a slope change rate using the least squares method. The charge / discharge amount LCA calculated as the slope change rate in this manner reduces noise and improves accuracy. As a result, the accuracy of the SOC-OCV characteristics estimated by the first learned model 122 (described later) and the SOH estimated by the second learned model 103 (described later) using the charge / discharge amount LCA can be improved.

[0048] The state variable processing unit 121 is as follows Figure 2 Each of the large number of intervals T calculated as shown, Figure 3As shown, the corresponding voltage (output voltage Vout) change (voltage change) is calculated. That is, the state variable processing unit 121 calculates the voltage change LEV(t) corresponding to the most recent large interval T1, calculates the voltage change LEV(t-1) corresponding to the past large interval T2-1, and calculates the voltage change LEV(t-2) corresponding to the past large interval T2-2.

[0049] The state variable processing unit 121 may calculate at least one of the voltage changes LEV(t), LEV(t-1), and LEV(t-2) as a slope change rate using the least squares method. In this case, the accuracy of the SOC-OCV characteristics estimated using the voltage change LEV by the first learned model 122 (described later) and the SOH estimated using the second learned model 103 (described later) can also be improved.

[0050] The state variable processing unit 121 can calculate at least one of the voltage changes LEV(t), LEV(t-1), and LEV(t-2) as a slope change rate for a predetermined unit bulk charge / discharge amount LCA or for the charge / discharge amount LCA(t) corresponding to a predetermined bulk interval of time using the least squares method. The voltage change LEV(t) calculated as the slope change rate in this manner improves accuracy by reducing noise. As a result, the accuracy of the SOC-OCV characteristics estimated by the first learned model 122 and the SOH estimated by the second learned model 103, described later, can be improved using the voltage change LEV(t) calculated as the slope change rate.

[0051] In the following description, when the charge and discharge amounts LCA(t), LCA(t-1), and LCA(t-2) are not particularly distinguished, they are described as the charge and discharge amount LCA.

[0052] In the following description, when the voltage change amounts LEV(t), LEV(t-1), and LEV(t-2) are not particularly distinguished, they are described as voltage change amounts LEV.

[0053] When the charge / discharge amount LCA and the voltage change amount LEV are not particularly distinguished, they are also recorded as “large amount of section data”.

[0054] For example, the state variable processing unit 121 may calculate data on the voltage change amount per unit of the bulk charge and discharge amount corresponding to the most recent bulk interval as the bulk interval data.

[0055] Return the description Figure 1In addition to calculating the voltage change LEV for each large interval T corresponding to the unit large charge / discharge amount LCA, the state variable processing unit 121 also calculates the actual time and voltage change SEV(t) for each small interval corresponding to the unit large charge / discharge amount SCA(t). The state variable processing unit 121 may also use past values ​​corresponding to times t-1 and t-2, for example, for the charge / discharge amount SCA and voltage change SEV.

[0056] The state variable processing unit 121 may, for example, define the period from the current time t until a predetermined current accumulation smaller than that corresponding to the large-volume interval is obtained as a small-volume interval. Alternatively, the state variable processing unit 121 may define the period during which a predetermined unit current accumulation (charge and discharge amount) smaller than that corresponding to the large-volume interval T is performed as a small-volume interval. The small-volume interval may be, for example, several seconds to several tens of seconds.

[0057] When the charge / discharge amount SCA(t) and the voltage change amount SEV(t) are not particularly distinguished, they are described as “small amount of interval data”.

[0058] As described above, the small amount interval is set to obtain a period during which each current integrated value reaches a predetermined unit small amount charge / discharge amount SCA. Therefore, the length of time of each small amount interval may be different.

[0059] The number of past small intervals set by the state variable processing unit 121 is not limited to one, and may be one or more.

[0060] In the example of this figure, the power change SEV(t) for one small interval is input. Similar to the large interval T described above, two small intervals can be set, or two small intervals that are separated in time can be set and the power change SEV(t) for each of the set small intervals can be input. Discontinuity can be achieved by setting gaps between the small intervals. When three or more small intervals are set, it can be set so that two small intervals separated in time are continuous, while two small intervals separated in other time periods are discontinuous. The small intervals can also be set based on a specified unit time.

[0061] The state variable processing unit 121 calculates the voltage data V(t) and power data P(t) corresponding to the current time t. The state variable processing unit 121 may set the output voltage Vout at the current time t as the voltage data V(t). The state variable processing unit 121 may calculate the power data P(t) based on the output current Iout and output voltage Vout at the current time t.

[0062] In the following description, large-volume interval data, small-volume interval data, voltage data V(t), and power data P(t) calculated by the state variable processing unit 121 are described as state variable processed data unless they are particularly distinguished.

[0063] The state variable processing unit 121 outputs state variable processing data at the current time t, which is updated every time a predetermined time passes. Therefore, the state variable processing data becomes time series data obtained at each predetermined time.

[0064] First learned model 122 is generated through machine learning using sample data corresponding to state variable processing data and the SOC-OCV characteristic as teaching data. The model takes the state variable processing data as input and outputs an estimated SOC-OCV characteristic. First learned model 122 can output characteristic identification information identifying the estimated SOC-OCV characteristic as the SOC-OCV characteristic estimation result.

[0065] The first learned model 122 is configured as an RNN (recurrent neural network). Based on this, the intermediate layer of the first learned model 122, which is an RNN, can be configured as an LSTM (long short-term memory) or a GRU (gated recurrent unit). Alternatively, the first learned model 122 can be configured as a CNN (convolutional neural network). The first learned model 122 can treat the estimation of SOC-OCV characteristics as either a regression problem or a classification problem.

[0066] The following description uses the example of a case where the first learned model 122 is configured as an RNN with an LSTM in the intermediate layer, and the estimation of SOC-OCV characteristics is handled as a regression problem. In this case, the state variable processing data output by the state variable processing unit 121 is input to the first learned model 122 as LSTM blocks.

[0067] The SOC-OCV characteristic shows the correlation between the SOC and OCV of the secondary battery as the state of the secondary battery.

[0068] Figure 4 A specific example of an SOC-OCV characteristic is shown. In this figure, the horizontal axis represents SOC and the vertical axis represents OCV. This figure shows curves C1 to C5 corresponding to five different SOC-OCV characteristics. For example, multiple SOC-OCV characteristics can be set as estimation candidates corresponding to the first learned model 122, and characteristic identification information can be assigned to each of the SOC-OCV characteristics set as estimation candidates.

[0069] In the example of this figure, for the same SOC, curve C1 has the highest OCV value, followed by curves C2, C3, C4, and C5, in that order. As a relative relationship, for example, when comparing two SOC-OCV characteristics corresponding to curves C1 and C2, the SOC-OCV characteristic corresponding to curve C1 is a higher characteristic, while the SOC-OCV characteristic corresponding to curve C2 is a lower characteristic.

[0070] The characteristic identification unit 123 inputs the state variable processing data corresponding to the current time t into the first learned model 122. The characteristic identification unit 123 obtains characteristic identification information output by the first learned model 122 in response to the input state variable processing data. The characteristic identification unit 123 outputs the obtained characteristic identification information as characteristic identification information CID(t) corresponding to the current time t.

[0071] The second learned model 103 is a machine-learned model using sample data corresponding to the input data Din(t) and the characteristic identification information CID(t) and the SOH as teaching data. The second learned model 103 outputs the SOH in response to the input data Din(t) input by the degradation state estimation unit 104.

[0072] The input data Din(t) includes a large amount of interval data (charge and discharge amount LCA(t), LCA(t-1), LCA(t-2), voltage change LEV(t), LEV(t-1), LEV(t-2)), a small amount of interval data (charge and discharge amount SAC(t), voltage change SEV(t)), voltage data V(t), and characteristic identification information CID(t).

[0073] The input data Din(t) may also include power data P(t).

[0074] The second learned model 103 is configured as an RNN (recurrent neural network). Based on this, the intermediate layer of the second learned model 103, which is an RNN, can be configured as an LSTM (long short-term memory) or a GRU (gated recurrent unit). Alternatively, the second learned model 103 can be configured as a CNN (convolutional neural network). The second learned model 1103 can treat the estimation of SOH as either a regression problem or a classification problem.

[0075] In the following description, the second learned model 103 is configured as an RNN with an LSTM in the intermediate layer, and the estimation of SOH is treated as a classification problem. In this case, the state variable processing data output by the state variable processing unit 121 is input to the second learned model 103 as an LSTM block.

[0076] The degradation state estimation unit 104 inputs input data Din(t) corresponding to the current time t into the second learned model 103. The degradation state estimation unit 104 obtains the SOH value output by the second learned model 103 in response to the input data Din(t). The degradation state estimation unit 104 outputs the obtained SOH value as an estimated SOH value Dout.

[0077] Figure 5 An example of the relationship between charge and discharge capacity and voltage change is shown. This figure shows four curves, C11, C12, C21, and C22, corresponding to different SOH estimated values. Curves C11 and C21 represent the case where the secondary battery capacity is the same, α (Ah), while curves C12 and C22 represent the case where the secondary battery capacity is the same, β (Ah), which is less than α (Ah).

[0078] For example, the estimated SOH value also corresponds to a specific SOC-OCV characteristic. Therefore, the four curves C11, C12, C21, and C22 also correspond to specific SOC-OCV characteristics. As a relative relationship, the SOC-OCV characteristics corresponding to curves C11 and C12 can be distinguished as high characteristics, while the SOC-OCV characteristics corresponding to curves C21 and C22 can be distinguished as low characteristics.

[0079] In the example of this figure, as shown as the intersection IS, there is a portion where the curve corresponding to the high-characteristic SOC-OCV characteristic and the curve corresponding to the low-characteristic SOC-OCV characteristic intersect.

[0080] exist Figure 6 , three curves C31, C32, and C31-1 are shown, illustrating the relationship between the charge and discharge amount and the voltage change. In this figure, curves C31 and C32 correspond to different SOHs, and are curves obtained when the corresponding secondary battery starts operating at a 100% SOC. On the other hand, curve C31-1 corresponds to the same SOH as curve C31, but is a curve obtained when the secondary battery starts operating at a 50% SOC. In the example in this figure, curves C31 and C32 do not overlap. However, in the case of curve C31-1, which corresponds to the same SOC-OCV characteristics as curve C31 but has a different SOC at the start of operation, it overlaps with curve C32. In other words, there are cases where overlapping sections occur between multiple curves depending on the SOC at the start of operation.

[0081] Depending on fluctuations in the state variable measured by the state variable measuring unit 101 , curves corresponding to different SOHs may overlap or intersect.

[0082] In this embodiment, the degradation state estimation unit 104 inputs the input data Din(t) to the second learned model 103, including multiple charge and discharge amounts LCA(t), LCA(t-1), and LCA(t-2) corresponding to each of the multiple (three) large intervals (T1, T2-1, and T2-2), and three voltage changes LEV(t), LEV(t-1), and LEV(t-2). Specifically, the second learned model 103 receives, in addition to the short-term history (lookback) of the small interval data, the long-term history (large interval data) of the charge and discharge amounts and the voltage changes as input feature quantities.

[0083] Therefore, even when the curves of the SOH estimated values ​​serving as candidate estimates overlap or intersect, as described above, second learned model 103 uses the charge / discharge amounts and voltage changes corresponding to a large number of temporally distinct intervals for estimation. This allows it to distinguish between the overlapping and intersecting curves and identify the SOH. In this case, second learned model 103 increases the number of input parameters between histories within a single LSTM block. By increasing the number of input parameters for the histories, it is expected that the breadth of past information learned will be expanded. Consequently, the accuracy of the SOH estimated value Dout(t) output by second learned model 103 can be improved.

[0084] The state variable processing data input by the characteristic recognition unit 123 to the first learned model 122 also includes a plurality of large interval data corresponding to the charge and discharge amount and a plurality of large interval data corresponding to the voltage change amount, which are the same as those contained in the input data Din(t).

[0085] By inputting such a large amount of interval data, the first learned model 122 can also perform estimation using the charge and discharge amounts and voltage change amounts corresponding to a large number of intervals that are different in time. Figure 5 In the case where the curves corresponding to the high and low SOC-OCV characteristics intersect, as shown in the example, the curve interval can be referenced over a long period of time rather than instantaneously. This allows for accurate discrimination of the SOC-OCV characteristics under high and low characteristics, thus avoiding misjudgment. This improves the accuracy of the SOC-OCV characteristics estimated by the first learned model 122.

[0086] The input data Din(t) inputted by the degradation state estimation unit 104 to the second learned model 103 includes the characteristic identification information CID of the SOC-OCV characteristic estimated by the first learned model 122 as described above. In other words, the input data Din(t) includes information on the SOC-OCV characteristic.

[0087] By using input data Din(t) including information on the SOC-OCV characteristic, it is no longer necessary to use the estimated SOC value or the estimated internal resistance value when estimating the SOH using the second learned model 103. By not using the estimated SOC value or the estimated internal resistance value, in this embodiment, the second learned model 103 can estimate the SOH without being affected by errors in the estimated SOC value, thereby achieving improved accuracy.

[0088] The characteristic identification information CID included in the input data Din(t) for estimating the SOH is not fixedly set to correspond to the target secondary battery 200. Rather, the characteristic identification information CID corresponds to the result estimated by the first learned model 122 using state variable processing data obtained based on the state variables measured for the target secondary battery 200. Therefore, in this embodiment, the characteristic identification information CID corresponding to the target secondary battery 200 can be estimated with a certain level of accuracy or higher, regardless of changes in the specifications of the secondary battery 200. Therefore, the degradation state estimation device 100 of this embodiment can appropriately estimate the SOH for a wide variety of secondary batteries 200.

[0089] The second learned model 103 of this embodiment can estimate the SOH using the input data Din(t) that does not include the characteristic identification information CID, if, for example, a certain accuracy condition is satisfied.

[0090] Reference Figure 7 The flowchart of FIG. 1 will describe an example of a processing procedure executed by the degradation state estimation device 100 in association with the estimation of SOH. The processing in this figure is started every time a predetermined time has passed.

[0091] The state variable measuring unit 101 measures the state variables of the secondary battery 200 (step S100). The state variables of the secondary battery 200 are the output current Iout and the output voltage Vout corresponding to the current time t.

[0092] In the pre-processing unit 102 , the state variable processing unit 121 calculates state variable processing data corresponding to the current time t using the output current Iout and the output voltage Vout measured in step S100 (step S102 ).

[0093] State variable processing data includes large-scale interval data, small-scale interval data, voltage data V(t), and power data P(t). Large-scale interval data includes charge and discharge amounts LCA(t), LCA(t-1), and LCA(t-2), and voltage variations LEV(t), LEV(t-1), and LEV(t-2). Small-scale interval data includes charge and discharge amounts SCA(t) and voltage variation SEV(t).

[0094] The characteristic recognition unit 123 inputs the state variable data calculated in step S102 to the first learned model 122 (step S104 ).

[0095] First learned model 122 outputs characteristic identification information as an estimated result of the SOC-OCV characteristic, based on the state variable data input in step S104. Characteristic identification unit 123 acquires the characteristic identification information output by first learned model 122 and outputs the acquired characteristic identification information as characteristic identification information CID(t) corresponding to the current time t (step S106).

[0096] The degradation state estimation unit 104 inputs the input data Din(t) to the second learned model 103 (step S108 ).

[0097] The second learned model 103 estimates the SOH according to the input data Din(t) input in step S108. The degradation state estimation unit 104 outputs the SOH estimated by the second learned model 103 as the SOH estimated value Dout(t) corresponding to the current time t (step S110).

[0098] The use of the secondary battery 200 of this embodiment is not particularly limited. For example, the secondary battery 200 can be installed in a vehicle to drive the vehicle. The secondary battery 200 can also be installed in a residence, office building, etc. The secondary battery 200 can be included in a power transmission network such as a smart grid.

[0099] A degradation state estimation system can be configured in which the functions of the degradation state estimation device 100 of this embodiment are distributed across multiple devices. For example, such a degradation state estimation system may include a terminal device equipped with a secondary battery for degradation estimation, and a cloud server communicatively connected to the terminal device. The terminal device may transmit the state variables of the secondary battery, measured by the function of the state variable measurement unit 101, to the cloud server. The cloud server uses the received state variables to output an estimated SOH value using the functions of the pre-processing unit 102, the second learned model 103, and the degradation state estimation unit 104.

[0100] A program for implementing the functions of the degradation state estimation device 100 described above may be recorded on a computer-readable recording medium, and the computer system may read and execute the program recorded on the recording medium, thereby performing the processing of the degradation state estimation device 100 described above. Here, "causing a computer system to read and execute the program recorded on the recording medium" includes installing the program on the computer system. The "computer system" herein is assumed to include hardware such as an operating system and peripheral devices. A "computer system" may also include multiple computer devices connected via a network, including communication lines such as the Internet, a wide area network (WAN), a local area network (LAN), or dedicated lines. A "computer-readable recording medium" refers to removable media such as floppy disks, magneto-optical disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into the computer system. Thus, the recording medium storing the program may also be a non-transitory recording medium such as a CD-ROM. Recording media also include internal or external recording media accessible from a distribution server for distributing the program. The code of the program stored on the distribution server's recording medium may differ from the code of the program in a format executable by a terminal device. That is, as long as it is downloaded from the delivery server and can be installed in a form that can be executed in the terminal device, the form in which it is stored in the delivery server does not matter. It is also possible to divide the program into multiple parts and combine them in the terminal device after downloading at different times, and the delivery servers that transmit the divided programs are different. Moreover, the "computer-readable recording medium" is set to include a recording medium that retains the program for a certain period of time, such as a server in the case of sending a program via a network, and a volatile memory (RAM) inside a computer system that becomes a client. The above program can also be used to implement a part of the above functions. Moreover, it can also be a so-called difference file (difference program) that can realize the above functions by combining with a program already recorded in the computer system.

[0101] While specific embodiments of the present invention have been described above, the present invention is not limited to these embodiments at all, and various modifications and substitutions can be made without departing from the spirit of the present invention.

Claims

1. A secondary battery degradation state estimation system, wherein: The secondary battery degradation state estimation system includes: a state variable measuring unit that measures state variables including an output current and an output voltage of the operating secondary battery at each predetermined timing; a pre-processing unit that outputs input data calculated based on the state variables measured by the state variable measuring unit; as well as a degradation state estimating unit that estimates the degradation state of the secondary battery in operation using the input data output by the pre-processing unit and a learned degradation state model; The pre-processing unit includes a state variable processing unit, which sets a unit charge and discharge amount, and the state variable processing unit sets a recent interval closest to the current time and one or more past intervals before the recent interval as a desired interval obtained based on the unit charge and discharge amount used in calculating the input data, and calculates the voltage change amount obtained based on the state variable as interval data for each of the set intervals, and outputs the calculated interval data by including it in the input data.

2. The degradation state estimation system according to claim 1, wherein: The state variable processing unit sets a period during which a current integrated value obtained by integrating the output current measured by the state variable measuring unit reaches a predetermined value as the desired interval.

3. The degradation state estimation system according to claim 1, wherein: At least one of the calculated charge and discharge amount per desired interval and the calculated voltage change amount per desired interval is a slope change rate calculated using a least squares method.

4. The degradation state estimation system according to claim 1, wherein: The degradation state model is constructed as an RNN, i.e., a recurrent neural network.

5. The degradation state estimation system according to claim 4, wherein: The middle layer of the RNN is constructed as LSTM (Long Short-Term Memory) or GRU (Gated Recurrent Unit).

6. The degradation state estimation system according to any one of claims 1 to 3, wherein: The degradation state model is constructed as a CNN, i.e., a convolutional neural network.

7. A secondary battery degradation state estimation system, wherein: The secondary battery degradation state estimation system includes: a state variable measuring unit that measures state variables including an output current and an output voltage of the operating secondary battery at each predetermined timing; a pre-processing unit that outputs input data calculated based on the state variables measured by the state variable measuring unit; as well as a degradation state estimating unit that estimates the degradation state of the secondary battery in operation using the input data output by the pre-processing unit and a learned degradation state model; The pre-processing unit includes a state variable processing unit, which sets a large charge and discharge amount as a unit charge and discharge amount and a small charge and discharge amount as a unit charge and discharge amount that is smaller than the large charge and discharge amount, and the state variable processing unit sets the most recent large interval and the most recent small interval closest to the current moment and one or more past large intervals and past small intervals that are earlier than the most recent large interval and the most recent small interval as the desired intervals obtained based on the large charge and discharge amount and the small charge and discharge amount used in the calculation of the input data, calculates the voltage change amount obtained based on the state variable as large interval data and small interval data for each of the set large intervals and the small intervals, and outputs the calculated large interval data and the small interval data by including them in the input data.

8. The degradation state estimation system according to claim 7, wherein: The degradation state model is constructed as a CNN, i.e., a convolutional neural network.

9. A method for estimating a degradation state of a secondary battery, wherein: The computer in the degradation state estimation system performs the following processing: measuring state variables including output current and output voltage of the operating secondary battery at each predetermined timing; Output input data calculated based on the measured state variables; estimating a degradation state of the secondary battery in operation using the output input data using a learned degradation state model; A unit charge and discharge amount used in the calculation of the input data is set as a desired interval obtained based on the unit charge and discharge amount, a most recent interval closest to the current moment and one or more past intervals preceding the most recent interval are set, and for each of the set intervals, a voltage change amount obtained based on the state variable is calculated as interval data, and the calculated interval data is included in the input data and output.

10. A storage medium storing a program, wherein: This program causes the computer in the degradation state estimation system to perform the following processing: measuring state variables including output current and output voltage of the operating secondary battery at each predetermined timing; Output input data calculated based on the measured state variables; estimating a degradation state of the operating secondary battery using the output input data using a learned degradation state model; A unit charge and discharge amount used in the calculation of the input data is set as a desired interval obtained based on the unit charge and discharge amount, a most recent interval closest to the current moment and one or more past intervals preceding the most recent interval are set, and for each of the set intervals, a voltage change amount obtained based on the state variable is calculated as interval data, and the calculated interval data is included in the input data and output.

Images