Training model correction system for updating battery state model parameters

Abstract

The invention discloses a training model correction system for updating battery state model parameters. According to the training model correction system, an artificial intelligence machine learning algorithm is used for learning a linear change rule that the battery state data of short-time charging and discharging of the battery changes along with time, so that a battery health state estimation model is trained. The training model correction system uses the battery health state estimation model to estimate battery health state data after the battery is charged and discharged for a longer time, and the battery health state data is used as battery health state estimation data. The training model correction system calculates an error value between the battery health state estimation data and the actual battery health state data so as to correct the battery health state estimation model.

Description

Technical Field

[0001] The invention relates to a battery, and in particular to a training model correction system for updating battery state model parameters. Background Art

[0002] Batteries are widely used in all aspects of people's daily lives. Whether it is for new energy vehicles using batteries or backup power supplies for wireless communication base stations, battery life prediction has become a key factor in the development of various industries. In particular, rapid and accurate prediction of battery life is the core of safety technology and also a difficult problem currently faced. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a training model correction system for updating battery state model parameters in view of the deficiencies of the prior art, comprising a segment data sampling module, an estimation model training module, a battery state estimation module, a battery usage detection module, an estimation error calculation module and a training model correction module. The segment data sampling module is configured to determine a battery voltage range of each battery charge when the number of times the battery is charged and discharged reaches a specified number of times of charge and discharge, and divide the battery voltage range into a plurality of battery voltage segment ranges. The segment data sampling module is configured to sample one of the battery voltage segment ranges from the plurality of battery voltage segment ranges when the battery is charged each time. The specified number of times of charge and discharge is less than a rated number of cycles of the battery. The segment data sampling module is configured to sample only the state data of the battery in one or part of the plurality of charge and discharge time segments as a battery segment sampling data. The estimation model training module is connected to the segment data sampling module. The estimation model training module is configured to collect multiple charging time lengths within a battery charging time range corresponding to the battery voltage segment range sampled by the segment data sampling module and multiple values ​​of battery capacity data when the battery is discharged multiple times as a battery segment sampling data into a training set. The estimation model training module is configured to use an artificial intelligence machine learning algorithm to learn the linear change rules between the multiple values ​​in the training set to train a model, and define the model as a battery health state estimation model. The battery state estimation module is connected to the estimation model training module. The battery state estimation module is configured to use the battery health state estimation model to estimate the battery state data that changes linearly with the charging time within the battery charging time range after any number of charge and discharge cycles of the battery as battery health state estimation data. The battery use detection module is configured to detect the battery state data at at least two time points during the subsequent battery use process as battery use state data. The estimation error calculation module is connected to the battery state estimation module and the battery use detection module. The estimated error calculation module is configured to calculate the error value between the battery health state estimation data and the battery actual state data to output the initial model estimation error data. The training model correction module is connected to the estimated error calculation module and the estimated model training module. The training model correction module is configured to correct the battery health state estimation model according to the initial model estimation error data to form a battery health state estimation correction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0004] Figure 1 It is a block diagram of a training model correction system for updating battery state model parameters according to a first embodiment of the present invention.

[0005] Figure 2FIG. 4 is a block diagram of a training model correction system for updating battery state model parameters according to a second embodiment of the present invention.

[0006] Figure 3 This is a graph of battery voltage versus charging time after charging of a battery established by a battery state prediction system based on interval linear data modeling according to a second embodiment of the present invention.

[0007] Figure 4 This is a graph of battery voltage versus battery capacity after charging, established by the battery state prediction system based on interval linear data modeling according to the second embodiment of the present invention.

[0008] Figure 5 The graph is a graph of a battery capacity versus charging time curve and a correction curve thereof established by a battery state prediction system based on interval linear data modeling according to a second embodiment of the present invention.

[0009] Figure 6 The graph is a prediction curve of battery capacity versus battery charging cycle number and a correction curve thereof established by a battery state prediction system based on interval linear data modeling according to the second embodiment of the present invention.

[0010] Figure 7 The graph is a graph of a prediction curve of battery health status data versus battery charging cycle number and a battery health status estimation curve established by a battery health status prediction system based on interval linear data modeling according to the second embodiment of the present invention.

[0011] Figure 8 A graph showing a battery health state estimation curve, a health state prediction curve and a correction curve of battery health state data versus battery charging cycle number established by a battery state prediction system based on interval linear data modeling according to a second embodiment of the present invention. DETAILED DESCRIPTION

[0012] The following is an explanation of the implementation of the present invention through specific embodiments. Those skilled in the art can understand the advantages and effects of the present invention from the contents disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and the details in this specification can also be modified and changed in various ways based on different viewpoints and applications without departing from the concept of the present invention. In addition, the drawings of the present invention are only for simple schematic illustration and are not depicted according to actual sizes. It is stated in advance.

[0013] See also Figure 1, which is a block diagram of a training model correction system for updating battery state model parameters according to a first embodiment of the present invention. The training model correction system according to the first embodiment of the present invention comprises a segment data sampling module 10, an estimation model training module 20, a battery state estimation module 30, a battery use detection module 40, an estimation error calculation module 50 and a training model correction module 60.

[0014] The estimation model training module 20 is connected to the segment data sampling module 10 , the battery state estimation module 30 and the training model correction module 60 . The estimation error calculation module 50 is connected to the battery state estimation module 30 , the battery use detection module 40 and the training model correction module 60 .

[0015] During the test phase, the number of charge and discharge cycles of the battery reaches a specified number of charge and discharge cycles, which is less than a rated number of charge and discharge cycles of the battery. The rated number of cycles is the maximum number of charge and discharge cycles when the battery is brand new.

[0016] During the test phase, the segment data sampling module 10 determines a battery voltage range of each charge of the battery when the number of times the battery is charged and discharged reaches a specified number of times, divides the battery voltage range into a plurality of battery voltage segment ranges, and samples only one battery voltage segment range from the plurality of battery voltage segment ranges when the battery is charged each time.

[0017] Next, the estimation model training module 20 converts the battery voltage segment range (eg Figure 3 The battery charging time range (e.g., the battery charging voltage data on the vertical axis of the curve diagram framed by the dashed line) corresponds to Figure 3 Multiple charging time lengths within the charging time data on the horizontal axis of the curve graph framed by the dashed line as shown and multiple values ​​of the battery capacity data when the battery is discharged multiple times are collected into the training set as a battery segment sampling data.

[0018] Next, during the actual battery usage phase, the estimation model training module 20 uses an artificial intelligence machine learning algorithm to learn the linear change rules between multiple charging time lengths within the battery charging time range collected in the training set and the battery capacity during multiple discharges to train a model, and define the model as a battery health status estimation model (i.e., the initial model).

[0019] Next, in the actual use stage of the battery, the battery status estimation module 30 uses a battery health status estimation model trained by the estimation model training module 20 to estimate the battery status data that changes linearly with the charging time within the battery charging time range after the battery undergoes any number of charge and discharge cycles, as a battery health status estimation data.

[0020] In the actual battery use stage, the battery use detection module 40 detects the battery status data at at least two time points during the subsequent battery use (charging and discharging), for example Figure 5 As shown, the battery capacity at two or more charging time points is obtained as battery usage status data.

[0021] Next, the estimation error calculation module 50 calculates the error value between the battery health state estimation data estimated by the battery health state estimation module 30 using a battery health state estimation model (ie, initial model) and the battery actual state data to output an initial model estimation error data.

[0022] The training model correction module 60 corrects the battery health state estimation model (i.e., the initial model) constructed by the estimation model training module 20 according to the initial model estimation error data from the estimation error calculation module 50 to form a battery health state estimation correction model (i.e., the final model).

[0023] See also Figures 2 to 8 ,in Figure 2 FIG. 4 is a block diagram of a training model correction system for updating battery state model parameters according to a second embodiment of the present invention.

[0024] Compared with the first embodiment, the training model correction system of the second embodiment of the present invention further includes a segment characteristic curve construction module 70, an estimation curve construction module 80, a practical model establishment module 90, a battery usage status estimation module 100, a practical curve construction module 110 and a correction curve construction module 120.

[0025] First, in the test phase, the segment data sampling module 10 obtains multiple data of battery charging time and battery voltage when the number of times of charging and discharging of the battery reaches a specified number of charging and discharging times (less than a rated charging and discharging cycle number of the battery).

[0026] In the test phase, the segment characteristic curve construction module 70 can obtain the battery charging time and battery voltage data (or / and battery charging time and battery capacity data) obtained from the segment data sampling module 10 when the battery is charged and discharged for a specified number of times, as shown in the following figure. Figure 3 As shown, a battery curve is established on a battery voltage time variation curve graph.

[0027] In the test phase, the segment data sampling module 10 samples one of the battery voltage segment ranges from a battery voltage range divided into multiple battery voltage segment ranges during each charging and discharging process of the battery for a specified number of times. The segment characteristic curve construction module 70 can be based on a battery voltage segment range sampled by the segment data sampling module 10, as shown in the following example. Figure 3 or Figure 4 In a battery voltage time variation curve diagram, one of the curve segments of a battery curve is framed by a dotted line frame as a battery segment characteristic curve.

[0028] The estimation model training module 20 generates a voltage segment corresponding to the battery voltage segment range sampled by the segment data sampling module 10 (or the segment characteristic curve construction module 70 as shown in FIG. Figure 3 Multiple charging time lengths within a battery charging time range (corresponding to a battery segment characteristic curve enclosed by a dotted box on a battery voltage time variation curve diagram), and multiple values ​​of battery capacity data when the battery is discharged multiple times are collected into a training set as a battery segment sampling data.

[0029] Next, the estimation model training module 20 uses an artificial intelligence machine learning algorithm to learn a battery section sampling data collected in the training set to train a battery health state estimation model (ie, an initial model).

[0030] In the actual use stage of the battery, the battery state estimation module 30 uses a battery health state estimation model trained by the estimation model training module 20 to estimate the battery health state data that changes linearly with the charging time length within the battery charging time range after any number of charge and discharge cycles as a battery health state estimation data. The above battery health state estimation data can be expressed as:

[0031] SOH=(Q max / Q n )·100%,

[0032] Among them, SOH represents the health status data of the battery, Qmax refers to the maximum amount of electricity that can be discharged when the battery is fully charged after charging and discharging, and Qn is the rated capacity of the battery.

[0033] In the actual battery usage phase, the estimated curve building module 80 can build, for example, Figure 5 A battery segment characteristic curve in the training set shown in FIG. 1 is on a battery capacity time variation curve graph, and the battery health status estimation data can be used to construct a battery capacity time variation curve graph as shown in FIG. Figure 5 or Figure 6 A battery health state estimation curve of a battery health state estimation model (ie, initial model) shown is connected to a battery segment characteristic curve constructed based on the training set to form a linear curve with the battery segment characteristic curve.

[0034] On the other hand, during the actual battery use phase, the battery use detection module 40 detects the battery status data at at least two time points during the subsequent battery use (charging and discharging), for example, obtaining Figure 5 The battery capacity at the two charging time points indicated by the dotted circles in FIG. 1 is used as the battery usage status data. The practical model building module 90 uses an artificial intelligence machine learning algorithm to learn the data obtained from the battery usage detection module 40 (such as Figure 5 The rules for how the battery capacity changes with the length of the battery charging time are used to construct a battery usage estimation model.

[0035] The battery usage state estimation module 100 uses a battery usage estimation model established by the practical model establishment module 90 to estimate the battery health state data after any charge and discharge cycle of the battery as a battery usage estimation data, including SOH.

[0036] In the actual battery usage phase, the utility curve construction module 110 can construct a battery usage estimation curve on the battery capacity time variation curve graph according to the battery usage estimation data obtained from the battery usage status estimation module 100, such as Figure 5 The battery usage status data enclosed by circles in the figure are connected to form a dotted line, which is close to Figure 5 A battery actual state curve constructed based on actual battery data is indicated by a solid line outside the training set.

[0037] The utility curve building module 110 can use a battery usage estimation model to estimate the battery usage estimation data and the corresponding cycle number to generate a battery usage estimation model. Figure 7 and Figure 8 As shown, a battery usage estimation curve CU21 is constructed on a battery health status change curve diagram.

[0038] During the actual battery usage stage, the estimated error calculation module 50 can calculate the error value between a battery health state estimation data (i.e., data of a battery health state estimation curve) estimated by the battery state estimation module 30 using a battery health state estimation model (i.e., the initial model) and the battery usage estimation data (i.e., data of a battery usage estimation curve) estimated by the battery usage state estimation module 100 using a battery usage estimation model, so as to output an initial model estimation error data.

[0039] Next, in the actual battery use phase, the training model correction module 60 can correct a battery health state estimation model (ie, initial model) into a battery health state estimation correction model (ie, final model) according to the initial model estimation error data obtained from the estimation error calculation module 50.

[0040] In the actual use stage of the battery, the correction curve building module 120 can estimate the correction model (i.e., the final model) according to the battery health status to Figure 5 or Figure 6 As shown, a final correction prediction curve (i.e., the final curve) is constructed on a battery capacity time variation curve graph, which is located between a battery health status estimation curve of battery health status estimation data estimated using an initial model and a battery usage estimation curve of battery usage estimation data estimated using a battery usage estimation model, and is close to an actual curve established based on the actual battery health status data.

[0041] The prediction model training module 20 can establish a battery linear prediction model based on a battery health state estimation data estimated using a battery health state estimation model (i.e., initial model) or based on a battery health state estimation correction model (i.e., final model) and the corresponding number of cycles. The battery state estimation module 30 uses this battery linear prediction model to predict a future battery health state data.

[0042] The estimated curve construction module 80 can be based on a battery health state estimation data estimated using a battery health state estimation model (i.e., the initial model) or based on a battery health state estimation correction model (i.e., the final model) and the corresponding cycle number, such as Figure 7 and Figure 8 A battery health state estimation curve CU11 for establishing a battery linear prediction model is shown on a battery health state change curve diagram.

[0043] The prediction curve construction module 80 can further establish a battery linear prediction curve CU12 on a battery health state change curve diagram according to the future battery health state data of a specified number of cycles predicted by a battery linear prediction model, and connect a battery health state estimation curve CU11.

[0044] The battery usage detection module 40 detects battery status data, such as SOH. The practical model building module 90 builds a battery usage linear prediction model according to the battery status data detected by the battery usage detection module 40 .

[0045] The utility curve building module 110 can use the battery linear prediction model and the corresponding cycle number as follows: Figure 7 and Figure 8 As shown, a battery usage linear prediction curve CU22 is constructed on a battery health status change curve diagram.

[0046] The battery usage state estimation module 100 uses the battery usage linear prediction model to predict the health state data of the battery in future use as the battery future use battery health state data.

[0047] The estimation error calculation module 50 can calculate the error value between a future battery health state data estimated by the battery state estimation module 30 using a battery linear prediction model and a future battery health state data predicted by the battery usage state estimation module 100 using a battery usage linear prediction model, so as to output a final estimation error data. The training model correction module 60 corrects a battery linear prediction model into a battery state prediction correction model according to the final estimation error data.

[0048] The correction curve building module 120 can be based on a battery state prediction correction model such as Figure 7 and Figure 8 As shown, a battery health state prediction correction curve CU4 is established on a battery health state change curve diagram, located between the above-mentioned battery linear prediction curve CU12 and the above-mentioned battery usage linear prediction curve CU22, and is close to the data of a battery actual health state curve CU3 established based on the battery actual health state data.

[0049] In summary, the present invention provides a training model correction system for updating battery status model parameters, which can be used to quickly and accurately predict future battery (health) status data when the battery undergoes more charge and discharge cycles.

[0050] The contents disclosed above are only preferred feasible embodiments of the present invention, and are not intended to limit the scope of the present invention to the claims. Therefore, all equivalent technical changes made using the description and drawings of the present invention are included in the scope of the present invention.

Claims

1. A training model correction system for updating battery state model parameters, characterized in that: The training model correction system for updating battery state model parameters includes: a segment data sampling module, configured to determine a battery voltage range of each charge of the battery when the number of times of charging and discharging of the battery reaches a specified number of charge and discharge times, divide the battery voltage range into a plurality of battery voltage segment ranges, and sample one of the battery voltage segment ranges from the plurality of battery voltage segment ranges when the battery is charged each time, wherein the specified number of charge and discharge times is less than a rated cycle number of the battery; an estimation model training module, connected to the segment data sampling module, configured to collect multiple charging time lengths within a battery charging time range corresponding to the battery voltage segment range sampled by the segment data sampling module and multiple values ​​of battery capacity data when the battery is discharged multiple times as a battery segment sampling data into a training set, use an artificial intelligence machine learning algorithm to learn the linear change rules between the multiple values ​​in the training set to train a model, and define the model as a battery health status estimation model; a battery state estimation module, connected to the estimation model training module, configured to use the battery health state estimation model to estimate the battery state data that changes linearly with the charging time within the battery charging time range after any number of charge and discharge cycles of the battery as battery health state estimation data; A battery usage detection module configured to detect battery status data at at least two time points during subsequent battery usage as battery usage status data; an estimated error calculation module, connected to the battery state estimation module and the battery use detection module, configured to calculate an error value between the battery health state estimation data and the battery actual state data, so as to output an initial model estimation error data; as well as The training model correction module is connected to the estimation error calculation module and the estimation model training module, and is configured to correct the battery health state estimation model according to the initial model estimation error data to form a battery health state estimation correction model.

2. The training model correction system for updating battery state model parameters according to claim 1, characterized in that: The training model correction system for updating battery state model parameters further comprises: The segment characteristic curve construction module is connected to the segment data sampling module and is configured to construct a battery segment characteristic curve on a battery capacity time variation curve graph based on the battery segment sampling data, wherein the vertical axis data of the battery capacity time variation curve graph is the battery capacity included in the battery segment sampling data, and the horizontal axis data of the battery capacity time variation curve graph is the charge and discharge time of the battery.

3. The training model correction system for updating battery state model parameters according to claim 2, characterized in that: The training model correction system for updating battery state model parameters further comprises: The estimation curve construction module is connected to the battery state estimation module and is configured to construct a battery health state estimation curve on the battery capacity time variation curve diagram based on the battery health state estimation data, and connect the battery segment characteristic curve to form a linear curve with the battery segment characteristic curve.

4. The training model correction system for updating battery state model parameters according to claim 3, characterized in that: The training model correction system for updating battery state model parameters further comprises: A practical model building module, connected to the battery use detection module, configured to learn the rule that the battery use status data changes with the number of battery charge and discharge cycles, so as to build a battery use estimation model; as well as A battery usage state estimation module, connected to the practical model establishment module, configured to use the battery usage estimation model to estimate the battery state data when the battery undergoes any number of charge and discharge cycles as a battery usage estimation state data; The estimated error calculation module calculates the error value between the battery health status estimation data and the battery usage estimation status data to output an initial model estimation error data.

5. The training model correction system for updating battery state model parameters according to claim 4, characterized in that: The training model correction system for updating battery state model parameters further comprises: The utility curve construction module is connected to the battery usage status estimation module and is configured to construct a battery usage estimation curve on the battery capacity time variation curve diagram according to the battery usage estimation data.

6. The training model correction system for updating battery state model parameters according to claim 5, characterized in that: The training model correction system for updating battery state model parameters further comprises: The correction curve construction module is connected to the training model correction module and is configured to estimate the correction model based on the battery health status to construct a final correction prediction curve on the battery capacity time change curve diagram.

7. The training model correction system for updating battery state model parameters according to claim 6, characterized in that: The battery state estimation module uses the battery health state estimation model or the battery health state estimation correction model to estimate a battery health state data when the battery undergoes any charge and discharge cycle based on the charging time when the battery is in use; Among them, the estimated curve construction module constructs a battery health status estimation curve on a battery health status change curve graph based on the battery health status data, wherein the vertical axis data of the battery health status change curve graph is the battery health status data and the horizontal axis data is the battery charging time or the number of charge and discharge cycles.

8. The training model correction system for updating battery state model parameters according to claim 7, characterized in that: The battery state estimation module uses the battery health state estimation model or the battery health state estimation correction model to estimate the battery health state data when the battery undergoes any charge and discharge cycle, thereby estimating the battery health state data when the battery undergoes any charge and discharge cycle.

9. The training model correction system for updating battery state model parameters according to claim 8, characterized in that: The prediction model training module estimates the battery health status data and the corresponding cycle number of the battery when the battery undergoes any charge and discharge cycle based on the battery state prediction module to establish a battery linear prediction model, and the battery state prediction module uses this battery linear prediction model to predict future battery health status data.

10. The training model correction system for updating battery state model parameters according to claim 9, characterized in that: The utility model establishment module establishes a battery usage linear prediction model according to the battery status data detected by the battery usage detection module, and the battery usage status estimation module uses the battery usage linear prediction model to predict the battery health status data of a battery in the future; The estimated error calculation module calculates the error value between the battery's future battery health status data and the future battery health status data to correct the battery linear prediction model to form a battery status prediction correction model.

Images