An input feature construction method suitable for data-driven SOC estimation

Abstract

The application relates to an input feature construction method suitable for data-driven SOC estimation. The SOC data-driven estimation model input feature construction comprises the following steps: modeling and construction of an open-circuit voltage input feature, construction of an open-circuit voltage curve smoothing input feature, construction of an end voltage curve smoothing input feature, fusion construction of the open-circuit voltage curve smoothing input feature and the end voltage curve trend feature, and SOC input feature construction based on an ampere-hour integral equation. After the input feature is determined, sampling processing is carried out, and the SOC is estimated at the sampling point. Through the construction of the effective input feature, the SOC data-driven modeling is promoted, and the working condition adaptability and estimation precision of the SOC data-driven estimation model can be improved.

Description

TECHNICAL FIELD

[0001] The application relates to an input feature construction method, in particular to an input feature construction method suitable for data-driven SOC estimation, and belongs to the technical field of lithium ion battery performance evaluation. BACKGROUND

[0002] Lithium ion batteries have become the main power source of electric vehicles due to their good performance, and effective state of charge (SOC) estimation is of great significance to the safe driving and thermal management of electric vehicles. Data-driven SOC estimation methods have attracted widespread attention because they do not need to consider the complex reaction mechanism inside the battery and do not need to balance the accuracy and complexity of the battery model. However, in practical applications, there are fewer measurable parameters of the battery, which is not conducive to the establishment of the SOC data-driven estimation model. In addition, the strong nonlinear characteristics of the measurable parameters make the SOC data-driven estimation model established under a specific environment no longer applicable under other temperature environments, resulting in poor SOC estimation results.

[0003] At present, most data-driven SOC estimation methods still use only measurable voltage, current and temperature as input features of the SOC data-driven estimation model. For example, the SOC estimation method based on RVM proposed in Chinese invention patent applications CN116068412A and CN113933725A, in which the model input is: voltage, current and temperature variables; Chinese invention patent CN112782594B takes environmental temperature, discharge current, terminal voltage and internal resistance as input of the SOC data-driven estimation model; Chinese invention patent CN106154168B takes voltage, current, temperature and historical SOC as input feature quantities to establish the SOC data-driven estimation model. The above-mentioned patent methods all model and predict under a specific environment, and even with limited measurable input features, good SOC estimation results can be obtained. However, the traditional input features (voltage, current, etc.) used in the above-mentioned patents make the SOC data-driven estimation model established have poor working condition adaptability. SUMMARY

[0004] To solve the problems in the background art, the application provides an input feature construction method suitable for data-driven SOC estimation, which promotes SOC data-driven modeling by constructing effective input features, and can improve the working condition adaptability and estimation accuracy of the SOC data-driven estimation model.

[0005] To achieve the above-mentioned purpose, the application adopts the following technical scheme: an input feature construction method suitable for data-driven SOC estimation, comprising the following steps:

[0006] Step 1: SOC data-driven estimation model input feature construction

[0007] 1.1 Modeling and construction of open-circuit voltage input feature

[0008] Voltage difference between terminal voltage and open-circuit voltage based on reaction mechanism inside the battery The sum of voltage change caused by ohmic internal resistance and polarization voltage, i.e.,

[0009]

[0010] where k represents discrete time, U t represents terminal voltage, U OCV represents open-circuit voltage, I represents random discharge current, R0 represents ohmic internal resistance, U p represents polarization voltage,

[0011] A data-driven model between terminal voltage, random discharge current and voltage difference is established using a data-driven algorithm as follows:

[0012]

[0013] where, represents data-driven algorithm, M represents the data-driven model established between input features U t , I and

[0014] The final expression of open-circuit voltage input feature is constructed as follows:

[0015]

[0016] where OCV e represents open-circuit voltage curve constructed based on data-driven, represents the predicted after data-driven model M;

[0017] 1.2 Construction of open-circuit voltage curve smoothing input feature

[0018] The open-circuit voltage curve constructed based on data-driven is smoothed, and the average value of open-circuit voltage in window N at current time is taken as the open-circuit voltage smoothed and constructed at current time, and the expression is as follows:

[0019]

[0020] where, represents open-circuit voltage curve smoothed and constructed, represents open-circuit voltage parameter value in window size N; ​

[0021] 1.3 Construction of smoothed input feature of terminal voltage curve

[0022] Similarly, the terminal voltage curve is smoothed, and the expression is as follows:

[0023]

[0024] In the formula, represents the smoothed terminal voltage curve, represents the terminal voltage parameter value in the window size N;

[0025] 1.4 Fusion construction of smoothed input feature of open-circuit voltage curve and trend feature of terminal voltage curve

[0026] For a terminal voltage curve of a battery in a period of time , the relationship curve between time and terminal voltage is fitted, and the fitting equation of the relationship curve is set as:

[0027]

[0028] In the formula, x represents time, y represents terminal voltage, and are coefficients of the fitting equation,

[0029] The trend feature of the terminal voltage curve is:

[0030]

[0031] In the formula, represents the trend curve of the terminal voltage curve, and T represents time,

[0032] The smoothed input feature of the open-circuit voltage curve and the trend feature of the terminal voltage curve are fused to construct a new input feature, and the expression is as follows:

[0033]

[0034] In the formula, represents the fusion curve of the smoothed open-circuit voltage curve and the trend curve of the terminal voltage curve;

[0035] 1.5 Construction of SOC input feature based on ampere-hour integral method equation

[0036] The SOC based on the ampere-hour integral method equation is constructed as an input feature for SOC modeling, that is:

[0037]

[0038] In the formula, SOC(k) represents the SOC at the current time, SOC(k-1) represents the SOC at the previous time, and I(k) represents the current at the current time. represents the sampling time, C cap represents the capacity of the battery;

[0039] Step two: processing of the input features constructed based on the feature points

[0040] Based on any one of the five input features constructed in step one, the size between the selected feature value at the current time and the selected feature value at the previous time is judged in real time during the construction process, and the smallest feature value in the storage space is recorded and updated continuously, so as to sample the selected input feature during the discharging process and extract and construct the input feature with a monotonically decreasing trend. Meanwhile, the training data set used in the SOC data-driven estimation model is processed in the same way based on the sampling standard of the selected input feature.

[0041] Compared with the prior art, the beneficial effects of the present application are as follows: the present application determines the input features of the SOC data-driven estimation model through the modeling and construction of the open-circuit voltage input feature, the construction of the open-circuit voltage curve smoothing input feature, the construction of the terminal voltage curve smoothing input feature, the fusion construction of the open-circuit voltage curve smoothing input feature and the terminal voltage curve trend feature, and the SOC input feature construction based on the ampere-hour integral equation, then further samples the constructed input features based on the monotonous change trend of the features, and estimates the SOC at the sampling point, which provides the input features conducive to modeling, provides a basis for obtaining good SOC estimation results, is suitable for different application environments, improves the working condition adaptability of the SOC data-driven estimation model, can be constructed and obtained online in real time, is conducive to the online deployment of the model, does not require a large amount of training data, is easy to establish the mapping relationship between the constructed input features and the SOC, is more conducive to practical application, can provide a basis for range prediction, aging state estimation and the like, and is conducive to the safe driving of electric vehicles. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 is a flowchart of the present application;

[0043] Figure 2 is a principle diagram of the present application after smoothing processing of the open-circuit voltage curve and the terminal voltage curve. DETAILED DESCRIPTION

[0044] The technical solutions in the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0045] As Figures 1-2 shown, an input feature construction method suitable for data-driven SOC estimation, the specific process is combined with Figure 1 shown, comprising the following steps:

[0046] Step one: construction of SOC data-driven estimation model input features

[0047] 1.1 Modeling and construction of open circuit voltage input features

[0048] The open circuit voltage has a clear mapping relationship with the SOC, which is generally characterized by a high-order polynomial. Therefore, once the open circuit voltage of the battery is obtained, the SOC can be obtained based on the relationship between the open circuit voltage and the SOC. However, the open circuit voltage, like the SOC, is difficult to establish a mapping relationship between the voltage, current and other variables with strong nonlinear characteristics and the open circuit voltage through a data-driven method. In the SOC data-driven estimation model, the SOC at the current time changes based on the SOC at the previous time, so the SOC curve has obvious rising or falling time series characteristics over a long period of time. However, the current of the input feature is random, and the voltage of the input feature is also based on the response of the random current. Based on the reaction mechanism inside the battery, the response equation of the internal terminal voltage of the battery can be summarized as:

[0049]

[0050] In the formula, k represents the discrete time, U t represents the terminal voltage, U OCV represents the open circuit voltage, I represents the random discharge current, R0 represents the ohmic resistance, and U p represents the polarization voltage. Therefore, from the above formula, it can be seen that the terminal voltage fluctuates obviously with the random discharge current, and the random fluctuation characteristics of the terminal voltage curve also bring adverse disturbance information to the modeling of the SOC.

[0051] According to the above formula, the voltage difference between the terminal voltage and the open circuit voltage is the sum of the voltage change caused by the ohmic resistance and the polarization voltage, that is:

[0052]

[0053] It can be seen that it is easy to establish a mapping relationship between the voltage difference and the random discharge current based on data-driven. Therefore, the data-driven model between the terminal voltage, the random discharge current and the voltage difference is established using a data-driven algorithm as follows:

[0054]

[0055] In the formula, represents a data-driven algorithm, M represents The established input feature U t , I and between the data-driven model.

[0056] Therefore, by converting the data-driven model between the established terminal voltage, random discharge current and open circuit voltage into a data-driven model between the established terminal voltage, random discharge current and voltage difference, the influence of the random fluctuation characteristics of the terminal voltage and random discharge current on the modeling of the open circuit voltage is eliminated, and the modeling accuracy and modeling effect are improved.

[0057] The final open circuit voltage input feature expression is constructed as follows:

[0058]

[0059] In the formula, OCV e represents the open circuit voltage curve constructed based on data driving, represents the predicted after the data-driven model M;

[0060] 1.2 Construction of open circuit voltage curve smoothing input feature

[0061] Compared with the traditional voltage curve with obvious random fluctuation characteristics, the random fluctuation characteristics of the open circuit voltage input feature constructed based on step 1.1 are greatly reduced, which is more beneficial to the modeling of SOC. However, the open circuit voltage curve constructed based on data driving inevitably produces random estimation errors, which may affect the modeling results of SOC. Therefore, in order to suppress the influence of random estimation errors on the modeling of SOC, the open circuit voltage curve constructed based on data driving needs to be smoothed to construct a more smooth open circuit voltage curve.

[0062] As shown in Figure 2 , set the window N, and take the average value of the open circuit voltage in the current time window N as the current time smooth constructed open circuit voltage, the expression is as follows:

[0063]

[0064] In the formula, represents the smooth constructed open circuit voltage curve, represents the open circuit voltage parameter value in the window size N. After the above smoothing processing, the influence of the random estimation errors caused by the open circuit voltage curve constructed based on data driving in step 1.1 on the modeling of SOC can be suppressed to a certain extent;

[0065] 1.3 Construction of terminal voltage curve smoothing input feature

[0066] As shown in Figure 2As shown, similarly, the end voltage curve is smoothed, and the expression is as follows:

[0067]

[0068] In the formula, represents the smoothed end voltage curve, represents the end voltage parameter value in the window size N. After the above smoothing, the influence of the random fluctuation characteristics of the end voltage curve on the SOC modeling is reduced, and the role of the original end voltage curve in the SOC modeling is maximized;

[0069] 1.4 Fusion construction of open circuit voltage curve smoothing input feature and end voltage curve trend feature

[0070] Based on the open circuit voltage input feature constructed in step 1.1, for the end voltage of the battery, if the current is continuously discharged, the end voltage will tend to the discharge cutoff voltage, so from the overall discharge process, the entire end voltage curve presents a downward trend, but the random characteristics of the current and the influence of the internal polarization characteristics of the battery cause the entire end voltage curve to present a random fluctuation trend, which will also affect the SOC modeling effect. Therefore, by extracting the trend curve of the entire end voltage curve, the influence of the random fluctuation characteristics of the end voltage on the SOC modeling is eliminated.

[0071] For the end voltage curve of the battery in a period of time , the relationship curve between the fitting time and the end voltage is fitted, and the fitting equation of the relationship curve is set as:

[0072]

[0073] In the formula, x represents time, and y represents end voltage, and are the coefficients of the fitting equation.

[0074] Then the end voltage curve trend feature is:

[0075]

[0076] In the formula, represents the trend curve of the end voltage curve, and T represents time.

[0077] The open-circuit voltage curve smoothing input feature constructed in step 1.2 characterizes the average lithium ion concentration of the positive and negative electrodes in the battery with the electrode overpotential, while the above-mentioned terminal voltage curve trend feature characterizes the overall trend change of the lithium ion concentration of the positive and negative electrodes in the battery during the entire charging and discharging cycle. Both of the constructed features characterize the relationship between the relevant characteristics in the battery and the SOC from different angles. Therefore, in order to consider and construct the average lithium ion concentration of the positive and negative electrodes in the battery with the electrode overpotential and the overall trend change of the lithium ion concentration of the positive and negative electrodes during the entire charging and discharging cycle, a new input feature is constructed by fusing the open-circuit voltage curve smoothing input feature and the terminal voltage curve trend feature, and the expression is as follows:

[0078]

[0079] In the formula, represents the fusion curve of the trend curve of the smoothed open-circuit voltage curve and the terminal voltage curve;

[0080] 1.5 SOC input feature construction based on ampere-hour integral method equation

[0081] When the initial value of SOC is inaccurate, the SOC calculated by the ampere-hour integral method equation has a large error. However, even if affected by the inaccurate initial value of SOC, the overall trend of the SOC curve calculated based on the ampere-hour integral method equation has a strong correlation with the accurate reference SOC curve, and the change trend is beneficial to the modeling of SOC. In addition, the SOC curve calculated based on the ampere-hour integral method equation has no random fluctuation characteristics, which will suppress the influence of the input feature constructed based on the terminal voltage curve in the modeling of SOC. Therefore, the SOC based on the ampere-hour integral method equation is constructed as the input feature for SOC modeling, that is:

[0082]

[0083] In the formula, SOC(k) represents the SOC at the current time, SOC(k-1) represents the SOC at the previous time, I(k) represents the current at the current time, represents the sampling time, C cap represents the capacity of the battery;

[0084] Step two: processing of the input feature constructed based on feature point sampling

[0085] For the five input features constructed in step one, in order to further reduce the influence of random fluctuation characteristics of the input features on the modeling of SOC, an input feature processing strategy based on feature point sampling is set based on the change trend of the constructed features. Based on the five input features constructed in step one, on the basis of any input feature, the size between the selected feature value at the current time and the selected feature value at the previous time is judged in real time during the construction process, and the smallest feature value in the storage space is recorded and updated continuously, so as to sample and process the selected input feature during the discharging process and extract and construct the input feature with a monotonically decreasing trend.

[0086] At the same time, based on the sampling standard of the selected input feature, the same processing is performed on the remaining input features constructed in step one. In addition, the training data set used in the SOC data-driven estimation model is also processed in the same way.

[0087] Embodiment

[0088] This embodiment verifies the five constructed input features under the New European Driving Cycle (NEDC) working condition at different temperatures (-10℃, 0℃, 10℃, 25℃ and 45℃), and the selected battery type is ternary lithium ion battery, and the nominal capacity of the battery is 57 Ah. Table 1 is the Pearson correlation coefficient value between the constructed input features and SOC at different temperatures.

[0089] Table 1 Pearson correlation coefficient value between the constructed input features and SOC

[0090]

[0091] From the above experimental results, it can be seen that the correlation between the constructed input features and SOC is very strong, so it can be used to establish a better SOC data-driven estimation model.

[0092] It is obvious to those skilled in the art that the present application is not limited to the details of the above exemplary embodiments, and the present application can be implemented in other embodiments without departing from the spirit or essential characteristics of the present application. Therefore, the embodiments should be regarded as exemplary and non-limiting, and the scope of the present application is defined by the appended claims rather than the above description, and therefore all changes falling within the meaning and scope of the equivalent conditions of the claims are intended to be included in the present application. Any reference signs in the claims should not be regarded as limiting the claims involved.

[0093] Furthermore, it should be understood that although the specification is described in terms of embodiments, not every embodiment includes every feature or implementation described herein. The specification can include implicit combinations of explicitly mentioned features and / or implicit combinations of implicitly mentioned features. Such combinations are also expressly included within the scope of the specification and an embodiment.

Claims

1. An input feature construction method suitable for data-driven SOC estimation, characterized by comprising the following steps: Step 1: Construction of SOC data-driven estimation model input features 1.1 Modeling and construction of open-circuit voltage input feature The voltage difference between the terminal voltage and the open circuit voltage based on the reaction mechanism inside the battery is the sum of the voltage change due to the ohmic internal resistance and the polarization voltage, i.e.: where k represents a discrete time, U t represents an open circuit voltage, I represents a random discharge current, R0represents an ohmic internal resistance, U OCV represents an open circuit voltage, I represents a random discharge current, R0represents an ohmic internal resistance, U p represents a polarization voltage, The data-driven model between terminal voltage, random discharge current and voltage difference is established using data-driven algorithm as follows: wherein represents a data-driven algorithm, M represents the established input feature U t , I and a data-driven model between The final construction of open-circuit voltage input feature expression is as follows: In the formula, OCV e represents an open-circuit voltage curve constructed based on data driving, represents a predicted value of the open-circuit voltage after passing through the data-driven model M ; 1.2 Construction of open-circuit voltage curve smoothing input feature The open-circuit voltage curve based on data-driven is smoothed, and the average value of open-circuit voltage in the current time window N is taken as the smoothed open-circuit voltage at the current time, and the expression is as follows: wherein represents the smoothed open circuit voltage curve, represents the open circuit voltage parameter value in the window size N; 1.3 Construction of terminal voltage curve smoothing input feature Similarly, the terminal voltage curve is smoothed, and the expression is as follows: wherein denotes the smoothed end voltage curve, denotes the end voltage parameter value in the window size N; 1.4 Fusion construction of open-circuit voltage curve smoothing input feature and terminal voltage curve trend feature for the end voltage curve of the battery over a period of time a curve of relationship between the fitting time and the end voltage, the fitting equation of the curve of relationship being set as: where x represents time and y represents the terminal voltage, and are coefficients of the fitted equation, The terminal voltage curve trend feature is: wherein a trend line representing the terminal voltage curve, T represents time, The new input feature is constructed by fusing the open-circuit voltage curve smoothing input feature and the terminal voltage curve trend feature, and the expression is as follows: In the formula, represents a fusion curve of the trend curve of the open-circuit voltage curve and the terminal voltage curve after smoothing. 1.5 SOC input feature construction based on ampere-hour integral method equation The SOC based on ampere-hour integral method equation is constructed as the input feature for SOC modeling, that is: In the formula, SOC(k) represents the SOC at the current time, SOC(k-1) represents the SOC at the previous time, I(k) represents the current at the current time, represents the sampling time, C cap represents the capacity of the battery; Step two: processing of input feature constructed based on feature point sampling Based on the five input features constructed in step one, any input feature is taken as the basis, and the size between the selected feature value at the current time and the selected feature value at the previous time is judged in real time during the construction process, and the minimum feature value in the storage space is recorded and updated constantly. The selected input feature is sampled and processed during discharge, and the input feature with monotonic decreasing trend is extracted and constructed, and at the same time, the training data set used in the SOC data-driven estimation model and the remaining input features constructed in step one are processed in the same way.

Images