CN110687459A

CN110687459A - Soc estimation method

Info

Publication number: CN110687459A
Application number: CN201910966433.5A
Authority: CN
Inventors: 杨文伟; 许霞; 李宏
Original assignee: Shanghai Industrial Utechnology Research Institute
Current assignee: Shanghai Industrial Utechnology Research Institute
Priority date: 2019-10-12
Filing date: 2019-10-12
Publication date: 2020-01-14
Anticipated expiration: 2039-10-12
Also published as: CN110687459B

Abstract

The invention relates to a soc estimation method, which comprises the following steps of 100: collecting battery history parameters; step 200: inputting the battery history parameters to the LSTM; step 300: the LSTM outputs an estimated SOC; the method is characterized in that the historical parameters of the battery comprise charging times F, which are counted once after the battery is exhausted and recharged again; the full charge SOC corresponds to the charging times F; the battery use time delta T corresponding to the charging times F is the time from the battery exhaustion, the battery operation to the full SOC exhaustion after the battery is charged again to the full SOC; torque TQ and/or speed V within the battery age Δ T. The time recursive neural network of the LSTM is utilized to accurately estimate the SOC, and the method has the advantages of time saving, labor saving and accurate estimation. The battery pack can be applied to the field of portable battery use, such as the fields of new energy automobiles, Internet of vehicles, Internet of things and the like.

Description

Soc estimation method

Technical Field

The invention relates to a SOC estimation method for a total broadband, in particular to an SOC estimation method suitable for various energy storage batteries.

Background

When the price of crude oil rises and global environmental problems lead the rapid development of novel battery energy storage systems. Lead-acid batteries, nickel-metal hydride batteries, nickel-cadmium batteries, lithium ion batteries and the like are the most commonly used batteries in the industry at present. The battery has the advantages of high working battery voltage, little pollution, low self-discharge rate and high power density. The battery is widely applied to pure electric vehicles or hybrid electric vehicles due to its portability.

Estimating SOC (state OF charge) is a basic requirement for using a battery, and SOC is a very important parameter OF a control strategy, so accurate estimation OF SOC can protect the battery, prevent overdischarge, improve battery life, and enable applications to make a reasonable control strategy to save energy. However, batteries are chemical energy storage sources, no means to directly obtain such chemical energy values, and battery models are limited and there is uncertainty in the parameters, making accurate estimation of SOC very complex and difficult to implement. The existing SOC estimation technology mainly comprises (i) direct measurement: the method uses physical battery characteristics such as the voltage and impedance of the battery. (ii) Book-keeping estimation: the method uses the discharge current as an input and integrates the discharge current over time to calculate the SOC. (iii) The self-adaptive system comprises the following steps: the adaptive system is self-designed, can automatically adjust SOC according to different discharge conditions, and various SOC estimation adaptive systems exist in the prior art. (iv) The mixing method comprises the following steps: the hybrid model integrates the three SOC estimation methods, and the advantages are obtained respectively.

However, since the chemical energy characteristics of various energy storage batteries are not consistent, for example, the SOC and the OPEN CIRCUIT VOLTAGE (OCV) of a lithium battery lead-acid battery are approximately in a linear relationship, but the SOC and the OCV of a lithium battery are not in a linear relationship, the above various estimation methods need to be adjusted from time to time, and are time-consuming and labor-consuming, and the existing SOC estimation technology does not consider the influence of the number of times of battery usage and other factors on SOC estimation, which results in structural defects of the estimation methods.

Disclosure of Invention

The invention provides a SOC estimation method, which inputs all historical parameters such as the use times of a battery into a long-SHORT TERM memory model LSTM (Long AND SHORT TERM memory), utilizes a time recursive neural network of the LSTM, AND accurately estimates the SOC.

The invention provides a soc estimation method, which comprises the following steps: step 100: collecting battery history parameters; step 200: inputting the battery history parameters to the LSTM; step 300: the LSTM outputs an estimated SOC; the method is characterized in that the historical parameters of the battery comprise charging times F, which are counted once after the battery is exhausted and recharged again; the full charge SOC corresponds to the charging times F; the battery use time delta T corresponding to the charging times F is the time from the battery exhaustion, the battery operation to the full SOC exhaustion after the battery is charged again to the full SOC; torque TQ and/or speed V within the battery age Δ T.

Preferably, the estimated SOC is derived from the following equation,

the number of charging times Fn is equal to n, f (n) is the weight of SOCn, g is the number of neurons in LSTM, and K is the relationship between the weight W of each neuron in LSTM and SOCn that best meets the predictor torque TQ.

Preferably, the estimated SOC is derived from the following equation,

the number of charging times Fn is equal to n, f (n) is the weight of SOCn, g is the number of neurons in LSTM, and K is the relationship between the weight W of each neuron in LSTM and SOCn that best meets the estimation factor rate V.

Preferably, the estimated SOC is derived from the following equation,

where f (n) is the weight of SOCn, g is the number of neurons in LSTM, and K is the relationship between the weight W of each neuron in LSTM and the torque TQ of SOCn that best meets the predictor.

Preferably, the estimated SOC is derived from the following equation,

where f (n) is the weight of SOCn, g is the number of neurons in LSTM, and K is the relationship between the weight W of each neuron in LSTM and the rate V of SOCn that best fits the predictor.

Preferably, step 100 includes calculating the full SOC from the battery time Δ T and the current I.

Preferably, the battery age Δ T is associated with a load operation recording module.

Preferably, the battery history parameters further include a non-operation parameter, where the non-operation parameter refers to a change of the battery parameter when the load does not operate.

Preferably, the non-operational parameter includes a battery elapsed time Δ T ', and a voltage change Δ U' or a current change Δ I 'within the battery elapsed time Δ T'.

Preferably, the non-operational parameter outputs a non-operational loss Δ SOC'.

Still another soc estimation device, including historical data collection module, LSTM module and display module, its characterized in that, historical data collection module includes: the charging frequency F collecting module is used for recording the charging frequency F of the battery, wherein the charging frequency F is counted once after the battery is exhausted and recharged; the full-charge SOC collection module is used for recording a full-charge SOC, and the charging times F correspond to the full-charge SOC; the battery use time delta T collecting module is used for recording battery use time delta T, and the battery use time delta T is the time from the battery exhaustion to the battery operation to the full SOC after the battery is charged again to reach the full SOC; a torque TQ collection module and a rate Vcollector module to record an open circuit torque TQ and/or a rate V over the battery age Δ T.

Preferably, the charging frequency F collecting module is a counting circuit, and the counting circuit records the charging frequency F.

Preferably, the battery time Δ T collecting module is a timing circuit, and the timing circuit records the battery time Δ T.

Preferably, the full SOC is calculated from the battery time Δ T and the torque TQ.

Preferably, the full SOC is calculated from the battery time Δ T and the rate V.

Preferably, the historical data collecting module further collects non-operation parameters, and the non-operation parameters refer to changes of battery parameters when the load does not operate.

Preferably, the historical data collection module is associated with a load operation recording module.

Preferably, the battery age Δ T collection module records battery age Δ T' when the torque TQ collection module records activity or the rate vtoller module records activity, and the load operation recording module records inactivity; during the battery elapsed time Δ T ', the torque TQ collection module records the torque change Δ TQ ', and/or the rate Vcollection module records the rate change Δ V '.

Preferably, the full SOC collection module collects the torque variation Δ TQ 'and/or the rate variation Δ V', the battery elapsed time Δ T ', and calculates the non-operation loss Δ SOC'.

Preferably, the load operation recording module is a vehicle mileage recording module.

The SOC estimation method provided by the invention inputs all historical parameters such as the battery use times AND the like into a long-SHORT TERM memory model LSTM (Long AND SHORT TERM memory), AND utilizes a time recursive neural network of the LSTM to accurately estimate the SOC, thereby having the beneficial effects of time saving, labor saving AND accurate estimation. The battery pack can be applied to the field of portable battery use, such as the fields of new energy automobiles, Internet of vehicles, Internet of things and the like.

Drawings

FIG. 1 is a schematic diagram of the SOC estimation method of the present invention;

FIG. 2 is a schematic diagram of an SOC estimation apparatus according to the present invention;

FIG. 3 is a flow chart of the SOC estimation method of the present invention.

Detailed Description

The following describes in detail a specific embodiment of the soc estimation method provided by the present invention with reference to the accompanying drawings.

In the drawings, the dimensional ratios of layers and regions are not actual ratios for the convenience of description. When a layer (or film) is referred to as being "on" another layer or substrate, it can be directly on the other layer or substrate, or intervening layers may also be present. In addition, when a layer is referred to as being "under" another layer, it can be directly under, and one or more intervening layers may also be present. In addition, when a layer is referred to as being between two layers, it can be the only layer between the two layers, or one or more intervening layers may also be present. Like reference numerals refer to like elements throughout. In addition, when two components are referred to as being "connected," they include physical connections, including, but not limited to, electrical connections, contact connections, and wireless signal connections, unless the specification expressly dictates otherwise.

The portable battery is widely applied to the fields of new energy automobiles, internet of vehicles, internet of things and the like, and in use, a battery use strategy based on full-charge SOC is particularly important, the full-charge SOC refers to the electric quantity of a battery which is exhausted and charged again to reach a full-charge state once, and how to estimate the next full-charge SOC (estimated SOC)_i+1) The electric quantity becomes an important basis for making a battery use strategy.

The invention provides a soc estimation method, as shown in fig. 1 and 3, comprising the steps of:

step 100: collecting battery history parameters;

step 200: inputting the battery history parameters to the LSTM;

step 300: the LSTM outputs an estimated SOC;

adaptive systems include Back Propagation (BP) neural networks, Radial Basis Function (RBF) neural networks, fuzzy logic methods, support vector machines, fuzzy neural networks, and Kalman, which may be automatically tuned in a constantly changing system. Since batteries are affected by many chemical factors and have non-linear SOC, adaptive systems provide a good solution for SOC estimation. The integrated long-SHORT term memory model LSTM (LONG AND SHORT TERMMEMORY) neural network serving as the self-adaptive system has good nonlinear mapping, self-organizing AND self-learning capabilities AND time recursion, can determine the relation AND the problem of each parameter in SOC estimation when being applied to complex SOC estimation, wherein the relation between an input AND a target is nonlinear, AND the SOC is predicted by using each parameter of a battery based on the LSTM neural network. Therefore, the LSTM time recursive neural network is utilized, and the method has the advantages of accurately estimating the SOC, saving time and labor and accurately estimating.

The battery history parameters comprise charging times F, the number of times F is counted once after the battery is exhausted and recharged, as is known, energy storage chemical substances in the battery can change in physical properties along with the fact that the charging times and the service time of the battery are normal, and the SOC estimation technology does not have structural defects of estimation results due to the fact that an estimation method for the charging times is not considered.

The battery history parameters also comprise a full charge SOC corresponding to the charging times F;

the battery history parameters also comprise a battery use time delta T corresponding to the charging times F, which is the time from the battery exhaustion to the battery operation to the full SOC after the battery is charged again to reach the full SOC;

the battery history parameters also include the torque TQ and/or the rate V over the battery age Δ T.

In this embodiment, as shown in fig. 1, there are g neurons in the LSTM, which are respectively associated with the above battery history parameters, the input layer is the above battery history parameters, and the output layer is the estimated SOC_i+1。

In the present embodiment, the estimated SOC is obtained by the following equation (1),

where the number of charges Fn ═ n, f (n) is the weight of SOCn, g is the number of neurons in LSTM,

k is the relationship between the neuron weights W and SOCn in LSTM that best meet the predictor torque TQ.

In this embodiment, the estimated SOC can also be obtained from the following equation (2),

wherein the number of charging times F_nN, f (n) is the weight of SOCn, g is the number of neurons in LSTM, and K is each of LSTMThe neuron weights W and SOCn most closely match the relationship of the predictor velocity V.

With regard to equations (1) and (2), as the number of battery charges increases, the number of charges F becomes the most dominant factor affecting the full SOC, and therefore the SOC is estimated by taking the fitting relationship SOCn between the number of charges F and the full SOC as the base budget_i+1(ii) a f (n) representing each full SOC (the battery power in the last full state) and the estimated SOC_i+1The magnitude of the fitted weight relationship, in terms of existing battery maintenance and energy storage material replacement possibilities, in combination with data, loss of physical and chemical performance of the battery is an irreversible process, i.e., there is always f (f:)_n)>f(_n-1) Such a relationship. The number of neurons in LSTM, g, K, is the most consistent estimated SOC selected by LSTM choose_i+1The functional relationship between the factor torque TQ or the rate V and the weight W of each neuron will jointly determine the estimated SOC_i+1The size floats. Wherein the best fit estimation of the estimated SOC in equation (1)_i+1The factor of (1) is the torque TQ, the best fit estimated SOC in equation (2)_i+1Factor V.

In another embodiment, the estimated SOC is given by equation (3),

In another embodiment, the estimated SOC may also be derived from equation (4) below,

In equations (3) and (4), as the number of times of charging the battery increases, Δ T becomes a shadow during the battery useThe main factor of the full-charge SOC is influenced, so the fitting relation SOCn between the battery time delta T and the full-charge SOC is taken as a basic budget to estimate the SOC_i+1(ii) a f (n) representing each full SOC (the battery power in the last full state) and the estimated SOC_i+1The magnitude of the fitted weight relationship, in terms of existing battery maintenance and energy storage material replacement possibilities, in combination with data, loss of physical and chemical performance of the battery is an irreversible process, i.e., there is always f (f:)_n)>f(_n-1) Such a relationship. The number of neurons in LSTM, g, K, is the most consistent estimated SOC selected by LSTM choose_i+1The functional relationship between the factor torque TQ or the rate V and the weight W of each neuron will jointly determine the estimated SOC_i+1The size floats. Wherein the best fit estimation of the estimated SOC in equation (3)_i+1The factor of (4) is the torque TQ, the best fit estimated SOC in equation (4)_i+1Factor V.

Note that the SOC is estimated_i+1Can be estimated by one of the above equations (1), (2), (3) or (4) alone or by at least two of the equations (1), (2), (3), (4) together, without the inventors being limited thereto.

In this embodiment, step 100 includes calculating the full SOC from the battery time Δ T and the current I.

In other embodiments, particularly in the field of new energy vehicles, the battery age Δ T is associated with a load operation logging module.

In the field of new energy vehicles, the battery history parameters further comprise non-operation parameters, and the non-operation parameters refer to changes of the battery parameters when the load does not operate. Namely, the battery parameter changes during abnormal leakage of the battery when the vehicle is not in operation.

Thus, the non-operation parameter outputs the non-operation loss Δ SOC ' from the battery elapsed time Δ T ' and the voltage change Δ U ' or the current change Δ I ' within the battery elapsed time Δ T '.

As shown in FIG. 2, the present invention also provides a soc estimation device, which comprises a historical data collection module, an LSTM module 20 and a display module 30.

The historical data collection module comprises a charging frequency F collection module 11 for recording the charging frequency F of the battery, wherein the charging frequency F is counted once after the battery is exhausted and recharged, as is known, along with the normal charging frequency and the normal use time of the battery, the energy storage chemical substances in the battery can change in physical properties, and the SOC estimation technology does not consider the estimation method of the charging frequency to cause the structural defect of the estimation result.

The historical data collection module further comprises a full charge SOC collection module 12 to record a full charge SOC, the number of charges F corresponding to the full charge SOC.

The historical data collection module further comprises a battery use time delta T collection module 13 for recording the battery use time delta T, wherein the battery use time delta T is the time from the battery exhaustion, the recharging to the full-charge SOC and the battery operation to the full-charge SOC exhaustion.

The historical data collection module also includes a torque TQ collection module 14 and/or a rate V collection module 14 to record the torque TQ and/or rate V over the battery age Δ T.

In this embodiment, the charging frequency F collecting module 11 is a counting circuit, and the counting circuit records the charging frequency F.

In this embodiment, the battery age Δ T collecting module 13 is a timing circuit, and the timing circuit records the battery age Δ T.

In the present embodiment, the full SOC is calculated from the battery time Δ T and the torque TQ.

In another embodiment, the full SOC is calculated from the battery time Δ T and the rate V.

In other embodiments, especially in the field of new energy vehicles, the historical data collecting module further collects non-operation parameters, wherein the non-operation parameters refer to changes of battery parameters when a load does not operate a function.

Preferably, the battery age Δ T collection module also records a battery elapsed time Δ T' when the torque TQ collection module 14 records activity or the rate vtip collection module 14 records activity, and the load operation recording module records inactivity;

preferably, the torque TQ collection module 14 also records the torque change Δ TQ ' and/or the rate V collection module records the rate change Δ V ' during the battery elapsed time Δ T '.

The foregoing is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, various modifications and decorations can be made without departing from the principle of the present invention, and these modifications and decorations should also be regarded as the protection scope of the present invention.

Claims

1. A soc estimation method, comprising:

step 100: collecting battery history parameters;

step 200: inputting the battery history parameters to the LSTM;

step 300: the LSTM outputs an estimated SOC;

wherein the battery history parameters include,

the charging frequency F is counted once after the battery is exhausted and is charged again;

the full charge SOC corresponds to the charging times F;

the time delta T of the battery corresponding to the charging frequency F is that the battery is exhausted and is recharged once

After the full SOC, the time when the battery runs to the full SOC is exhausted;

torque TQ and/or speed V within the battery age Δ T.

2. The SOC estimation method according to claim 1, wherein the estimated SOC is obtained by the following equation,

3. The SOC estimation method according to claim 1, wherein the estimated SOC is obtained by the following equation,

wherein the number of charging times F_nWhere n, f (n) is the weight of SOCn, g is the number of neurons in LSTM, and K is the relationship between the weight of each neuron in LSTM W and SOCn that best matches the estimated factor rate V.

4. The SOC estimation method according to claim 1, wherein the estimated SOC is obtained by the following equation,

5. The SOC estimation method according to claim 1, wherein the estimated SOC is obtained by the following equation,

6. The SOC estimation method according to claim 1, wherein step 100 includes calculating the full SOC from the battery time Δ T and the current I.

7. The soc estimation method according to claim 1, wherein the battery age Δ T is associated with a load operation record module.

8. The soc estimation method as claimed in claim 1, wherein the battery history parameters further include non-operation parameters, and the non-operation parameters refer to the variation of the battery parameters when the load is not operating.

9. The soc estimation method of claim 9, wherein the non-operational parameters include a battery elapsed time Δ T ', and a voltage change Δ U' or a current change Δ I 'within the battery elapsed time Δ T'.

10. The SOC estimation method of claim 10, wherein the non-operational parameter outputs a non-operational loss Δ SOC'.