TWI461718B - Battery power test method - Google Patents

Description

Battery test method

The invention relates to a test method, in particular to a battery power test method.

At present, the battery power test method mainly includes an open circuit voltage method and an improved coulomb method. The open circuit voltage method uses the principle that the terminal voltage value will decrease with the discharge of the battery when discharging, and the battery power is known by measuring the open circuit voltage of the battery, but the dynamic residual battery capacity cannot be effectively estimated. The improved Coulometric method, because the battery is continuously discharged, the total amount of electrochemical reaction is reduced, so it is impossible to provide accurate battery power during the discharge process.

In addition, there is an ampere-hour integral method, which takes the terminal voltage and load current of the battery in a single-chip system timing, and calculates the accumulated consumption current to estimate the residual battery power. According to the charge and discharge characteristics of the battery, the relative value of the battery power that can be used is estimated. However, due to the complicated internal electrochemical reaction of the battery, the battery terminal voltage and the battery power may have a nonlinear relationship during the discharge process. Therefore, the ampere-hour integral method cannot effectively test the remaining energy of the battery.

Accordingly, it is an object of the present invention to provide a battery power test method.

Therefore, the battery power test method of the present invention is applicable to a battery having a test battery parameter set, including a first correction step and a training step. And a test step. The first correcting step corrects a plurality of training battery parameters and power corresponding groups to obtain a plurality of corrected training battery parameters and power corresponding groups. The training step trains a classifier according to the modified battery parameters and the power corresponding group. The test step inputs the test battery parameter set into the classifier to obtain the test battery power of the battery.

The effect of the present invention is that, by the correction step and the analysis capability of the classifier, after the training process is completed, the parameters of the battery are input to the classifier to obtain an accurate test battery power.

The above and other technical contents, features and advantages of the present invention will be apparent from the following detailed description of the preferred embodiments of the invention.

Referring to FIG. 1 and FIG. 2, the first preferred embodiment of the battery power test method of the present invention is applicable to a battery 9 having a test battery parameter set, and is implemented by the battery power test system 2 as shown in FIG. The battery 9 can be, for example, a rechargeable battery of an electric motor vehicle or the like. As shown in FIG. 1 , the first preferred embodiment of the battery power test system 2 includes a sensing device 201 , a first correcting unit 202 , a classifier 203 , a sensing device 211 , a classifier 203 , and a display device . 213. The above components 201, 202, and 203 relate to the operation flow of a training phase, and the components 203, 211, and 213 are related to the operational flow of a test phase.

The training stage is directed to a specific type of battery 8. The vehicle manufacturer needs to sense a plurality of training battery parameters and power corresponding groups by using the sensing device 201 in the R&D and production process, and then correct the supply by using the first correcting unit 202. Training the battery parameter and the power corresponding group (first correction step S1) to obtain a plurality of modified training battery parameters and power corresponding groups, and then training the classifier 203 according to the modified training battery parameters and the power corresponding group ( Training step S2). As for the test phase, when the locomotive rider rides the electric motor vehicle on the road, the induction device 211 of the electric motor vehicle obtains the test battery parameter set of the battery 9 in the electric motor vehicle, and inputs it into the classifier 203 installed in the vehicle. The classifier 203 then proceeds to a test step S3 to calculate the test battery power displayed on the display device 213 of the electric motor vehicle.

As shown in the first modification step S1 of the training phase of FIG. 2, the first correction unit 202 corrects the training battery parameters and the power corresponding group to obtain the corrected training battery parameters and the power corresponding group. Each of the training battery parameters and the power corresponding group includes a training current I _train , a training voltage V _train and a corresponding training battery power SOC _train , and each modified training battery parameter and power corresponding group includes the For training current I _train , a modified training voltage V _train ' and the corresponding training battery SOC _train .

In the preferred embodiment, the training battery parameters and the power corresponding group have a total of 500 groups. The training current I _train and the training voltage V _train in each group for the training battery parameter and the power corresponding group are received by the sensing device 201 at intervals of the CNS-D3029 driving cycle standard. Current and voltage, and the corresponding SOC _train for training battery is a percentage of the remaining capacity after subtracting the amount of electricity discharged at that time from the rated capacity of the battery.

Next, the sensing device 201 transmits each of the received training currents I _train , the training voltage V _train and the corresponding training battery power SOC _train to the first correcting unit 202 . Then, the first correcting unit 202 adds a correction function f(I _train ) to the training voltage V _{train of the} same group according to each training current I _train to obtain the corrected training voltage V _train ' . In the preferred embodiment, the function is for multiplying the training current I _train by a constant c, which can be expressed by the following equation: V _train '=V _train +f(I _train )=V _train +c‧I _train

The optimum value of the constant c may vary depending on the type of the battery, but since the training data is usually pre-normalized, the optimum value of the constant c is usually between 0.1 and 0.9; It has been experimentally observed that the battery 8 used in the preferred embodiment has the best training effect when the constant c is 0.525.

Furthermore, between the training battery parameters and the corrected training voltage V _train ' in the power supply corresponding group, since the corresponding training voltage V _train is received by the sensing device 201 at a certain interval, A prioritized relationship over time. Value of each corrected value for training voltage V _train 'will again after the previous correction of at least two adjacent training supply voltage V _train' re-training for correcting the correction voltage V _train 'itself.

For example, the supply voltage V _train training is 51.53V, this time for training is 1.7A current I _train, after the training for the correction corresponding voltage V _train 'is 52.4225V.

In this embodiment, the reference interval is set to 5 data, and then the training voltage V _train ' is applied to the first two and the last two corrections in the sequence, which are 52.55825V, 52.2905V, 52.4225V, 52.4815V, 52.17375. V, taking the average value can get 52.3853V, so the last set of training battery parameters and power supply corresponding group is (1.7, 51.53; 81.52%), the corresponding corrected training battery parameters and power corresponding group are (1.7 , 52.3853; 81.52%).

Referring to FIG. 1, FIG. 2 and FIG. 3, as shown in the training step S2 of the training phase, the classifier 203 is trained according to the modified training battery parameters and the power corresponding group. In the preferred embodiment, the classifier 203 is a probability-like neural network model, including an input layer, a hidden layer, and an output layer. The hidden layer includes a plurality of neurons (taking five neurons as an example). ), each neuron corresponds to a weight directly determined by a simple algebraic operation.

The plurality of sets of the training current I _train obtained by the first correcting unit 202, the corrected training voltage V _train ' and the corresponding training battery power SOC _train are configured for training The battery parameter and the power corresponding group will become a training set (Training Set) for inputting the classifier 203 to determine the weights. For example, in the first correcting step S1, the first correcting unit 202 corrects and obtains five sets of corrected training battery parameters and power corresponding groups, respectively (I ₁ , V ₁ '; SOC ₁ ), ( I ₂ , V ₂ '; SOC ₂ ), ( I ₃ , V ₃ '; SOC ₃ ), ( I ₄ , V ₄ '; SOC ₄ ) and ( I ₅ , V ₅ '; SOC ₅ ). As shown in FIG. 3, the hidden layer has five corresponding neurons, having weights H ₁ , H ₂ , H ₃ , H _{4 ,} and H ₅ , respectively, assuming the test battery parameters to be input in the test step S3. The group is ( I _x , V _x ), and the corresponding test battery power is SOC _x . After the training step S2, the five weights are respectively set to: , net ₁ = (I _x -I ₁ ) ² +(V _x -V ₁ ) ²

, net ₂ = (I _x -I ₂ ) ² +(V _x -V ₂ ) ²

, net ₃ = (I _x -I ₃ ) ² +(V _c -V ₃ ) ²

, net ₄ = (I _x -I ₄ ) ² +(V _x -V ₄ ) ²

, net ₅ = (I _{x -} I ₅ ) ² + (V _x - V ₅ ) ² The test battery power SOC _{x obtained} is:

Where σ is 10 ^-3 , which is a preset value of a probability type neural network, and I _x , V _x , and SOC _x are unknown variables in the training step S2.

Since this weight setting can be completed in a simple algebraic relationship, the calculation time required for this step is extremely short, and it takes a long time compared to other types. The neural network of the learning or iterative process, the probability-like neural network used in the present invention not only has a short training time, but also the method can be used to control the wafer implementation by the micro-processing, and the cost is lower and more efficient.

Next, as shown in the test step S3 of the test phase, the test battery parameter set is input into the classifier to obtain the test battery power of the battery. Since the classifier 203 has been constructed after the training step S2, the instantaneous current I _test and the instantaneous voltage V _test received by the sensing device 211 are used as the test battery parameter group for inputting the classifier. After 203, the test battery power SOC _{test is obtained} and output to the display device 213.

Wherein H ₁ , H ₂ , H ₃ , H ₄ and H ₅ are: , net ₁ = (I _test -I ₁ ) ² +(V _test -V ₁ ) ²

, net ₂ = (I _test -I ₂ ) ² +(V _test -V ₂ ) ²

, net ₃ = (I _test -I ₃ ) ² +(V _test -V ₃ ) ²

, net ₄ = (I _test -I ₄ ) ² +(V _test -V ₄ ) ²

, net ₅ = (I _test -I ₅ ) ² +(V _test -V ₅ ) ²

Due to the characteristics of the probability-like neural network, in the test step S3, only the instant current I _test and the instantaneous voltage V _{test need} to be substituted into I _x and V _x in the training step S2, respectively, and the test battery at that time can be obtained instantly. The SOC _test is very suitable for immediately obtaining the residual power of the battery at the time of discharge.

Referring to FIG. 4 and FIG. 5, a second preferred embodiment of the battery power testing method of the present invention further includes a second correcting step S4 between the training step S2 and the testing step S3, and the system 2 further includes a Second correction unit 212. As shown in FIG. 5, the second correcting unit 212 is configured to correct one of the battery parameter groups of the battery 9 received by the sensing device 211 as the test battery parameter set. The instant battery parameter set includes the instantaneous current I _test and the instantaneous voltage V _test , and the test battery parameter set includes the immediate current I _test and a test voltage V _test ' . The second correcting unit 212 first adds a correction function f(I _test ) to the instantaneous voltage V _test according to the instantaneous current I _test to obtain the test voltage V _test ' , and then inputs the classifier 203. In the preferred embodiment, the function multiplies the instantaneous current I _test by a constant d , which can be expressed by the following equation: V _test '=V _test +f(I _test )=V _test +d‧I _test

The constant d varies depending on the battery model, but is usually between 0.1 and 0.9, which is 0.525 in the preferred embodiment. Due to the functions of the first correcting unit 202 and the second correcting unit 212, the test battery power is further closer to the real situation.

In summary, the main theoretical basis of the probability-like neural network used in the present invention is based on the Bayesian decision-making, using a multi-layer network architecture of a probability-like neural network, even for a sparse sample space, for the wrong information. It is quite tolerant, so even if the discharge of the battery is a non-linear model, the probability-like neural network cooperates with the first modification step S1 and the second correction step S4 of the present invention, which is not only efficient but also low in cost, and The battery power at the time is correctly tested during the discharge of the battery, so that the object of the present invention can be achieved.

The above is only the preferred embodiment of the present invention, and the scope of the invention is not limited thereto, that is, the simple equivalent changes and modifications made by the scope of the invention and the description of the invention are All remain within the scope of the invention patent.

S1‧‧‧First Amendment Step

S2‧‧‧ training steps

S3‧‧‧ test procedure

S4‧‧‧ second amendment step

2‧‧‧Battery power test system

201‧‧‧Induction device

202‧‧‧First Correction Unit

203‧‧‧ classifier

211‧‧‧Induction device

212‧‧‧Second correction unit

213‧‧‧ display device

8‧‧‧Battery

9‧‧‧Battery

I ₁ to I ₅ ‧‧‧ for training current

V ₁ ' to V ₅ '‧‧‧corrected for training voltage

H ₁ to H ₅ ‧‧ ‧ weights

SOC ₁ to SOC ₅ ‧‧‧ for training battery power

I _x , V _x ‧‧‧ for test battery parameter set

SOC _x ‧‧‧Test battery power

1 is a system block diagram showing a battery power test system for implementing the first preferred embodiment of the battery power test method of the present invention; FIG. 2 is a flow chart illustrating a first preferred embodiment of the method of the present invention Figure 3 is a schematic diagram showing a probability-like neural network; Figure 4 is a flow chart illustrating a second preferred embodiment of the method of the present invention; and Figure 5 is a system diagram illustrating the operation of the battery of the present invention A battery power test system of a second preferred embodiment of the test method.

S1‧‧‧First Amendment Step

S2‧‧‧ training steps

S3‧‧‧ test procedure

Claims (8)

A battery power test method is applicable to a battery having a test battery parameter set, the method comprising the following steps: a first correcting step, correcting a plurality of training battery parameters and a power corresponding group to obtain a plurality of corrected Training battery parameters and power corresponding groups, wherein each of the training battery parameters and the power corresponding group includes a training current, a training voltage, and a corresponding training battery power, and each modified battery parameter and power corresponding group Include the training current, a modified training voltage, and the corresponding training battery power, wherein the training voltage is corrected according to the training current to obtain the corrected training voltage; a training step, according to the training step The modified battery parameter and the power corresponding group are trained to train a classifier; and a test step is performed, and the test battery parameter group is input into the classifier to obtain the test battery power of the battery. The battery power test method according to claim 1, wherein the first correcting step is to calculate the corrected training voltage V _train ' according to the following expression: V _train '=V _train +c‧I _train Where V _{train is} the training voltage, c is a constant, and I _{train is} the training current. According to the battery power test method of claim 2, wherein in the first correcting step, the training battery parameter and the corrected training voltage V _train ' in the battery corresponding group have a time The prioritized relationship, after each correction, is used to correct the value of the training voltage V _train ' itself after the corrected training voltage V _train ' with at least two temporally adjacent corrected training voltages V _train ' . According to the battery power test method of claim 1, further comprising a second correcting step between the training step and the testing step, correcting one of the battery immediate battery parameter sets as the test Battery parameter group. According to the battery power test method of claim 4, the instant battery parameter set includes an instantaneous current and an instantaneous voltage, and the test battery parameter set includes the instantaneous current and a test voltage. The battery power test method according to claim 5, wherein the second correcting step corrects the instantaneous voltage according to the instantaneous current to obtain the test voltage. According to the battery power test method described in claim 6, wherein the second correcting step is to calculate the corrected test voltage V _test ' according to the following expression: V _test '=V _test +d‧I _test Where V _{test is} the instantaneous voltage, d is a constant, and I _{test is} the instantaneous current. According to the battery power testing method described in claim 1, wherein the classifier is a probability-like neural network model, including a plurality of neurons, each of which corresponds to a weight directly determined by a simple algebraic operation. .